Re: st: Sample size calculations

[Nick Cox <njcoxstata@gmail.com>]

Regressions are not called "multivariate" if there is only one
response. Perhaps several decades ago having several variables in a
dataset was enough for people to talk about multivariate problems, but
terminology has changed. See also -help mvreg-.

That aside, what you are trying to do here? Sample size calculations
are for planning data collection, but it seems that you have the data.

Data on frequency of -ethnicity- is directly obtainable by e.g. -tabulate-.

If you are seeking some answer to the questions of the form "how large
would be my sample size need to be for something to be significant at
some level?" then you can experiment directly. Use -expand- or
weighting to see what answers would be with a larger dataset. Or
sample randomly to get a smaller dataset.

Nick

On Thu, Mar 14, 2013 at 2:01 PM, Amal Khanolkar <Amal.Khanolkar@ki.se> wrote:

> I plan to run a set of multivariate linear regressions models which will help analyze ethnic differences in my outcome of interest.
>
> regress Y i.ethnicity age i.education height bmi bloodpressure
>
> I have 14 different ethnic groups.
>
> As some of my ethnic groups are fairly small I would like to calculate the sample size needed for each of the fourteen ethnic groups. I have a rough idea of the expected effect or parameter p in each ethnic group.
>
> I've only calculated sample sizes for a two group comparison using the sampsi command in Stata.
>
> Does anyone know how I could go about testing expected sample sizes for several categories of the exposure variable?
>
> I would set other parameters needed to calculate sample size as follows: alpha=0.05, power (0.90), expected effect for most ethnic groups varies between -2.85 to -7.65
>
*
*   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/