How to choose a statistical test?

Quantitative Data

Categorical Data

Comparing means between groups or a group and a specific value

Independent samples (all involved samples)

Independent samples

Dependent samples

Two Samples

Three or more Samples

Dependent samples (all involved samples)

One Sample

How big is the sample?

Sample size 50 - 300

Sample size > 300

Sample size < 50

Graphical methods:

  1. Histogram
  2. box plot
  3. Q-Q plot

Normality test

Normality test

Really strong normality test

Shapiro-Wilk test (Up to 2000)

Is it normally distributed?

Yes

No / have doubts

Bootstrap methods

How big is the sample?

Sample size 50 - 300

Sample size < 50

Sample size > 300

Normality test for each sample

Normality test for each sample

Normality test for each sample

Really strong normality test

Really strong normality test

Evaluate the behavior in the tails of the distribution

Really strong normality test

Shapiro-Wilk test

Anderson-Darling test

Not as strong as "Shapiro-Wilk test"

Kolmogorov-Smirnov test with Lilliefors correction

Are all samples normally distributed?

Yes

No / have doubts

Testing homogeneity of variances for each sample

Levene's test

If at least one sample is not normally distributed

Mann-Whitney U test

Are the dispersions homogeneous?

Yes

No / have doubts

Two Sample t-Test

Welch's t-test

Shapiro-Wilk test

Evaluate the behavior in the tails of the distribution

Anderson-Darling test

Assess skewness and kurtosis

D'Agostino K^2 test

Not as strong as "Shapiro-Wilk test"

Kolmogorov-Smirnov test with Lilliefors correction

Evaluate the behavior in the tails of the distribution

Assess skewness and kurtosis

Not as strong as "Shapiro-Wilk test"

Kolmogorov-Smirnov test with Lilliefors correction

D'Agostino K^2 test

Anderson-Darling test

Shapiro-Wilk test (Up to 2000)

Are all samples normally distributed?

Yes

No / have doubts

Testing homogeneity of variances for each sample

If at least one sample is not normally distributed

Mann-Whitney U test

Bartlett's test

Are the dispersions homogeneous?

Yes

No / have doubts

Known population variance?

Yes

No

Welch's t-test

Two Sample t-Test

Two Sample z-Tests

Normality test

Not as strong as "Shapiro-Wilk test"

Kolmogorov-Smirnov test with Lilliefors correction

Is it normally distributed?

No / have doubts

Bootstrap methods

Evaluate the behavior in the tails of the distribution

Really strong normality test

Yes

Shapiro-Wilk test

Anderson-Darling test

One-sample t-Test (Student's t-test)

Evaluate the behavior in the tails of the distribution

Anderson-Darling test

Really strong normality test

Shapiro-Wilk test

Assess skewness and kurtosis

D'Agostino K^2 test

Not as strong as "Shapiro-Wilk test"

Kolmogorov-Smirnov test with Lilliefors correction

Known population variance?

No

One-sample t-Test (Student's t-test)

Yes

Evaluate the behavior in the tails of the distribution

Anderson-Darling test

Wilcoxon signed-rank test

One-sample z-Test

Assess skewness and kurtosis

D'Agostino K^2 test

Not as strong as "Shapiro-Wilk test"

Kolmogorov-Smirnov test with Lilliefors correction

How big is the sample?

Sample size < 50

Normality test for each sample

Evaluate the behavior in the tails of the distribution

Anderson-Darling test

Are all samples normally distributed?

Yes

Testing homogeneity of variances for each sample

Levene's test

Are the dispersions homogeneous?

Yes

No / have doubts

Welch's ANOVA

No / have doubts

If at least one sample is not normally distributed

Really strong normality test

Shapiro-Wilk test

Not as strong as "Shapiro-Wilk test"

Kolmogorov-Smirnov test with Lilliefors correction

one-factor ANOVA

Kruskal-Wallis test

Are there significant differences between the groups?

Yes

No

Tukey's test (Tukey HSD)

Are there significant differences between the groups?

Yes

No

Games-Howell test

Are there significant differences between the groups?

Yes

No

Dunn's method

Sample size 50 - 300

Normality test for each sample

Evaluate the behavior in the tails of the distribution

Anderson-Darling test

Really strong normality test

Shapiro-Wilk test

Not as strong as "Shapiro-Wilk test"

Kolmogorov-Smirnov test with Lilliefors correction

Assess skewness and kurtosis

D'Agostino K^2 test

Sample size > 300

Normality test for each sample

Evaluate the behavior in the tails of the distribution

Anderson-Darling test

Really strong normality test

Shapiro-Wilk test (Up to 2000)

Not as strong as "Shapiro-Wilk test"

Kolmogorov-Smirnov test with Lilliefors correction

Assess skewness and kurtosis

D'Agostino K^2 test

Are all samples normally distributed?

Yes

Testing homogeneity of variances for each sample

Bartlett's test

Are the dispersions homogeneous?

Yes

one-factor ANOVA

Are there significant differences between the groups?

Yes

Tukey's test (Tukey HSD)

No

No / have doubts

Welch's ANOVA

Are there significant differences between the groups?

Yes

Games-Howell test

No

No / have doubts

If at least one sample is not normally distributed

Kruskal-Wallis test

Are there significant differences between the groups?

Yes

Dunn's method

No

Two Samples

How big is the sample?

Sample size 50 - 300

Normality test for each sample

Really strong normality test

Shapiro-Wilk test

Evaluate the behavior in the tails of the distribution

Anderson-Darling test

Assess skewness and kurtosis

D'Agostino K^2 test

Not as strong as "Shapiro-Wilk test"

Kolmogorov-Smirnov test with Lilliefors correction

Sample size < 50

Normality test for each sample

Evaluate the behavior in the tails of the distribution

Anderson-Darling test

Are all samples normally distributed?

Yes

No / have doubts

Really strong normality test

Shapiro-Wilk test

Not as strong as "Shapiro-Wilk test"

Kolmogorov-Smirnov test with Lilliefors correction

Sample size > 300

Normality test for each sample

Really strong normality test

Shapiro-Wilk test (Up to 2000)

Evaluate the behavior in the tails of the distribution

Anderson-Darling test

Assess skewness and kurtosis

D'Agostino K^2 test

Not as strong as "Shapiro-Wilk test"

Kolmogorov-Smirnov test with Lilliefors correction

paired t-test

Wilcoxon signed-rank test

Three or more samples

How big is the sample?

Sample size 50 - 300

Normality test for each sample

Really strong normality test

Shapiro-Wilk test

Evaluate the behavior in the tails of the distribution

Anderson-Darling test

Assess skewness and kurtosis

D'Agostino K^2 test

Not as strong as "Shapiro-Wilk test"

Kolmogorov-Smirnov test with Lilliefors correction

Sample size < 50

Normality test for each sample

Evaluate the behavior in the tails of the distribution

Anderson-Darling test

Are all samples normally distributed?

Yes

No / have doubts

Really strong normality test

Shapiro-Wilk test

Not as strong as "Shapiro-Wilk test"

Kolmogorov-Smirnov test with Lilliefors correction

Sample size > 300

Normality test for each sample

Really strong normality test

Shapiro-Wilk test (Up to 2000)

Evaluate the behavior in the tails of the distribution

Anderson-Darling test

Assess skewness and kurtosis

D'Agostino K^2 test

Not as strong as "Shapiro-Wilk test"

Kolmogorov-Smirnov test with Lilliefors correction

repeated measures ANOVA (RM ANOVA)

Checking sphericity

Mauchly's test

Is sphericity confirmed?

Yes

No / have doubts

Greenhouse-Geisser correction

Friedman test

Are all samples normally distributed?

Yes

No / have doubts

repeated measures ANOVA (RM ANOVA)

Friedman test

Checking sphericity

Mauchly's test

Is sphericity confirmed?

Yes

No / have doubts

the violation of sphericity is not too significant

the violation of sphericity is too significant

Huynh-Feldt correction

Greenhouse-Geisser correction

Wilcoxon signed-rank test