Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: Using -sampsi- for clustered observations


From   Steven Samuels <sjhsamuels@earthlink.net>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: Using -sampsi- for clustered observations
Date   Fri, 27 Feb 2009 17:24:22 -0500

--

Ted, the answer depends on the purpose of your study: descriptive, comparative? What is being compared o? Will the factor defining the comparisons (age, exposure) differ between families, between individuals in the same familie, both? Are you interested in comparing means or comparing slopes/ changes in the repeated measurements? Do you have choices on the number of people to take within families, or, do you have to take all eligible? Is the number of repeated measurements related to the outcome (e.g. censored because of worsening condition). Do you plan on multilevel modeling, or do you just need to adjust standard errors for family differences?

Of other concerns--how were the families selected? Did you identify eligible subjects according to some criterion and then take their families? Or was there recruitment (hopefully random sampling!) of families.

I don't think that there is a general answer to your question, and simulation might be the best approach.

In any case, the Joanne Garrett's -sampclus- ("search sampclus, all") and the accompanying paper(http://econpapers.repec.org/article/ tsjstbull/y_3A2001_3Av_3A10_3Ai_3A60_3Asxd4.htm) may help.

-Steve

On Feb 27, 2009, at 4:40 PM, Ted White wrote:

Hi, all. I need to do a power calculation for a study that will contain two levels of clustering; there will be repeat dichotomous outcomes for individuals who will be clustered within families. While the documentation shows to use post() to specify the number of repeat outcomes in a dataset, I'm left with two questions:

1) What if the number of repeat observations will vary by respondent? I have a postulated distribution for the number of observations that will take place.

2) How to handle nesting by family in the power calculation?

If this isn't handled by -sampsi- it seems that I could do a series of simulations. But I'd feel pretty silly doing that if there's a simpler method that I don't know about.

Thanks!

Ted White
Yale University Center for Interdisciplinary Research on AIDS
*
*   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/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index