This is from a paper on markers of endotoxemia in horses post abdominal surgery and the impact of two different NSAIDs on endotoxemia development. They dichotomized the marker sCD14 and compared the percentage of subjects in each treatment group. According to their analysis the only “statistically significant” difference is 48 hours post surgery with a relative risk of 3.231 (0.8632–12.94) 0.0437[*]. How can you get a p value less than .05 when the confidence interval encompasses the null? Would the one-sided chi-squared test that they mention be the appropriate test for relative risk?

did they fit a model to obtain the risk estimates? Prof Robert Newcombe’s book would help: Confidence Intervals for Proportions and Related Measures of Effect Size He has been active on this forum before

edit: to answer your Q, if they dont model then likely they used cochran mantel haensel, i assume that is what the chi-square is from

1 Like