_mi_miss not found

[Lukas Arnold]

Hi everyone,
I want to do MI imputation but always get the error: _mi_miss not found. It occours even at the point of mi register variables.

I use these commands:

misstable sum, gen(miss_)
tab1 miss_cultcap miss_econcap miss_mj
mi set wide
mi register regular female Postmaterialism country_pp_gdp_ten
mi register imputed cultcap econcap mj
mi impute mvn cultcap econcap mj, add(10) noisily

Error:
(passive variables rel_red abs_red unregistered because not in m=0)
(imputed variables gini_disp gini_mkt unregistered because not in m=0)
_mi_miss not found

The variables, listed in the error (rel_red, abs_red, gini_disp gini_mkt) are not even in my data set (I droped them before imputation)

Do you have any ideas why this happens?

Thank you in advance! Any recommendation would be helpful!


PS: all STATA files are up to date

[daniel klein]

Start all over from the original dataset. Type

use filename_of_original_dataset
mi query

Stata should respond with

. mi query
(data not mi set)

If you see anything else, you have some left-over characteristics from a previous run of mi commands; in that case, show the results of

char list _dta[]

(after making sure it does not contain confidential information).

[Lukas Arnold]

thank you, Daniel!
here are the results of char list _dta[] (and I can already see that there are these two variables!

. char list _dta[]
_dta[ReS_Xij_n]: 1
_dta[ReS_Xij_long1]: y_
_dta[ReS_Xij_wide1]: y_2003 y_2007 y_2011 y_2015 y_2019
_dta[ReS_i]: Country_Name Country_Code Indicator_Name Indicator_Code
_dta[ReS_ver]: v.2
_dta[ReS_j]: year
_dta[ReS_str]: 0
_dta[ReS_Xij]: y_
_dta[_lang_c]: default
_dta[_lang_list]: default
_dta[_mi_pvars]: rel_red abs_red
_dta[_mi_ivars]: gini_disp gini_mkt
_dta[_mi_M]: 100
_dta[_mi_marker]: _mi_ds_1
_dta[_mi_update]: 2003480417
_dta[_mi_style]: wide
_dta[ReS_Xij_wide2]: x2_1 x2_2
_dta[ReS_Xij_long2]: x2_

[daniel klein]

It seems you already have 100 imputations in your dataset. If you are sure that there aren't, then you could try typing

mi extract 0

which should leave you with the original, unimputed dataset. You can then start over imputing missing values again.

[daniel klein]

By the way,

mi impute mvn cultcap econcap mj, add(10) noisily

is probably not what you want. If you have variables without missing values (typically the regular variables), you should include those as predictors in the imputation model. Otherwise, the respective associations will be biased towards zero.