Imputation not using all obs

[Robyn Jones]

Hi, I am using stata v18 on mac OS Ventura 13.4.1

I am using a dataset with a total of 8410 observations, when I check missing data I get the following results:

misstable summarize agew4 sex ethnicity notmarried adl wealth education mmw5 ssupport4

Variable | Obs=. Obs>. Obs<.
-------------+--------------------------------+----
ssupport6 | 3,780 4,630
adl | 673 7,737
wealth | 208 8,202
education | 171 8,239
--------------------------------------------------

I run the imputations using the following code and the result returns the model results but with only 7297 observations. What is the reason for this please?

global modelb agew4 i.sex i.ethnicity
global modelc $modelb i.education i.wealth i.notmarried i.adl

mi set mlong
mi register regular sex agew4 ethnicity notmarried mmw5
mi register imputed ssupport4 education wealth adl
mi impute chained (regress) ssupport4 (logit, augment) adl (ologit, augment) wealth education = mmw5 sex agew4 ethnicity notmarried, add(45) rseed(54321)

mi estimate, post: nbreg mmw5 ssupport4 $modelc [pweight=w4scwt], irr

Thank you

[Robyn Jones]

I have used:

mi estimate, esample(check): nbreg mmw5 ssupport4 $modelc [pweight=w4scwt]

looking at this check variable I can identify participants where the imputations/observation isn't being used – some with fully observed data and some without, I can't seem to find a pattern for why they would be excluded? are there any known problems or explanations?

[Robyn Jones]

Here is an example of the imputation for one participant who is excluded from the mi estimate:

age	ethnicity	exp	outcome	wealth	education	sex	spouse	limitation	_mi_m	esample check
84	1.white	.	3	1	NVQ1/CSE other grade equiv	male	0	.	0	0
84	1.white	1.693354	3	1	NVQ1/CSE other grade equiv	male	0	none	1	0
84	1.white	1.976703	3	1	NVQ1/CSE other grade equiv	male	0	none	2	0
84	1.white	1.389476	3	1	NVQ1/CSE other grade equiv	male	0	1+	3	0
84	1.white	2.325524	3	1	NVQ1/CSE other grade equiv	male	0	none	4	0
84	1.white	1.723726	3	1	NVQ1/CSE other grade equiv	male	0	none	5	0
84	1.white	1.579414	3	1	NVQ1/CSE other grade equiv	male	0	none	6	0
84	1.white	1.26626	3	1	NVQ1/CSE other grade equiv	male	0	none	7	0
84	1.white	2.209294	3	1	NVQ1/CSE other grade equiv	male	0	none	8	0
84	1.white	1.510552	3	1	NVQ1/CSE other grade equiv	male	0	1+	9	0
84	1.white	1.900252	3	1	NVQ1/CSE other grade equiv	male	0	none	10	0
84	1.white	2.055556	3	1	NVQ1/CSE other grade equiv	male	0	none	11	0
84	1.white	1.226345	3	1	NVQ1/CSE other grade equiv	male	0	1+	12	0
84	1.white	1.704704	3	1	NVQ1/CSE other grade equiv	male	0	none	13	0
84	1.white	1.754613	3	1	NVQ1/CSE other grade equiv	male	0	1+	14	0
84	1.white	1.66927	3	1	NVQ1/CSE other grade equiv	male	0	none	15	0
84	1.white	1.270409	3	1	NVQ1/CSE other grade equiv	male	0	none	16	0
84	1.white	2.162107	3	1	NVQ1/CSE other grade equiv	male	0	none	17	0
84	1.white	1.880571	3	1	NVQ1/CSE other grade equiv	male	0	none	18	0
84	1.white	1.348033	3	1	NVQ1/CSE other grade equiv	male	0	none	19	0
84	1.white	0.7318558	3	1	NVQ1/CSE other grade equiv	male	0	none	20	0

[daniel klein]

We need to see more complete output -- wrapped inside CODE delimiters, please. It is nearly impossible to read the tables as posted now. In particular, we need to see the complete output of

misstable summarize

before imputation for all variables used in either the imputation or the analyses model. That includes the weights (which should probably be in the imputation model, too).

We also need to see the output (including any notes) from

mi impute

and (at least the header of) the results table from

mi estimate

Edit:

By the way, please show us the code that you have really used! Some of the variable names in #3 do not appear in any of the code shown in #1. Also, the esample() option of mi estimate requires flong or flongsep style; according to #1 you used mlong. You are wasting our time and yours if you do not show us what you have typed, exactly, and what Stata did in response, exactly.

[Robyn Jones]

Hi Daniel,

I have identified the issue now but thanks for the advice on how to post! Sorry that it is not formatted correctly and unreadable – I am constantly working on trying to figure it out myself so there are things that had changed which I forgot to mention.

Thanks again!

[daniel klein]

I have identified the issue now

Then please tell us more as others may find themselves in a similar situation and will benefit from your work.