(Note: Prints best in landscape mode)
One Variable:
Number of Samples |
||||
Level of Measurement |
One Sample |
Two Related Samples (Paired) |
Two Independent Samples |
More than Two Samples |
Nominal Data (Categories or Names) |
c 2 Goodness of Fit TestBinomial Test Poisson Test |
--- |
c 2 Goodness of Fit Test |
c 2 Goodness of Fit Test |
Ordinal Data (Ranks) |
Kolmogorov-Smirnov Test Wilcoxon Test |
Sign Test Wilcoxon Test |
Median Test Kolmogorov-Smirnov Test Shapiro-Wilks Test Kolmogorov-Smirnov-Lilliefors Test Mann-Whitney U Test |
Kruskal-Wallis Test |
Interval or Ratio (Continuous) Note: Tests in this row of cells are parametric. Assumptions must be met. |
One Sample t Test |
Paired Sample t Test |
Means: Welch's Approximate t Two Sample t Test Variances: F test Levene O'Brien Brown-Forsythe Bartlett |
Means: Welch's Approximate ANOVA ANOVA Variances: Fmax Levene O'Brien Brown-Forsythe Bartlett |
Notes: The table lists, cumulatively downward, the tests appropriate for each level of measurement. Tests for lower levels of measurement may be applied to higher levels of measurement but these lower level tests generally have less power.
Two Variables:
Nominal Data:
1. To test independence of two variables: c2 Contingency Test
Ordinal Data:
2. To test association or whether two items vary together: Spearman's or Kendall's rank correlation.
Interval/Ratio Data:
3. To test association or whether two items vary together: Correlation.
4. To test prediction, causality or the functional dependence of one variable on another: Linear Regression
Assumptions:
All parametric tests require normality at the population level in some measure being tested.
When pooled variances are used in parametric tests, sample variances being pooled must be roughly equal.
All tests, both parametric and non-parametric, assume random sampling, with individuals within a sample being independently selected.