using MICE on panel data

[Dr. Iqra Yaseen]

hello STATA. I have a panel data for 45 developed countries for 25 years. the countries are paired to make bilateral combination. my estimation model has 2 dependent variables as tradeflow_imf_o (continuous variable) and tradeflow_imf_d (continuous variable). my independent variables are disputes (categorical variable), outcome (factor variable), timeframe (factor variable) and rta (dummy variable). i have missing data for tradeflow_imf_o, tradeflow_imf_d and rta, although 85% of the dataset has completed data for these 3 variables. the missing pattern for these 3 variables by using

misstable pattern tradeflow_imf_dtradeflow_imf_o rta

the result i get is


Missing-value patterns
(1 means complete)

Pattern

Percent

1

2

3

87%

1

1

1

10

1

1

0

2

0

0

0

<1

1

0

0

<1

1

0

1

<1

0

0

1

<1

0

1

0

100%

Variables are (1) rta (2) tradeflow_imf_d (3) tradeflow_imf_o


the pattern seems to be non-monotone, although i am not sure about it. plus, i am assuming the data to be MCAR.

the MICE code i used is

mi set mlong
mi xtset year
mi register imputed tradeflow_imf_d tradeflow_imf_o rta
mi register regular disputes outcome timeframe
mi impute chained (regress) tradeflow_imf_d tradeflow_imf_o (logit) rta = disputes outcome timeframe, add(20) rseed(random_number)

although after a successful code run, the results i get are undesired. the variables tradeflow_imf_d and tradeflow_imf_o impute very low negative values although their value should be above 0 and positive.

please correct my code.

thank you

[Rich Goldstein]

first, you should review the following FAQ which might be relevant: https://www.stata.com/support/faqs/s...and-mi-impute/

second, when you impute using regress you are specifically allowing for the chance that imputed data may be outside some range (in your case, <0); you might want to use method "pmm" instead - see the help file

help mi impute chained

[Dr. Iqra Yaseen]

as per your advice Rich Goldstein, i edited my code to include

pmm

. the modified code is

 mi impute chained (pmm, knn(10)) tradeflow_imf_d tradeflow_imf_o (logit) rta = disputes outcome timeframe, add(20) rseed(6755)

but the result i get is

Conditional models:
               rta: logit rta tradeflow_imf_d tradeflow_imf_o disputes outcome timeframe
    tradeflow_im~d: pmm tradeflow_imf_d i.rta tradeflow_imf_o disputes outcome timeframe , knn(10)
    tradeflow_im~o: pmm tradeflow_imf_o i.rta tradeflow_imf_d disputes outcome timeframe , knn(10)

Performing chained iterations ...
convergence not achieved
convergence not achieved
logit failed to converge on observed data
error occurred during imputation of tradeflow_imf_d tradeflow_imf_o rta on m = 1
r(430);

but when i exclude tradeflow_imf_d from MICE code, i get the desired results.

what may be the possible cause of code crash after including tradeflow_imf_d.

[Rich Goldstein]

sorry, but I know nothing about your data - maybe someone else can help you

[Dr. Iqra Yaseen]

thanks Rich Goldstein for at least replying. hope for someone to help me out.