I'm not sure this is on-topic for the Evidence-Based-Health mail list. My
apologies for going off on a tangent.
Dr. Bob Phillips writes:
>We're coming across difficulties in choosing the correct way of
>producing confidence intervals, especially for studies where the
>results are small.
I assume you mean that that the sample sizes are small. You don't say
exactly how small, though.
>What we've got are two opinions - I think - one holding out for the
>use of the Newcome-Wilson score and one urging the use of exact
>intervals based on a binomial distribution.
This is also a bit vague. Is your outcome variable binary (e.g., live/dead)?
That requires a different type of confidence interval than a continuous
outcome measure like cholesterol level.
In general, the use of exact methods requires specialized software and
cannot be done in a spreadsheet like Excel. The confidence interval has no
easily expressed algebraic form that you can type into a spreadsheet cell.
Even if you are lucky enough to find an obscure function in Excel that works
you still have problems. Excel does not have all of the graphical and
diagnostic tools that you would need to provide a good quality data
analysis.
>I'd love to know which is the best (or any others) given that we
>need to calculate them using Excel.
What is best depends on what your objective is. If you are requiring the
work to be done in Excel, that implies to me that the results are not
intended for any serious use, such as publication in a peer reviewed
journal. If the results aren't important enough to invest some time and
energy in a professional quality statistical analysis software system, then
you might as well use the basic formulas for confidence intervals that are
found in any introductory statistics textbook.
If money is an issue, there is plenty of free software that would produce
better quality data analyses than Microsoft Excel. The Centers for Disease
Control and Prevention in the United States have a nice piece of software
called Epi-Info, but there are many others that also produce good and free
statistical software. There are even some nice quality Java applications on
web pages that would do better than Excel.
If the issue is one of training (e.g., my data analyst can only work with
Excel), then perhaps that fact should encourage you to find a better trained
data analyst.
I don't mean to be too harsh about Excel. Keep in mind that it was built for
financial applications and it does those very well. It is better than most
spreadsheets, and I use it quite a bit for very simple statistical
calculations. It also is wonderful for outlining a variety of scenarios as
you might need in a sensitivity analysis. In addition, it can serve as a
useful teaching tool for some things.
But using Excel for advanced statistical analysis is like pounding in nails
using a screwdriver.
Steve Simon, [log in to unmask], Standard Disclaimer.
STATS: STeve's Attempt to Teach Statistics. http://www.cmh.edu/stats
P.S. There are some add-on packages that you can supplement Excel with, but
if you are going to the trouble of buying something, why not buy something
that was designed for data analysis from the ground up.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|