How teffects nnmatch deal with mix covariates (numerical and multilevel) ?

[Mauricio Carvalho]

Hi everyone,

I am studying how "some" treatment (for the sake of confidentiality I am not allowed to explain exactly what this treatment is) can effect interest rate. i have two kinds of covariates, numerical and categorical ones. For categorical variables I have something like, for example, x = {0,1,2} and m = {1,2,3,4,...39}. I have tried to use -teffects nnmatch- using operator "i." and without "i." for such a variable and the results are different. So, my question is how the algorithm works? How it´s calculate the distance in the nearest neighbor algorithm ?

To be more clear,

M

and

Prz

are continuous variables and

Emp d NUTS cl c1 cod

are categorical variables with Emp being the only dummie (i.e, it is 0 or 1) the other ones are multilevel.

Then I run the two commands described below

teffects nnmatch  (TAA M Prz i.Emp i.d i.NUTS i.cl i.c1 i.cod) (treat), atet  biasadj(M Prz) vce(robust)

and

teffects nnmatch  (TAA M Prz Emp d NUTS cl c1 cod) (treat), atet  biasadj(M Prz)  vce(robust)

As I said, the results are quite different. So, how the distance showed below is calculated in this two different cases (with and without "i." operator)?

Thank you very much for your attention!
Kind Regards,
Mauricio.

[Mauricio Carvalho]

My questios is, more precisely, with respect to the vector of covariates x_i = { x_i1, x_i2 , ...} because in the case of mixed covariates how am I supposed do calculate the distance between, say, x_i = { 1000, 120, A, insurance} and x_j = { 3000, 30, B, manufacturing } ? Is it more appropriated to use - teffects psmatch - in this case ?

Again, thank you very much!
Kind Regards,
Mauricio.