I have the following question:

I collected the number of different symptoms at two time points (Baseline BL, Follow-up FU) in two groups (Control Group CG, Intervention Group IG). So, there is a between-subjects factor (Group; CG, IG) and a within-subjects factor (Time Point; BL, FU). So, I have a pre-post design with two independent groups. I would like to compare the frequencies of individual symptoms between groups and time points. Each individual can either have the symptom (1) or not (0).

For example: Fatigue CG at BL: 20 CG at FU: 18 IG at BL: 20 IG at FU: 10

Which test could I apply here? As far as I know, the McNemar test is only for purely dependent data, while the Chi-Square test and Fisher's test assume independence of the data.

Thank you in advance for your help.

  • $\begingroup$ If the outcome is count of symptoms, maybe multilevel poisson regression with random intercept of subject? $\endgroup$
    – Sointu
    Jan 17 at 12:54
  • $\begingroup$ thank you for your comment. My problem is that I believe that ultimately my straightforward question—whether the number of reports (agreement) of the symptom differs significantly over time between the two groups—may no longer be meaningfully addressed if, for example, I calculate a multilevel poisson regression with random intercepts. $\endgroup$ Jan 17 at 13:05

1 Answer 1


It seems you're already on the right track by considering McNemar's test, Chi-Square test, and Fisher’s Exact Test. Let me try to help refine this approach with a focus on the specific design of your study.

First, you could start off by using McNemar’s test to analyze changes within each group (CG and IG) from Baseline to Follow-up. Then, use the Chi-Square or Fisher’s Exact Test to compare the groups at each time point separately. Alternatively, you could also use Moderated Regression Analysis (for this one you might need to make some adjustments to your data), or Generalized Estimating Equations, or Mixed-Effects Logistic Regression (i.e., Multilevel Modeling or Hierarchical LM).

Lastly, if you can get a larger time period sample for your data (if empirical), this would enhance your temporal structure and possibly enable you to do a Vector Error Correction Model, this will allow you to examine the long-term trends between your variables of interest and will enable you to do cointegration testing. After you figure out the cointegration order you can do pairwise testing to figure out which variables are related in the long run and also Granger Causality testing, which uses the Chi-Square statistic. Here are some open-access sources you can check out for further information on these methods: https://usq.pressbooks.pub/statisticsforresearchstudents/chapter/moderation-assumptions/ https://www.statsmodels.org/stable//gee.html https://stats.oarc.ucla.edu/r/dae/mixed-effects-logistic-regression/ https://spureconomics.com/vecm-estimation-and-interpretation/


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.