Structural Equation Modeling (SEM) Discussion > Sample size
Hey Erin,
Sorry for the delay in my response. To answer your question as simply as possible, it is not generally a problem if you have different rates of missingness in your SEM data, as long as the data isn't systematically missing in a severe way (e.g. as long as the highest scores of your mediator variable aren't all missing, more so than your other variables). In theory, SEM's use of Maximum Likelihood Estimate (MLE) should alleviate a lot of your problems with differing rates of missing data, as long as your data is reasonably missing at random (MAR).
August 27, 2012 |
Jeremy Taylor
Jeremy Taylor
Hi all---
I'm hoping this is a relatively simple question that I'm just not thinking logically about (because it's late and I'm overly stressed out by my dissertation). For my dissertation, I am using SEM (at least, partially). The large sample size issue has already be established. Now that I'm in data collection, however, I'm finding that I may have a different sample size for my mediator than for my IV and DV. For example, the n for my IV and DV could each be 200-250, but the n for my mediator could be, say, 175. My first question is basically is this an issue, specifically the unequal sample sizes? And how do I treat it? Or do I exclude the data that does not reflect my mediator, and only use those extra participants to run my other analyses?
Any input, guidance, help would be greatly appreciated!!! Thanks!