topic How to use known Std for mean using aggregated data in proc mixed in Statistical Procedures

How to use known Std for mean using aggregated data in proc mixed

thowald — Mon, 09 Sep 2019 07:18:23 GMT

I have aggregated data for group i including number of observations ,mean, Std, + 1 independent variable. I want to estimate a linear random effect model with group as random effect like mean = independent variable + group. As said before, since I deal with aggregated data mean and Std (and so standard error SE) for mean is known. How can I put in the known Std or equivalently SE for the mean into SAS proc mixed in a simple way?

I did in Winbugs but would like to compare the results with SAS proc mixed or any other SAS proc in SAS 9.4 TS 1M6.

Any hint very welcome

Thomas

Re: How to use known Std for mean using aggregated data in proc mixed

pau13rown — Fri, 13 Sep 2019 06:16:37 GMT

use proc glm and a random statement with 'inverse' option in the model statement to incorporate SE as weights?

although anne whitehead says in her book on meta-analysis, section 4.2.4 on weighted LS regression, regarding SAS: "It should be noted that the standard error and test statistics displayed for the intercept parameter are incorrect for the required model, because they assume that var(εi) = σ2/wi, where σ2 is to be estimated from the data, instead of equal to 1. This will also be the case for other statistical packages."

i'm not sure if this is helpful ....

Re: How to use known Std for mean using aggregated data in proc mixed

thowald — Thu, 19 Sep 2019 11:33:35 GMT

Thanks for your suggestion. Since the SE depends on SD as well as N, how do I include N i.e. the number of observations each mean is based on into the model?