Hello Bikram,

One appropriate test would be McNemar Chi-squared which accounts for the pre-post data structure (note: this is calculated differently that the usual Pearson Chi-squared which does not accommodate this).

After the intervention we would want to test if the pretest effect is different from the posttest effect. We fit an ARIMA model to the pretest data and use it to forecast for the posttest period. Take a difference between the posttest data and their forecasts. Model these differences for the transfer function. The result is the intervention model. If the model is statistically significant, the difference between the pretest and the posttest data is significant.

My approach would be with a log-linear model with a random effect for individual unit. You might start with these keywords in your search: log-linear mixed model, generalized linear mixed model, generalized hierarchical model, binomial generalized mixed model. Best of luck with your analysis.

Caleb A. Aldridge A think a log-linear model might be a bit complex for what amounts to a 2 x 2 table. That is why I recommended Mark R Speechley's suggestion of the McNemar Chi-squared test, which as he says is specifically designed for pre- post- comparisons.

Bikram Adhikari McNemar's Chisquare as you have a binary outcome and two repeated measures. In the case of multichotomous outcome, you can possibly look at the Test of Marginal Homogenity, and in the case of multiple repeated measures, Cochran's Q test might be useful.

David L Morgan you're probably right but the original post (Bikram Adhikari) didn't specify if there was only one measurement before and after for each group (2 x 2 table). I wanted to offer up an approach if there were multiple measurements on each group (control and treatment) in each time period (before and after) that wasn't necessarily a time series, in which case Ette Etuk approach could be used. If not, as you said, McNemar's test that Mark R Speechley suggests should get the job done.

Pearson chi-square approach