surveyselect question

[kurofufu]

I used proc surveyselect to create control and test groups of equal size. But I was not able to make the average of some key variables tp be similar between two groups. What options can I use to remedy this situation? If I keep changing the seed number until two comparable groups are obtained, then this seems not random sampling anymore. Does SAS have an option to select two groups of equal size with all key variables having similar distribution?

[PGStats]

Hi. To my knowledge, there is no provision in surveyselect to select covariate-balanced samples. For the single covariate case there are relatively simple methods to balance samples. Three methods are compared in . Find a more general discussion in the paper http://www.stat.ubc.ca/lib/FCKuserfiles/STAT527A.Minimization.Final.pdf

PG

[kurofufu]

Thanks! This saves me a lot of time.

[kurofufu]

Any comments on this paper for my question?

http://support.sas.com/resources/papers/proceedings10/213-2010.pdf

Second, can we use propensity matching here?

Suppose we have 2000 members available for creating a control and test groups.

1. Randomly select one members from them

2. Use propensity score matching to find a matching member

3 Repeat step 1 and 2 until exhaust all members

ps. SAS paper for propensity score matching: http://www2.sas.com/proceedings/sugi26/p214-26.pdf