MICE method for missing data

[Sabine Dicks]

Hi everyone. I am working on my thesis. I have 3203 observations in my dataset. I am estimating the chance of a radicalized individual turning violent based on the PIRUS dataset (https://www.start.umd.edu/data-tools...d-states-pirus). This is an opensource cross-sectional data set thus data is not missing at random. Beckers (2021) also used this method to deal with missing data when he used the same dataset. All variables in the dataset are binary except for age.

However, I do not really know how to do this.

I now used these commands:

mi register imputed Group_Membership
mi register imputed Radical_Beliefs
mi register imputed Marital_Status
mi register imputed Student
mi register imputed Employment_Status
mi register imputed Work_History

mi impute chained (logit) Group_Membership Radical_Beliefs Marital_Status Student Employment_Status Work_History = Violent Gender Radicalization_Far_Left Radicalization_Far_Right Radicalization_Islamist, add(10)

When I do this I am getting 27082 observations when I summarize the data instead of 3203 (inital number of observations). How do I solve this?

Kind regards,
Sabine

[Carlo Lazzaro]

Sabine:
if you add complete datasets to your original one, no wonder that the number of observations increased.
If this is your question, there's nothing to fix, as this is the way multiple imputation works.

[Rich Goldstein]

In addition to Carlo's helpful reply (yes, your new data set (under the assumption below) will be 11 times larger than the original data set) - your numbers do not match this, but I am assuming that whatever variables you tried to -summarize- had missing data

I'm guessing that your data structure here is now "flong"; in that case, use the new variable _mi_m and the original data have a 0 for this variable

for what estimation commands, including descriptives, work correctly with -mi-; see

help mi_estimation##estimation_command

I don't understand at all the following from your original post: "This is an opensource cross-sectional data set thus data is not missing at random."

also, as per the FAQ, complete citations can be helpful but just giving author and year is close to worthless