Data analysis as insight: reply to Morrison and

I begin by underscoring a fundamental point of agreement between Morrison and Weaver and me: that data analysis should entail designing whatever set of techniques optimizes insight into whatever question the experiment was designed to address, rather than im-plementing some rote set of pre-ordained rules and regulations. Against this shared philo-sophical backdrop I then (1) reiterate problems with hypothesis testing and (2) address some of the quite pertinent issues that Morrison and Weaver raise with respect to compu-tation of confidence intervals. Morrison and Weaver (1994) end their commentary by noting that "...programs to plot results...are not in themselves an adequate substitute for thought " (manuscript p. 13). Although this assertion sounds like a truism, it is not. Rather it is a legitimate response to one aspect of a trend about which numerous psy-chologists (including me) have become in-creasingly concerned over the years—that of data analysis as no-brainer. I believe, as I have asserted elsewhere, that this trend is largely rooted in an almost exclusive reliance on hy-pothesis-testing procedures which, as they are modally used, amount to reducing a rich and complex data set to an impoverished and unin-formative string of significant/nonsignificant decisions. Throughout Morrison and Weaver's re-marks is woven the theme that such gratuitous information reduction is unacceptable—that data analysis should be carried out and re-ported in such a way as to optimally exploit the information carried by a data set. For ex-ample, they note that if confidence intervals are reported, there is no need to make the standard homogeneity of variance assumption; one can individualize confidence intervals for The writing of this article was supported by an NIM