Many thanks to all those who took the time to respond to my query about a
nomogram - one respondent suggested I should should summarise the responses,
so here goes.
1. A number of respondents directed me to the work of Doug Altman, in
particular the nomogram published in BMJ 1980; 281:1336-1338 (referenced in
Medline as page 2336-2338, but this is incorrect) and Practical Statistics
for Medical Research. 1991, p456-460.
2. Several respondents also mentioned a nomogram published by Young et al.
Sample size nomograms for negative trials. Ann Intern Med 1983; 99: 248-251.
The above present their tool slightly differently and require some minor
arithmetic on the part of the user. Both can be used for continuous or
dichotomous data, and the Altman nomogram can be used with studies that have
different sample sizes in each arm. I suspect it is a matter of taste as to
which is preferable, although the Altman nomogram uses a layout I am more
familiar with.
3. Other respondents mentioned their favoured biomed stats texts, including
Glantz SA. Primer of Biomedical Statistics. McGraw Hill, and Norman &
Streiner. 1994. Biostatistics; The bare essentials. St Louis: Mosby.
4. Finally Rob Herbert suggested that it might be better to concentrate on
teaching the calculation of confidence intervals rather than power and
offerred a simple calcuIation. In order not to misrepresent him I will quote
him directly:
"I have been thinking about this issue for several years - what is the best
way to help students EASILY determine the statistical power of a study. Over
the past few years I have moved away from talking about statistical power -
now I encourage
students to calculate confidence intervals instead. I find it's easier for
students to understand CIs, and easier for them to get an answer. My efforts
have been directed towards teaching them easy short cuts for calculating
CIs. For example, when comparing the difference b/w 2 means, I encourage
students to generate quick and dirty CIs by taking the observed
difference b/w means and adding and subtracting (3SD/sqrt(n)). [Here SD is
the average SD of the 2 groups, and n is the average sample size of the 2
groups). It works well, and it's easy enough to do in your head (if only
roughly)! I'm still working on a good short cut for dichotomous data."
Many thanks for everybody's assistance.
regards
Andrew Jull
Clinical Nurse Consultant
Auckland Hospital
NEW ZEALAND
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|