594 followers
@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