I have run multiple imputations on my data and need to now export a final dataset that I can then calculate the mean of each imputed variable. How would one do this?

I can't pool the data straight into my analysis because of the way my data is set up:

My dependent variables are measures of participants response time and accuracy when reading different words (20 words in total, so 20 accuracy scores and 20 response time scores) for which I have around 150 observations. This is where the missing data is..

However my independent variables are measures of word factors e.g. word length, word frequency. So I have 20 words each with those variables.

So my end goal is to run a multiple regression to see the predictive value of word factors for response time and accuracy. (E.g. can word length predict variance in response time of that word)

However this is my issue: I have to first find the mean value of the 150 observations (with the imputations) for response time for each word. I then input these means into a clean data set with my independent variables. And then use that dataset in my regression model.

So I can't directly pool the imputed data into the model. I need to first average those variables. But how do I find the mean of the imputed variables when I have 30+ imputation cycles?

....wondering if this is even possible now or if I should revert to se deletion


1 Answer 1


If you have $M$ imputations done, then you want to create $M$ analysis ready datasets, one for each of the multiple imputations. Each of the $M$ datasets is based on processing the observed data (goes into each dataset) and for dataset $m$ the imputation values for imputation $m$. Then, you analyze each of the $M$ datasets and combine the analysis results (e.g. combining some estimated regression coefficients and their SEs using Rubin's rule).

  • $\begingroup$ Thank you Björn. So just to make sure I understand (my stats knowledge is very basic sorry). If I run 10 imputations, I will then export 10 complete data sets. For each data set I can then find the mean of each variable and input this in a new dataset with my predictor variables. So I'll have again 10 new datasets which have the mean response times for each word (the imputed data). I can then run 10 separate regressions for each set and use Rubins rule to combine the analysis? And by using the mean variable values of the 10 imputed datasets I don't risk losing any variance or anything? $\endgroup$
    – Paige Cox
    Commented Apr 23, 2022 at 9:43
  • 1
    $\begingroup$ I'm not sure that I understand the full analysis you intend to do with various different datasets. However, if it's clear what you would do, if you had no missing data, then you basically do the same thing M times for each of the M datasets with imputations (including for any subsequent steps you do). In the end you get M different estimates +- SE of what you are interested in and combine those. The extra-variation across the M estimates reflects the uncertainty about the imputed data (and results in higher SEs than if you had the same amount of complete data). $\endgroup$
    – Björn
    Commented Apr 23, 2022 at 13:28
  • $\begingroup$ @PaigeCox Stef van Buuren explains how to proceed in detail in his freely available Flexible Imputation of Missing Data. There's some subtlety in how to get the pooled variance; see this page for a start. $\endgroup$
    – EdM
    Commented Apr 23, 2022 at 13:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.