@daveschester 45% isn't too bad. However, this influences how many imputed datasets you create. Graham et al. (2007) showed that when 50% of data is missing, you need to impute about 40 datasets to retain as much power as just using FIML. https://t.co/wIfT
@daveschester This seems completely appropriate for multiple imputation. Just select a large enough number of imputed datasets. See https://t.co/xzm8Sccrda
Looking into this today, how many sets of complete data needed for multiple imputation, https://t.co/B1VkLvSkFi great discussion but simulation was done with normally distributed variables, nonetheless, take home message was inline with https://t.co/HmsX1L