If you have issues viewing or accessing this file contact us at NCJRS.gov.

, r ,
(, lfil(',l v""~'

101287
U,S, Department of Justice
Nationallnstllute of Jusllce
This document has been reproduced exactly as received from the
person or organization originating It. POints of view or opinions stated
In thIS document are those of the authors and do not necessarily
represent tho official pOSition or policios of the National Institute of
JuStico.

Permission to reproduce this c~od matorlUl has been

granted by
Public Domain/NIJ
US- Department of. ,Tustice ,

to the National Criminal Justico Reference Service (NCJFiS).

f'urther reproduction outSide 01 tho NCJRS system roqUires pormis-

sian of the ~t ow nor

ATTACHMENT

Testing Cohen's Routine Activity Theory:

An Analytic Scheme
George E. Capowich

NCJfte,
JUL f,9 \985

A. c (-\ U 181XiO N G
INTRODUCTION

The routine activity theory, as set forth by Lawrence E. Cohen,

Marcus Felson, and their ~olleagues, posits a causal relationship between
changes in ecological structures and fluctuations in national crime
trends since the end of World War II. Cohen, et ale argue that this
perspective satisfactorily explains temporal changes in crime rates that
confound other theories. Relying on official sources of reported crime,
Cohen and Felson (1979) maintain that the predictions of many
sociological theories were ~ot realized during the 1960s, when crime rose
throughout much of the decade despite widespread economic prosperity. On
the other hand, the increase in crime rates slowed during the 1970s while
the country experienced economic recessi0ns and rising unemployment.
The routine activity perspective has its conceptual roots in
social ecology. Cohen and his associates specifically base the theory on
the ecological conception of a human community as an organization of
various relations among inhabitants (including groups of inhabitants) of
the community. These relations are defined and become evident by
patterns of activity performed in space and overtime. The authors of
routine activity theory extend this notion of interd~pendent human
interactions to criminology by reasoning that illegal activities
performed by certain members of the population depend upon patterns of
legal activities exhibited by others in the community. In other words,
the variations and characteristics of legitimate routines result in
situational exigencies that may inhibit or facilitate the likelihood that
certain crimes will occur. The temporal and spatial organization of
routine, legal activities therefore influences the location, frequency,
and distribution of crimes.
The main thesis of the routine activity approach is that
variations in the daily activities (e.g., work and leisure behavior) of
individuals increase or decrease the likelihood that conditions necessary
for the occurance of a victimization will converge in time and sp~ce.

The three conditions that must be met for a victimization to occur,

according to Cohen and Cantor (1980), are the presence of motivated
offender(s), the availability of suitable target(s}, and the absence of
capable guardians for the target(s}. Mere convergence of these factors
does not, of course, guarantee that a crime will take place. Rather,
victimization is conceptualized as a stochastic process dependent on the
spatial and temporal organization of human behavior that permits
interaction of the three minimal conditions.
Considerable evidence supports the proposition that situational
opportunities encountered by offenders underlie a large percentage of
crimes. Within this context, routine activity theory hypothesizes that
five mediating variables account for the bivariate associations
frequently found in studies of the relationship between crime and
demographic characteristics. These mediating factors, which are actually
risk factors that affect the likelihood of victimization, are exposure,
guardianship, proximity to potential offenders, target attractiveness,
and definitional properties of the crime. By taking criminal motivation
as a given, the framework focuses on the interplay of these risk factors
and considers how trends and changes in social conditions affect the

2
convergence of a victimization event1s minimal components. It is
therefore possible to construct an explanation of temporal trends in
victimization rates.
The formulation of this theory has prompted several research
efforts to test and evaluate it hypotheses. Although previous studies
have largely supported routine activity theory, important issues
regarding the perspective1s efficacy remain unaddressed due to the foci
of these studies. These assessments of the routine activity framework
can be grouped according to methodological emphasis. One group of
research employs highly aggregated temporal measures of changes in social
structure and social production patterns to account for long-term
('1~47-l974) fluctuations in Uniform Crime Report (UCR) figures for
certain crimes. The second group of studies analyzes data from the
National Crime Survey in cross-sectional tests of particular
hypotheses. Despite the soundness of both groups of studies, and the
support their res:llts lend to the theory1s propositions, these studies do
not adequately test certain aspects of theory.
Much of the theoretical development and empirical assessment of
routine activity theory concentrates largely on a macro-structural
level. Cohen and Felson (1979), in a seminal article, explain clearly
that their interests li~ in the examination of how the structural aspects
of social organization which influence occurrences of crime fit into the
overall organization of a human community. The assumption which
underlies this theoretical approach is that aspects of the ways in which
members of a human population organize and interact to achieve legitimate
social ends influence (positively and negatively) the opportunities for

3
certain crimes to occur within that population. The result is that
longitudinal tests of the theory have not specifically addressed the
household or individual level manifestations of the causal relations set
for.th in the theory. This research is designed to fill that void.
Chapter 1 contrasts in general the structural emphasis of
opportunity frameworks with the motivationa1 focus of other sociological
theories that concentrate on offender characteristics and dispositions
toward crime. It is shown that these two broad theoretical guides
approach the study of crime from radically different starting points •
.
Opportunity theories assume that individual motivation toward crime is
constant and focus on how the environmental context in which crime takes
place affects opportunities for crime. On the other hand, theories that
concentrate on offender behavior seek explanations for the likelihood
that individuals will be disposed toward committing a crime.
Chapter 2 discusses the ecological foundation for routine
activity theory and outlines the theory. Following a critical review of
research that tests the framework, the chapter closes with a general
description of the eXisting shortcomings addressed by current work.
Three conceptual features of the theory--its crime-specific nature, the
individual level focus of its main thesis, and the causa1 nature of its
hypotheses--are identified as deserving special attention in the
discussion of specific design issues.
Chapter 3 expands upon material raised in the previous chapter
and discusses directly the implications of the perspective's
crime-specific focus for the design and conduct of t~is research. The
practical and theoretical considerations that influenced the decision to

4
limit this inquiry to residential burglary are discussed, and research
that investigates patterns and correlations of burglary is reviewed.
This examination suggests a need to direct specific attention toward the
causal dynamics of the burglary incident. This is precisely the focus of
this project. In addition, the chapter introduces the question of
reciprocal causation in connection with routine activity theory. Within
this context, reciprocal causation refers to a phenomenon where changes
in lifestyle are presumed to be affected causally by victimization, as
well as the opposite. Sufficient evidence eXists, it is argued, to
suggest probative value in exploring the effects that the introduction of
feedback hypotheses have on the analysis of a routine activity model of
residential burglary.
The final chapter discusses the panel design of the dataset
employed in this research and explains the analytic strategy, including
methods to be used and variables available for analysis. Special
attention is given to the problem of censored data (loss of information
on units in the sample) often associated with panel designs. Specific
remedies for eliminating or minimizing detrimental design effects are
proposed. The chapter concludes by summarizing the overall importance of
the direction taken in this research.

5
CHAPTER 1

Conventional wisdom among scholars, policymakers, and the public

holds that crime has risen at an unprecedented rate during the last two
decades. This presumed rise in crime prompted considerable political and
academic discussion resulting in a variety of programs, research efforts,
and theoretical explanations. The National Commission on the Causes and
Prevention of Violence (1970) alludes to the puzzling configuration the
data exhibit when it asks rhetorically, "Why ••• have urban violent crimes
increased substantially during the past decade when the conditions that
are supposed to caus€ violent crime have not worsened---have, indeed
generally improved?" (p.33). While carefully noting counter-trendS and
gaps that persist between minority subgroups and the larger society, the
Commission goes on to state:
The Bureau of the Census ••• states that most
indicators of well-being point toward progress in
the cities since 1960. 1 ••• [TJhe proportion of
b1acks in cities who completed high school rose
from 43 percent in 1960 to 61 percent in 1968;
unemployment rates dropped significantly between
1960 and 1968; the median income of families in
cities rose by 16 percent between 1959 and
1967 ••• and the median family income of blacks in
:ities increased from 61 percent to 68 percent of
the median white family income during the same
period. Also during the same period the number
of persons living below the lega1ly-defined
poverty level in cities declined from 11.3
million to 8.3 million. 1 (p.33)
As UCR statistics illustrate, the paradoxical circumstances
cited in the violence commission's Final Report are not limited to
violent crime. Property crimes, which account for the vast majority of

6
offenses, increased at a similar rate to that for violent crimes despite
economic expansion and falling unemployment. Aggregately, vio1ent crime
rose 104.9 percent during the 1960s while property crime grew by 94.1
percent. The annual rate fluctuations for each UCR index category show
clearly that they rose moderately during the first five years of the '60s
with a much larger incr ase recorded in the following five year period
(UGR, 1982). These increases occurred despite general economic expansion
and a steadily declining unemployment rate. Many authors consider it
puzzling indeed that one component in the quality of life (crime) rises,
thereby detracting from lite's desirability while most other ingredients
in a quality of life index (e.g.; real income, poverty, unemployment)
that supposedly influence crime signal overall improvement.
Although dramatic increases in the crime rate have been
documented consistently by UCRs, it has been widely argued that UCRs do
not reflect patterns of real crime but, rather, the organizational
processes and structures of the reporting agencies (Kitsuse and Cicourel,
1963; Wolfgang, 1963; Robinson, 1966). By this time, the long list of
objections and problematic issues associated with the UCRs is well known
making detailed discussion unnecessary; the list includes, among other
issues, agency bias, variation in reporting practices; patterns of
non-reporting, imprecise and broad crime categories, crude rate
computations, definitional variation, and inadequate offense information
(see Hindelang, 1974 for a complete review).

There is additional research that suggests the longitudinal

trends eXhibited by UCR figures may be proportional to real crime trends

7
in spite of the many difficulties associated with UCR. Studies reveal
consistently that seriousness of offense is the best predictor of whether
a crime will be reported to the police (Hindelang) 1976; Nettler, 1978).
It seems reasonable then that some portion of the upward trend in UCR
index crimes is the result of an increased volume in these crimes.
Changes in organizational processes and reporting practices of law
enforcement agencies are sometimes cited as the cause for the large
increases in crime (Black, 1970; Maltz, 1977). Although changes in
departmental policies probably contribute to changes in rates of reported
crime, it Seems doubtful t~at this is a major factor in the longitudinal
character of UCRs. Although McCleary, Nienstedt, and Erven, (1982)
demonstrate that official estimates of crime are functions of
organizational structures and practices, these authors emphasize that
their time series analysis of crime rates does not suggest that
longitudinal trends in those figures are inaccurate. They conclude to
the contrary that, in the absence of additional evidence; confidence in
UCR trends is warranted.
Social theor'y of the period could not satisfactorily explain the
magnitude of crime increases experienced in t.he 19605, and explaining
this phenomenon became a high prio,'ity on the social research agenda.
Investigations of these crime trends focus ott a variety of social,
political and economic variables resulting in an explosion in the
professional literature. A complete review of this research is
unnecessarily lengthy but a brief discussion will be helpful. The early
~tudies in this genre of research analyzed bivariate relationships
bab/een UCR! and a variety of demouraprlic variables (see, for eXilmple,

8
Sagi and Wellford, 1968; Ferdinand, 1970; Wellford, 1973). Perhaps the
best known of these studies, The Task Force Report of the President's
Commission on Law Enforcement and the Administration of Justice (19b7),
atttibutes half of the 1960-65 UCR increase to changes in age, sex, race
and other factors associated with urbanization. The Commission also
attributes a large part of the UCR rate to unemployment. This finding
was based on a simple bivariate relationship, however: young black males
who suffer disproportionately high rates of unemployment and are also
Misproportionately represented in the offender population. Correlation
does not necessarily convey causality, of course, so this and other major
findings in the Report are suspect.
Most of the early research suffers from this same methodological
flaw, which might account for the consistent findings of the literature.
The early studies had a strong influence on theory and policy
nevertheless. Criminologists in the late 160s and '70s concentrated on
dispositional or motivational constructs and crime prevention programs of
this period stressed offender motivation and rehabilitation (Clarke,
1980; Lewis and Salem, 1981). The predictions of the early research were
not borne out, however. When the economy approached and surpassed full
employment in the '60s, for example, crime rates did not drop; they
rose. Similarly, when the baby boom generation matured in the '70s,
crime rates continued to rise, albeit more slowly.
These types of considerations prompted some scholars to express
dissatisfaction with traditional theoretical frameworks that guide the
scientific study of crime. Gould (1969), for example, argues that
criminological research stagnated and urges researchers to move away from

9
- --- -- - - -----~~-

their' reliance on of"';lnrler motivation in their search for explanations of

crime. Instead, he advocates that structural and Situational factors
affecting criminal opportunities be explored. A synopsis of the
the.oretical outlook that concentrates on individual dispositions toward
criminality follO\~s which v-l111 c1at~ify the distinction between
motivational and structural approaches.
Theoreticul perspectives donrln.1nt in the eat'ly 19tius posited tl

motivational construct emanating from dysfunctional social conditions

that lead to deviance. This is evident in ~ter'tQr.ls (1938) general theory
.
that norm violations resu1t from the poor integration of cultural values
and structural opportunities. In other words, the means available for
people to pursue socially induced goals of economic success and
self-fulfillment are not available equally to all segments of society.
The result of this ends-weans discrepancy is stress that ;n turn prompts
individual deviance in an effort to alleviate the tension.
Other scholars, focussing primarily on Merton's lack nf
specificity regarding particular deviant adaptations, sought to expand
this general strain framework. Cloward and Ohlin (1964) apply the theory
to specific delinquent reactions of theft, drug use, and aggression.
A1though they ddopt the hypothes is t:lat strcd n and its presumed
frustration provide the mot1vatirnal impetus for crime, more was
necessary; motivation alor::, they argue, ;s insufficient to explain
criminality. Successful adaptation through delinquency also requires the
opportunity to learn and use the illegitimate means et~loyed in a
reduction of stress. They tl1erefor-e present a mixed model that combines

10
elements of cultural deviance theory within a strain framework and argues
for a subculturally supported delinquent response when legitimate
opportu!,ities are blocked and the opportunity exists (through the
subculture) for exposure to illegitimate means and criminal roie models.
In contrast, Albert Cohen (1955) expands the basic strain
formulations in a slightly different direction due, in part, to his
skepticism of the emphasis strain theorists place on the social class of
delinquents. If initial tendencies toward delinquent activity originate
with juveniles l discontent and frustration about their lower class
positions in society, then'middle class delinquency and norm violations
among the upper classes are not explained. Cohen continues by arguing
that perceived status discontent, not dissatisfaction with class
position, provides a motivational impetus for delinquent reactions to
strain. The youth is unable to achieve SUCCt-S by reaching socially
va1ued goals and the enhanced status it conveys and so adopts the values
held within a delinquent subculture in order to achieve the desired
status. Although dominant norms are repressed they are not eradicated;
consequently, violations of these norms result in anxiety that is
countered by a defense mechanism Cohen refers to as reaction formation.
This response is characterized by an exaggerated repudiation of the norms
$0 that the delinquent is persuaded to hate what is actually desired.
Reaction formation, therefore, accounts for the nonMutilitarian,
hedonistic nature of subcultural delinquency that Cohen felt traditional
strain theory left unexplained.
If it is true that science requires theory to challenge the
data, then the motivational perspective falls short. The general crime

11
trends and indicators of social well-being already discussed are, to some
extent, problematic for such a theoretical framework. Furthermore,
evidence from research in tangential areas of interest suggesting
cultural value consensus and disconfirming the pressure of discrepant
goal and aspiration orientations (e.g., Gold, 1963) undermines the
presumed efficacy of a motivational foundation in the study of
delinquency. Gibbons (1971) took note of such evidence and the r~lative

stagnation under which etiological research in criminology labored,

calling for new ways to look at criminality. New ground needed to be
broken for new and more promising hypotheses to emerge. Some scholars
(e.g., Ferdinand, 1966; Clinard and Quinney, 1967) explore the heuristic
value of typologies while others (e.g., Briar and Piliavin, 1965; Liebow,
1967; Gould, 1969) investigate situational or structural factors that may
induce or inhibit crime.
Although crime rose during the 1970s, the rate of increase
slowed; this occurred despite the onset of economic recession and riSing
unemployment. The rate of reported violent crimes rose by 32.5 percent
between 1970 and 1975 (compared to an 81.6 percent increase from 1965 to
1970) and by 20.6 percent during the 1975-1980 period. In comparable
time periods, property crime climbed by 32.6 percent (1970-1975) and 10.8
percent (1975-1980). This same pattern of declining rates of increase in
the 1970s characterizes all index crimes. Reacting to methodological
shortcomings of earlier work, researchers moved away from
motivational-dispositional theories toward ecological theories,
especially theories relating to criminal opportunity and victim
activity. Rather than focus on offenders' disposit~ons, opportunity

12
perspectives assume that criminal motivatiQn is constant and concentrate
instead on situational factors which contribute to the likelihood of
victimization. Bittner and Messinger (1980) comment that this area of
theoretical development received an important impetus with the
introduction of lifestyle variables by Hindelang and his colleagues in
their analyses of victimization patterns. Routine activity theory
emerges from this school of thought and represents a major theoretical
advance. Several authors note the complementary nature of the
motivational and structural perspectives, and suggest that combining
features from each may lead to more powerful explanations of crime. Two
works (Land and Cohen, 1983; Sampson, 1983) specifically argue for a
synthesis of routine activity theory and other sociological frameworks in
the explanation of crime rates. However, these efforts toward
integration may be premature in the case of routine activity theory
because crucial elements of the perspective remain untested. This
research project will exp10re some of the important gaps in the
theoretical development of routine activity and provide necessary tests.

13
CHAPTER 2

A Routine Activity Theory of Crime:

Ecological Basis and Conceptual Outline

Ecology as a Research Framework

Ecology is traditionally defined as a type of research method
that encompasses principles and assumptions which govern the study of
organisms within a specified environment. Ecologists focus on the
spatial and temporal interrelationships between organisms (including
groups of organisms) and the environment that comprises an ecosystem or
subsystem of particular interest. Emey'y Iol.rd Trist (1973) point out that
an ecologist has a characteristic viewpoint regardless of the discipline
or subject under study. The common thread is the belief that a pattern
of life in a defined habitat constitutes one system of
interdependencies. When one group in a population changes in some way,
ecology assumes that the structural regularities which make the pattern
recognizable are a result of the mutua1 influences each group exerts,
directly or indirectly, on all other groups as well as the effects of all
groups on their shared environment. Viewed in this manner, a habitat's
life patterns constitute a net of multiple, reciprocal influences which
operate constantly.
The structure of this net of interrelations, the focus of
attention for ecologists, is manifested in the cycles, r!1ythms; and tempo
that characterize spatial and temporal developments in a habitat (Hawley,
1950). Social ecologists concentrate on the study of interrelations

14
between a human population and its environment. Ecological studies in
the social sciences typically focus attention on the network of relations
in the human community by directing attention to the distribution (e.g.,
geographic) of human populations, human group characteristics (e.g ••
demographics, voting behavior) and their interactions, and the products
(e;g., social, political, economic organizations) generated by those
populations (Gibbs and Martin, 1959).
This approach has been applied to the study of crime by many
researchers. In fact, ecological studies of crime can be traced to the
relatively early days of European criminology as represented by the works
of Quetelet (1843) and Pike (1876). Ecology has been used widely by
American criminologists as well to study many aspects of crime and
delinquency (e.g., Shaw, McKay, Zorbaugh, and Cottrell, 1929; Boggs,
1965; Turner, 1969; Riess, 1976). The modern ecological crime literature
is vast and diverse, ranging from Harries' (1974) geographic description
of crime's distribution to Fox's (1982) application of sophisticated
predictive forecasting methods. l
Although there is considerable agreement among scholars about
the importance and potential value of an ecological approach in
criminology (Schuessler, 1962; Bittner and Messinger, 1980), serious
questions exist regarding limitations associated with ecological
inquiries. Hirschi (1969) explicitly doubts the wisdom of basing
criminological theory on ecological data and Wilks (1967) discusses in
detail some of the methodological difficulties and inferential
limitations often associated with ecological studies of crime.

15
Many of the inferential difficulties usually discussed in
connection with ecological studies in the social sciences result from
aggregation bias. Analytic dangers associated with the ecological
fallacy are well documented since RObinson's (1950) demonstration that it
;s inappropriate to draw inferences about individual behavior from group
level data. Relationships observed at the group level do not necessarily
hold for the indiViduals in the group; the direction and magnitude of an
association between variables can change significantly as one moves
between levels. Of course, it is equally hazardous for a researcher to
infer group characteristici solely from the compilation of individual
data. Most social phenomena of interest to researchers are multi-level
in nature, however. Group properties have manifestations at the
individual level and affect those who populate the group just as
individual attributes contribute to the texture and character of the
group. For this reason, specific attention must be given to the
complexities presented by the multi-level character of the subject under
scrutiny.
In the case of routine activities, and their relation to crime,
cross-level implications seem clear and straightforward. Daily routines
are synonymous with styles of life. 2 Since patterns of activity are
largely structured by work and leisure, and people with similar
lifestyles tend to cluster, it ;s possible to characterize aspects of
lifestyle aggregately. Researchers often employ measures or social
indicators which reflect a group's character. For example, Cohen,
Felson, and Land (1981) form an aggregate index that partially measure

16
household exposure to the opportunity for household property crimes with
the following ratio computed from U.S. Labor Department statistics:
no. of working women with husbands + no. of non-husband/wife households
Total Households

As a macro-level social indicator (Cohen, et~. refer to this

as a household activity ratio), however, this ratio does not necessarily
reflect the variability of the individual households that comprise the
group. By household exposure the authors refer to the household's state
of unguardedness, yet such a measure is insensitive to household-level
manifestations of this activity ratio. One illustration is provided by
households where the husband and wife both work outside the home, yet the
household might contain other occupants (e.g., older offspring, extended
family members) who are home during working hours and are capable
guardians for the property. Moreover, patterns of change in work or
leisure routines within households (and, therefore, changes in exposure)
can not be detected by such a measure. The household-level form and
variation exhibited by this particular component of lifestyle may not be
the same as that observed at the group level. Furthermore, relationships
that are observed at the group level (e.g., the association between
household activity and the likelihood of certain victimizations) could
take on a different complexion when analyzed at the household level.
The consequences of this multi-level character for social
research are troublesome if the investigator seeks cross-level
generalizations. If a researcher's data and analytic strategy

17
concentrate only on one level, then conclusions are, for the most part,
limited to that level. Nettler (1978) notes that most ecological studies
in criminology are interested in generalizing about individuals, but rely
on aggregate data because relevant information on individuals is often
unavailable or prohibitively expensive to collect. It is not that Cohen
and his associates are unconcerned about the role of individuals; several
of their hypotheses are stated in terms of individuals and their
households (see Cohen and Felson, 1979; Cohen and Cantor, 1980; Cohen &
Felson, 1981). But their many measures do not explore the
household-level ramifications of the longitudinal relations they posit.
There are, of course, ways of overcoming the analytic and
inferential constraints associated with ecological studies. Langbein and
Lichtman (1978) present some solutions that rely on computational methods
and the manipulation of procedures used to group the data. These are
designed specifically for use when only aggregate data are available, but
generalization to the individual level is desired. Alternatively, many
social ecologists point to the value of combining survey data with
ecological traits as a fruitful way for generating cross-1evel
understanding of social phenomena (Allardt, 1969; Scheuch, 1969; Moore
and Golledge, 1976). This not only provides individual-level assessments
of observed macro-level relations, but broadens the theory's informative
value. According to Allardt (1969), the notion of informative value
pertains to the empirical content of scientific statements. By combining
survey and aggregate data, researchers increase the empirical content of
hypotheses because the new level of data increases the number of ways a
theoretical statement can be falsified. This implies, of course, a more

18
rigorous, scientific test. But the greater the variety of events that a
theory explains, the higher its informative value. The ability to draw
conclusions across levels of aggregation, in turn, broadens the scope of
possible generalizations.
Most of the research completed thus far that tests routine
activity theory relies heavily on analyses of aggregated data (Cohen and
Felson, 1979; Cohen, Felson, and Land, 1980). Little work has been done
that examines the household-level implications of propositions set forth
by routine activity theory. Cohen and Cantor1s (1981) study of burglary
employs data from the Natibnal Crime Survey, and Sampson (1983) uses the
NCS to draw conclusions relevant to the theory from his investigation of
structural density and crime. Neither of these stUdies nor other
research explores the longitudinal character of household level
relationship5 as displayed by individual members of the home. This
research will fill that need by concentrating on a longitudinal
assessment of routine activity theory at the household level using data
from the National Crime Panel. Before moving forward with the details of
this project, however, it ;s necessary to consider the broad conceptual
framework outlined by the theory. The discussion begins with an
explanation of the general ecological basis for the perspective, proceeds
to outline the theory1s basic structure, and concludes with a review of
previOUS research that tests the theory.
The Ecological Community
Cohen derives his perspective, in part, from Hawley1s (1950)
Human Ecology: A Theory of Community Structure. In this work, Hawley
uses spatial and temporal aspects of human organization to explain

19
interactions between people and environments. Drawing on ecological
biology, Hawley describes the human community as an organization of
symbiotic and commensalistic relations defined by the patterns of
activities performed in space and over time. These spatial and temporal
patterns represent structures which are manifestations of interactive
relationships that operate wit~in the human community. This is quite
similar to the web of life construct employed by Charles Darwin in his
work on evolution. This (web of life) concept is, of course, identical
to the idea presented earlier about a network of relationships. Ali
life, Hawley explains, ;s inevitably dependent upon other surrounding
life. The result is an interdependent habitat constituting a set of
stimuli to which organisms respond and adapt. As the community's
division of labor evolves, the argument continues, patterns of
interdependence and adaptation will change. Early ecological studies of
crime (e.g., Shaw ~~., 1929) and geographic studies (e.g., Harries,
1974) concentrate on spatial dimensions, ignoring temporal
considerations. Routine activity theory is explicitly longitudinal,
however. The assumed interdependence of diverse community activities is
related specifically to criminology within this perspective, and ;s used
as a theoretical foundation for explaining the temporal fluctuations in
post-World War II crime rates.
A point of clarification ;s appropriate before continuing
further. While human activity within an ecological perspective is
related to and dependent upon the structural milieu, strict determinism
is neither intended nor implied. Factors linked to habitat are by no
means the sole determinants of behavior. Rather, structural

20
"I
characteristics exert an influence on human actions that ranges from
permissive to restrictive. Within criminology, concentration on
ecological structures that either facilitate or inhibit the occurrence of
crime overcomes, in part, a limitation inherent in theories that focus
only on personal attributes of offenders (Schuessler, 1962). Offenders'
characteristics (e.go; demographics) are usually associated with the
disposition or propensity to commit some criminal offense and can not
address directly the aspects of social organization which affect
opportunities for certain crimes independent of criminal inclinations.
·
Explanations of crime are thereby broadened and enriched when
opportunity~related variables are brought to bear.
The Ecology of Routine Activities and Criminal Opportunities
As noted earlier, ecological theories are not new to
criminology; indeed the roots of ecological approaches to the study of
crime date to the early nineteenth century. However~ the particular
conceptualizations contained in Cohen's opportunity perspective are new
in two ways. One, mentioned in the previous section~ is the reliance on
temporal as well as spatial variations within a human community context.
The second is treatm~nt of routine activities as ecological structures.
Cohen and Felson (1979) extend the intardependent character of community
life to occurrences of crime by reasoning that illegal activities must
depend on legal activities; spatial and temporal structures of routine
activities should, therefore, influence the location, frequency and
distribution of illegal acts.
The main thesis of routine activity theory is that variations in
the daily activities of individuals increase or decrease the probability

21
of convergence in time and space of three components which are minimally
necessary for a victimization to occur: (1) motivated offender(s), (2)
suitable target(s), and (3) the absence of capable guardian(s) for those
targets (Cohen and Cantor, 1980). Although these three conditions are
necessary for a victimization, they are not sufficient; convergence does
not guarantee a victimization. The mere presence of an opportunity for
crime does not mean that other factors can not influence the outcomea
Whether a motivated offender acts upon an available opportunity might
depend, for example, upon the subjective perception of that offender. As
Clarke (1984) explains, ev~n a motivated offender might not act upon an
opportunity if he fails to recognize it subjectively as such for some
reason. Victimization is viewed within an opportunity perspective as a
stochastic process dependent on the spatial and temporal allocation of
human activities. The spatial and temporal distributions of human
behavior combine to either permit or impede the interaction of those
components necessary for a crime to occur.
Data are available which illustrate concretely how changes in
the pattern of particular, legitimate routine activities influence
opportunities for certain criminal victimizations. Surveys that explore
how people use their time show a consistent pattern of change in the
allocation of routine activities from home to non-home settings since
World War II (dedrazia, 1962; Szalai, 197J; Kuic, 1931). A few details
of this aggregate shift are especially pertinent to a consideration of
how social changes alter specific opportunities for criminality. The
surveys report that free time resulting from reduced work hours is often
occupied by second and third jobs taken, in part, to pay for the purchase

22
of various household goods. If we consider this together with the
increased participation of married women in the labor market (1975 U.S.
Census as cited by Cohen and Felson, 1981), implications for the
structure of opportunities for certain illegitimate activities become
clear. More people working more jobs with greater frequency will
increase the amount of time people spend away from their homes which, in
turn, decreases the availability of capable guardians for the household
and its property. Furthermore, some of the household items purchased
with the additional income earned at those extra jobs will increase the
supply of available targets:
The saliency of effects that changes in routines exert on
opportunities for crime ;s clearer in light of the cons;derdble evidence
that suggests the importance of factors oth~r than economics and offender
characteristics in accounting for crime trends (Gould, 1969; Gibbons,
1971; Sparks, 1977; Hindelang, Gottfredson, and Garofalo, 1978; Cohen and
Felson, 1981). For property crimes, these studies support the
proposition that situational opportunities encountered by offender~

underlie a large percentage of these crimes. If married female work

force participation, the value of household items, and the amount of time
spent away from home increase between two pOints in time, all else equal,
the opportunities for household related property crimes will also
increase. These changes in routine activities create greater opportunity
for household property victimizations by decreasing the availability of
capable guardians for the home and increasing simultaneously the
suitability of th~ target. Therefore, the probability of convergence for
the three necessary components of a household victimization increases.

23
Biographical accounts (e.g.) Letkemann, 1973; Klockars, 1974) and
interviews with convicted burglary offenders (Heppetto, 1974) demonst.rate
the importance of these behavior patterns for law breakers who regularly
seek the most vulnerable and attractive target dvailable. These three
studies also illustrate the relevance of opportunity-related variables to
different types of offender&. A professional thief in Letkemann's study
emphasizes that an attractive target's vulnerability is regularly
assessed by watching routines (e.g., opening and c'osing times, customer
traffic, security arrangements) associated with the business before
deciding when and how to pr'oceed with crime. Vincent Swaggi, Klockars'
professional fence, gives the example of purposively seeking street
intersections in the city where traffic congestion is most likely to
enhance the opportunity for "boosting" the contents of delivery trucks.
Many of the housebreakers Reppetto studies explain that they often check
a residence for occupants before attempting a burglary. For example, one
burglar explains that he first rings the doorbell. If someone responds
he poses as a door-to-door salesperson or asks for a ficticious person
before leaving to find a more vulnerable target. If the doorbell is
unanswered, howeve~, he moves forward with his plan to break and enter.
These examples provide germane illustrations of the interdependent
adaptations inherent in the ecological concept of a human community. A
segment of the population involved in ~egitimate activities modifies its
routines (e.g., gets a second job, spouse takes a job outside the home)
in response to complex personal, social, cultural, and economic forces.
These changes then prompt reactions by a different subgroup of the
population (those involved in certain illegitimate activities), thereby

24
influencing the rate and pattern of criminality. Such a set of
circumstances serves as a criminologically relevant description of human
symbiosis.
This general outline of symbiotically interdependent subgroups
within the human popUlation forms the foundation for the criminal
opportunity perspective set forth by routine activity theory. Changes
w~th;n spheres of legitimate activity (i.e., daily routine activities)
precipitate changes in the likelihood of criminal victimizations by
altering situational factors which either enhance or inhibit the
opportunity for motivated bffenders to commit crimes. By explicitly
seeking explanations for longitudinal trends in ct'iminal victimization,
Cohen and his cOlleagues posit specifically that changes in the daily
routines of legitimate actors account for subsequent changes in the crime
rate.
A Routine Act;vitx Theorx of Crime
A primary contention of the routine activity appro~ch is that
societal-level changes in production activities and consequent
alterations in routine activity patterns act together to influence
opportunities for certain types of victimizations. Trends in the
production of durable goods (e.g., miniaturization in electronics) affect
a property crime target's suitability for victimization by furnishing
items which are valuable, accessible, and portable (Cohen and Felson,
1979). In addition, Cohen and Felson (1981) note that changes in typical
activity patterns associated with more women in the labor market and the
tendency to spend more time away from home prompt changes in the type 1
tempo, and location of routine activities. These types of social changes

25
affect the structure of daily routines and influence particular crime
rates by influencing the target's state of guardedness and exposure.
Lifestyle and Victimization
Routine activities are consequences of, and largely synonymous
with, styles of life. Lifestyle;s a seminal concept in routine activity
theory and criminal opportunity research in general. Hindelang, at al.
(1978) conceptualize lifestyle as equivalent to daily routine act1vities
and define lifestyle formally as: "a characteristic way of distributing
one l $ time ••• among the common social roles of adult life--those of
worker, parent, spouse, citizen, ••• and user of leisure time
(Hav;ghurst, 1961, p. 333 as cited by Hindelang, n~.~ 1978 p.
244-245). Hindelang and his associates expand this definition slightly
to include juveniles and then consider how different styles of life and
their antecedents are associated with differences in exposure to those
circumstances with a high ri:k of becoming the victim of rape, assault,
robbery, or personal larceny (see Chapter 11).
Within this lifestyle context, Cohen, Kleugel, and Land (1981)
posit that five mediating variables explain the bivariate patterns
usually found in the study of the relation between crime and race, age,
or income. Four of these variables are risk factors a3sociated with
different lifestyle activities: (1) exposure, (2) guardianship, (3)
proximity to potential offenders, and (4) target attractiveness. The
fifth involves nefinitional properties of the crime.
These five variables explain specifically the theoretical
mechanisms that connect social structure, routine activities, and
opportunities for certain crimes. Exposure and attractiveness refer to

26
the potential crime target's suitability for victimization and
guardianship alludes to the target's vulnerability. The proximity
variable refers to the supply of motivated offenders capable of taking
advantage of the opportunity to commit a crime presented by a vulnerable
and exposed target. The fifth factor is listed by Cohen et a1. because
it establishes the relevant context within which the first four
opportunity variables interact. A brief example will clarify this last
point.
Crime, when used to label behavior or occurrences, refers to
many diverse phenomena. Labelling an event a crime implies something
about how society views the occurrence and tells us how a community
chooses to react to these circumstances. It tells us very little,
however, about the act itself. Aggravated assault and burglary are both
UCR index crimes but involve targets, behaviors, and circumstances that
are very different. Situational opportunities which affect the
1ikelihood of each crime are also quite likely to be different.
Activities and circumstances that combine to ncrease the
likelihood that someone will be the victim of a serious assault do not
necessarily affect the chances for other criminal victimizations in the
same way. In fact, it is easy to think of situations where the same
circumstances can enhance the opportunity for one type of crime while
decreasing the probability that another crime will take place. For
example, if the employed spouse in a one-career faw~ly stops working for
some reason (i.e., layoffs), all else equal, the household is less likely
to be a target for burglary (according to routine activity theory)
because of the additional guardian for the property and the increased

27
activity around the home. The opportunity for aggravated assault between
the married couple is enhanced, however, if the frequency of serious
personal injury arising from familial violence is considered. The same
o~currence (being laid off) inhibits the chances of a burglary
victimization at that residence, yet enhances the opportunity for serious
spouse abuse. Exposure to burglary is decreased while exposure to
serious injury at the hands of a spouse is increased.
By taking criminal motivation as a given and focussing on the
interplay of risk factors as they affect certain crimes, one can consider
how trends and changes in ~ocial conditions and lifestyle affect the
frequency with which the minimal components of victimization converge.
It is therefore possible to construct an explanation of temporal trends
in victimization rates.
Cohen, Felson, and Land (1980) make two explicit assumptions
relevant to a consideration of how variations in routine lifestyle
activities of legitimate actors affect criminal opportunities:
(1) Offenders prefer targets with fewer
guardians.
(2) Persons related to an individual by
secondary group ties, or persons who do
not share a stable relationship and do
not themselves have norm~enforcing role
obligations, are less likely to act as
guardians for each other or their
property than persons involved in
primary group relations.
Given these assumptions, they derive a theorem:
Decreased (population) density in physical
locations that are sites of primary group
routine activities produces an increase in
criminal opportunity.

28
Regardless of target, then, routine activity theory hypothesizes
that opportunities for certain crimes will increase as patterns of daily
routines shift from sites characterized by primary group relations (e.g.,
immediate family) to situations where people are related to one another
in a less intimate manner. Stated more generally, this perspective
m~intains that opportunities for certain crimes will vary as a function
of changes in the patterns of daily routines.
Formal Defin1tions and Assumptions
It is now possible to relate the structure of routine activit'y
theory and the necessary definitions as set forth by Cohen and his
colleagues. They seek to explain temporal trends in direct-contact
predatorY crimes. These crimes are n ••• illega1 acts in which someone
definitely and intentionally takes or damages the person or property of
another ••• [and violations] ••• involving direct physical contact
between at least one offender and at least one person or object which
that offender attempts to take or damage ll (Cohen and Felson, 1979:
p. 589). The general framework links aspects of routine activities with
criminal opportunity structures.
Definitions. 3 The definitions for the intervening risk factors set
forth in the theory and used in this research are:
Exposure: the physical visibility and accessibility
of persons or objects to potential offenders at given
time or place.
Proximity: the physical distance between areas where
potential targets for crime reside and the areas where
relatively large populations of potential offenders
are found.

29
v' "\" ,~. I
Guardianship: the effectiveness of persons (e.g.,
housewives, neighbors~ pedestrians, private security
guards, law enforcement officers) or physical security
measures (e.g., burglar alarms, locks, barred windows)
in preventing violations from occurring, either by
their presence alone or by some indirect action.
Target Attractiveness: the material or symbolic
desirability of a target to potential offenders, as
well as the perceived invulnerability of a target to
illegal intrusion (i.e., the weight and size of
property that discourages its theft and the physical
capacity of persons to resist attack). Furthermore
target attractiveness is differentiated on the basis
of whether the motivation to commit a crime is
primarily instrumental (i.e., the act is a means of
acquiring something one desires or needs) or
expressive (i.e.~ the act of attacking a person or
stealing property'is the only reward sought in doing
so) .
Definitional Properties of Specific Crimes: the
features of specific crimes that act to constrain
strictly instrumental actions by potential offenders.
For example, many larcenies are less difficult to
commit and require less knowledge of victim routine
activities than do burglaries (see Assumption 5 in the
next section). Such constraints limit the ability of
potential burglary offenders to consistently act
against targets that would maximize their economic
gain, thus requiring them to seek out less attractive
targets. By comparison, crimes motivated by
expressive needs (e.g., aggravated assault during an
argument) are less constrained.
Assumptions. There are five assumptions about links between the risk
factors defined above and the likelihood of criminal victimization.
Exposure: all else equal, increased exposure leads to
an increase in victimization risk.
Guardianship: all else equal, offenders prefer
targets that are less well-guarded to those that are
guarded more closely. Therefore, the greater the
guardi'inship, the less the risk of criminal
victimization.
Proximity: all else equal, the closer the residential
proximity of potential targets to relatively large
populations of motivated offenders, the greater the
risk of criminal victimization.

30
Attractiveness: all else equal, if a crime ;s
motivated by instrumental ends, the greater the
attractiveness of a target, the greater the risk of
victimization.
Properties of Crimes: the strength of the partial
effects exerted by exposure, guardianship, and
proximity on victimization risk depends upon the
degree to which properties of crimes them3elves
constrain strictly instrumental action. Specifically,
the more a criminal action is constrained as strictly
instrumental, the stronger will be the effects of
exposure, guardianship, and proximity on victimization
risk relative to the effect of target attractiveness.
Empirical Tests of Routine Activity Theory
A discussion of the research literature pertaining to the
routine activity theory must be prefaced by a comment on the
~

crime-specific design of these studies. Since the type of crime both

establishes the context within which the other risk variables operate and
shapes interpretations by defining the relevance of certain daily
activities to certain crimes, the type of crime must be represented in
the study design. This is usually accomplished either by studying one
type of crime or by conducting separate analyses for each crime. Of
course, this applies to the present research and will be considered in
the next chapter's discussion of this researcher's choice to study
burglary victimizations.
The relevant point here is that studies testing the routine
activity perspective present findings and conclusions stated in terms of
individual crime categories. Strictly speaking, a study focussing solely
on burglary might be expected to concentrate only on results that relate
to burglary. The purpose of this brief review is more general, however.
Attention is given in this section to the overall performance of a
routine activity model rather than crime-specific details of the

31
results. Of course, accuracy requires that conclusions and relationships
be discussed in terms of the crime to which they apply. But the main
concern is not a comparison across crime types of the association between
an. opportunity-related variable (e.g., income) and victimization. The
relevant question is whether the relationships are of a type and
direction that are consistent with the routine activity framework. This
will provi de an overa 11 picture of how we 11 the theory s urvi ves tes ts of
its major propositions.
The relatively few studies that bear directly on these questions
.
can be divided into two categories. One involves longitudinal stUdies
that address the overall performance of a routine activity model by
analyzing how well the perspective accounts for long term trends in crime
data. The second category includes specific cross-sectional tests of
hypotheses generated by the theory. The discussion begins with the
latter group of research efforts.
Cross-Sectional Research
Research designed specifically to evaluate certain theoretical
components of routine activity theory is largely supportive. The partial
coefficients used to test the effects of risk variables hypothesized as
mediators between demographic characteristics and victimization for the
most part confirm such relationships. Cohen and Cantor (1980), in their
study of personal larceny (using NCS data from July 1975 to June 1976),
obtain coefficients of partial determination that indicate a negligible
effect for race. Somewhat unexpected, however, is the finding that age
has a strong main effect that is associated negatively with the risk of
personal larceny, followed in magnitude by major household activity
(categories are employed, unemployed, keep house). The number of people

32
in the household displayed a negative relation with personal larceny,
while income is related positively. These results lead the authors t6
the substantive conclusion that those with yearly incomes of $20,000 or
greater, those between sixteen and twenty-nine years old, those who live
alone, and the unemployed all face an above average risk of being
personal larceny victims. In contrast, people who are fifty years of age
or older and those who keep house rather than work have a below average
risk of becoming a victim of personal larceny.
The authors present detailed analyses suggesting that some
interesting interactions affect the risk of personal larceny
victimization. Individuals who are between sixteen and twenty-nine years
old and in the middle income category ($10,000 to $19,000 per year) have
relatively high chances of being victimized. Older citizens, by
comparison, are less likely to become victims if they either fall in the
lowest income category or keep house as their major household activity.
here is a greater risk for all those in the lower income categories if
they keep house or work than if they are unemployed. Lastly,
victimization risk is highest for those who live alone when they are
unemployed compared to the other major activity categories.
In a subsequent study of residential burglary, Cohen and Cantor
(1981) seek the characteristics of individuals and aspects of their
lifestyles that are associated differentially with the risk of burglary
victimization. This study relies on the same data set as the study on
personal larceny. Stated in summary fashion, the authors conclude that
the types of people heading households that have a higher than average
risk of becoming burglary victims include: central city residents
(central city/non-central city dichotomy), the youngest (three

33
categories: ages 16-29, 30-49, and 50 or above), those in the highest or
lowest income groups (four categories: income below $7,500;
$7,500-$14,999; $15,000-$24,999; $25,000 or more), the nonwhite, and
those whose households are unoccupied frequently. In comparison, homes
are least likely to suffer a burglary when the head of household is
older, has an income in one of the two middle categories, lives outside a
central city, and when the home is occupied relatively often. Additional
analysis, again utilizing coefficients of partial determination, reveals
that age of household head is the strongest predictor of burglary
victimization followed in order of importance by area type, income,
househo1d occupancy, and race.
There are also significant interactions among variables
associated with higher or lower than average risks of burglary
victimization. The characteristics related to a lower than average risk
in order of magnitude include: over 49 years old with less then $7,500
in annual income, age 30 to 49 with an income between $15,000 and
$24,999, age 16 to 29 with an income of $15,000 to $24,999, age 30 to 40
in either the $7,500 to $14,999 or over $24,999 income brackets, and age
16 to 29 with an income of $25,000 or more. The significant interactions
that relate to a higher than average risk are, listed in order of
magnitude: those 30 to 49 years old with incomes below $7,500, those
over 49 with incomes between $15,000 and $24,000, 16 to 29 year olds that
have incomes less than $7,500, and those 50 or older with incomes of more
than $24,999 or between $7,500 and $14,999.
These results are mainly supportive of the theory. Cohen and
Cantor (1981) offer a post hoc interpretation for the parabolic shape of
the relation between income and the risk of burglary. Since this

34
association retains its U-shape inside and out of central city areas, the
authors contend plausibly that this reflects the operation of two types
of burglars. For one group, proximity to targets may be the most
important factor. A disproportionate number of these offenders might
live in low income areas of central ~ities and outlying areas. Because
th1s group lacks mobility, potential offenders concentrate on situational
opportunities encountered within (or near) their own neighborhoods to
burglarize homes of the poor. In contrast, a second group of more
professional burglars, for whom target affluence is a major
.
consideration, may victimize richer households.
In addition to questiuns of household affluence, other results
suggest the importance of victim's routines near or away from home in
accounting for certain vicimizations. Although not among the strongest
predictors analyzed, households more frequently occupied by capable
guardians are less likely to be burglarized than homes that are occupied
less often. This points directly to the importance of household exposure
postulated by the theory. .
Further evidence of exposure's importance can
be seen in the result that age is the strongest predictor of both
burglary and larceny. Cohen and Cantor (1981) cite survey data from the
U.S. Census that show young people tend to spend more time away from home
than do older people. This would explain the lower rates of burglary
victimization among older citizens; they are home more often than the
young and able to better guard their property.
The negative relation between age and personal larceny can al~o

be explained by the different patterns of activity displayed by the young

and old. Personal larceny is a crime that tends to occur away from
home. One expects, therefore, that older people who tend to remain at

35
home will be exposed to the risk of personal larceny less often than the
young. Another of the variables found to be associated with a higher
risk of larceny (Cohen and Cantor. 1980) bolsters this interpretation.
The major activity variable showed that those who spend more time away
from home (e.g., the unemployed looking for work) are subject to a higher
risk of personal larceny than those who remain home comparatively more
often.
Longitudinal Studies
In contrast to the studies just reviewed, research that tests
the overall temporal stabiiity of the structural factors hypothesized by
routine activity theory relies on analyses of aggregate data--UCR crime
rates and various types of census data~ This genre of research employs
measures of social structures presumed to influence the convergence in
time and space of the three components necessary for a victimization.
The emphasis is on accounting for long term fluctuations in rates of
reported cime by using measures of social conditions as criteria
variables.
Many other efforts to model crime trends focus on economic
factors (e.g.) Brenner, 1976) or demographics (Fox, 1979). Traditional
variables such as unemployment and poverty perform less than
satisfactorily, however, when used as predictors in forecasts of crime
rates. Cohen and Felson (1981) show that measures of poverty,
unemployment, and age structure do not account for changes in the UCR
reporting rates for robbery, burglary, larceny, and auto theft between
1947 and 1974 (see Table 1, p. 145 and Table 2, p. 147). None of those
variables reach a .05 significance level. Moreover, the direction of
each relationship is opposite of that which might be expected. The

36
authors argue that if poverty and unemployment are useful in explaining
longitudinal crime trends, crime will vary inversely with their poverty
ratio. This measure is the ratio of the income of the bottom fifth of
the population to median income; it increases as the economic conditions
improve for those in the lowest stratum of wage earners relative to the
higher income groups. On the other hand; crime is expected to vary
directly with the unemployment rate.
Both independent variables, however, aChieve coefficients with
signs indicating relations with crime that are opposite from the
predicted direction. The ~esults show a direct association between each
of the property crimes and the poverty ratio, and a negative relation
between each crime and unemployment. These results suggest that property
crimes increase as the relative income of the poorest improves and they
decrease as unemployment rises. Substantively, these findings are more
supportive of the opportunity perspective than the more traditional
approaches in criminology which argue that crime is expected to increase
as poverty worsens. But the weak associations (neither the poverty ratio
nor the unemployment rate reach a 5 percent significance level) suggest
the need to specify such models differently. This conclusion is
reinforced, the authors note, by the fact that they are able to accept
the null hypothesis of no autocorrelation of residuals only when
forecasting trends in burglary_ Well-specified equations permit the
acceptance of the null.
There are several studies that respecify longitudinal models so
they perform better and forecast more accurately. SpeCifically, this
research uses measures of the major opportunity"'related variables
(proximity, guardianship, target attractivness, exposure, and properties

37
of the crime) in longitudinal assessments of crime rates in place of or
together with the standard variables which reflect economic conditions
and attributes of those thought to be the most likely offenders.
Cohen and Felson (1979) present the results of a time series
analysis in which they evaluate the effects of household activity on
trends in the rates of five UCR index crimes (non-negligent homicide,
.
forcible rape, aggravated assault, robbery, and burglary) from 1947 to
1974. They construct a household activity ratio which estimates the
proportion of U.S. households that are expected to be most highly exposed
to the risk of victimizati6n in a given year because of a greater
distribution of activity away irom the home. Routine activity theory
asserts that crime and this activity ratio are directly related.
The results show a statistically significant, positive
association between the ratio and each crime rate. The robustness of
this relation is evident in the fact that the relationship persists
whether unemployment and population age structure variables are entered
into the equations as controls. The strength and direction of the
relationships lead the authors to conclude that household activity is an
important explanatory variable for temporal fluctuations in each of the
five crime rates, and suggest that routine activities influence
opportunities for crime. The likelihood that households or their members
will be targets for one of these crimes increases as members of those
households spend more of their time away from home.
Coh~n and Felson (1981) extend these findings by formulating
three other substantive variables. They construct measures of married
female workforce participation from U.S. labor statistics, the incidence
of people living alone from U.S. Census Bureau data, and the presence of

38
lightweight durable goods from national consumer spending patterns.
These variables are employed as social production functions to gauge the
effects changes in these social functions exert on trends in the
reporting rates of robbery, burglary, auto theft, and larceny over $50.
Through a series of steps to construct the pertinent indices, and the
addition of age and unemployment variables along with the number of
automobiles per capita, the researchers II ••• operat;onalize [and
analyze] the impact of offender, target, and guardianship trends upon
crime rates in terms of changes in the age t and routine activity
structure and, hence, the ~riminal opportunity structure" (p. 148).
Their analysis is conducted in two stages. In the first,
stochastic equations which model crime trends in robbery, burglary, and
auto theft between 1947 and 1972 are estimated. Larceny is also
modelled; but for 1947 through 1970 due to limitations imposed by the
data. In general, the equations perform quite well; relationships are
found to be of the size and type expected. Relatively modest increases
in social indicators studied (e.g., supply of lightweight goods) account
for large increases in burglary and larceny. (See Tables 3-6 1n Cohen
and Felson, 1981: p. 149-153 for details).
The second stage of analysis, viewed by the authors as the
major test of the theory, involves ex post forecasts of the crime
rates. 4 Results of the first stage estimation of trends in robbery,
burglary and auto theft are used to forecast trends in these crime rates
for 1973 through 1975. For larceny, the early estimates are used to
forecast for 1971 and 1972. (This is because the UCRs stopped recording
data on larcenies of $50 or more).

39
-l
I

From an opportunity perspective, the results of these forecasts

are encouraging and show the important role played by opportunity-related
variables in the prediction of crime trends. (See Table 7 and 8, Cohen
and Felson; 1981, p. 156"157). The three-year average error in
forecasting rates are 4.6 and 1.9 percent for robbery, 3.8 and 5.5
percent for burglary, and 1.5 and 2.0 percent for auto theft. Two-year
average errors in forecasting larceny rates are also rather low: 1.3,
1.5, 1.7, and 2.5 percent. Each of the percent figures cited as errors
represents the discrepancy between each rate forecasted from an equation
and the rate actually recor~ed. Two equations are estimated for each of
the first three crimes and four equations for larceny. With a mean
forecasting discrepancy of 4.9 percent, and no pattern of systematic
forecasting errors, the authors are justifiably satisfied with the
overall performance of their models.
Evaluating the Eviden~

Cohen and Fe1son (1981) argue that the accumulated evidence

provides important confirmation of the theory. They point appropriately
to some of the early cross-sectional studies that stress the importance
of opportunity-related variables in general (e.g., Gould, 1969) and the
lifestyle variable in particular (e.g., Hindlelang, et al., 1978) in the
explanation of criminal victimization patterns. The cross-sectional
tests of routine activity hypotheses (e.g., Cohen and Cantor, 1980; 1981)
support the idea that the structure of daily routine contributes to
certain crimes by influencing the risk factors of exposure, guardianship,
proximity to off8nders, and target attractiveness associated with
particular types of offenses. However, the cross-sectional design of

40
these studies does not permit a satisfactory test of the thesis that
changes in the character of routines over time account for longitudinal
trends in victimization r1tes.
Lastly, the longitudinal studies reviewed in the foregoing
section confirm the perspective's postulate that social changes since the
end of World War II have increased the occurrence in time and space of
motivated offenders, suitable targets, and the absence of effective
guardians, thereby contributing to increases in crime rates. However,
this last body of research does not address adequately the theory's main
thesis that temporal variations in the daily activities of individuals
increase or decrease the probability that components necessary for a
criminal victimization will converge and influence crime trends (Cohen
and Felson, 1979; Cohen and Cantor, 1980).
It is not surprising that Cohen and his associates focus on
macro-level concerns in their longitudinal assessments of the theory.
Their main interest involves macro-sociological issues. This point is
made clear by Cohen and Felson (1979) when they state:
Although details about how crime occurs are
intrinsically interesting, the important analytic task
is to learn from these details how illegal Jctivities
carve their niche within the larger system of
activities. (p. 592)
The authors continue immediately after this passage to compare their
research with other efforts by scholars to link criminal activities with
the larger social and economic structures. Given this focus of
attention, the authors undertake to show that their approach is
consistent with what is known about micro-level relations; they
deliberately eschew longitudinal tests at that level in favor of
macro-level assessments.

41
The use of econometric models to analyze and forecast crime
trends (Erlich, 1973; Land and Felson, 1976; Cohen, Falson, and Land,
1980; Cohen and Felson, 1981) ;s particularly germane. Econometric
models are explicitly longitudinal and well suited to a study of crime
trends. Econometric models commonly employ highly aggregated measures in
attempts to analyze societal trends and relations. National crime rates
and the macrodynamic social indicators used by Cohen ~ll' in their
research (e.g., female labor participation and the supply of lightweight
durable goods) are illustrative examples. This poses a significant
methodological problem, however. Decker and Kohfeld (1982) criticize Fox
(1978) for ignoring regional variation in UCR trends. This criticism
applies to all analyses of national UCRs, of course. Fox (1982) responds
correctly that, aggregation bias notwithstanding, the aggregation
criticism ;s relative. If a model purports to explain national trends,
then aggregation may be appropriate. However, aggregate data and
macro-level indicators of social functions do not permit adequate tests
of the individual-level thesis set forth by routine activity theory.
Although Cohen and his associate rely heavily on analyses of
aggregate data (e.g., Cohen and Felson, 1979; Cohen; Felson, and Land,
1980), their most recent work (Cohen, Kleugel, and Land, 1981) analyzes
NCS victimizations. The difference between the two approaches is focus.
In the first case, they focus on macrodynamic trends across the post-war
era; the time frame is decades. In the second case, they focus on tests
of specific hypotheses; assuming system equilibrium, they examine causal
relationships at one point in titne. Generally, the opportunity model of
victimization, with its emphasis on structural risk factors, seems to
provide a pm'lerful explanatory framework, analytic difficulties aside.

42
- ----------~-

The research proposed here will strike a middle ground. Analyses will be
explicitly longitudinal~ focussing on the effects of changes in routine
activity of individuals on opportunities for burglary victimization. But
of necessity, the analysis will cover only the period from 1977 to 1981
for which appropriate NCS victimization data are available.
There are, of course, important issues which have been ignored
in an attempt to provide an overview of the routine activity
perspective. The material discussed thus far involving the theory
characterizes it as a framework which explains crime-specific
10ngitudinal trends in crim'inal victimizations at the social and
individual level. As such it represents a major departure from
traditional ecological approaches which, for the most part, concentrate
on much higher aggregation levels. John Laub's (1980) work on rural and
urban patterns of crime, and the city-level investigation by Gibbs and
Erikson (1976) are representative of the macro-level analyses generally
associated with ecological inquiries. The divergence of routine activity
theory from that tradition raises substantive and theoretical issues that
pertain to important questions including the crime under study, the
analytic techniques employed, and the nature of the data set to be
analyzed. These are all important considerations for meeting the major
objective of this study: to test the causal relationships between
changes in household activity structures and burglary victimization that
routine activity theory posits. These design issues receive individual
discussion in the next two chapters.

43
NOTES

1. For an excellent historical review of the early ecological crime

literature, refer to Wilks (1967).
2. The concept of lifestyle plays a central role in the routine activity
framework and will be discussed more fully in a later section.
3. The following formal definitions and statements of assumptions rely
heavily on Cohen, et al. (1981, p. 507-509).
4. If a researcher does not have very current social data, it is .
difficult to make accurate forecasts into the near future. The way Cohen
and Felson choose to avoid this problem is by going back a few years to
forecast crime rates which are already reported. Comparisons between the
known rates and those provided by the statistical analysis permits
evaluation of the forecast model being employed. This procedure is known
as ex post forecasting.

44
CHAPTER 3

Substantive and Theoretical Issues

It is, of course, axiomatic to assert that the design of

scientific studies testing theoretical frameworks must be guided by the
structure set forth in that theory. Such congruence is crucial fo~

adequate, powerful assessments. Chapter 2 concludes by characterizing

the routine activity theory as a perspective that purports to explain
longitudinal fluctuations in specific criminal victimizations as causal
functions of changes in the daily routines of potential victims. These
three conceptual features of the theory--its crime-specific nature, the
causal hypotheses, and its individual (household) focus--have important
implications for the design of this research. The second two features
relate directly to the type of sample necessary for an adequate test of
the theory, and the nature of analytic procedures a~propriate for
assessing the major tenets of routine activvity and for supporting causal
interpretations. The issues of sample design and the analytic course of
this research are discussed in Chapter 4.
The issues of immediate concern in this chapter relate to the
crime specific applicability of this criminal opportunity perspective.
The choice of a particular crime through which to test the assertions set
forth under routine activity is not completely arbitrary; limitations of
the available data, modelling concerns, the character of previous
research, and the structure of the theory impinge, to various degrees, on

45
that choice. This chapter begins by explaining this author's decision to
focus on residential burglary, and then reviews recent literature that
pertains to the patterns, characteristics, and analysis of burglary.
Crime-Specific Application of Routine Activity Theory
Aggregation bias can be a problem when crimes are grouped into
categories. Designations such as IIserious,1I IIproperty,1I or II violent ll can
mask potentially important variability that exists among the individual
crimes that comprise each broader category. It is difficult to think of
a variable that uniformly affects any set of generally defined
.
victimizations. This is particularly true of variables relevant to
criminal opportunity structures in general and the routine activity
theory specifically. It is quite likely that a routine activity theory
of burglary involves different structures and relationships than a
routine activity theory of robbery or assault. The definitional elements
of the crime influence differentially which structural variables are
relevant and, perhaps, the direction of their interrelationships. For
example, it seems reasonable that, if this opportunity theory of
victimization is correct, a shift in routine activities precipitated by
the loss of employment by the one person in a married household who works
outside the home, decreases the risk of burglary victimization (since
effective guardianship of the home improves), while increasing the risk
of becoming the victim of familial violence (because the married couple
will presumably spend more time together). Changes in routine activity
patterns that result from this particular shift in employment status
alter certain criminal opportunity structures such that the likelihood of
spousal assault is increased, but the chances of becoming a burglary
victim decrease.

46
Hindelang, Gottfredson and Garofalo (1978) make an analogous
pOint in their discussion of individual interactions and physical
surroundings. Suburban shopping centers are often established by
bU$iness people who wish to locate their enterprises away from the high
crime of inner cities. While it might be true that such areas have
comparatively
. low rates of crimes involving personal contact (e.g., rape
or robbery), the large congregation 0f unattended vehicles is amenable to
higher rates of crimes such as auto theft or larceny that do not involve
personal contact bet\~een victim and offender. Clearly, the same set of
circumstances and structur~l array is not expected to affect the rates of
all crimes uniformly or even similarly.
Researchers who examine extant relationships between structural
factors and crime rates consistently report differential effects across
crime categories. Decker, OIBrien, and Shichor (1979) report wide
variation in the magnitude of associations between indices of urban
structure and various contact and non-contact victimizations of
juveniles. Concentrating on violent crimes, Block (1979) states that
although proximity to poor and middle class residents is the best
predictor of victimization among the neighborhood characteristics
considered, the range of variation in both the rates and types of crime
is much larger in low proximity neighborhoods than those with a high
proximity. He concludes that different structural characteristics seem
to interact with and affect certain crimes selectively. Similarly,
sCholars who examine the specific influence of structural density on
crime find that the magnitude of associations between victimization rates
and density varies with the type of crime (e.g., Sampson, 1983). If, as
these results suggest, elements of crimes interact differentially with

47
ecological structures, then it is reasonable to expect individual crimes
to relate differentially to patterns of change in those structures.
There is little reason to expect the effects of changes in ecological
structures to be equal or similar across victimization types. This pOint
is obviously crucial to the opportunity perspective, since changes in
crime rates are attributed to changes in routine activities. An accurate
assessment of the preCise relationship between these structural factors
and victimization requires that crimes be studied individually.
Residential Burglary
The decision to limit this inquiry to residential burglary is guided
by theoretical and practical considerations. Property crimes such as
residential burglary differ from personal crimes (e.g., robbery and
assault) in that the former lacks personal interaction between
offender(s) and victim(s). Considerable research indicates that who the
victim is, the relationship that exists between the offender(s) and
intended victim(s), and some actions (e.g., self-protective measures)
taken by the victim profoundly influence the occurrence and ultimate
outcome of crimes against persons (see, Wolfgang, 1958; Amir, 1971;
Hindelang, Gottfredson, and Garofalo, 1978; Toch, 1980). The apparent
influence of interpersonal contact in crimes against persons requires
that the nature of that interaction be considered explicitly in
explanations and analysis of personal victimizations. The definitional
elements of property crimes, on the other hand, make it possible to
eliminate the confounding effects presented by a victim-offender
confrontation, thereby simplifying the process of multivariate
model-building, testing hypotheses, and interpreting the results.

48
Of the predatory property crimes available for analysis,
residential burglary is an appealing choice. Compared to the other
non-contact crimes, residential burglary ;s usually ranked as more
serious than either larceny or auto theft, independent (to some degree)
of the amount stolen (Conklin and Bittner, 1973). From 1973 to 1978, the
NCS data show that household larcenies outnumbered burglaries by
ap~roximately 1.3 to 1.0 (Hindelang, Gottfredson, and Flanagan, 1981);
yet, Repetto (1974) finds burglary to be the qUintessential residential
crime in the minds of potential victims despite the greater frequency
with which larcenies occur •. The cliche about a home as a family's castle
is not taken lightly by most people. This might explain the
psychological intrusion victims often feel after a burglary violates the
sanctity of their homes (Lavrakas, 1981). People expect safety within
their homes and expect others to honor the privacy and security inherent
in another's residence.
Finally, there is a large body of research that describes the
victim, offender, target, and environmental characteristics associated
with residential burglary. Such a descriptive foundation is useful in
its own right and is an essentia1 ingredient for further theoretical
development. Firm descriptive underpinnings rnust precede the formation
of unambiguous concepts which, in turn, provide the blocks upon which
scientific theory is built (Durkheim, 1938; Gibbs, 1972; Turner, 1978).
Previous research that outlines the various attributes of residential
burglaries will be reviewed. The major studies in this area are
discussed in terms of three themes relevant to this study: offense
patterns, offender characteristics, and structural correlates.

49
Corre1ates of Residential Burg~

Research on the ecology of crime concentrates on the search for

explanations of crima's non-random distribution through the population.
Structural features are presumed to account for some of the differences
that distinguish between areas with low and high crime rntes.
Specifically, this section examines studies that analyze the relation
between burglary rates and geographic and population variables. The
different structural aspects associated with variations in the level of
burglal"Y victimizations are of particular interest~ Since routine
activity theory hypothesizes that changes in social structure prompt
changes in people's lifestyles, and that these changes in routines, in
turn, influence opportunities for criminal victimization, the ways
different structural arrays affect burglary rates are important and
relevant.
Scarr (1973) adopts an ecological approach by correlating 1970
census data with official burglary statistics and victimization data for
the census tracts he studies in the Washington, D.C. metropolitan area.
The research explores spatial patterns of home burglaries in an effort to
uncover the structural characteristics which facilitate such incidents.
He concludes that it is not possible to distinguish between high and low
rate areas by aggregate social indicators (e.g., poverty, educational
level) in the suburban locations; these indicators are useful in urban
areas, however, consistent with previous research (e.g., Boggs, 1965).
More recent eco1ogica1 tesearch presents additional evidence
which confirms that particular physical characteristics (city size,
percent owner-occupied housing, and geographical location) and population
variables (ethnicity and median fami1y income) are in fact correlated

50
significantly with the burglary rate (Dunn, 1974; Booth, Johnson, and
Choldin, 1977; Sampson, Castellano and Laub, 19B1)~ Byrne (1983)
confirms that these variables explain a moderate amount of variance, and
expands the inquiry by examining different size cities. Stronger
relationships are achieved between burglary and both sets of variables
(physical characteristics and population variables) for smaller than for
larger cities. It seems that different ecological variables affect
burglary rates differently depending on the size and nature of the
population aggregate.
The usefulness of' examing how different influences exerted by
structural factors affect burglary is evident in a related body of
research. Stafford and Gibbs (1980) offer a hypothesis posing an
interaction between a population ratio (city/SMSA) and a city attraction
variable referred to as dominance. The former variable represents
potential attraction while the latter represents actual attraction.
Therefore, dominance reflects the pull of people into a city to utilize
the economic and social facilities. This phenomenon is measured by the
percentage of SMSA retail sales accounted for by a particular city within
the metropolitan area (city retail sales/SMSA retail sales). Stafford
and Gibbs argue that:
If a city has a high population ratio and high
dominance •.• its gravitational potential
(attraction to nonresidents) will be •••
manifested in a high crime rate. (p. 655)
They find dominance to be correlated negatively with property crime, but
"city crime rates are directly related to the interaction of the
population ratio and dominance, even after various racial and economic
characteristics of cities are introduced" (p. 662). Focussing only on

51
correlates of burglary, however, neither dominance nor the interaction of
dominance and population ratio is significantly correlated with the
criterion. This suggests that these variables are relatively unimportant
in an explanation of burglary; perhaps burglary is not a direct function
of the flow of people into cities~ As we shall see in a subsequent
section, burglary in general and residential burglary in particular are
typically committed by young offenders operating fairly close to th~ir

own homes. It is not very likely that young offenders who often lack
mobility will be drawn to the opportunities of a central city.
Therefore, it seems reasonable that the residential burglary rate of a
given area will be largely unaffected by that area's transient population.
Family Income
Macro-level research provides mixed results concerning thp
relationship between income and burglary; the magnitude and direction of
correlations often vary. For example~ Schuessler (1962) reports a slight
inverse correlation (r = -.15) between average income and burglary.
Similarly, Harries (1974) reports that poverty (percent of all families
below low income level) is negatively associated with burglary. Cohen
and Fe 1son (1981) and Sampson, Caste 11 ano and Laub (1981) reach the same
conclusions, but other researchers find stronger associations. Flango
and Sherbenou (1976), for example, conclude that poverty is the key to
explaining intercity variation of burglary rates. Quinney (1966)also
observes that median family incomes and burglary are much closely
related. He reports a direct assocation in rural areas (r = '1-.34), a
much weaker relation in cities (r ~ +.03), and no relationship in SMSA
areas (r ::: ~OO). In contrast) Jones (1976) eliminates income from his

52
stepwise regression analysis of burglaries because he fails to achieve
unstandardized regression coefficient that is at least twice as large as
its standard error.
Although some disagreement exists on the nature of the
relationship between aggregate measures of income and burglary, it is
fair to say that the weight of evidence indicates an inverse association
is generally reported with some variation in the magnitude. This
presents an interesting discrepency with other research (see, e.g.,
Gillespie, 1977; Cohen, et ~., 1980; Skogan and Maxfield, 1981) that
shows households in low and high income brackets as the most frequently
victimized; on the individual level, Cohen ~~. (1980) attribute the
inconsistent findings to inadequate multivariate controls. The need for
appropriate controls notwithstdnding, the discrepency might also be the
result of distortion introduced by macro-level indicators. Median income
or average income for a certain area might be too insensitive to detect
the U-shaped distribution of burglary victimizations across income
categories.
Braithwaite (1979), in his review of this literature, notes the
inconsistency that characterizes research on the criminogenic properties
of poverty. In a separate analysis of crime rates in 193 American
cities, he finds that poverty (operationalized in both absolute and
relative terms) does not add to the explained variance after controlling
for an area's size. Income in~q~ality (as measured by the GIN!
coefficient), however, does augment the variance explained for each of
the seven index crimes. This 1 ;iie of research has been pursued by other
reseachers, again with inconsistent results. Jacobs (1981) reports that
income inequality is a good positively related predictor of burglary

53
rates in the 195 largest SMSAs for 1970 whereas Carroll and Jackson
(1983) disclose a negative association with burglary in 93 cities for the
same year.
The inconsistency of the ecological research in this area is a
topic for further research; perhaps, as Carroll and Jackson (1983)
suggest, it is an issue which must be addressed via specifically
designed, longitudinal investigations. Regardless of the perspective
adopted, and thr different methodologies employed, a consistent
identifiable strain which emerges is that individual crimes react
differently to disparate patterns of change in societal structures
(Gibbs, 1965; Wilks, 1967; Dunn, 1974; Laub, 1980; Cohen, et al., 1980).
Moreover, there is persuasive evidence that generalizations about crime
drawn from one population aggregate will not remain constant for other
aggregates of different sizes (Schuessler and Slatin, 1964). These
points, considered along with the relative inconclusiveness of the
ecological literature, suggest the necessity for exploring specific
dynamics associated with criminality.
Offense Patterns
Scarr (1973) analyzes residential and commercial burglary in
Washington, D.C. and two of its suburban areas (Fairfax County, Virginia
and Prince George's County, Maryland). He utilizes correlational
analysis to differentiate residential from non-residential offenses and
reports differences in the patterns of the two offenses. Residential
burglaries occur more frequently than commercial burglaries and are more
likely to occur during the day on weekdays. The finding that residential
burglary occurs more often than commercial offenses, while accurate,
requires elaboration. Pope (1975) probes this point in more detail and

54
reports that the rate of commercial burglary is actually the higher of
the two based upon opportunity-specific rates computed for each type of
burglary. The rates are based on the number of targets at risk and
provide a clearer idea of the relative frequency with which these two
crimes occur. Most burglar1e~ of both types involve theft of moderately
valued items ($100 to $500) (Pope, 1975) including, presumably, household
merchandise that is easily converted into cash or to personal use. The
intercity variation in the relative proportions of commercial and
residential burglaries reported by Byrne (1983) further underscores the
need for a crime-specific analysis because the two types of burglary
involve similar activities within different contexts. It specifically
points to the need for opportunity-specific analyses of burglary
incidents.
A work quite similar to Scarr1s is C1arke 1s (1972) study
conducted in Charlotte, North Carolina. Also using census tract data,
residential burglaries are found to be primarily daytime and weekday
phenomena that ocur during the daylight hours. In addition, they occur
predominately in low income areas. C1arke 1s study is different
geographically than Scarr1s; both studies concentrate on urban areas, but
Washington, D.C. and its suburbs have a higher total population and are
more highly urbanized than the Charolotte metropolitan area.
Nevertheless, the results of both studies are compatible.
Conklin and Bittner (1973) concentrate their efforts on
burglaries in one suburb over a one-year period and report results that
are largely consistent with the other studies. They notice little

55
monthly variation in the offense rate, most losses were in the moderate
range; and residential burglaries are most likely to occur during the day
in the middle of the week.
Reppetto (1974) investigates spatial patterns of residential
burglaries and robberies in the Boston metropolitan area by tapping data
derived from many different sources. He employs official reports~

surveys of victims and nonvictims, observations of physical security

measures, and interviews with 97 convicted burglars. Besides confirming
the results found by others regarding time of occurrence and nature of
the theft, the diverse and ~ich data allow Reppetto to provide greater
insights that suggest trends and areas for further inquiry.
Victimization rates and income are related positively regardless of
race. The only difference in victimization patterns that exists between
blacks and whites is that blacks are more likely to SUffer multiple
victimizations. Those mast likely to be burglarized are the young,
single, better educated, and those who more frequently leave their homes
unoccupied. Lastly, the actual experience of victimization often
prompted a response in that those who were victimized were more likely to
enhance their home1s security by installing alarms or additional locks.
Hindelang (1976), in his analysis of 1972 victimization data in
eight cities, finds that the rate of residential burglaries is higher for
renters than homeowners. His findings an time of occurrence and amount
of loss are generally in line with the other stUdies although precise
comparisons are not possible since Hindelang used a different breakdown
for his time categories: 6 A.M. to 6 P.M. (35 percent occurred) and 6
P.M. to 6 A.M. (46 percent occurred).

56
Offender Characteristics
Most inquiries into the characteristics of burglary offenders
have been limited to those apprehended. Routine activity theory posits
one risk factor (proximity to offender population) that is concerned
directly with offender types. Assuming comparability of official arrest
data and the offender information elicited in victim surveys (see,
Hindelang, 1974; 1977), and the similarity between victimization data and
official accounts of burglary incidents (Pope, 1976), personal
characteristics of arrested burglars will provide a fairly accurate
picture of this offender population.
Burglary research undertaken by the Santa Clara Criminal Justice
Pilot Program (1977) includes an examination of offender characteristics
in San Jose, California. The report indicates that 81 percent of those
apprehended were males while only 6 percent were female. Whites
comprised 51 percent of the group, Mexican-Americans accounted for 26
percent, and blacks for 9 percent. Of those arrested, 58 percent were
adults and 38 percent were under 18 years old. Forty-six percent of the
time a single offender was caught, whereas more than one was arrested in
54 percent of the cases.
Chimbos (1973) also reports that the vast majority of
apprehended burglars are male. In fact, all the females that are in his
sample from a Canadian city had been arrested along with at least one
male. Of those incarcerat~d, 57 percent had acted with one or more other
persons with an overall mean age of 17.
Repetto's (1974) results after interviewing 97 adjudicated
burglary offenders and supplementing that information with checks of
official records prove rather interesting. Residential burglars tend to

57
be young and non-white -- quite consistent with the other studies. He
went beyond the basic demographic data, however, to provide additional
insight into the burglary offender and the crime. Seventy~three of those
interviewed stated that they engage in some planning to a certain degree
prior to the offense, and all specifically seek unoccupied dwellings.
Careful attention is given to the rhythms associated with the travel and
work patterns of household occupants. Furthermore, offenders who are
young and non-white tend to commit burglaries in own neighborhoods, with
apparent affluence one of the foremost factors influencing the choice of
a burglary target.
Pope (1977a; 1977b) conducts a more extensive analysis of
offender attributes using prediction attribute analysis and cluster
analysis to examine the characteristics of those apprehended for burglary
in six California jurisdictions over a one-year period. Although the
specific aim of Pope's work is to explore interrelationships between
burglary incidents and those apprehended for burglary, his findings
regarding offender characteristics are relevant. Juveniles were more
likely to commit offenses during daylight in their own neighborhoods,
although fema1es tend to do so away from their homes. Both sexes of
juveniles were found to act in the company of others.
Synthesis of Findings
The conSistency of the results reported by those who the
distribution and characteristics of residential burglaries makes it
possible to summarize concisely the picture that emerges. Burglaries of
homes tend to take place during the daylight hours of weekdays, at houses
that qre relatively unguarded (e.g., because of work patterns), and when
a theft accompanies a break-in, it usually involves items of moderate

58
value taken from relatively affluent households. Those committing the
offenses tend to be young (under 18 years old) black males acting with
one or more r7 their peers within their own neighborhoods. Hindelang's
(1976) investigation of these issues shows an apparent interaction
between household affluence and the tendency of burglars to concentrate
on targets to which they have easy access. He notes the overall positive
relation between the burglary rate and household income (within racial
groups), but contrasts this with the data showing blacks in the lowest
income group with a rate of burglary victimization higher than white
families in all income categories except the highest. This suggests that
target affluence is of secondary importance to accessibility since black,
comparatively poor households are chosen as targets for burglary instead
of the richer white households because the former are more accessible to
the population of burglary offenders. At the very least, this indicates
that, given a choice, burglars will select targets that are (or appear to
be) more attractive.
These data clearly suggest the importance of variables relevant
to the opportunity structure of burglary. Factors affecting the
potential risk of residential burglary victimization include such things
as the age, marital status, employment status, and income of the
residents, and proximity of the dwelling to the offender-prone
populaton. These empirical regularities lead Hindelang (1976) to
conclude that factors other than offender characteristics account for
crime trends and hold much promise as a guide for further research. This
is particularly true, Hindelang reports, of lifestyle variables that seem
to occupy an important position in the explanation of victimizations.

59
Hindelang, et al. (1978) explore these patterns further with
regard to personal crime victimizations and specifically ask why these
personal characteristics are so often associated with higher rates of
victimization. They argue that the patterns of victim attributes which
result in higher likelihoods of criminal victimization are factors that
might be associated with differences in personal styles in life. These
lifestyles are related differentially to various opportunity structures
that either inhibit or facilitate the occurrence of a crime. They
continue to develope this lifestyle/exposure hypothesis into an
empirically grounded theoretical model explaining personal victimization
(see Chapter 11, 1978). The key concept for Hindelang et !l.'s model and
the one proposed by Cohen and his colleagues is, of course, lifestyle.
Both rely on patterns of life cycles to explain the distribution of
victimization rates and Cohen et!l. carry the idea further by
attributing temporal variations in crime trends to changes in the usual
life rhythms (daily routine activities).
There is an issue related to the lifestyle-victimization
association that has until recently escaped attention in the literature~

yet could potentially influence the theory's structure and certainly has
methodological implications. The question of reciprocal effects deserves
specific attention in a routine activity model of residential burglarly.
The ommission of feedback hypotheses in this context is arguably a
significant oversight in the work completed thusfar on the theory. If
occurrences of criminal victimizations causally influence changes in
routine activities of persons' lives (e.g., decisions to move;
decisions to work), then the conceptual basis of opportunity theory must
be reconsidered.

60
The causal model of routine activity would have to be altered
from its present undirectional form. This major change would have
important methodological ramifications. It is well known, of course,
that estimated parameters within a causal system can assume different
values in a nonrecursive model as opposed to one that is recursive.
Estimation by ordinary least squares, for example, provides unbiased
estimates only if the dependent variable is related to the independent
variables recursively; similarly, cross-sectional data provide reliable
parameter coefficients only when instrumental variables are properly
specified a prior; (He;se,' 1975). Moreover, there are examples in the
research literature that show recursive tests can mistakenly indicate an
assymmetric causal structure when the actual relationship is reciprocal,
or even in a direction opposite of that suggested by a recursive analysis
(Thornberry and Chr; stenson, 1984). In short, if feedback hypotheses can
not be ruled out, causal models in the routine activity literature are
biased due to m;sspecification.
Cohen and Felson (1979), in an early exposition of the routine
activity theory, argue that the location, frequency and distribution of
illegal acts are determined to some degree by the spatial and temporal
structures of legitimate activitiPJ. As the patterns of legal, routine
activities change, therefore, so will the patterns of crime. This is an
intuitively appealing, unidirectional causal relation that explains the
association between lifestyles and the distribution of criminal
victimizations. There are theoretical and empirical justifications,
however, for suggesting that such a representation of the causal flow
between styles of life and victimization is oversimplified.

61
Social researchers in many disciplines have learned that
unidirectional presumptions of causal direction are often unrealistic
given the nature of many social phenonomena. In education, for example,
it is clear that one can not investigate adequately the relationship
between student performance and teacher expectations unless the
interrelationships of these two concepts are taken into account
(Humphreys and Stubbs, 1977) Similarly, recent research indicates that
the association between unemployment and criminal activity is better
explained by a model that incorporates a reciprocal causal structure than
by one which posits a unidirectional affect. (Thornberry and
Christenson, 1984). The assumption here is that the study of crime and
lifestyle falls within this category; an appropriate causal model should
test explicitly for reciprocal effects.
Sufficient evidence can be found in the literature to warrant a
nonrecursive approach in this analysis of the relationship between
routine activities and burglary victimization. Residential burglary,
though devoid of victim-offender confrontation, is likely to contain a
sUbstantial fear component. As raised earlier, the public views burglary
as the prototypical residential crime, and those whose homes are
burg1arized often express strong feelings of psychological intrusion and
personal violation. Dubow (1979) suggests that these feelings arise
because residents are conscious of the potential the incident had for a
confrontation with the offender. Skogan and Maxfield (1981) mention how
the loss of treasured; irreplaceable items, or the financial strain
imposed by the crime can intensify fear. Other factors undoubtedly act
to induce or reinforce such fears. Regardless of the reasons, to the
extent that people react to these fears, peop1e might be expected to

62
change some portion(s) of their daily routines in order to cope with
their fear and reduce the perceived chances for a subsequent
victimization.
Some research on the fear of crime suggests that individuals
react tn various ways to their concern. This;s reflected by the
willingness of people to employ assorted home protective measures
(Lavrakas, 1981), as well as in a wide range of other behaviors (see,
Skogan and Maxfield i 1981). A different, though related, set of findings
sheds additional light in this area. Dubow and Emmons (1981) indicate
that successfu 1 community c'rime prevention programs are most often part
of multi-issue citizen groups and must compete with other community
oriented projects for resources. This is noteworthy because it places
crime firmly among, not separate from, other community problems. People
apparently react to a fear of crime on a community wide basis for the
same reason they mObilize against other problems: they perceive a threat
to their environment.
Although indicative of behavioral reactions to a fear of crime,
the actual behaviors cited are rather trivial. The fact that people
install deadbolt locks or participate in neighborhood watch programs in
response to their fear of crime does not permit the inference that fear
of crime prompts changes in routine activties or lifestyles. The
inference is reasonable only if lifestyle or routines are defined so
broadly that they encomp~ss every behavioral aspect of people's lives, no
matter hO\'1 minor or inconsequential. Such extreme generality is not very
helpful analytically, however.
Additional reasons exist for suggesting a more substantial
impact on lifestyle activities by the fear of crime, however. Fear in

63
general often elicits avoidance behavior. Fear of crime can, for
example, interfere with common social interactions and a1ter lifesty1es
(Hartnagel, 1979; Wilson, 1975). In the case of burglary in particular,
Dubow (1979) reviews evidence that shows burglary victims sometimes
react, because of their fear, to a burglary incident by becoming afraid
to stay home alone. The manifestations of such a reaction will differ'
according to the type of individual, the nature of the crime, and the
level of fear, but substantial change in lifestyle activities seems
implicit. It is quite plausible that, depending on individual
circumstances, a burglary v~ctim might alter work hours or even change
jobs in order to be home the same time as at least one other household
occupant. Other kinds of activities might also be rearranged or
curtai1ed. Alternatively, individuals who live alone might seek a
roommate or begin spending more time in the homes of others. This
evidence, though scant, ;s sufficient to suggest probative value in
further exploration of feedback loops in a model of routine activities
and burglary victimization.

64
-- - --- ---------------~--

CHAPTER 4

Data and MethodoJ.Q.Ql

The conceptual organization of the routine activity theory

requires a dataset with specific properties. The data must accommodate
the longitudinal character of the household level relationships
hypothesized by the theory and provide non-crime information that is
sufficient to allow analyses of the routine activities affecting the
crime targets' (in this case the residence) exposure and vulnerability to
burglary. One of the practical benefits reaped from selecting
residential burglary as the crime of interest is that the routines of
interest revolve around the home. The national crime sample consists of
household addresses with information collected on the characteristics and
major activities of all household members. Although the survey does not
document behavior at work or elsewhere in public, sufficient details on
routine behaviors are avai1ab1e to allow an assessment of the ways
particular behaviors affect household vulnerability to burg1ary.
Moreover, the dataset must permit the tracking of changes over time in
the structure of those routine activities, since such shifts presumably
affect temporal variations in victimization rates. A longitudinal data
file constructed from the houselhold-level information collected by NCS
can accommodate these needs.
The National Crime Survey
The NCS is a record of personal and property victimizations that
includes: rape, robbery, assault, burglary (commerical (until 1977);

65
residential), larceny (personal and household) and auto theft.
Victimization surveys were initiated as an alternative to UCR figures
which are system-dependent and able to furnish su~mary information on
most crimes. In contrast, the NCS collects extensive detail about
selected crimes by interviewing a sample of households. The use of
nationwide victim surveys was first advocated by the 1967 Presidential
Commission on Law El'lforcement and the Administration of Justice;
following pre-tests and pilot studies the survey was implemented in two
forms: city samples and a national sample. Though the files share a
common history, they have different designs. Since this research
analyzes the national data, only the design and structure of that fil~

will be discussed. Detailed treatment of issues peculiar to the

city-level sample can be found elsewhere (e.g., Hindelang, 1976; Garofalo
and Hindelang, 1977).
Although the national survey began in 1973, the Inter-University
Consortium for Political and Social Research (ICPSR) at the University of
Michigan in Ann Arbor, Michigan states that data in the form necessary
for this research are available only since 1977 and processing is now
complete through 1981. The entire data file maintained by the ICPSR
archives consists of approximately 1370 variables arrayed in incident,
person, and household files (see rCPSR, 19B1 for a complete listing). Of
course, much of that file is irrelevant to the present undertaking, so a
reduced file was received from ICPSR which wil', in turn, be trimmed
further to those variables most pertinent to the research task at hand.
Before detailing the precise variables subject to analysis, however, some
of the more general aspects of the dataset are considered.

66
-- ~.----------

Samp1e and Design

The national sample tape obtained from ICPSR contains
approximately 79 t 963 household units in the sample with about 158,963
people in those homes. Units are selected in a stratified multistage
cluster sample. The first sampling stage consists of primary sampling
units (PSUs) formed from all counties in the country. The resulting PSUs
are then grouped into 220 strata by combining areas which shdre similar
characteristics (e.g., geographic region, population density, nonwhite
population). There are 156 remaining strata referred to as
self-representing areas that represent single locations that are certain
to be a part of the sample. One area is chosen from each of the 220
non-self-representing strata based on a probability proportionate to that
area's population.
The other sampling stages insure a self-weighted probability
sample of all housing units within each of the selected geographic
areas. Under such a procedure, each sample unit has the same initial
probability of selection. One thousand-nine hundred-seventy enumeration
districts l are first chosen proportionately according to population,
and segments of clusters containing four households are included in the
sample from within the d~stricts. To insure that the sample reflects
changes in building patterns, procedures are designed to include housing
units built after 1970. A sample is drawn routinely from among permits
authorizing new construction. In those areas that do not issue such
building permits, area segments are sampled in order to detect recent
construction. It is true, of course, that these pro~edures account for
only a small portiQI1 of the total sample; hO\~ever, their contribution
increases steadily as the sample develops over time.

67
There is an analytica1ly relevant problem associated with the
NCSls sampling scheme ~hich must be mentioned: statistical theory
generally assumes a standard random sample. This assumption is, of
course, violated in the national NCS data. Many authors (e.g.,
Lazerwitz, 1968; Kerlinger, 1973) point out that cluster sampling
introduces a sampling error component into the final sample because of
variation in the actual size of the final sample, increased variances,
and enhanced homogeneity within individual clusters. Traditional
5tatistical tests are likely to Y'ield inaccurate estimates of the
standard error and therefore will be misleading. Appropriate adjustments
for correcting the biased estimates of standard error are available
(Lazerwitz, 1968: p. 301-308).
The entire national sample is structured using a panel design.
This feature is especially appropriate for a test of routine activity
theory for the two reasons raised earlier. The panel design accommodates
tests of the perspectivels longitudinal hypotheses and focusses on the
individual level since each panel consists of household addresses and the
individuals who occupy themo The sample of households is divided into
six rotation groups, and following an initial interview, each group is
interviewed every six months for a maximum period of three years. Six
panels are designated within each rotation group and a different panel is
interviewed each month throughout the six months. Initial interviews at
a household are known as bounding interviews and are not used to prepare
estimates of victimization rates. Rather, the information gathered is
used to establish a time reference boundary in order to avoid the
duplication of previously reported events. Summaries of the bounded
events are given to the interviewers so they are better able to detect

68
forward telescoping by respondents. The NCS employs a panel rotation
scheme in order to replenish the sample as households reach the end of a
three year period and exit the sample. Table 1 i11ustrates how one
complete sample is replaced by a second one over three hypothetical
interview years. Each sample has six rotation groups and six panels
comprise a rotation group. In the body of the table, panel numbers are
opposite the months in which they are interviewed, and directly below
their rotation group; panel numbers in parentheses denote bounding
interviews.
NCS File Composition
The NCS data base consists of three separate records of
information. The household file furnishes detail about household
characteristics (e.g., income, number of members) and characteristics of
the surrounding neighborhood (e.g., population composition, housing
patterns) as compiled by Census Bureau in 1970. Information about all
household members over the age of twelve is contained in the person
record. Demographic details and information about work, schooling s or
other major activities are collected from the individual whenever
oossible or a proxy when a certain individual is unavailable. Lastly,
the incident file includes data compiled from reports of property and/or
personal victimizations mentioned during the interviews.
The entire file is quite large and contains much information
that is not relevant to this research. Consequently, the ICPSR archives
were asked to supply a reduced longitudinal file from the household,
incident, and person level records. In an effort to facilitate analysis,
the file was transformed by ICPSR from its hierarchical pattern to a more
conventional rectangular (flat) file. Even after that was completed by

69
the archive personnel, additional data management and manipulation were
necessary. The data arrived separately in two files -- household and
incident, and person. Since it is essential that the data be arranged
and matched correctly according to household, appropriate household unit,
it was necessary to uniquely identify the sample households and sort each
file accordingly before merging the separate records. While a household
identification number is assigned during the data processing phase by
ICPSR, confidentiality requires that those numbers be destroyed. It was
necessary to rely on seven control variables (group sample designation,
psu, segment number, checi digit, serial number, segment type, and
household number) generated by the Census Bureau to identify sample
units, thereby insuring integrity of the data. By sorting the individual
files according to those variables before concatenation, a rectangular
file was merged such that the longitudinal information from both files is
matched with the correct housing unit.
NCS Data Collection
There are two basic sets of questions asked of respondents by
NCS interviewers: the basic screen questionnaire and the crime incident
report (sample copies in Appendix A). The first questionnaire (p. 1-4 in
Appendix A), in addition to the administrative record it provides,
supplies data on household characteristics (Items 1-13b on p. 1 of Basic
Screen Questionnaire), information about each member of the household
twelve or more years old (Items 14-28e on p. 2 and p. 4 of Basic Screen
Questionnaire), and rUdimentary information concerning household crimes
or victimizations sufferred by individuals living in the home.
Preliminary indicatons of possible crimes committed against a household's
property or members come from the household and individual screen

70
questions (Items 29-48). If crimes are uncovered by these screen
questions, then a crime incident report is completed (p. 5~8). One such
report is completed for each crime incident reported to the interviewer.
There are certain circumstances, however, where several
incidents are classified as a series incident on one incident report.
This classification is used only as a last resort and only when three
conditions are met:
1) The incidents must be of the same type with
similar details.
2) There must be a minimum of three incidents in the
series.
3) The respondent must not be able to recall dates
and other details well enough to report the
incidents separately.
Series incidents are problematic for several reasons (see,
Hindelang, 1976; Hindelang and Garofalo, 1977). They will be excluded
from analysis in this research primarily because they represent a rather
small portion of the victimizations considered. Of the 39,591 household
property crimes reported by the respondents in this sample, only 1,218
(3.1 percent) are recorded as series incidents. (The breakdown for
residential burglaries is not yet known). Even if we consider that each
of those actually represents a minimum of three crimes, series incidents
would still account for less than 10 percent of the incidents in the
sample. Considering that no sdtisfactory way has ye!t been devised to
overcome the analytic difficulties presented by series incidents
(Garofalo and Hindelang, 1977), their inclusion seems unwise as well as
unnecessary.
The interviewing practices followed by NCS field workers are
quite well established and well documented (see, U.S. Bureau of the

71
~. ". ,'r' . f •
Census, 1975). Before an interviewer calls on a sample household, a
letter is sent to the occupants informing them of the survey and
explaining that an interviewer will contact them shortly. As soon as
possible after the initial contact by mail, an interviewer visits the
home and interviews as many of the occupants as possible. QUL~~ions

pertaining to the household as a whole are asked once of a household

.
respondent who could reasonably be expected to provide informed answers.
These respondents are also asked to furnish individual information for
the members of their households who are 12 and 13 years old. All other
people in the household are asked the individual screen questions
directly. Every attempt is made to minimize the number of
non-interviews. Of couse, some missing interviews (e.g., unoccupied
units) are unavoidable. But in those instances in which data can not be
gathered because certain individuals are not home, return visits are
scheduled for more convenient times. Return visits are also made to
those units where no one was home initially or the occupants were
uncooperative.
Interviewers receive extensive training and the Census Bureau
maintains quality control checks through direct observation of the field
workers. The accuracy of completed interviews is verified by
reinterviewing a sample of household units and comparing the two sets of
answers. Graham (197ba, 1976b) reports that quality checks generally
show that interviewers follow the established field procedures and little
inconsistency is found when the two sets of responses are analyzed.
Moreover, there are additional efforts to assure a high degree of quality
and consistency. Garofalo and Hindelang (1977) explain that the
centralized data processing involves a clerical edit of all materials

72
submitted from the field. As the data are transferred to a machine
readable format, all the work of operators is verified until an
acceptable degree of proficiency is achieved. Subsequent checks are then
continued periodically. Once the data are compiled, computer editing
tests for internal consistency. Errors found through this process are
checked and appropriate corrections are made. Discrepancies which remain
unresolved are coded as missing values.
NCS Methodological Issues
Beyond the internal checks and experimental assessments of NCS
accuracy, there are other methodological issues that deserve attention.
Although the victimization data were generally welcomed as an alternative
measure of crime, a few warned the research c~mmunity not to view the
figures uncritically or as a panacea. Levine (1976); for example, points
out that incentives for overreporting may exist in the NCS; moreover, he
seems to reflect a basic distrust of exclusive reliance on
observational/self-report data of any kind. Consequently, he argues that
multiple indicators of crime levels and patterns should be used rather
than one type of data. Although his advice might seem prudent in
abstract terms, it probably cannot be realized in practical settings.
Research that specifically investigates the comparability of different
measures of crime (e.g., Hindelang, 1974; Hindelang, 1977; Hirschi, et
~., 1981) shows consistently that UCRs, victimization surveys, and
self-report questionnaires each supplies valid indications of criminality
within the behavioral domain each measure taps. All three are to some
degree imperfect and are not universally interchangeable. The

73
appropriate pOint is not which measure is better or that no measure is
adequate; the question ;s which data are best suited to address the
research question(s) under investigation.
Given that the NCS's national sample is the most appropriate
data for testing the routine activity theory, there are certain
methodological background issues which are relevant. Before nationwide
intervieNing for the NCS commenced in 1972, the Census Bureau conducted
numerous tests to evaluate and refine the proposed procedures. As a
result, significant methodological problems were eXamined ~hich bear on
the survey's reliabilty and validity.
One of the first issues identified in the development of NCS was
the ability of respondents to rememb~r incidents (Ennis, 1967). It was
presumed that variation in this regard would depend, in part, on the
length of a reference period over which a respondent would be asked to
recall. Reverse record checks were used to determine the length of time
which would produce the most reliable data. In their summary of the
results from three pilot projects (\~ashington, D.C., Baltimore and San
Jose), Garofalo and Hindelang (1977) report that for simply determining
whether a victimization occurred, there is no appreciable difference
between a six and twelve-month reference period. However, if we are
concerned about victims recalling accurately the quarter in which the
incident took place (an important characteristic for reliable panel
data), a shorter time period is dramatically superior. The six-month
period was adopted by the NCS.
Studies utilizing the reverse record checks also addressed other
methodological questions that are pertinent to this analysis of the
national sample. Telescoping refers to phenomena of memory in which

74
respondents mistake the time of victimizations. Forward telescoping is
one kind of memory defect that involves respondents who report
victimizations as taking place during the six·month reference period when
the crime actually occurred before the reference period. This can
obviously inflate victimization estimates. The Census Bureau instituted
bounding to conteract the effects of telescoping on the accuracy of the
data. Bounding is a strategy whereby information collected in the first
interview of a household is not used as part of the dataset. Instead,
the results of bounding interviews are used as tools to reduce bias that
is introduced by forward t~lescoping. By comparing a summary of the
victimizations recalled during the bounding interview with those crimes
reported when the household is recontacted, interviewers are able to
detect crimes Which occurred prior to the reference period. Comparisons
of the victimization rates obtained with and without bounding demonstrate
that unbounded estimates are far greater than those recorded when
bounding is used. Tables 3 and 4 constructed by Garofalo and Hindelang
(1977, p. 28) show that the differences in rates are statistically
Significant for both personal crimes (completed and attempted rapes,
robberies, and assaults) and crimes against property (completed and
attempted burglaries, robberies, and auto thefts). Bounding clearly
reduces the inflation of victimization that results from forward
telescoping.
A second type of telescoping which potentially affects responses
on the NCS is internal for~~rd telescoping. This occurs when a
respondent reports a victimization as occurring more recently in the
reference period than it did in actuality. Evidence of internal
telescoping in the national sample is inferred from Ennis's (1967)

75
results showing a tendency for reported victimizations to cluster in
those months of the reference period closest to the month of the
interview (Garofalo and Hindelang, 1977). The problem here is slightly
different than with forward telescoping. Internal telescoping mistakes
the month in which a particular victimization incident occurs, not the
placing of an inr,ldent in the correct reference period. Since this
research focusses on the effects changes in routines at time one have on
victimization probabilities six months hence, the precise month of
occurrence is not necessary. It is sufficient to know whether the
.
househoid suffered a burglary victimization in a given reference period.
Another difficulty involves the degree of trust one can place in
respondents I answers. Reverse record checks allow researchers to
determine what proportion of victimizations known to the po1ice are
actually reported to NCS interviewers. The findings in the three pi lot
project cities show that responses for all crimes tested (assault,
robbery, rape, burglary, and larceny) are quite good; they are
particularly encouraging for burglaries. Table 1, constructed by
Garofalo and Hinde1ang (1977), shows that the percentage of burglary
victims known to police who reported the crime in the victimization
surveys was 88 percent in Washington, BG percent in Baltimore, and 90
percent in San Jose.
Another issue of importance is the question of panel bias.
There are two forms of panel bias which can lead potentially to analytic
complications that are pertinent to this research. Panel bias refers
generally to analytic and interpretive complications that arise from the
panel design. By definition, persons in panel samples remain members of
the sample sufficiently long for data to be collected at more than one

76
point in time. Repeated contacts by researchers that intrude upon the
privacy and time of those in the sample introduce the possibility that
respondents will tire of their involvement in the study. This
motivational fatigue might prompt some 1n the sample to seek ways of
alleviating the inconvience that comes from repeated inquiries by social
researchers. To the extent this type of adaptation influences respondent
participation or, assuming continued participation, the answers that
respondents furnish, the data are biased.
A second complication attributable to the panel format is
censored data. The rotation of panels in and out of the sample results
in artificial time constraints, imposed on both ends of the time span
over which the data are collected. Panel members exiting the sample for
some reason (e.g. II relocation to new home) are not tracked after 1eaving
the sample, thereby resulting in the precipitous censoring of data. In
addition, the panel rotation scheme censors data on both the right and
left sides of the samp1e ' s time line; that is, information is terminated
on rotation group panels that leave at the end of the three-year period
(right-hand censoring).
The difficulties presented by panel bias affect the design and
conduct of this research to different degrees and require explicit
consideration. They will be discussed in the following sections,
beginning with response bias precipitated by loss of respondent
motivati~n and the resulting panel attrition.
Motivational Fatigue in the NCS Crime Panel
The members of each household may be interviewed up to seven
times over three and one half years; the bounding interview followed by
one intervie\'1 every six months for three years. This lengthy period,

77
along with the length of typical interviews and the correspondence or
rescheduling efforts that might be necessary, provide ample reasons for
people to tire of their involvement in the survey. Such motivational
fatigue can become manifest in two ways. Members of households might
refuse to cooperate altogether by not answering even the basic screen
questions, or they could become less willing to report victimizations
since a positive response to that screen question triggers the lengthy
critical incident report. Garofalo and Hindelang (1977) cite two studies
which address these issues. Woltman and Bushery (1975) report that
noninterview rates increas~ with the length of time a household is in the
sample; but the effects are too slight to suggest systematic bias. It
appears that respondents in the panel sample do not regularly choose to
opt out of the survey because of the length of time they must remain in
the sample,
This does not necessarily imply, of course, that the panel
design is not introducing bias; respondent fatigue could result in lower
reporting rates. Woltman and Bushery (1977) investigate this question by
comparing victimization rates of respondents interviewed for the second
time with those interviewed the third time; third-time respondents with
tht.se interviewed for the fourth time, and so forth. They state that
victimization rates decline steadily with the length of time respondents
are in the sample. However, the only statistically significant
difference ;s found between the respondents interviewed for the second
time and those for the third time. Although respondents· productivity
drops the longer they are in the sample, these results do not suggest
widespread effects across the sampl~ period.

78
,I
Censoring of Panel Data
The results of previous research suggesting panel attrition may
not be an extensive problem do not imply that researchers can proceed
with analyses unconcerned about potential bias. Nearly all analytic
models assume that right-hand censoring is independent of the occurrence
of events; that is, sample units are not lost selectively because of
their increased or decreased likelihood of experiencing the event of
interest (Allison, 1982). Although this is a necessary assumption, it
may not always be rea1istic. Ii'\dications of an inverse (though generally
insignificant) association between length of time in an NCS panel and
respondents I reporting rates necessitate some attention.
A straight forward way for assessing whether victimizations
influence panel attrition is with a control group comparison. The data
permit a comparison between households that leave the sample following a
victimization and those that are lost from the sample without a preceding
victimization. If a significant effect of a victimization event upon
mortality is detected in the sample, analyses and conclusions can be
adjusted accordingly.
Left-hand censoring of the data is an altogether different
problem. This refers generically to the lack of information on the
history of members in replacement panels. When households enter the
sample bounded in the first irterview, nothing is known about their
victimization history prior to their inclusion in the sample. Research
suggests that personal, household, and ecological characteristics
interact so as to cluster the risks of housrhold victimizations. For
example, the NCS city sample for 1972 to 1974 shows that the likelihood
for aactl type tlf household victimization is considerably greater for

79
those households that experience another household crime (Hindelang, et
21., 1978). Whatever the nature and shape of the factors that account
-
for this type of victim-proneness, bias is introduced to the extent that
these cllaracteristics affect victimization(s) reported initially by new
sample members.
One aspect of this project's design is intended to neutralize
the effects of left-side censoring. This research concentrates on the
causal relations between burglary victimizations and lifestyle
c~aracteristics of household members. By analyzing victimization events
that occur after some change in lifestyle, a starting pOint for the
analysis is defined which does not arbitrarily truncate the data. This
approach is justified theoretically because the main thesis of routine
activity is that changes in lifestyle patterns influence the
opportunities for certain crimes. This design is similarly applicable to
an assessment of reciprocal causation. The starting point for a
consideration of how victimization(s) affect lifestyle (e.g., moving) is
defined by the occurrence of a victimization event.
This approach to counteracting this type of bias is also well
grounded methodologically. Parameter estimates are affected adversely by
censoring from the left because there is no substantively meaningful
event which accounts for the entry of units into the sample (Al1ison f
1982). Since the focus here is the analysis of victimizations that occur
after some change in daily routines, the starting point is meaningful
theoretically.
Panel Data Availa~le for Analysis
It is unfortunate, t!lough not surprising, that a panel sample
for all the years that victimization surveys have been conducted is not

80
~urrently available for analysis. Practical and administrative factors
combined to limit the number of waves of data to f~ur, and the period of
years ttlose waves cover begins with the third quarter of 1977 an~ ends
after the second quarter of 1981. The ICPSR Archives have an on-going
program continually making more data available on public access tapes for
researchers. At the time this request for the NCS crime pan~l was made,
processin8 had been completed through the second quarter of 1981 ..
Archive personnel also explained that the process of providing access to
the NCS panel began with the mid-1977 data. Any attempt at the time of
this reseacher's request (July 1984) to retrieve data from earlier years
would have been both costly and time-consuming.
Similarly, the number of waves in which the data are arrayed is
also the result of practical limitations. The maximum number of waves
for a given household is six, excluding the bounding interview. ICPSR
staff explained that the hierarchical structure of the dataset
facilitates subsetting of individual files (household, person, and
incident), but matching information across files by waves is considerably
more cumbersome. Data pertaining to specific waves are distributed
throughout the files and must be extracted one wave at a time. ICP$R
rerues~ed that the number of Waves necessary for the study be specified
along with the variables of interest. Four waves were decided ~rorl in an
effort to baiance time and cost constraints with design and analytic
concerns.
The NCS sample of victimizations to be analyzed is the best
availabie data source For investigating the dynamics associated with the
relations between routine activities and residential burglary. Powerful
assessments of the causal relations postulated in Cohen ~ al.ls work

81
---------~

I
I

require an analytic strategy that produces reliable estimates in light of

the expected intercorrelations (i.e., reciprocal causation) and the
inherent problems (e.g., autocorrelation) associated with detecting and
analyzing change over time. The discussion now turnS to the NCS
variables availabe for analysis and the fundamental analytic strategy.

Anal,i'tic
..- Plan
Major Hyputheses
Besides the assumptions and definitions discussed in Chapter 3,
several routine activity hy~otheses applicable to residential burglary
gui de thi s research. It is probably worth restati ng the theory I s major
thesis since it guides the approach this research takes. Cohen and
Cantor (1980) explain that the probability of motivated offender(s),
suitable targets(s), and capable guardians(s) converging in time and
space is increased or decreased as a function of variations in the daily
activities of individuals.
Within the routine activity framework, Cohen, Kleuga1, and Land
(1981) present several multivariate relationships that are relevant to
this research:
Income has opposing effects on burglary victimizations.
Increases lead to lower risk through exposure, guardianship, and
proximity; but also lead to increased target attractiveness and
a higher risk.
Age is invers j related to risk because of lifestyle.
Nonwhites have higher risk3 of burglary due to proximity.
Holding lifestyle and proximity constant; income has a direct
effect on burglary victimization.
Holding lifestyle, proximity, and income constant, race and age
hav& no direct effects on r1sk of victimization.

82
Cohen and Cantor (1980) give the general hypothesis derived from these
considerations as: the greater the proportion of routine activities
centered n~ar the home, the lower the risk of burglary victimization.
Opportunities for' residential burglaries vary according to four
risk factors (exposure, guardians~ip, proximity, and target attractive-
ness). The interplay of these factors influences whether the three
minimal components of victimization will intersect at a particular time
and place. Various daily routines that individuals follow have different
consequences for the exposure, guardianship, proximity, and
attractivent!ss attributed to a tar'get. This in turn r'esults in d'lfferent
victimization rates. It follows, then, that temporal changes in the
daily routines will result in longitudinal fluctuations of the
interaction among the risk fdctors. This will, to the extent that the
theory can be confirmed, account for differences in victimization over
time. Variables that record changes in relevant routines are therefore
necessary.
Before reviewing those variables available in the NCS, a few
gene~~l comments are in order. The variables are displayed in Tables 2-4
and are sometimes discussed individually. All statements concerning
bivariate relationships and descriptions of the influence a variable
presumably exerts on certain risk factors should be read with the
statement "all else being equal" implied. In addition, it is apparent
that some of the variables are redundant. They are included as
safeguards. Complications such as missing data can obviate the use of
certain variables and alternative indicators might be necessary.
Moreover, all variables measuring some aspect of a phenomenon do not
necessarily perform well empirically and judgments must be made after

83
---- --- -----~.~-----~

analysis about the use of particular variables. This was the situation
faced by Cohen and Cantor (1981) when their major activity variable did
not function nearly as well as age when they were used as measures of
activities away from home. Theoretical and empirical considerations
persuaded the authors to rely on age in that case.
The emphasis throughout has been on measuring change in the
routines and charactel'istics of households and their occupants that
translate into enhanced or reduced opportunities for residential
burglary. In light of Cohen and Cantor's hypothesis on activity around
the home, the term change has a specific meaning. It refers specifically
to changes that suggest a shift in the locus of that activity. Relative
dispersion over time of activity away from the household will presumably
affect an increase in the ri5k residential burglary, while changes that
result in a greater concentration of activity around the home result in a
lower risk of victimization (Cohen and Felson, 1981).
Variables Available for Analysis
Several variables in the NCS permit a longitudinal assessment of
how changes in routine activities around the home influence opportunities
fOI· burglary victimization. Data on the household and information on
some routines of hOJsehold occupants will provide specific insight into
how changes in household routines and characteristics affect (over time)
dwelling exposure, its state of guardianship, and its attractiveness as a
burglary target. Before discussion of the housef101d variables contained
in each of the four waves, a comment is in order about some neighborhood
characteristics contained in the dataset.
The NC$ collects neighborhood characteristics summary data
describing neighborhoods of household addresses contained in the panels.

84
- --- ----------

For purposes of the survey, neighborhoods are not identified, but

represent various types of areas populated by approximately 4,000
people. These data are taken from the 1970 Census on populations and
their housing. Because of this, the neighborhood information does not
vary across the sample's time period. One partiaLl set of these variables
from the first wave is contained in the data and will prove usefu1 •
.
t1any of these variables provide neighborhood measures of
~tructural effects hypothesized as important by the theory (see Table
2). Variable 1044 is an i~dicator appropriate for exploring the theorem
that a" ••• decrease [p0Pulafion] ••• in physical locations that are sites
of primary group routine activities ••• produces an increase in
[opportunities for] ••• property crime violations" (Cohen, Felson, and
Land, 1980, p. 99). This variable may also supplement variables 1050,
1051, 1075, and 1086 in an effort to address a point first raised by
Reppetto (1974) when he observed that neighborhoods with the lowest
burglary rates tend to be those with a strong territorial identifica-
tion, such as that found in areas with high ethnic concentrations. Cohen
and Cantor (1981) expand this slightly and propose that one key to a
reduction of burglary may be the tighter social organization of
neighborhoods which translates into a greater degree of guardianship.
Although they maintain that their analysis supports this
proposition indirectly, it is unclear how they reach that conclusion when
their variables include income, age, race, and major activity of the
households' head, and the type of area (central city, other). These five
neighborhood variables provide an opportunity to explore this area more
fu lly by exam; ni n9 differences in burg 1ary rates among nei ghborhoods with
indications of varying degrees of social cohesion. For example, an area

85
-- -- -~----~--------

with a predominantly Hispanic population that owns their homes might be

expected (all else equal) to have a lower burglary rate than a
neighborhood with a heterogeneous population living in rental property.
The income variables (1069-1071) and the poverty measure (1074)
pertain to the consistent finding that burglary is distributed
parabolically according to income. Neighborhoods at the low and high
sides of an income scale would be expected to have similar rates of
burglary, with both higher than the middle income groups. Variable 1072
relates to the ar3ument that social and income inequality influences
victimization rates independently of lifestyle considerations because of
differential access to social resources (Cohen, Kleugal, and Land,
1981). Variable 1054 is the population age classification used in the
NCS that is closest to the offense-prone age. Neighborhoods with high
concentrations of youths mIght be expected to suffer higher burglary
rates than areas with an older population.
The specific effects of these particular neighborhood features
can not be analyzed longitudinally, since the 1970 values are static
across sample years. However, cross-sectional comparisons among
neighborhoods can supply conformation or refutation of some of the
previous research. Moreover, their inclusion in a routine activity
causal model could furnish in.lght into how these characteristics
interact with household routines to influence burglary victimizations.
Incident and Household Variables. Variables from the household,
person, and incident file are arrayed in panel form. These comprise the
core of information on which the proposed analyses will concentrate. The
number of persons that occupy dwellings will of course vary rather widely
among households, as will the number of victimizations. Although the NCS

86
collects personal information on a11 residents, and records every
victimization reported to the interviewer~ practical considerations
necessitated that only households below a certain number of occupants and
incidents be included in the final sample. Descriptive frequencies and
discussions with ICPSR Archive staff let to the decision that households
with a maximum of five occupants or four victimizations per wave would be
part of the final sample.
Table 3 lists household characteristics compiled by the NCS that
are useful in a longitudinal t£~t of routine activity theory. Variable
1010 is an important indica~or in two respects. It is a household
sequence number that records whether people interviewed at one time
(e.g., Wave #2) are the same as those interviewed the previous time (Wave
#1). This enables the researcher to know when members of a household
change, thereby ensuring that individuals and households are matched
correctly over time. In addition, this variable together with 1022 and
1024 provide measures of household stability. The frequency of moving
associated with a household Is a good \~ay to assess reciprocal
hypotheses. If residential moving is associated with victimization, then
crime's effect on changes in lifestyle are better understood, and changes
that may occur in a routine activity model of burglary (because of
estimation with feedback loops in the causal scheme) will enlighten our
understanding of the theory.
Variables 1030 and 1031 furnish measures of household size.
Although not a variable which other studies have considered, it seems
reasonable that the number of people occupying a hOuse will influence
guardianship and exposure of the property. For example, large househr.lds
would be better guarded than small ones by virtue of a greater

87
probability that someone will be home at a given time. Similarly, large
numbers of occupants suggest implicitly a greater likelihood that some
routine activities will take place around the residence. Family income
(1034) is consistently found to be a significant predictor or burglary
(Chapter 3) and is included as a measure of target attractiveness.
Increases and decreases in income over the sample period, with the
,
changes they imply for attractiveness, should be important for thetest
of routine activity theory.
Variables 1096 through 3195 in Table 3 apply to the head of a
.
particular household. In addition to basic demographics, certain of
these variables are expected to convey pertinent information about the
structure of activities around the home. Age (1096) as Cohen and Cantor
(1981) report, has strong implications for property exposure. Older
residents tend to spend more time at home than do young people. Since
burglars prefer unattended property, homes occupied by older people are
likely to be less vulnerable to burglary than houses of the young.
Changes in marital status (1097) are likely to affect opportunities for
burglary as well. Households with one-person heads will, depending on
other factors, present greater opportunities for burglary than those with
heads who are married or cohabit~ting with another adult. Consideration
of marital status in conjunction with other variables present some unique
possibilities for testing the theory. For example, if a married person
who did not work previously takes a job (variable 1103), the decrease in
time spent near the home and the increased income have direct
implications for the property's state of guardedness and its
attractiveness. Other combinations of change among these variables can

88
of course be evident in the data. The point to be emphasized is that
they represent useful indicators for assessing the impact of lifestyle
changes on burglary victimization.
The last five variables in Table 3 (3191-3195) are taken from
information solicited from victims as part of the incident
questionnaire. Since the head of a household is defined by NCS as the
victim of a residential burglary, they are included here with other data
on household heads. Work-related behavior ;s an important determinent of
the routines most people follow. Changes in a household headls work
patterns are likely to influence victimization. Perhaps the most unique
aspect of the dataset employed in this research is the inclusion of
detailed information on individuals who share housing. Table 4 lists the
variables available for those who occupy a household (a maximum of five
for each unit). Besides the personal attributes these items reflect
(2007-2013, 2034), data are provided that document work behavior
(2020-2033, 2035) and classify other major activities (2018). Such
information present the opportunity to research the texture of household
related behavior in detail. Relatively subtle changes that affect the
frequency of behavior around the house are indicated and their effect on
burglary victimization over time can be assessed.
Panel Methods
The analysis is made possible by recent advances in statistical
panel methods (Kessler and Greenberg, 1981; Joreskog and $orbom, 1981).
A minimal panel model, functional form not withstanding, is

89
-- ~- ----~ -- ~- ---~------ -
~~~~ l

where Bt is burglary v'ictimization in the t!!}. year, At is some

routine activity (work or leisure behavior, homeownership, etc.), Z is
some static exogenous variable (neighborhood characteristics, e.g.), and
u~ is an error term. Since victimization may affect some, but
certainly not all, activities (feedbac~ loops apply here), the minimal
model would be expanded to:

Exogenous instruments sufficient to identify the model are available.

The emphasis in the panel analysis to be undertaken is on change
over time. Such change is crucial if we are to infer the causal
re1ations set forth by the theory. There are, of course, several
analytic techniques appropriate for analyzing causal associations.
Prominent examples include cross-sectional analysis, trend studies, time
series, and continuous time series. While serious questions exist
regarding the efficiency of the first, trend studies are best suited for
assessing change in one variable, and continuous time series methods have
limited application in social science research (Kess1er and Greenberg;
1981). Although similarities exist between panel analysis and time
series, the former is best suited given the nature of the data. More
than two waves of data is generally considered fortuitous for panel
methods, while measurements taken at four points in time comprise an
inadequate design for time series analysis (see, Markus, 1979). A panel

90
study collects repeated observations on a number of sample units at two
or more points in time, thereby furnishing cross-sectional as well as
temporal inform3tion.
To a cetain e~tent, the panel analysis will be empirical.
Theory is not strong enough to specify the lags in equation (2). Some
models address this issue (e.g., Balkin, 1979; Balkin and McDonald,
1981), but Cohen and his associates, as discussed previously, ignore this
significant issue. The results could point the way toward modification
of the theory and augment existing knm'lledge of the relationship between
routine activities and cri~e. The larger task is confirmatory> however.
The panel model is inlerpreted as any structural equation model and,
except for differences in estimation (in this instance~ full information
maximum likelihood), it is subject to the same limitations. By
incorporating lagged endogenous variables into equations (1) and (2), the
model explicitly accounts for structured change (that is, change that can
be predicted from initial scores); omitted variables are implicitly
represented; and regression artifacts are controlled (Kessler and
Greenberg, 1981: Chapters 3 and 4). But the most rigorous test of
routine activity hypotheses concerns the longitudinal stability of
structural effects. The effects of activ.ties on victimization (and vice
versa) are expected to be equal across the 1977 ... 1981 study period. The
theory ;s viable to the extent that its structural effects are stable.
The multiwave (four) deSign of this particular panel study
presents additional advantages for this particular theoretical test. As
Heise (1969) states, the inability to distinguish between measurement
error and instability (change) in a two-wave model make it difficult at
best to study the exact nature of causal dynamics operating within that

91
system. Furthermore, two-wave designs can be identified only if both
cross-lagged and contemporaneous effects can be estimated by fixing their
values (Kessler and Greenberg, 1981). Otherwise the model is
underidentified and its parameters cannot be estimated. However, it is
possible to identify a model consisting of three or more waves by
imposing consistency constraints on the parameters. [Detailed
discussions along with algebraic representations are found in Greenberg,
Kessler~ and logan (1979) and in Kessler and Greenberg (1931: p.
33-46)]. Considered generally, such constraints involve the assumption
that certain structural ef~ects are constant between the first and last
waves. This is certainly a much weaker assumption than fixing specified
values in order to reduce the number of parameters that must be
estimated. In that way it is possible to identify the model under
consideration. Stated substantively in terms of a routine activity
model, one would expect the effects of daily activities on victmization
(and vice versa) to be equal across the 1977-1981 period.
Besides their increased complexity, structural equation mOdels
present speCial estimation problems. The choice of an estimation
procedure, of course, depends on the data being analyzed and the model
being estimated. The routine activity mode1 under consideration involves
a discrete dependent variable (victimization/non-victimization) which
requires special consideration. In such cases, Hanusheck and Jackson
(1977) explain,
••• we are interested in choice behavior or the
occurrence/non-occurrence of a particular event.
When this is the case, we are not interested in
estimating the value or numerical size of the

92
- -------"----~-------------

dependent variable. Instead, we are interested in

analyzing the underlying probability of a given event
or choice; more specifically~ how a series of
exogenous variables influences the underlying
probabilty (po 215).
Stated another way, it is not the probabilities themselves which are of
interest, but whether or not a specific event takes place under various
circumstances. Problems Bre exacerbated in cases, such as this, where
the model contains jOintly dependent variables and error terms are
correlated with the endogenous variables. Since these conditions produce
biased and inconsistent results, alternative procedures such as maximum
1ike 1i hood are generally recommended (Hanushek and Jackson, 1977; Draper
and Smith, 1981). The maximum likelihood technique for estimation
involves chOOSing the underlying parametel~s of a given distribution that
maximize the observed results. By relying on estimates of the known
distribution f parameter values Bre selected that would give the highest
probability of obtaining that particular outcome. Statistically,
according to Draper and Smith (1981), maximizing this likelihood is the
same as minimizing the sum of absolute errors. This is in contrast to
least squares which seeks to minimize the sum of squares of errors.
Joreskog and Sorbom (1981) state that the maximum likelhood principle is
based on the assumption that the observed variables display a multinormal
distribution, 'thereby becoming most precise in large samples.
Panel Model Estimation
.........-..- ----......

There are various techniques available for computing the

parameters contained in a causal panel model. Two cornplelllentiu'y
procedures will be employed in this research. The first is log linear
analysis. It is an approach that is well suited to the study of
categorical variables and is often suggested as a procedure for analyzing

93
panel data and estimating causal parameters (Markus, 1979). There are
two major approaches to the log linear analysis of cross tabulated data.
One is the general model that does not distinguish between independent
and dependant variables in the exploration of mutual associations among
variables. The second, a special case of the first, is the logit model
in which a dependent variable is designated. The first model analyzes
expected cell frequencies as a function of all the variables whereas the
latter dissects the expected odds as a function of the other independent
variables (Knoke and Burke, 1980). Since this research has a clear
dependent variable, the logit model is applicable.
The focus of the log linear (logit) analysis is on direct
effects exerted by the independent variables and on interactions among
them. Other research (e.g., Cohen and Cantor, 1981) uses this technique
because it permits the multivariate capability of analyzing
simultaneously the effects of lifestyle and demographic variables on
victimization. In fact 10git models involving categorical variables are
frequently described as analogs to linear regression models employing
continuous data (Bishop, Fienberg, and Holland, 1975; Knoke and Burke,
1980). The analytic logic that underlies this a"proach will be
informative.
One typical way of proceeding with a log odds analysiS is to
begin with a full or saturated model. This is a model of the data that
contains the same number G? effects as the degrees of freedom contained
in the cross-classified table. Theoretical or empirical considerations
could lead the researcher to conclude that particular terms in the model
represent insignificant or trivial effects. Under such conditions, the
appropriate term(s) are dropped from the model and the new model is

94
estimated. The odds resulting from the revised calculations will likely
differ from the odds associated with the saturated model. A standard chi
square test can then be used to test for significant differences in the
effect parameters generated by each model. If the difference is
insignificant, that term may be excluded from consideration. This
sequential procedure continues until as parsimonious a model as possible
is developed without sacrificing goodness of fit to the data. The
extension of this general analytic strategy to the analysis of change
over time in a panel sample is detailed by Bishop, et ale (1975: Chapter
7) and by Knoke and Burke (1980: p. 49-53).
Log linear analysis, like all statistical techniques, has its
shortcoming~ and limitations. Of the limitations raised in the
professional literature, the one which is most problematic for this
project is the relative inability of log linear models to deal adequately
with the adverse consequences of measurement error. Markus (1979)
acknowledges this difficulty and covers two methods developed as
potential tools for copying with its effects. After reviewing each
adjustment procedure, he concludes that the use of one is unwieldly and
hindered considerably by the lack of some measure for goodness of fit.
The other technique is useful, according to Markus, only under rather
restrictive circumstances. Fortunately, an alternative is available.
LISREL (Linear Structural Relations) is a computer program,
currently available in its sixth version, which was developed by Karl
Joreskog for estimating the parameters of causal models which are
confounded by measurement error. Joreskog and Sorbom (1981) explain
that, in its general form, the LISREL model consists of a measurement
model and a structural equation model. The former describes how

95
hypothetical constructs are measured in terms of the measured variables
and the latter specifies the causal model. The parameters are estimated
by employing a full information, maximum likelihood approach to the
analysis of relationships represented in the structural equations. The
procedure enables the researcher to fit the theoretical model to the data
and then test relevant hypothesis via a chi-square statistic that is
generated. The LISREL estimates are able to accommodate correlated error
and interdependence among variables. This is particularly important
given our longitudinal panel sample and the presumed reciprocity expected
between victimization and some lifestyle variables. LISREL estimates the
correlations within the context of the entire model and adjusts the
parameter estimates in light of the covariation.
Research Contribution
The combination of the unique dataset and analytic strategy
outlined permit a crucial test of the routine activity theory which has
heretofore not been conducted. The theory represents a significant
development in the study of opportunity structures that influence
criminal victimization. Most assessments of routine activity theory have
either been flawed seriously (e.g., macro-level indicators or
misspecified models) or only suggestive of the perspective's
applicability. This research is designed to overcome past shortcomings
by specifically testing the longitudinal, incident-level model set forth
by the theory and, perhaps, suggest new directions for development of the
theory and criminal/opportunity research in general.
The focus of this research is clearly theoretical. That is, the
purpose is to provide a crucial, longitudinal test of a criminological
theory that purports to break new ground in the scientific study of

96
crime. Results from this test will furnish an original contribution to
the development of routine activity theory because the questions this
project pursues have been overlooked in the professional literature. The
results of this research might suggest alterations to the perspective and
will also advance the theoretical understanding of crime in general.
This is not meant to ignore potential policy implications that
flow from the theory. The routine activity theory characterizes crime as
a natural consequence of ecological change so the results of this
research will enlighten the predictive utility of the theory. If
longitudinal hypotheses are supported, then the structural effects
associated with the theory's hypotheses will be accepted as part of the
theory. Thp. results can undoubtedly influence future attempts to
forecast crime and this may be seen as a definite policy benefit.
The operational consequences of this research go beyond
forecasting, however. The routine activity theory points directly toward
aspects of crime prevention policies that are not ordinarily considered
by criminal justice agencies. Housing and employment patterns, for
example, usually play no role in criminal justice policy. The routine
activity perspective deals explicitly with these policy areas for crime
prevention planning. Systematic inquiry into the precise nature of the
relationship between these factors and crime trends will also enhance
certain police practices. For example, knowledge of changes in a
community's housing patterns would permit more informed decisions
regarding personnel deployment. Also, efforts to design and organize
community crime prevention efforts could be augmented significantly by
knowledge of how routine activities in that community affect crime rates
and patterns.

97
Notes
1. Enumeration districts are defined as small geographic areas with
well-defined boundaries established for the 1970 census that contain 250
households spread throughout a PSU.

98
REFERENCES
Allardt, E. Aggregate analysis: The problem of its informative
value. In M. Dog~n &S. Rokkan (Eds.) Quantitative ecoloqical
analysis in the social sciences. Cambridge: the M.I.T. Press, 1969.
Allison, P. D. Discrete-time methods for the analysis of event
, hi stori es. In S. Leinhardt (Ed.) Sociological methodology 1982. San
Francisco: Jossey-Bass, 1982.
Amir, M. Patterns of forcible rape. Chicago: University of Chicago
. Press, 1971.
Baker, M.H., Nienstedt, B.C., Everett, R.S. and McCleary, R. The impact
of a crime wave: Perceptions, fear, and confidence in the police.
Law and Society Review, 1983, ~, 319-335.
Balkin, S. Economic factors and the rate of crime. Land Economics,
1978, 48, 193-196. .
Balkin, S. and McDonald, J. The market for street crime: An analysis of
victim-offender interaction. Journal of Urban Economics, 1981, 64,
154-164.
Biderman, A., Johnson L., McIntyre, J. and Weir, A. Report on a pilot
study in the Distict of Columbia on victimization and attitudes
Toward Law Enforcement. President1s Commision on Law Enforcement and
Administration of Justice, Washington, D.C.: G.P.O., 1967.
Bishop, Y.M.M., Fienberg, S.E., and Holland, P.W. Discrete mUltivariate
analysis: Theory and practice. Cambridge: M.I.T. Press, 1975.
Bittner, E. and Messinger, S.L. Editors, Criminology review yearbook
Volume 2. Beverly Hills: Sage, 1980.
Black, D.J. Production of crime rates. American Sociological Review.
1970, ~, 733-748.
Block, R. Community, environment and violent crime. Criminology, 1979,
ll., 46-57.
Boggs, S.L. Urban crime patterns. American Sociological Review, 1965,
30, 899-908.
Booth, A., Johnson, D.R. and Choldin, H.M. Correlates of city crime
rates: Victimization surveys versus official statistics. Social
Problems, 1977,1[, 189-197.
Briar, S. and Piliavin, I. Delinquency, situational inducements and
ct.:nmitment to conformi ty. Soc; a1 Prob 1ems, 1965, 11, 35-45.
Byrne, J.M. Ecological correlates of property crime in the United
States: A macro-environmental analysis. Paper presented at the 35th
annual convention of the American Society of Criminology, Denver,
Colorado, November, 1983.

99
Cantor, D. and Land, K. Unemployment and crime rates in the post World
War II United Staes: A theoretical and empirical analysis. Paper
delivered at the 35th annual meeting of the American Society of
Criminology, Denver, Colorado, November, 1983.
Carroll, L. and Jackson, P.I. Inequality, opportunity and crime rates in
central cities. Criminology, 1983, n, 178-194.
Chimbos, P.O. A study of breaking and entering offenses in Northern
City, Ontario. Canadian Journal of Criminology and Corrections,
1973, 1[, 316-325.
Clarke, R.V. Opportunity-based crime rates. The difficulties of further
refinement. British Journal of Criminology, 1984, 24 (1), 74-B3.
Clarke, R.V.G. Situational crime prevention: Theory and practice.
British Journal of Criminology, 1980, 20, 136-147.
Clarke, S.H. Burglary and 'larceny in Charlotte-Mecklenburg: A
description based on police data. Charlotte: Mecklenburg Criminal
Justice Planning Council, 1972.
Clinard, M.D. and Quinney, R. (Eds.) Criminal behavior systems. New
York: Holt, Rinehart and Winston, 1967.
Cloward, R.A. and Ohlin, L.E. Delinquency and opportunity. New York:
Free Press, 1964.
Cohen, Albert K. Delinquent boys. New York: Free Press, 1955.
Cohen, L.E. and Felson, M. Modeling crime trends: A criminal
opportunity perspective. Journal of Research in Crime and
Delinqeuncy, 1981, ~, 138-163.
Cohen, L.E. and Cantor, D. The determinants of larceny: An empirical
and theoretical study. Journal of Research in Crime and Delinquency,
1980, lI, 149-159.
Cohen, L.E. and Cantor, D. Residential burlary in the United States:
Life-Style and demographic factors associated with the probability of
victimization. Journal of Research in Crime and Delinquency, 1981,
~, 113-127.

Cohen, L.E. and Felson, M. Social change and crime rate trends.
American Sociological Review, 1979, 44, 588-60B.
Cohen, L.E., Felson, M. and Land, K.C. Property crime rates in the
U.S.: A macrodynamic analysis, 1947-1977 with ex ante forecasts for
the mid-1980s. American Journal of SociolQgl, 1980, 86, 90-118.
Cohen, L.E., Kleugal, J.R., and Land, K.C. Social inequality and
criminal victimization. American Sociological Review, 19B1, 46,
505-524.

100
Conklin, J.R. and Bittner, E. Burglary in a suburb. Criminology, 1973,
11, 206-232.
Decker, S.H. and Kohfe1d, C.A. Fox re-examined: A research note
examining the perils of quantification. Journal of Research in Crime
and Delinquency, 1982,12, 110-119.
Decker, D.L., O'Brien, R.M. and Shichor, D. Patterns of juvenile
victimization and urban structure. In William H. Parsonage (Ed.)
Perspectives on victimo10gy, Beverly Hills: Sage, 1979.
deGrazia, S. Of Time, work, and leisure. New York: The Twentieth
Century Fund, 1962.
Dogan, M. and Rokkan, S. (Eds.) Quantitative ecological analysis
in the social sciences. Cambridge: The M.I.T. Press, 1969.
Draper, N.R. and Smith, Jr. H. Applied regression analysis (Second
Edition). New York: Jbhn Wiley, 1981.
Dubow, F. Reactions to crime: A critical review of the literature.
Washington, D.C.: G.P.0.,1979.
DuBow, F. and Emmons, D. The community hypothesis. In Dan A. Lewis
(Ed.) Reactions to crime. Beverly Hills: Sage, 1981.
Durkheim, E. The rules of sociological method. Glencoe, Illino;s: Free
Press, 1938.
Dunn, C.S. The Analysis of Environmental Attribute/Crime Incident
Characteristic Interrelationships. Unpublished Ph.D. dissertation,
SUNY at Albany, 1974.
Emery, F.E. and Trist, E.L. Towards a social ecology: Contextual
appreciation of the future in the present. New York: Plenum Press,
1973.
Ennis, P.H. Criminal victimization in the United States: A report of a
national survey. The PreSident's Commission on Law Enforcement and
Administration of Justice. Washington, D.C.: G.P.O., 1967.
Erlich, I. Participating in illegitimate activities: A theoretical and
empirical investigation. Journal of Political Economy, 1973, ~,
521-565.
Ferdinand, T.N. Typologies of ee1inguency. New York: Random House, 1966.
Ferdinand, T.N. Demographic shifts and criminality. British Journal of
Criminology, 1970, lQ, 169-175.
Flango, V.E. and Sherbenou, E.L. Poverty, urbanization, and crime.
Criminology, 1976, li, 331-346.
Fox, J.A. Forecasting crime data. Lexington: Lexington Books, 1979.

101
-- - - -- - - - - - - - - - - - - l

Fox, J.A. Re-examining some perils of quanitifaction in the econometric

study of crime: A reply to Decker and Kohfeld. Journal of Research
in Crime and Delinquency, 1982,12, 122-131.
Garofalo, J. and Hindelang, M.J. An introduction to the national crime
survey- Washington, D.C.: G.P.O. 1977.
Gibbons, D. Some observations on study of crime causation. American
Journal of Sociology, 1971, 77, 262-278.
Gibbs, J.P. Sociological theory construction. Hinsdale, Illinois:
. Dryden Press, 1972.
Gibbs, J.P. and Erikson, M.L. Crime rates of American cities in an
ecological context. American Journal of Sociology, 1976, g, 605-620.
Gibbs, J.P. and Martin, W.T. Toward a theoretical system of human
ecology. Pacific Sociological Review, 1959, 2, 29-36.
i -

Gillespie, R.W. Economic factors in crime and delinquency: A critical

review of the emeirica1 evidence. Unpub1. final report submitted to
NILECJ, 1975.
Gold, M. Status forces in delinquent bays. Ann Arbor: Institure for
Social Research, University of MiChigan, 1963.
Gould, L. The changing structure of property crime in an affluent
society. Social Forces, 1969, ~, 50-59.
Graham, D. Res onse errors in the national crime surve ~ ei ht
~ive largest cities, Mimeographed paper. Washington,
U.s. Bureau of the Census, 1976a.
Graham) 0, National crime surve evaluation of errors in coverage and
content. (Mimeographed paper. Washington, D.C.: U.s. Bureau of
the Census, 1976b.
Greenberg, D.F., Kessler, R.C. and Logan, C.H. A panel model of crime
rates and arrest rates. American Sociological Review, 1979, 44,
843-850.
Hanushek, E.A. and Jackson, J.E. Statistical methods for social
scientists. New York: Academic Press, 1977.
Harries, K.D. The geography of crime and justice. New York: McGraw
Hill, 1974.
Hartnagel, T.F, The perception and fear of crime: Implications for
neighborhood coheSion, social activity, and community affect. Social
Forces, 1979, 1§., 176-
Heise, O.R. Causal analYs.ll- Ne~'1 York: Joim ~~iley and Sons, 1975.
Hawley, A. Human ~co10gy: A Theory of Community Structure. New York:
Ronald, 1950.

102
Hinde1ang, M.J. The Uniform Crime Reports revisited. JournaL2.L
frimina1 Justice, 1974, 1, 1-18.
Hinde1ang, M.J. Criminal victimization in eight american cities.
Cambridge: Ballinger, 1976.
Hindalang, M.J. Race and involvement in common law personal crimes.
American Sociological Review, 1978, 43, 93-109.
Hindelang, M.J., Gottfredson, M.R. and Flanagan, T.J. §ourcebook of
criminal justice statistics - 1980. Washington, D.C.: G.P.O., 1981.
Hindelang, M.H., Gottfredson, M.R., and Garofalo, J. Victims of person.li..
crimes: An empirical foundation for a theory of personal
victimization. Cambridge: Ballinger, 1978.
Hirschi, T. Causes of Delinquency. Berkeley: University of California
Press, 1969.
Humphreys, L. G. and Stubbs, J. A longitudinal analysis of teacher,
student expectation and student achievement. Journal of Educational
Measurement, 1977, 14, 261-270.
Inter-University Consortium for Political and Social Resaarch, National
crime surveys 1973-1979 (ICPSR Edition). Ann Arbor: The University
of Michigan, 1981.
Jones, E.T. Crime change patterns in American cities. Journal of
Criminal Justice, 1976, i, 333-340.
Joreskog, K.G. and Sorbom, D. LISREL V Analysis of linear structural
relationships by maximum likelihood and least squares methods.
Chicago: National Education Resources, 1981.
Kerlinger, F.N. Foundations of behavioral rezearch, Second Edition, New
York: Holt, Rinehart and Winston, 1973.
Kessler, R.C. and Greenberg, O.F. Linear panel analysis: Models of
quantitative change. New York: Academic Press, 1981.
Kitsuse, J.I. and Circourel, A.V. A note on the uses of official
statistics. Social Problems, 1963, 11, 131-139.
K1ockars, C. The Professional fence. New York: The Free Press, 1974.
Knoke, David and Burke, Peter J. Log linear models. Sage University
Paper series on Quantitative Applications in the Social Sciences.
Beverly Hills: Sage, 1980.
Xuic, V. Work, Leisure and Culture. Review of Politics, 1981, 43 (3),
436-465.

103
Land, K.C. and Felso~\ M. A general framework for building dynamic
macro social ina;~ator models including an analysis of changes in
crime rates and police expenditures. American Journal of Sociology,
1976, ~, 405-421.
Langbein, Laura Irwin and Lichtman, Allan J. Ecological inference. Sage
University Paper on Quantitative Applications in the Social Sciences,
Series Number 08-010. Beverly Hills: Sage, 1978.
Laub, J.H. Criminal Behavior and the Urban-Rural Dimension. Unpublished
Ph.D. disseration, State University of New York at Albany, 1980.
Lavrakas, P.J. On households. In D. A. Lewis (Ed.). Reactions to
Crime. Beverly Hills: Sage, 1981.
Lazerwitz, B. Sampling theory and procedures. In Hubert M. Blalock, Jr.
and Ann B. Blalock (Eds.) Methodology in social research. New York:
McGraw Hill, 1968.
Letkemann, P. Crime as work. Englewood Cliffs: Prentice Hall, 1973.
Levine, J.P. A critique of victimization surveys. Criminology~ 1976,
1.1., 307-330.
Lewis, D.A. and Salem, G. Community crime prevention: An analysis of a
developing strategy. frime and Delinquency, 1981, £I, 405-421.
Liebow, E. Tallts corner. Boston: Little, Brown, 1967.
Maltz, M.D. Crime statistics: A historical perspective. Crime and
Delinquency, 1977, ~, 32-40.
Markus, G.B. Analyzing panel data. Beverly Hills: Sage, 1979.
McCleary, R., Nienstedt, B.C., and Erven, J.M. Uniform crime reports as
organizational outcomes: Three time series experiments. Social
Problems, 1982, 29, 361-372.
Merton, R.K. Social structure and anomie. American Sociological Review,
1938, 1, 672-682.
Moore, Gary T. and Golledge, R. G. Environmental knowing:
Theories, research, and methods. Stroudst"lrg, PA: UOt'/den,
Hutchinson & Ross, Inc., 1976.
National Commission on the Causes and Prevention of Violence. The final
report: To establish justice to insure domestic tranquility, New
York: Bantam BoOkS, 197U.
Nettler, G. Explaining crime, Second Edition, New York: McGraw Hill,
1978.
Pike, L.O. A history of crime in England. London: Smith and Elden,
1876.

104
Pope, C.E. Crime-specific analysis: The characteristics of burglary
incidents. Washington, D.C.: G.P.O., 1977.
Pope, C.E. Crima-specific analysis: An empirical examination of
bU\~glary offender charactistics. Washington, D.C.: G.P.G., 1977a.
Pope, C.E. Crime-specific analysis: An empirical investigation of
. burglary offense and offender characteristics. Washington, D.C.:
G.P.G., 1977b.
President's Commission on Law Enforcement and the Administration of
. Justice. Task force report: Crime and its impact an assessment.
Washington, D.C.: G.P.O., 1967.
Quetelet, A. A treatise on man. Edinburgh: Chambers, 1842.
Quinney, R.. Structural characteristics, population areas, and crime
rates in the United States. Journal of Criminal Law, Criminology,
and Police Science, 1966, 21, 45-52.
Reiss, A.J. Settling the frontiers of a pioneer in American
criminology: Henry McKay. In James F. Short (Ed.) Delinquency,
crime and society. Chicago: University of Chicago Press, 1976.
Reppetto, T.J. Residential crime. Cambridge: Ballinger, 1974.
Robinson, S.M. A critical review of the Uriform Crime Reports. Michigan
Law Review, 1966, 64, 1031-1054.
Robinson, W.S. Ecological correlations and the b~havior of individuals.
American Sociological Review, 1950, 15, 351-357.
Sagi, P.C. and Wellford, C.F. Age compositions and patterns of change in
criminal statistics. Journal of Criminal, Criminology, and Police
Science, 1968, ~, 29-36.
Sampson, R.L. Structural density and criminal victimization.
Criminology, 1983, Ki, 276-293.
Sampson, R.L., Castellano, T.C. and Laub, J.H. Juvenile criminal
behavior and its relation to nei hborhood characteristics. (Research
Monograph. Albany: Criminal Justice Research Center, 1981.
Santa Clara Criminal Justice Pilot Progralll, Burglary in San Jose,
Springfield, Virginia: U.S. Department of Commerce, 1977.
Scarr, H.A. Patterns of burglary. Washington, D.C.: G.P.O., 1973.
Scheuch, E.K. Social context and individual behavior. In M. Dogan and
S. Rokkan (Eds.) Quantitative ecological analysis in the social
sciences. Cambridge: The M.I.T. Press, 1969.
Schuessler, K. Components of variation in city crime rates. Social
Problems, 1962, ~ (4), 314-323.

105
Shaw, C.R., McKay, H.D., Zorbaugh, F., and Cottrell, L.S. Delinguency
areas. Chicago: University of Chicago Press, 1929.
Skogan, W.G. The victims of crime: Some material findings. In Anthony
L. Guenther (Ed.). Criminal behavior in social systems. Chicago:
Rand McNally, 1976.
Skogan, W.G. Dimensions of the dark figure of unreported crime. Crime
and Delinquency, 1977, ~, 41-50.
Skogan, W.G. and Maxfield, M.G. Coping with crime. Beverly Hills:
Sage, 1981.
Sparks, R. Criminal opportunities and crime rates. Paper delivered at
29th annual meeting of the American Society of Criminology, Atlanta,
1977 •
Stafford, M.C. and Gibbs, J.P. Crime rates in an ecological context:
Extension of a proposition. 'Soci a1 Sci e.nce Quarterly, 1980, .§l,
653-665.
Szalai, S. The use of time. The Hague, Morton, 1973.
Thornberry, T. and Christenson, R.L. Unemployment and criminal
involvement: An investigation of reciprocal causal structures.
American Sociological Review, 1984, 49, 398-411.
Tach, H. Violent men (Revised Edition). Cambridge: Schenkman
Publishing, 1980.
Turner, J.T. The structure of sociological theory (Revised Edition).
Homewood, Illinois: The Dorsey Press, 1978.
Turner, S. The ecology of delinquency. In Thorsten Sellin and Marvin E.
Wolfgang (Eds.) Delinquency selected stUdies. New York: John Wiley
and Sons, 1969.
U.S. Bureau of the Census. Interviewer's manual: National crime survey,
central cities sample. Washington, D.C.: G.P.O., 1975.
United States Department of Justice, Crime in the United States 1982.
Washington, D.C.: G.P.O., 1983.
Wellford, C.F. Age composition and the increase in recorded crime.
Crim;no~ogy, 1973, 11, 61-70.

Wilks, J.A. Ecological correlates of crime and delinquency_ In

President's Commission on Law Enforcement and Administration of
Justice, Task force report: Crime and its impact--an assessment.
Washington, D.C.: G.P.O., 1967.
Wison, James Q. Thinking about crime. New Yor~: Basic Books, 1975.
Wolfgang, M.E. Patterns of criminal homicide. New York: John Wiley,
1958.

106
Wolfgang, M.E. Uniform Crime Reports: A critical appraisal. University
of Pennsylvania Law Review, 1963, lll, 708-738.
Woltman, H. and Bushery, J. A panel bias study in the National Crime
Survey. Proceedings of the Social Statistics Section of the American
Statistical Association, 1975.
Wol~man, H. and Bushery, J. Update of the NC$ panel bias study.
Washington, D.C.: U.S. Bureau Of the Census, 1977.

107
APPENDIX A

TABLES
Table 1
NCS Rotation Group Structure

Tri'terv; ew First Sample Second Sample

Month 1 2 3 4 5 6 2 3 4 5 b
(Rotat;'on Group~

Jan. 1 1 1 1 1 1 ( 1)
Feb. 2 2 2 2 2 2 (2)
March 3 3 3 3 3 3 (3 )
April 4 4 4 4 4 4 (4)
May 5 5 5 5 5 5 (5)
June 6 6 6 6 6 . 6 (6 )
July 1 1 1 1 1 1 (1)
August 2 2 2 2 2 2 (2)
Sept. 3 3 3 3 3 3 (3)
Oct. 4 4 4 4 4 4 (4)
Nov. 5 5 5 5 5 5 (5)
Dec. 6 6 6 6 6 6 (6)
Jan. 1 1 1 1 1 1 ( 1)
Feb. 2 2 2 2 2 2 (2)
March 3 3 3 3 3 3 (3)
Apri 1
~~~
4 4 4 4 4 4
May 5 5 5 5 5 5
June 6 6 6 6 6 6 (6)
July 1 1 1 1 1 1 ( 1)
Aug. 2 2 2 2 2 2 (2)
Sept. 3 3 3 3 3 3 (3)
Oct. 4 4 4 4 4 4 (4)
Nov. 5 5 5 5 5 5 (5)
Dec. 6 6 6 6 6 6 (6)
Jan. 1 1 1 1 1 1 (1)
Feb. 2 2 2 2 2 2 (2)
March 3 3 3 3 3 3 (3)
Apr; 1 4 4 4 4 4 4 (4)
May 5 5 5 5 5 5 (5 )
June 6 6 6 6 6 6 (6)
July 1 1 1 1 1 1 (1)
Aug. 2 2 2 2 2 2 (2)
Sept. 3 3 3 3 3 3 (3)
Oct. 4 4 4 4 4 4 (4)
Nov. 5 5 5 5 5 5 (5)
Dec. 6 6 6 6 6 6 (6)

Panel members are shown in rows directly below their rotation group. The
panel members enclosed in parentheses indicate the bounded interview.

Source: ICPSR, 1981

Table 2
Static Neighborhood Characteristics from National Crime Survel

Variable Number a Variable Name

1011 Land Use
1012 Place Size
1013 Place Description
1025 Living Quarters
1044 Primary Individuals/Total Households
1050 Blacks/Total Population
1051 Spanish/Total Population
1054 Population 16-21 Years Old/Total Population
1069 Families Below $5000 Income/Total Population
1070 . Families Above $14,999 Income/Total Population
1071 Median Income
1072 GINI Coefficient
1074 Population Below Poverty/Total Population
1075 Owner Occupied Housing/Total Occupied Housing
1086 Vacant Housing Units/Total Housing Units

aVariab1e number is the one used in the NCS codebook (ICPSR, 1981). The
codebook also furnishes all details regarding the definitions and codes used
for each variable.
Table 3
Longitudinal NCS Variables on Household Characteristics

Variable Number Variable Name

1010 Household Number
1022 Household Status
1024 Tenure (owned, rented)
1030 Members 12 Years or Older
1031 Members Under 12 Years Old
1034 Family Income
[The following apply to the household's head]
1096 Age
1097 Marital Status
1098 Race
1099 Sex
1100 Education
11 01 Grade Completed
1102 Ethnicity
1103 Employment Status
3191 Employed at Time of Indcident
3192 Type of Job
3193 Occupation
3194 Industry
3195 Employee Class
Table 4
Longitudinal NCS Variables on Houehold Occupants

Variable Number Variable Name

2007 Relation to Head
2008 Age
2009 Marital Status
201 0 Race
2011 Sex
2013 Highest Grade Attended
2018 Major Activity
2020 Temporarily Absent from Job
2021 Looking for Work
2022 Could Not Take Job
2023 Never Worked
2026 Occupation
2033 Worked Last Week
2034 Ethnicity
2035 Employment Status

I L "L ,-,-",