Rachele Capocchi <[email protected]>:
One way is to run a regression. If you run a linear regression of
your outcome on year, you get the linear trend estimate, which you can
convert to a growth rate over the whole period using the mean of the
outcome at a variety of points in time (for example the central time
period), depending on the desired interpretation. If you regress the
log outcome on year, you get the exponential trend estimate, in
percent growth per year. Note that if the trend is neither linear nor
exponential, you may prefer a different transformation of the outcome,
or a different type of regression.
sysuse uslifeexp
su year
replace year=year-r(mean)
reg le year, r
nlcom _b[year]/_b[_cons]
poisson le year, r
nptrend le, by(year) nodetail
If you want a test rather than a CI, use the -test- command or
-nptrend-. Did you want a p-value on a test of the null hypothesis
that the trend is zero? Those p-values appear in the example above.
If you only have 5 data points, though, I don't think you should put
much trust in asymptotic approximations.
On Wed, Jun 4, 2008 at 5:54 AM, Rachele Capocchi
<[email protected]> wrote:
> Dear Statalisters,
>
> I have to do a trend analysis on the hospitalization rates in 5 years,
> between 2001 and 2005. I have the absolute numbers of hospitalization,
> the population of all years and I calculated the hospitalization rates
> yearly.
> I need to calculate the CI for the growth rate (ex. 12,5% - 13,5% growth
> rate from 2001 to 2005) and the p value.
> Which test I have to do ?
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
