I am working with unbalanced panel data at the individual-month level. I would like to match individuals in my data using CEM and then run a logistic regression using cem weights, with strata fixed effects, to estimate the effect of my treatment variable (treat_yn) on my dependent variable. Unfortunately, I do not have the rights to share the data even anonymized. However, below is the basic format of the data (collapsed to the individual level):
sl_no | district | village | individual_id | household_id | gender | age | hh_char1 | hh_char2 | treat_yn |
1 | d1 | v1 | ID_1 | HH_1 | M | 25 | 0.5 | 0.33 | 0 |
2 | d1 | v1 | ID_2 | HH_1 | M | 25 | 0.5 | 0.33 | 1 |
3 | d1 | v1 | ID_3 | HH_1 | F | 45 | 1 | 0 | 1 |
4 | d1 | v2 | ID_4 | HH_2 | M | 68 | 0.45 | 0.66 | 0 |
5 | d1 | v2 | ID_5 | HH_2 | M | 23 | 0.33 | 0.33 | 1 |
6 | d1 | v2 | ID_6 | HH_2 | F | 34 | 0.66 | 0.5 | 0 |
7 | d2 | v3 | ID_7 | HH_3 | F | 45 | 0.25 | 0.20 | 1 |
8 | d2 | v3 | ID_8 | HH_3 | M | 35 | 0.5 | 0.5 | 0 |
9 | d3 | v4 | ID_9 | HH_4 | F | 22 | 0 | 1 | 0 |
The cem command I am currently using is as follows:
cem district village age(25 32 40) gender hh_char1(1 1.5 2 3 4) hh_char2(0.01 0.25 0.3333333 0.5 0.75), treatment(treat_yn)
However, this does not ensure that individuals with the same household_id are NOT matched. Is there a way to achieve this in the cem options?
While there appear to be ways to force an exact match on a variable, I cannot find a way to force an exact 'no match'. Any leads on how to achieve this would be helpful.
0 Response to Coarsened Exact Matching - How to force cem to not match on a variable
Post a Comment