Re: st: Test for statistical significance

[Austin Nichols <austinnichols@gmail.com>]

Cornelius Nattey <cornelius.nattey@nioh.nhls.ac.za>:

Regress manganese blood levels on a dummy for month equal to November,
clustering by person (or higher level if using clustered survey data,
or if there is other clustering suspected).  You might want to use
-glm- with a log link for your regression, since manganese blood
levels must be nonnegative, especially if they exhibit right skew.
The easy interpretation in the latter regression would be in terms of
a percentage difference, rather than a difference in levels.

webuse nhanes2, clear
egen c=group(strata psu)
ren female nov
qui reg zinc nov [pw=finalwgt], cl(c)
loc d=cond(sign(_b[nov])==1,"higher","lower")
di "`: var lab zinc' is " abs(_b[nov]) " `d' in November"
qui glm zinc nov [pw=finalwgt], link(log) cl(c)
di "`: var lab zinc' is " abs(exp(_b[nov])-1) "% `d' in November"
nlcom exp(_b[nov])-1

On Tue, Aug 31, 2010 at 10:24 AM, Cornelius Nattey
<cornelius.nattey@nioh.nhls.ac.za> wrote:
> Dear All,
> I have two datasets of manganese blood levels taken in August and
> November 2005. They were from the same community but mainly different
> people in the community. There may have been some repeats but I do not
> know how many if there were any. I have treated them as independent and
> not paired tests.
>
> What test should be applied to see if there is a statistically sig
> difference? Is there a difference?
>
> Thanks very much
> Cornelius
*
*   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/