On Mon, 16 Jun 2003, Copeland, Laurel wrote:
> A pharmacist asked me if I could do a test for trend over time from this
> table:
>
> Year Pharm Clin Total Ratio
> 1997 400 44 444 4:1
> 1998 221 54 275 4:1
> 1999 187 69 256 3:1
> 2000 123 63 186 2:1
> 2001 74 42 116 3:2
> 2002 145 149 294 1:1
Without prejudice (meaning you can't sue me for defamation) I would worry
that what is expected is a mathematical sledgehammer to crush an answer
without worrying too much about the figures. There are techniques: to
those that have been mentioned, I'll add a runs test as it has least
assumptions.
But look at the data and consider the meaning. The annual totals vary by
4:1, and almost triple between 2001 to 2002. Unless the number of
treatments varied similarly - as might be the case if this relates to an
epidemic disease - then there is a very strange mechanism for reporting
adverse effects, with a very high variance. A trend test might come out
as significant, but another view might be that the 1997 total for
pharmacists was an outlier, and some new factor entered the system in
2002.
I'd worry about any "trend" claimed from a time series of 6 observations
as a statistical result. The term might be purely descriptive, but
calling it a trend invites the use of a forecasting model. The outcome is
seen on a regular basis in the behaviour of stockmarket investors.
R. Allan Reese Email: [email protected]
*
* 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/
