Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.

# Re: Re: st: Propensity score matching: confidence intervals

 From Nyasha Tirivayi To statalist@hsphsun2.harvard.edu Subject Re: Re: st: Propensity score matching: confidence intervals Date Wed, 27 Jul 2011 20:26:03 +0200

```Dear Ariel

I am using continuous outcomes. It is an N to 1 match. I now
understand your original suggestion of doing regressions after
matching. I will try it and see if I can get standard errors and
confidence intervals.

Thanks a lot!

N. Tirivayi

On Wed, Jul 27, 2011 at 6:33 PM, Ariel Linden. DrPH
<ariel.linden@gmail.com> wrote:
> First off, you have not been clear with us on what you are measuring. I
> suggested using regression manually (outside of PSMATCH2) and add in either
> robust or clustered standard errors. Your response was that you are using
> logistic regression.
>
> We cannot be expected to provide meaningful help if you do not provide us
> sufficient information.
>
> Are you using binary outcome variables or continuous? PSMATCH2 does not
> differentiate and will treat your outcomes using OLS regression.
>
> How many matched controls are you using for each treated case? If you have a
> N to 1 match, you would need to account for this by weighting the
> observations within each matched group).
>
> My original suggestion stands, and that is to perform the outcome
> measurement manually after PSMATCH2 identifies the matches for you (and the
> weights, assuming you have an N to 1 matching strategy). This will allow you
> to run your regressions  with the appropriate standard errors (accounting
> for the dependent nature of the data) .
>
>
>
> Date: Wed, 27 Jul 2011 00:41:21 +0200
> From: Nyasha Tirivayi <ntirivayi@gmail.com>
> Subject: Re: st: Propensity score matching: confidence intervals
>
> Thank you for the responses. I am bit confused. in the psmatch2 code,
> there is no option for vce or robust. psmatch2 appears to only allow
> bootstrapping, even though the software authors say its unclear if its
> valid. My core problem is obtaining confidence intervals.
> Bootstrapping gives me confidence intervals based on the normal
> distribution. I am interested in reporting the t-statistic. How can I
> obtain confidence intervals that correspond to the t-statistic?
>
> Secondly my t-statistic indicates strongly significant effect. After
> bootstrapping, the p-value (normal distribution) is marginally
> significant. Which p-value do I base my interpretation, the
> t-statistic or the one after bootstrapping?
>
>
> N.Tirivayi
>
>
>
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/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/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/
```