Wed, 03/07/2012 - 13:37 — sergio.epi

I'm trying to calculate the PABAK for one matrix, but the results found in R are different from those found in WINPEPI. I also used one MACRO in SPSS for this and the results is the same find in the R and different of the WINPEPI. For example:

(1) (2) (3) Total (1) 236 7 2 245 (2) 12 14 3 29 (3) 3 2 2 7 Total 251 23 7 281

In WINPEPI:

PABAK = 0.85

In the R and SPSS Macro:

PABAK = 0.92

This happens when the variables have three or more categories not ordered, when they have two categories (ordered or not) the results are the same in R, SPSS Macro and the WINPEPI.

- Login to post comments

## Comments

## Response from Winpepi's author

For your data, Winpepi's result was 0.85, whereas R and the SPSS macro's result was 0.92.

Winpepi's PAIRSetc Module B, which deals with unordered categories, reports that the PABAK for your data is 0.85. This coincides with the result calculated by the formula given by Bennett et al. (Public Opinion Quarterly 1954;18:303) for their S coefficient, which Byrt et al, who proposed the use of PABAK, say is the same as PABAK (Journal of Clinical Epidemiology 199;46:423).

Winpepi's PAIRSetc Module C, which is designed for ordered categories, also reported 0.85, ignoring the ordering.

This has now been corrected.Version 3.01 of PAIRSetc now reports a PABAK of 0.88 if linear weights are used, and 0.92 if quadratic weights are used. The latter value coincides with those reported by R and SPSS.I hope this is helpful.

## New version of WinPepi

New versions of the WinPepi (PEPI-for-Windows) programs and manuals have been issued, and can be downloaded free from here. The new release is version 11.20 of WinPepi. The latest versions of the programs are: COMPARE2, 2.69; DESCRIBE, 2.41; ETCETERA, 2.72; LOGISTIC, 1.47; PAIRSetc, 3.05; POISSON, 1.24; and WHATIS, 4.58.

LATEST ADDITIONS :Adjusted coefficient of individual equivalence [CIEA] for assessing agreement between observers or methods (PAIRSetc); synergy factor (ETCETERA).This relates to the topic of this thread..OTHER RECENT ADDITIONS :Diaz-Quijano's procedure for estimating risk ratios or prevalence ratios from a logistic regression analysis (LOGISTIC); measures of the similarity or dissimilarity of two distributions (proportion of similar responses [PSR] and area between curves [ABC]) (COMPARE2, PAIRSETC); calculation of correlation coefficient from paired t-test result (ETCETERA); bootstrap procedure for computation of a mean and its confidence intervals (DESCRIBE); standard normal cumulative distribution function WHATIS).