Keywords: Contingency Tables
The data comes from a small study in Western Australia of hypertension, alcohol, and obesity. This study was partly designed to mimic a previously reported U.S. study based on a larger sample. A log-linear interaction model is a convenient and effective way of investigating associations among the three variables. A prior-posterior analysis of this 3 x 2 x 4 contingency table using prior information from the previous study (Klatsky et al., 1977) may be appropriate. The previous study reported the general conclusion that alcohol intake and obesity were significantly and independently associated with hypertension (blood pressure). Although a few summary statistics were reported, the full data were not published. One difference between the two studies was in the definition of obesity categories.
The data is listed as follows: the first column (Obesity) contains a numerical value representing the level of obesity (1=low, 2=average, 3=high), the second column (BP) contains a numerical indicator of the presence of hypertension (0=no, 1 =yes). The next five columns are labelled with the levels of alcoholic intake of the subjects, in drinks per day. These columns contain the frequency of observations that have this level of intake, for each group of obesity level and hypertension presence.
Data file (tab-delimited text)
|Knuiman, M.W. and Speed, T.P. (1988) Incorporating Prior Information into the Analysis of Contingency Tables, Biometrics, 44, 1061-1071|