missing values in melogit

[Sam Honer]

Hello,

How does melogit deal with missing values?
I am running several melogit commands whereby I am accounting for clustering of individuals (it is a repeated measures design). So, I have melogit commands that look like this (increasing complexity).

 melogit suicide score if flw_up == 1|| ppt_id:, or

 melogit suicide score age i.sex_ i.employment i.ethnicity if flw_ == 1 || ppt_id:, or

I have a total of 800 participants who have var==1, but the number of groups in the output varies. For the first line of code the output indicates a total of 800 groups and for the second one 799.
I thought that melogit uses all available information and does not carry out listwise deletion as with other models, is this not the case? Am I missing something?


SH

[Clyde Schechter]

-melogit-, like all Stata regression commands uses listwise deletion to deal with missing values.

[Sam Honer]

Oh okay, seems like I may have misunderstood this before. Thanks for clearing it up - I thought that might explain why the number of groups was changing.

[Sam Honer]

Hi Clyde,

I am running melogit on a sample with longitudinal data whereby some people were followed up and others were not. 1000 people were included in the baseline study and then 600 of them were followed up.

melogit outcome_var x1 x2 x3 || subjectid, or

Yet, the melogit output says the number of groups is 998. Just wondering how this is the case if it does in fact carry out listwise deletion.

Thank you,
SH

[Clyde Schechter]

Listwise deletion refers to the removal of any observation where there is a missing value on any of the variables in the model. What you are referring to here is something different. The same person may have multiple observations in the data set corresponding to different times of observation. If there are missing values for some variables in some observations, those observations will be excluded from the estimation sample, but any other observations of the same person that have complete data will be retained in the estimation sample. So whole persons (groups, as far as -melogit- is concerned) are not deleted just because some time-period observations are partially or entirely missing. But any single observation is omitted if any variable is missing.

In this regard, -melogit- behaves like all other -me- and -xt- commands in Stata.

[Sam Honer]

So, in this sense, am I correct in my understanding with the -me- commands in Stata, both people who have been followed up and those who have not contribute information to the model and that this works well under the assumption that data are MAR?

Is it otherwise suggested to restrict the sample only to those who were followed up?

[Clyde Schechter]

The understanding you express in #6 is correct.

There is no reason to restrict the sample only to those who were followed up. In fact, doing so may introduce bias.

[Sam Honer]

Great, thanks a lot. Appreciate your input and quick responses!