Lu 30 mai 2011
Hello EpiData users,
I'm a new subscriber to this list.
I used recently EpiData Analysis for a matched case-control with a controls:case ratio of 38:19 or 2:1 (TAB DISEASE EXPOSURE COMMONID /O /t /ex).
In the results output, and in the online help, for the crude table, I could'nt find the type of test used, for example, which kind of Chi-square (I was able to deduced it was uncorrected), of odds ratio (maximun likelihoodestimate [MLE]?) and associated confidence interval (95%?), and Fishers exact test (bilateral?). In the adjusted table, the OR [MLE?] is provided with a CI (MLE and 95%?), followed by a Chi-square (unspecified kind) and another p value (which kind?). The provided p values are rounded at 3 digits after the decimal point. The last column provide p-exact values (Fishers?), and the one on the adjusted line seems to refer to another Chi-squared provided at the bottom of the table. Finally, the exact p value for the adjusted results (e.g.: 0.3140) doesn't match with the "probability of MLE >= value of MLE of OR (e.g. 6.10) if population OR = 1.0" that I obtain (e.g..: 0.00196) with the command MATCH COMMONID DISEASE EXPOSURE of Epi Info 6.04d DOS (which is probably equivalent to an exact test), and seem discordant with the values of the CI (e.g.: 1.62 to 34.17).
Using the cohort approach (TAB DISEASE EXPOSURE /RR /t /ex), I found similar lack of documentation about statistical tests.
I would suggest the following to the conceptors of EpiData: - provide some documentation about statistical tests used in Analysis (do any of the users have those answers?); - provide notes at the bottom of tables to eliminate any ambiguities about statistical methods; - provide p values in the ajusted table with 4 digits after the decimal point.
This software work pretty well, but need some ajustments. Keep on the good work.
Best regards,
Réjean
Réjean Dion, M.D. Médecin-conseil en santé publique (maladies infectieuses) Institut national de santé publique du Québec (INSPQ) / Laboratoire de santé publique du Québec (LSPQ) 20045, chemin Sainte-Marie, Sainte-Anne-de-Bellevue (Québec) Canada H9X 3R5 Tél.: (514) 457-2070 poste 325 Fax.: (514) 457-6346 Courriel / Lotus Notes: rejean.dion@inspq.qc.ca
Hi Réjean,
The current version uses the statistics recommended in the book
Statistics with Confidence: Confidence intervals and statistical guidelines, 2nd Edition Douglas Altman (Editor), David Machin (Editor), Trevor Bryant(Editor), Stephen Gardner (Editor) ISBN: 978-0-7279-1375-3 The help file references this book only with respect to Confidence Intervals. I agree that, in time, a more complete description of the statistical results is warranted. That's something we might want to work together on, to give Jens a hand--what do you think? This version of Analysis does not provide an option to increase the number of decimal positions shown with statistics, but more precise results are available from results variables such as $exact The command result shows all of the results accessible to the user. Jamie Hockin
On 2011-05-30, Réjean Dion wrote:
I would suggest the following to the conceptors of EpiData:
- provide some documentation about statistical tests used in Analysis (do
any of the users have those answers?);
participants (1)
-
epidata-list＠lists.umanitoba.ca