I have a question about effect sizes and how these may relate to whether the analysis is done across group of within the same group. So i have done a few glm models on the use of certain word orders in adults and children. I have set the structure as my dependent variable and speaker type and context as my independent variables. Most of my results come out as significant, but this is beyond the point.
Adult and context A is set as my intercept, and when this is compared to the child speaker I get a very low effect size (0.06), but when the usage of this structure is compared for the adults across the two contexts I get a much higher effect size (0.66). Now this happens also in another model, where I have inverted structure and context as my dependent/independent variables, and also for another language on which I ran the same tests.
So I am wondering, could the effect size be affected by the type of comparison: it is low when comparing adults to children, so groups containing different individuals, but it is much higher when the same group is compared on another factor? If not, what else could explain the distribution of my effect sizes? And if yes, why does this happen?