Hello,
I have used mi chained (on Stata 15.0) to create 100 imputations for my data set (11,192 participants). However, when trying to run my analysis on the imputed data, this is taking incredibly long (I currently have been waiting 5 hours for one mixed model to run and it still has not completed).
e.g. of a model:
mi estimate: mixed SDQ_parent_emotion_ i.Time c.Mean_PSAI_imputed if Sex ==1 || ID: Time , cov(exch)
I had chosen 100 imputations, as guidelines by White, Royston and Wood (2010) suggest that "m (number of imputations) should be at least equal to the percentage of incomplete cases". I have a lot of missing data which means that ~91% of participants have at least one of the variables missing. However, none of the individual variables themselves have more than 50% of participants with missing data. I'm a little unsure whether their guidelines refer to the % who are missing at least one variable, or the % of missing data in the variables themselves. If it is the latter, I thought I could perhaps reduce my number of imputations to 50 to try to speed things up a bit,
(In a test run I had done with only 5 imputations, the above mixed model ran in about 1 minute, so I don't think it is the model itself that is the problem - more the large number of imputations).
I wondered whether there was a way to speed up the running of my models, or alternatively whether I can reduce the number of imputations (but I do not know if this is statistically justified)?
Thanks,
Anna
I have used mi chained (on Stata 15.0) to create 100 imputations for my data set (11,192 participants). However, when trying to run my analysis on the imputed data, this is taking incredibly long (I currently have been waiting 5 hours for one mixed model to run and it still has not completed).
e.g. of a model:
mi estimate: mixed SDQ_parent_emotion_ i.Time c.Mean_PSAI_imputed if Sex ==1 || ID: Time , cov(exch)
I had chosen 100 imputations, as guidelines by White, Royston and Wood (2010) suggest that "m (number of imputations) should be at least equal to the percentage of incomplete cases". I have a lot of missing data which means that ~91% of participants have at least one of the variables missing. However, none of the individual variables themselves have more than 50% of participants with missing data. I'm a little unsure whether their guidelines refer to the % who are missing at least one variable, or the % of missing data in the variables themselves. If it is the latter, I thought I could perhaps reduce my number of imputations to 50 to try to speed things up a bit,
(In a test run I had done with only 5 imputations, the above mixed model ran in about 1 minute, so I don't think it is the model itself that is the problem - more the large number of imputations).
I wondered whether there was a way to speed up the running of my models, or alternatively whether I can reduce the number of imputations (but I do not know if this is statistically justified)?
Thanks,
Anna
Comment