Statistical Assumption Checklist

The Dissertation Compass | dissertationready.com

Before running any statistical test, verify that your data meets the required assumptions. Check each item below for the test(s) you are using. Document the results of each assumption check in your methodology or results chapter.

Study: Date:

Independent Samples t-Test

• Dependent variable is continuous (interval or ratio)
• Independent variable consists of exactly two categorical groups
• Observations are independent (no repeated measures)
• No significant outliers (check boxplots)
• Dependent variable is approximately normally distributed in each group (Shapiro-Wilk test, p > .05)
• Homogeneity of variances (Levene's test, p > .05; if violated, use Welch's t-test)

One-Way ANOVA

• Dependent variable is continuous
• Independent variable consists of three or more categorical groups
• Observations are independent
• No significant outliers (check boxplots per group)
• Dependent variable is approximately normally distributed in each group (Shapiro-Wilk per group)
• Homogeneity of variances (Levene's test; if violated, use Welch's ANOVA)

Paired Samples t-Test

• Dependent variable is continuous
• Observations are paired (same participants measured twice)
• No significant outliers in the difference scores
• Difference scores are approximately normally distributed (Shapiro-Wilk on differences)

Chi-Square Test of Independence

• Both variables are categorical (nominal or ordinal)
• Observations are independent
• Expected cell frequencies are 5 or greater in at least 80% of cells (if violated, use Fisher's Exact Test)

Pearson Correlation

• Both variables are continuous
• Observations are independent
• Linear relationship between variables (check scatterplot)
• No significant outliers (check scatterplot)
• Both variables are approximately normally distributed (bivariate normality)

Multiple Linear Regression

• Dependent variable is continuous
• Two or more predictor variables (continuous or categorical)
• Independence of residuals (Durbin-Watson statistic, value near 2)
• Linear relationship between each predictor and the DV (partial regression plots)
• Homoscedasticity (plot standardized residuals vs. predicted values -- look for random scatter)
• No multicollinearity (VIF values < 5, ideally < 3; Tolerance > 0.2)
• Residuals are approximately normally distributed (histogram and P-P plot of residuals)
• No significant outliers or influential cases (Cook's Distance < 1)

Logistic Regression

• Dependent variable is binary (two categories)
• One or more predictor variables (continuous or categorical)
• Observations are independent
• No multicollinearity among predictors (VIF values < 5)
• Linear relationship between continuous predictors and the log odds (Box-Tidwell test)
• Adequate sample size (minimum 10-15 cases per predictor variable)
• No significant outliers (standardized residuals < 3)

Mann-Whitney U Test (Nonparametric Alternative to t-Test)

• Dependent variable is ordinal or continuous
• Independent variable consists of two groups
• Observations are independent
• Distributions have similar shape in both groups (if comparing medians)