Question

When doing imputation on a dataset, should we do impuation for variable or the entire dataset?

Answer 1

Variables with missing values should usually be imputed jointly (i.e., all at once) in order to preserve the associations between them. Of course, you do not necessarily need to impute the whole of your 750 variables. Just include those variables in your multiple imputation model which are also involved in your analysis model, i.e. which are thought to affect you dependent variable. Some practical advice on how to select variables for your imputation model can be found in the attached paper.