using pscore - problems with syntax and final outcome

[antoniocapillo]

Hi all,

I have Stata 10, that is not compatible with PSMATCH2. That's why I am using Becker and Ichino (2002) (pscore, attnd, attnw, attr, atts, attk). I am not sur I am using the right syntax though. Does anyone know a good guidance for that?

Also, when I finally arrive to estimate the ATT, I would like to add an option that shows me the cofounders as well. Is that possible?

Thanks!

[Melissa Garrido]

All of these commands have associated help files (e.g. type "help pscore"). What kinds of problems with syntax/outcome are you having? If you post the code you are running along with the error messages, you're more likely to get a helpful answer from the forum.
What specific confounder information do you want displayed?

[Ariel Linden]

-pscore- will tell you whether the distribution of the covariates you use to estimate the propensity score are significantly different (treatment vs control) within strata. It is part of the output that is generated by the command. Psmatch2 works with v10, so why can't you use it?

[antoniocapillo]

All of these commands have associated help files (e.g. type "help pscore"). What kinds of problems with syntax/outcome are you having? If you post the code you are running along with the error messages, you're more likely to get a helpful answer from the forum.
What specific confounder information do you want displayed?

thanks Melissa. I fixed the syntax. In terms of cofounder, I would like to see the mean before and after matching - to show how the matching is making treatment and control comparable.

[Melissa Garrido]

Thank you for clarifying. To my knowledge, the commands you are using don't allow you to examine confounder balance after matching, unfortunately. If you specify the "detail" option after -attnd- or similar commands, you can see the mean of your outcome variable across the treatment groups, but that is it.

[antoniocapillo]

Thank you for clarifying. To my knowledge, the commands you are using don't allow you to examine confounder balance after matching, unfortunately. If you specify the "detail" option after -attnd- or similar commands, you can see the mean of your outcome variable across the treatment groups, but that is it.

Yes, that's what I managed to have. Not sure it makes sense, but could I put in the outcome variables position in turn each of the cofounders and replace it with the outcome variable in the model? As far as the treatment variable and the estimated propensity is the same, it should work, don't you think so?

[Melissa Garrido]

No, I'm sorry, I don't think that will work. You may wish to check with the authors of these commands for other suggestions.

[Sarah Edgington]

There's information in the helpfile for attnd about how to create a datafile with the indicators for which observations were used in the matching. You can then merge that indicator back to your original dataset and look at the means of your matched sample and compare those to the means for the entire sample before matching.

People have different opinions about what the best way to check for balance after matching is and not everyone agrees that comparing means is the best choice. However, if that's what you want to do, this is a pretty straightforward way to do it.

[gawantang]

So i have a question relating to merging the new data file with the indicator of whether or not the observation falls in the region of common support back into the original data file. Based on STATA's help info (below) I've created a new file and specified the variable in my original data set that uniquely identifies observations (iqid) but for some reason in the new file created all values for iqid are missing? Iqid is unique in the original data set.

xi: attnd promptACT_2wks expobabylowhigh if female==1, pscore(pscoreb12) comsup boot reps(10) dots matchvar(matchb12) matchdta(cm_matchbabylowhigh) id(iqid)
save, replace
use cm_matchbabylowhigh, clear
codebook iqid

So I can't merge back into my original data set.

Any ideas would be appreciated!
Grace

~~~~~~~~~~~~~~~~~~~~~

Referring to the attnd help file:
Furthermore, when a variable which identifies
observations exists in the dataset, users can specify a variable name for this ID variable using
the option id(varname) and this is saved as well.
id(varname) allows users to specify the variable name for a variable which identifies observations in the
data. This variable is then saved as part of newfilename.dta. This option enables users to merge
newfilename.dta by newvarname to their original dataset

[Sayoree Gooptu]

I am trying to estimate the average treatment effects too. However, I am getting different results for different commands. Using the pscore command, I find that the covariates in my model are balanced. Following this I use the attnd command. However, when I am using the psmatch2 command followed by pstest, it shows that some covariate means are significantly different in the treatment and control groups. Any suggestion would be really helpful.