st: RES: Propensity Score Matching

["Henrique Neder" <hdneder@ufu.br>]

Matthew

I do not know if this is an appropriate response but perhaps your covariates (age and region) do not discriminate much the values ​​of your dichotomous variable (black). The pscore command will begin with 5 blocks and will split the blocks if the test for the difference of propensity score means is not achieved (equality of propensity scores means in each block). So, it is unexpected that the algorithm ends in only one block. One possible answer for your so short common support region is that one of your groups has a much small interval of propensity scores, that is, there is only small values of the probability to belong in the reference group, say black. So, I propose that you verify the probit model results. I suspect that you will find some problems like, for example, some of your covariates has a great predictability and this may be an explanation for your unusual findings. Furthermore, I think (I am not so sure about this) that your covariates used in the matching !
 algorithm may approximately be the same as your control variables in your discrimination regression model. For the second part of your message, I prefer to use the psmach2 command because it is more appropriate do gather the matching results and allows to recover information about the matched individuals for your posterior regression.

Best regards

Henrique Neder

Universidade Federal de Uberlândia - Brasil     

-----Mensagem original-----
De: Matthew Mulley [mailto:MXM194@bham.ac.uk] 
Enviada em: terça-feira, 11 de fevereiro de 2014 14:35
Para: statalist@hsphsun2.harvard.edu
Assunto: st: Propensity Score Matching

Dear statalist,

I am currently doing some research into ethnic discrimination and would like to use the propensity score matching software to match the UK LFS merged sample using age and region before running regressions. However when I use the following command pscore

"pscore BLACK AGE REGION ,pscore(PS1) blockid(blockf1) comsup level(0.001)"

My region of common support estimated by the software is [.01604683, .01876305] with the final number of blocks being 1. It seems unusual that the common support region should be so low when I am only matching on two variables?

Furthermore after creating the propensity score regardless of what type of matching i.e. attnd etc. I use how do I get the software to create observations that are indicate they are matched so I can then run regressions using only the matched individuals in my sample?

Many thanks,

Matthew Mulley
University of Birmingham Student




*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/