Biostatistics Study NotesFrom concepts and derivations to interpretation

CHAPTER 08 · TOPIC 04

Appendix: Selecting a Statistical Test

Choose a statistical method from the outcome type, sample relationship, target hypothesis, and assumptions of the candidate test.

On this page
  1. Integrated Statistical-Test Selection Table
  2. Question 1: What Type of Outcome Variable?
  3. Question 2: How Are the Samples Related?
  4. Question 3: What Does H₀ Compare?
  5. Checks after Using the Table

Statistical methods are selected in the following order: the type of outcome variable, the relationship between the samples, and the population-level hypotheses to be compared. The table below is a starting point, and the final decision is made by reviewing the study design and assumptions for each method.

Integrated Statistical-Test Selection Table#

OutcomeOne sampleTwo independent groupsTwo paired time pointsThree or more independent groups
BinaryBinomialχ² / FisherMcNemarχ²
Approximately normal continuousOne-sample tIndependent t / Welch tPaired tANOVA / Welch ANOVA
Skewed continuous or outlier-proneSign / Signed-rankRank-sumSign / Signed-rankKruskal–Wallis
OrdinalSign / Signed-rankRank-sumSign / Signed-rankKruskal–Wallis

Question 1: What Type of Outcome Variable?#

TypeExamplesPrimary comparison
BinaryPresent/absent, positive/negativeProportion or event probability
ContinuousBlood pressure, weightMean, location, or distribution
OrdinalDisease stage, pain scaleRanks or distributional location

Coding categories as 0, 1, and 2 does not make them continuous measurements. Likewise, a significant normality test alone does not automatically require switching to a nonparametric method.

RelationshipDefinitionExample
One sampleOne group compared with a reference valueIs the median 4?
IndependentDifferent participants belong to different groupsTreatment and control groups
PairedRepeated or matched observationsBefore and after treatment

Question 3: What Does H₀ Compare?#

MethodPrimary null hypothesis
t / ANOVAThe specified means are equal
SignPositive and negative differences are equally likely
Signed-rankThe difference distribution is symmetric around zero
Rank-sumThe two independent groups have the same distribution
MedianThe proportions on both sides of the pooled median are equal
Kruskal–WallisAll independent groups have the same distribution
χ² independenceTwo categorical variables are independent

Methods described broadly as a "two-group comparison" can test different hypotheses. A rank test can be interpreted as comparing medians only under additional conditions, such as similarly shaped distributions that differ solely by a location shift.

Checks after Using the Table#

  1. See outliers, skewness, variance, and clusters in plots
  2. Write your research question and H₀ and check if your method answers the same question
  3. Check independence, pairing, expected counts, symmetry, homoscedasticity, ties
  4. Specify exact, permutation, and asymptotic methods and corrections for small samples.
  5. Report effect sizes and confidence intervals and do not reduce the analysis to p values alone