comparing by-subject with mixed-designs

Hi, I know a common critique for meta-analysis is the comparing apples and pears argument. I am still trying to include as many different study methods (the method will be a factor), but I am not sure if it statistically makes sense if I also include mixed-design studies when most others are done by subject. I aim to calculate r rather than d as most studies did ANOVAs. The problem I see is that the influence of the mixed-design studies on the overall effect size is bigger because the degrees of freedom depend on the n of both subjects and items. So far I have not found a good source how to deal with this, is there a common practice on how to set the results in proper relation or does it simply not make sense to compare different designs? Any help would be greatly appreciated