I used multiple imputation on a data set that had some missing values (I had to do this as the sample size was low so I couldn't just exclude the NAs). I know you can do

*read in dataset*
d <- mice(dat)
fit <- with(data = d, exp = glm(Condition ~ X1 + X2 + X3 + X4)) 
combine <- pool(fit)

But are you able to extract predicted values from this model? For example I want the probabilities of individuals being in the control or treatment condition. Normally this would be done like

predict(fit, d, type = 'response')

However I can't seem to make this work. Could I do a prediction for each dataset and average the predicted values? For example:

d <- mice(dat)
a <- complete(d, 1)
fit <- glm(Condition ~ X1 + X2 + X3 + X4, family = 'binomial')
predict(fit, a, type = 'response)

And do this for all datasets then take an average of the predicted values. Or would this logic be flawed and is there any other way to do it?

  • $\begingroup$ The idea is to pool the results, so you would pool the predictions from multiple models in some way, say average. $\endgroup$ – user2974951 Aug 21 '19 at 6:11

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.