Separated regression analysis vs control the covariate I am conducting a data analysis on an Epidemiology cross-sectional study.
Suppose the outcome variable is an binary variable for health status (1=health, 0= unhealth). And the exposue is infection at early year.  The hypothesis is that exposed to early infection can affect later health.
Another variable is gender which people which can affect the health status. There also a bunch of other variables that could be confounders.
The analysis we did pretty simple, we used a logistic regression analysis, we treat gender and other potential confounders as predictors and include some interactions such as gender*infection term in the model.
However, some professors (non-statisician)  insisted that we should do the analysis separately, one model for male and one model for female. 
I think separated model cannot even study the interactions.
My question is  what are the advantages to run regression analysis separately (one for male and one for female)?
Thanks 
 A: There almost never is an advantage to throwing away data, which is effectively what you do when you analzye males and females separately: you can't use information about males to inform your analysis of females, or vice-versa. Having more cases in an analysis typically provides more power to discern truly significant results. For example, if males and females have the same relation between exposure and outcome, then it's quite possible to have a situation in which neither gender-specific analysis shows a significant result while a combined analysis would, simply because of the numbers of cases that enter the significance calculations. Confidence intervals would be expected to be narrower for the combined analysis in any event
An exception might be if there is an inherent difference in the underlying biomedical issues between males and females, such as in breast cancer. But other than that, if your professors are concerned with differences between genders, the combined analysis provides a better way to test for such differences (and make possible interaction tests, as you note). Tests of gender differences come immediately from combined analysis, while comparing coefficients of models across the two separated data sets, while possible, is likely to provide less power for identifying true gender differences.
