Use cross-tabs, chi-square, Welch’s t-test and Mann–Whitney for distinct questions—and interpret more than significance.
03
Outcome: three unadjusted comparisons with assumptions, effect estimates and clinical context kept visible.
Seven-step investigationNow: 03 · Compare groups
Analysis checkpoint
Match one comparison method to one question.
Work through these six prompts in order. Each one supplies what the next one needs.
1 · Before you start
Have the overall cohort description from Step 02 and know which variable is the exposure group and which variable is the outcome for each comparison.
Words used on this page
Cross-tabulation
a table of counts for two categorical variables.
Independent observations
each participant contributes a separate observation, not a paired or repeated one.
Confidence interval
a range showing the estimate's statistical uncertainty.
p-value
how incompatible the data are with a specified null model; it is not the size or importance of an effect.
Welch's t-test
a comparison of two means that does not assume equal variances.
Mann–Whitney test
a comparison based on the ranks, and therefore distributions, of two groups.
2 · Why this comes now
Simple, unadjusted comparisons show the observed group patterns before effect measures and regression add more structure. The outcome scale determines the method.
3 · Do this now
Cross-tabulate medication review with six-month disability and inspect row percentages.
Compare mean quality of life with Welch's t-test and ranked medicine counts with Mann–Whitney.
For each comparison, record the group values, estimate or test result, confidence interval where available and p-value.
Same step in Stata and SPSS
Run Section 03 in Stata or SPSS. The command names differ, but both routes should produce the same 2 × 2 counts, group summaries and three unadjusted conclusions.
4 · Read this
Read the group values before the p-value. For the t-test use the unequal-variance result; for Mann–Whitney interpret a difference in ranked distributions unless the distribution shapes support a median interpretation.
5 · Ready to continue when
You can state what each method compares, report the observed values and uncertainty, and explain why statistical significance alone is not the finding.
Counts, row/column percentages and expected counts.
Do two independent groups differ in mean outcome?
Welch’s independent t-test
Group means, mean difference, 95% CI and outliers.
Do two independent groups differ in rank distribution?
Mann–Whitney / Wilcoxon rank-sum
Group distributions and the estimand; it is not automatically a test of medians.
Independence matters. These procedures do not cover paired
before-and-after data, repeated measurements or clustered samples.
Worked comparisons
Read the estimate, interval and context.
Chi-square
Disability risk is 22.5% with a review and 21.9% without. Pearson χ² = 0.033, p = 0.857.
Welch’s t-test
Mean EQ-5D-style scores are 0.750 and 0.751; mean difference −0.001, 95% CI −0.020 to 0.018.
Mann–Whitney
Medication count is higher in the review group: median 5 versus 3; p < 0.001.
Interpretation
The groups differ at baseline but not in the crude outcomes. That is a reason to examine confounding, not to hunt for another test.
Representative output · same teaching data in all three routes
Three unadjusted group comparisons
Outcome / method Estimate or test 95% CI p
Disability / chi-square 22.5% vs 21.9%; χ²=0.033 — 0.857
EQ-5D / Welch t-test reviewed − unreviewed=-0.001 -0.020 to 0.018 0.921
Medicines / Mann–Whitney medians 5 vs 3; W=37,897 — <0.001
What the output means
The disability proportions and mean quality-of-life scores are very similar before adjustment, and their tests provide no evidence of a crude group difference. Medication counts tend to be higher in the reviewed group. The Mann–Whitney result concerns the rank distributions; it is not automatically a confidence interval for a difference in medians.
Write the interpretation like this
At baseline, participants who received medication review took more medicines (median 5 versus 3; Mann–Whitney p<0.001). Crude six-month disability (22.5% versus 21.9%; χ²=0.033, p=0.857) and mean quality of life (difference −0.001, 95% CI −0.020 to 0.018) were similar between groups.
Stata
The three commands mirror the three questions: association between categories, difference in means, and difference in rank distributions.