Dealing with missing values

[johnstatistic]

I am analyzing the data of a cohort study.There are three sets of paired data (same measures on same subjects in 3 different point of time). The problem is that the third set of data have more than 50% missing values. my question is that, how can i deal with these missing? does MI help in these repeated measures? it seems that if I delete the missing I loose more than half of my data set. TNX.

 

[gdaem]

Multiple imputation can be used, yes. I would double check the assumptions of using MI first though.

You can base your MI off of means or a regression model. I prefer to use a regression model because it seems to be the most useful for what I have done, and gave me more consistent imputations. If you are using a repeated-measures design there may be a way to use MI to your advantage over the course of the study; you may be able to calculate missing values using MI regression method based off of previously-answered questions and the general model of that time period. Not sure how you would go about setting this up, but it may be worthwhile to dig in to a bit more.

 

[freezo]

Multiple imputation can be used, yes. I would double check the assumptions of using MI first though.

You can base your MI off of means or a regression model. I prefer to use a regression model because it seems to be the most useful for what I have done, and gave me more consistent imputations. If you are using a repeated-measures design there may be a way to use MI to your advantage over the course of the study; you may be able to calculate missing values using MI regression method based off of previously-answered questions and the general model of that time period. Not sure how you would go about setting this up, but it may be worthwhile to dig in to a bit more.

thanks.. its helpful for me also