Matching Samples

[Hardeep Randhawa]

Hi all,
I have cross sectional data and I am comparing wages between whites and ethnic minorities. I wanted to match the sample of whites to the sample of ethnic minorities (i.e for age as the average age of ethnic minorites is lower than whites) but I'm not sure how to do this in Stata. I've spent many hours researching into this but can't find a way to only keep the observations of whites that are matched with ethnic minority.
I would like to use the propensity score method and I've seen this in stata and used the command (which is teffects psmatch) however I don't know how to drop the observations that are not matched. I need the matched observations of whites to ethnic minorities to allow me to use oaxaca decompositions.
If anyone could help me I would be greatful

Regards,
Hardeep

[Hardeep Randhawa]

Anyone?

[Rich Goldstein]

your posting is confusing to me - using propensity scores means estimating a propensity of something (e.g., receiving a certain treatment) - what do you want to estimate the propensity of?

ignoring that issue; SSC contains several matching programs that allow you to know who is matched to whom - if someone is not matched that will be obvious and you can drop those cases - see, e.g., "vmatch"

[Hardeep Randhawa]

your posting is confusing to me - using propensity scores means estimating a propensity of something (e.g., receiving a certain treatment) - what do you want to estimate the propensity of?

ignoring that issue; SSC contains several matching programs that allow you to know who is matched to whom - if someone is not matched that will be obvious and you can drop those cases - see, e.g., "vmatch"

Thank you for the reply. I'm not 100% sure what I am meant to do. I am trying to replicate a paper and I'm looking at ethnic discrimination in terms of wages. I have 40,000 observations on white people and 1,000 on people from ethnic minority. When looking at the summary statistics of ethnic minorities, they are younger and located in one area. What I want to do is match the white sample to the sample of the ethnic minority. This is what the paper has done too, where a matched sample of whites is used to compare to ethnic minority groups using the propensity score approach. I just want to match the white sample on age and region to the ethnic minority sample. Then using this sub-sample of whites, I can run oaxaca decompositions.
The problem is I've never really studied propensity score method and I don't know how to use it, I would really value any help you give

[Marcos Almeida]

Hello Hardeep,

IMHO, propensity scores are deeply rooted in this principle: we have, say, 2 treatments whose outcomes we wish to compare, and we want the make the comparison as "fair" as possible. No doubt, "treatment" can mean a panoply of procedures and strategies. That said, it seems you lack this variable, i.e., it is not your study design, according to your short description.

Best,

Marcos

[Hardeep Randhawa]

Hello Hardeep,

IMHO, propensity scores are deeply rooted in this principle: we have, say, 2 treatments whose outcomes we wish to compare, and we want the make the comparison as "fair" as possible. No doubt, "treatment" can mean a panoply of procedures and strategies. That said, it seems you lack this variable, i.e., it is not your study design, according to your short description.

Best,

Marcos

Hi Marcos,
Thanks for your reply. Could I not use the dummy variable ethnicminority (where it takes value 1 if a person belongs to an ethnic minority and 0 if the person is white) as the treatment variable? As this would then mean I want to match the untreated (white sample) to the treated sample (ethnicminority)?

Thanks in advance

[Marcos Almeida]

Hi Hardeep,

I don't think so, unless under an animal experiment, and "treatment" would become something related to a genetic intervention.

Keep in mind that your query explicitly concerned propensity score matching. However, and considering yours is an observational study, you can match subjects under a case-control design. Finally, please don't forget the covariates are eventually "adjusted" when you perform some sort of regression under an appropriate model.

Best,

Marcos

[Adam Badenoch]

Hi Hardeep,

I am also a novice STATA user using teffects psmatch for the first time and want to know how to identify my matched controls do I can extract data about them. This is just a note to let you know that there is a second thread addressing this issue ( I've quoted the first post to help you find it).

Generating stubs has been suggested but we haven't found an answer that facilitates data extraction from the matched control group.

Hopefully consolidating our efforts in one place may help.

Dear Statalist,

I have a question regarding the treatment effects features in Stata 14. I am currently exploring it using the catanneo2 example dataset (Excerpt from Cattaneo (2010) Journal of Econometrics 155: 138-154) and run the model as described by Chuck Huber in the StataCorp_LP Youtube Chanel.

code:
teffects psmatch (bweight) (mbsmoke mmarried mage medu fbaby)


With Stata12 I used the user written psmatch2 *! version 4.0.11 22oct2014 E. Leuven, B. Sianesi which creates a number of variables for the convenience of the user: First and foremost _outcome_variable for every treatment observation stores the value of the matched outcome.

Is there any option that allows me to identifying which entities were matched e.g. mother 1 was matched with mother 47. Moreover can Stata14 store the value of each matched outcome e.g. store bweight of the nonsmoker mother 1 is in _bweight of the smoking mother47?

Thank you
Markus

Also, for what it's worth I don't see any problem in using your ethnicminority variable coded as 0 or 1 to identify your exposed vs unexposed groups. My understanding is that you should be able to use propensity score matching to compare these two groups ( just like you would be able to stratify or match them based on ethnicity using any other method)

Adam