When is it allowed to combine two samples? I have two samples (from two experiments) with different variances, and I want to combine these two samples with a factor (1 and 2 for each experiment), and then maybe do a multilevel multivariate analysis. 
Where can I find literature on the issue of whether I am allowed to combine these two samples with a factor? 
Or, is it better to do the analysis separately and do a meta-analysis on the results? 
 A: If your questionnaire instruments are the same and the two interventions are the same then pool the two samples and test each item for the two treatment groups.
You can obtain additional information to evaluate the similarity of the two groups. For example, you can include a term in your analysis for group or sample. If these are substantially different then you can conclude that one sample gives different answers, higher or lower. This doesn't mean you shouldn't pool the samples, only that you might report it. You might also examine the covariance matrices of the two samples. Again, being different doesn't mean you shouldn't pool the two samples.
You don't really have a sample survey of the population, as the term is usually used where the sampling uses complex methods such as stratification, cluster samples, unequal probabilities, etc. There is a large literature on how to combine two or more such samples for analysis, but you don't need that.
Even if your two samples are significantly different in some dimensions I would still pool them. They could differ for a number of unimportant reasons.
I think I would do all my analyses in a consistent way, either with a term in the model for the sample, or not.
