Response

Extract

Mr. Lee claims that a calculator program in our 1979 book ( 1 ) multiplied the correct confidence limits for an odds ratio by the quantity (R/1.96) (or the variance by the square of this quantity) and thus arrived at an incorrect result. This claim might alarm the thousands of users who have relied on and continue to use our programs. Fortunately, it is false, and there is no cause for alarm. In his checking, Lee should have consulted our book before claiming that the program was in error. The error to which he refers came from a user giving the program the wrong input, the odds ratio point estimate instead of a z-multiplier of 1.96, to obtain a 95% confidence interval. The program allows input of any z-multiplier to obtain confidence intervals at any desired level of confidence.

Mr. Lee evinces an underlying and inappropriate preoccupation with statistical significance ( 2 , 3 ) . If lack of statistical significance leads him to think that exposure to secondhand smoke in the workplace or during childhood has no effect on the risk of lung cancer, he needs to reconsider his reliance on statistical significance as a primary indicator of effect. Statistical significance testing leads to a dichotomous, nonquantitative interpretation that places statistical variation above more important quantitative concerns, such as bias from misclassified exposure and confounding. An appropriate meta-analysis would take into account the relevant biases and come to a quantitative conclusion that is not handicapped by the dichotomous thinking that characterizes statistical significance testing.