Multiple Imputation: dropping observations and/or variables after mi

[Rodolfo Hernandez]

Hello, I have a dataset in a long format with data as follows:

studyno	cost1	cost2	Csum	Ctotal	age	ageband	treatmentnumber
1	10	.	.	.	24	1	1
1	10	.	.	.	24	1	1
1	10	15	25	.	24	1	1
1	125	25	150	.	24	1	1
1	10	25	35	.	24	1	1
1	10	.	.	.	24	1	1
2	20	15	35	70	34	3	2
2	25	10	35	70	34	3	2
3	30	25	55	.	30	2	1
3	.	20	.	.	30	2	1
3	30	20	50	.	30	2	1
3	25	20	45	.	30	2	1
4	20	15	35	209	23	1	2
4	25	24	49	209	23	1	2
4	20	18	38	209	23	1	2
4	35	20	55	209	23	1	2
4	10	22	32	209	23	1	2

Csum = cost1 + cost 2 (by row); Ctotal = sum(Csum) by studyno

I use multiple imputation to impute 10 datasets. I want to estimate, say:

regress Ctotal teratmentnumber ageband

If I had complete data, after calculating Ctotal, I would just keep one row per studyno for Ctotal age ageband treatmentnumber.

How can I do this after running multiple imputation, please?

If I just drop the redundant observations (rows, not variables cost1...) I get the following message:

"estimation sample varies between m=1 and m=2; click here for details
no results will be saved "

Thank you very much in advance.
Kind regards,
Rodolfo Hernández

[Daniel Shin]

I see potential issues with what you're trying to do.

1) It seems like you have a long panel data and you have run multiple imputation on this data. I would caution that the method you may have chosen does not take your data structure into consideration and treats each row as an independent observation. This could lead to suboptimal imputations.
2) How are you trying to keep only one row per studyno? In your example, which of the studyno1 rows would you keep? My guess is that the Ctotal value is imputed from the other variables which seem to differ.

If you truly need only one row, I would recommend that you format your dataset such that you only have one studyno per row, then impute and estimate.