How to Choose the Right Statistical Test How to Choose the Right Statistical Test

Type of Analysis

Compare Means Relationship Categorical Data Survival Analysis Sensitivity/Specificity

Between X and Y

More Than Two Groups Correlation

Two Groups Linear Regression Multiple Regression Nonlinear Regression Kaplan-Meier ROC Curve
(Drug A vs Drug B (Strength of
(Slope) (More Than One X) (Dose Response)
vs Drug C) Relationship)

Same Subjects 100+ Nonlinear Model

Each Subject Pass Normality Reject Normality Response is Response is a Count
Measured Twice Plus User Defined
Measured Once Continuous
(Before/After) Models Available

Pairwise Differences Pairwise Differences Each Group Normal and Each Group Normal One or Both Groups Pearson Correlation Spearman Correlation Least Squares Poisson Regression Response Continuous Response Continuous Response is a Count
Normal Non-Normal Variances Equal Non-Normal Regression No Outliers With Outliers

Wilcoxon Matched Unpaired t test With

Paired t test Unpaired t test Mann-Whitney Test Least Squares Robust Regression Poisson Regression
Pairs Signed Rank Test Welch's Correction
(Parametric Test) (Parametric Test) (Nonparametric Test) Regression
(Nonparametric Test) (Parametric Test)

One Factor (Drug) Two Factors Three Factors

(Drug, Diet) (Drug, Diet, Time)

Each Subject Measured Each Subject Each Subject Measured Each Subject
More That Once Measured Once More That Once Measured Once

Pass Normality Reject Normality Pass Normality and Pass Normality Reject Normality Pass Normality Reject Normality Pass Normality Reject Normality
Variances Equal

Brown-Forsythe or Consider
Assume Sphericity Don't Assume Friedman Test ANOVA Kruskal-Wallis Test Assume Sphericity Don't Assume Consider ANOVA
Welch’s ANOVA Transformation
Sphericity (Nonparametric Test) (Parametric Test) (Nonparametric Test) Sphericity Transformation (Parametric Test)
(Parametric Test)

ANOVA With G-G

ANOVA or Mixed Model Correction or Mixed
ANOVA or Mixed Model ANOVA or Mixed Model
(Missing Data) Model (Missing Data)
(Missing Data) (Missing Data)
With G-G Correlation With G-G Correction 2 x 2 Table 2 x Y Table

Each Subject Measured Each Subject All Expected Not All Expected Chi-square Test
More That Once Measured Once Cell Counts at Least 5 Cell Counts at Least 5

Pass Normality Reject Normality Pass Normality Reject Normality Chi-square Test Fisher’s Exact Test

Assume Sphericity Don't Assume Consider ANOVA Consider

Sphericity Transformation (Parametric Test) Transformation

ANOVA With G-G

ANOVA or Mixed Model Correction or Mixed
(Missing Data) Model (Missing Data)
With G-G Correction