> David also noted that
> >"3. You "ski slope" bar charts have to be explained
> >somehow......<snip>but still the
> >cheapest practices spend usually about a third of the most
> >expensive."
> er - normal distribution? I think a situation where the variability
> disappears would be terribly suspicious in statistical terms.
I have a technical question about this.
I'm not clear whether these types of graphs are actually sampling
distributions. An example would be ranking practices in terms of
their antibiotic prescribing costs/1000 patients registered. So long
as data collection is for long enough (eg a year) and the event
you're collecting data about is common enough (antibiotic prescribing
probably is - 1000s of events, atypical antipsychotic prescribing
probably isn't - 10s of events), then what's collected isn't a
sample, it's the whole lot. For uncommon events, then I guess you
would need a lot longer data collection period to allow for chance
variation in presentation.
Would this mean that for common things like antibiotic prescribing
collected over a reasonable time (say a year), confidence intervals
calculated on the basis of it being a sampling distribution would be
inaccurate because they would be too big (no finite population
correction)? Or is there some subtlety about chance variation over
time that I'm missing?
I'd be grateful for any comments on this.
Bruce
Bruce Guthrie,
MRC Training Fellow in Health Services Research,
Department of General Practice,
University of Edinburgh,
20 West Richmond Street,
Edinburgh EH8 9DX
Tel 0131 650 9237
e-mail [log in to unmask]
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|