CODEWITHSUNDEEP
comparestatamatching
Daniela Parada
Daniela Parada

Reputation: 1

Matching observations based on a single variable or multiple variables on a single data set - stata

I need to match observations based on an index variable that measures home conditions, personal variables such as age, gender, education, etc. and year. My home index variable is numerical (from 0 to 103) and the personal characteristics are either dummies or categorical variables. For my analysis I need to match the most similar observations based on these variables. It is sort of a nearest neighbor match but without having a control or treatment group.

The dataset looks something like this.

indice_hogar anio mes directorio orden mujer nivel__educativo_cat trabaja
0 2018 08 4700731 1 1 4 1
0 2018 08 4700731 2 0 5 1
0 2018 11 4777752 1 0 5 1
37 2018 04 4605803 1 0 3 1
42 2011 07 2735691 1 1 4 1
42 2018 02 4545459 1 0 3 1
43 2018 12 4803694 1 0 5 1
44 2018 10 4747974 1 0 5 1
46 2018 05 4610096 1 0 3 1
47 2018 04 4598828 1 1 1 0
47 2018 08 4687722 1 0 1 0
48 2018 04 4592941 1 0 5 0
48 2018 06 4636177 1 0 3 1
50 2018 06 4645892 1 0 1 1
50 2018 06 4645892 2 1 4 1

For better understanding, I am using an IV that is the ability of the most similar person according to the index and to personal characteristics. That means I need to find the most similar observation to, for example, person A and then be able to take the match's abilities and use it for a regression.

I have not been able to create a code.

Upvotes: 0

Views: 1173

Answers (1)

Scott Susin
Scott Susin

Reputation: 11

Duplicate your dataset, and match the 1st copy to the 2nd using nnmatch.

* Duplicate the data set
gen byte treat = 1
gen nobs = _N
save temp, replace
replace treat = 0
append using temp

* Make a fake outcome variable to keep nnmatch happy
gen byte outcome = runiform()<.5

* nnmatch performs a nearest neighbor match, return the id of the matched cases as nnid
teffects nnmatch (outcome indice_hogar nivel_educativo_cat trabaja) (treat), gen(nnid)

* Unduplicate the data set
keep if treat == 0

* change nnid to point to the 1st copy of the data set, not the 2nd
replace nnid = nnid - nobs

Upvotes: 1

Related Questions