Multiple Imputation-Incomplete values need to be imputed

[Ginny Natale]

1. Im running an nbreg on CysC is the outcome. I have a key variables: Race, age, sex, diabetes, smoking, depression, and diagnosis.
2. This is my model: nbreg OCRP_ADJ i.DX i.reth agey ragender diab drink smokev cesd, irr has 2,476 listwise deletions
3. There are 2,476 values for CYSC that dont have a race value.
4. I want to impute the missing Race variable and impute the 2,476 values
5. When i run mi impute regress Stata skips over the 2,476 values I want to impute.

mi set wide
mi register imputed reth
mi impute regress reth OCYSC_ADJ agey ragender diab drink smokev cesd if (miss_OCYSC_ADJ==0), noconstant add(10) force


Univariate imputation Imputations = 10
Linear regression added = 10
Imputed: m=21 through m=30 updated = 0

------------------------------------------------------------------
| Observations per m
|----------------------------------------------
Variable | Complete Incomplete Imputed | Total
-------------------+-----------------------------------+----------
reth | 4769 2476 0 | 7245
------------------------------------------------------------------
(complete + incomplete = total; imputed is the minimum across m
of the number of filled-in observations.)

Note: Right-hand-side variables (or weights) have missing values;
model parameters estimated using listwise deletion.

.
end of do-file

[Rich Goldstein]

without a data example (see the FAQ and esp. the part about -dataex- please), it is impossible to be sure but here is my guess - the "missing" entries are extended missing values (".a", etc.); by design, these are not imputed; see the following which is from -help mi-:

If an imputed variable contains a value greater than . in m=0 --
it contains .a, .b, ..., .z -- then that value is considered a
hard missing and the missing value persists in m>0.

[daniel klein]

Some remarks:

• Race is usually not measured on an interval scale, hence, linear regression might not be appropriate. Perhaps mlogit would be a more appropriate choice?
• If reth is the only variable with missing values, you probably want: mi impute mlogit reth = OCYSC ...; note the equals sign.
• Why do you want conditional imputation, i.e., why do you specify an if condition?
• Why do you add the nocosnstant option?
• Why do you add the force option? You almost certainly do not want this. If you specify this option to ignore/suppress error messages, you should really better fix the underlying problem.

Best
Daniel