Scribd
Upload
14 views

BiostatisticsRx02_Inferential_Statistical_Test_Cheatsheet

This document serves as a guide for clinicians to select appropriate inferential statistical tests for calculating p-values, focusing on common tests for both parametric and non-parametric data. It covers various statistical tests for comparing differences, frequencies, time-to-event analysis, correlation, regression, and corrections for interim analysis. The guide emphasizes the importance of understanding independent, dependent, and confounding variables in the context of statistical testing.

Uploaded by

Khajan Mahtolia
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
14 views

BiostatisticsRx02_Inferential_Statistical_Test_Cheatsheet

This document serves as a guide for clinicians to select appropriate inferential statistical tests for calculating p-values, focusing on common tests for both parametric and non-parametric data. It covers various statistical tests for comparing differences, frequencies, time-to-event analysis, correlation, regression, and corrections for interim analysis. The guide emphasizes the importance of understanding independent, dependent, and confounding variables in the context of statistical testing.

Uploaded by

Khajan Mahtolia
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

Guide to Selecting a Test for Inferential

Statistics
ClinCalc Academy Biostatistics Cheatsheet
This document is designed to assist clinicians in selecting the most appropriate inferential
statistical test to calculate a p-value. By design, this guide is non-comprehensive, meaning only the
most common and relevant statistical tests have been included.

Terminology Used in this Document


Parametric vs. Non-Parametric Data
Some statistical tests rely on data being normally distributed (bell-shaped curve) – this is called
“parametric” data. If the data is not parametric (“non-parametric”), alternative statistical tests need to
be used.
Parametric Non-Parametric
(normally distributed) (not normally distributed)

(multimodal) (skewed) (not bell-shaped)

Independent, Dependent, and Confounding Variables


Independent Dependent Confounding
(explanatory) (response/outcome) (covariate)
Variable that is changed, Result or outcome of interest Impacts dependent variable;
manipulated or controlled that is measured may need to be accounted for
Ex: Atorvastatin 20 mg vs. placebo Ex: Incidence of heart attack or stroke Ex: Smoking rates between groups

© ClinCalc LLC. All rights reserved.


Page 1 of 3
Paired vs. Unpaired Samples
In paired samples, the patient acts as their own “control” – patients are exposed to each treatment arm
(independent variable) during the study period. A crossover or pre/post study design is most common.

In unpaired samples, the patient groups are independent and


unrelated to each other – each patient receives one version of the
independent variable usually in a “parallel” study design.

Statistical Tests to Compare Differences


These tests are used when the dependent (outcome) variable is a continuous variable. These tests
rely on comparing the mean or median/ordinal (ranked) values of the outcome.

# of Groups to Compare Outcome is Parametric Outcome is Non-parametric


(comparing means) (comparing rank/medians)
Parallel study design (independent samples)
Independent samples t-test Mann-Whitney U test
2
(AKA student’s t-test) (AKA Wilcoxon rank-sum test)
- One-way ANOVA
3 or more Kruskal-Wallis test
- With covariates: ANCOVA
Cross-over or pre/post design (paired samples)
2 (paired) Paired t-test Wilcoxon signed-rank test
3 or more (paired) Repeated measures ANOVA Friedman’s test

Statistical Tests to Compare Frequencies


These tests are used when both the dependent (outcome) and independent (explanatory)
variables are categorical (discrete). These tests rely on the observed vs. expected frequencies.

# of Groups to Compare Outcome is Categorical


Parallel study design (independent samples)
2 - Chi square (X2) test
- Fisher’s exact test
- With covariates: Cochran-Mantel-Haenszel test
3 or more Chi square (X2) test
Cross-over or pre/post design (paired samples)
2 (paired) McNemar’s test

© ClinCalc LLC. All rights reserved.


Page 2 of 3
Statistical Tests for Time-to-Event Analysis
Time-to-event analyses are used when study participants participate in the study for varying
amounts of time and not all participants experience the dependent (outcome) variable. Patients
who do not experience the outcome of interest are “censored”, which is a unique feature of this
type of analysis.

No Covariates Control for covariates


(confounding variables) (confounding variables)
Statistical test for hazard ratio
Log-rank test Cox proportional-hazards model
(HR) and p-value

Statistical Tests to Determine Correlation


Correlation describes the co-relationship between two continuous variables. Correlation is
described as an ‘r’ value (correlation coefficient) between -1 (negative correlation) and +1 (positive
correlation). A correlation coefficient of zero (0) indicates no relationship exists.

Variables are Parametric Variables are Non-Parametric


Statistical test for p-value and
Pearson’s correlation Spearman’s rank order correlation
correlation coefficient (r)

Statistical Tests for Regression


Like correlation, regression describes the relationship between two or more variables. Unlike in
correlation, regression can “predict” a dependent (outcome) variable and can be designed to
handle multiple independent and confounding variables.

Outcome is Categorical Outcome is Continuous


Statistical test for dependent Logistic regression Linear regression
variable coefficients and p-value (simple or multivariate) (simple or multivariate)

Statistical Tests to Correct for Interim Analysis


Each time an interim analysis is conducted, the risk of a type I error increases. There are a variety
of statistical methods that can be used to allow for interim analyses and adjust for an increased
risk of type I error. There is no “best” test to correct for multiplicity. Each test has advantages and
disadvantages.

• Bonferroni method • Haybittle-Peto method


• O’Brien-Fleming method • Pocock method

© ClinCalc LLC. All rights reserved.


Page 3 of 3

You might also like