I’m running CFA with lavaan, encountering 5% missing data among ordinal indicators. Should I implement WLSMV with listwise or pairwise deletion?
hey i think pairwise deletion is btr with WLSMV, since you keep more cases and reduce unused info. check your model fit tho, cause results can vary with each data set.
In my experience with CFA using lavaan, I have tended to favor listwise deletion when working with WLSMV for ordinal data. While it does reduce the effective sample size, it ensures that the estimates are based on consistent data across variables. I found that this method reduces the complexity in interpretation of fit indices, particularly when missingness isn’t patterned systematically. As always, I recommend checking your data’s missing pattern carefully to confirm this approach fits your specific context, especially with only a moderate amount of missing data.
Hey Ava_Books, really interesting scenario! I’m leaning a bit towards using pairwise deletion too, because it seems like a more flexible option when dealing with only 5% missing data, especially with ordinal indicators. It just feels more efficient to use available data rather than dropping cases across the board. But I also wonder how much the missing data mechanism (MCAR, MAR, etc.) might influence this decision. For instance, if the missingness has any sort of pattern, could that potentially bias our estimates when using pairwise deletion? Has anyone experimented with any robustness checks or even tried incorporating multiple imputation methods as a supplementary approach? It’s always fascinating to see how different strategies play out in terms of model fit and overall reliability. Let me know your thoughts or if you’ve encountered similar challenges, really curious to hear other experiences on this!
hey ava_books, if your missing data seems random, pairwise deletion could work. i’d also consider a quick multiple imputation check to see if results differ much. sometimes a simple simulation can help in choosing the best method. good luck!