No abstract

In 1940 Lindquist presented a discussion of an analysis of variance using the group mean as the unit of analysis. Unfortunately, most modern writers on statistical methodology have ignored this important topic and some of the best text-book book writers make the error of using the individual as the unit of analysis when the group mean would be more appropriate. The reluctance to use group mean as the unit of analysis is due in part to the belief that the resultant drop in the number of observations per treatment greatly reduces the probability of detecting a treatment effect. This concern is put in perspective in this paper and equations are presented to facilitate power estimates when the group mean is the unit of analysis.

Although findings from this research suggest that telephone-delivered, coping improvement group interventions have potential to facilitate the adjustment efforts of HIV-infected older adults, more rigorous evaluations of this intervention modality for this group are needed.

A Monte Carlo study was run to check the stability of canonical correlations, canonical weights, and canonical variate-variable correlations. Eight data matrices were selected from the literature for the canonical analyses, with the number of variables ranging from 7 to 41. The results showed that the canonical correlations are very stable upon replication. The results also indicated that there is no solid evidence for concluding that the components are superior to the coefficients, a t least not in terms of being more reliable. However, the number of subjects per variable necessary to achieve re1i:tbility in detecting the most important variables, using components or coefficients, was quite large, ranging from 42/1 to 68/1.

A recent survey of simulation studies concluded that an overwhelming majority of papers do not report a rationale for the decision regarding the number of Monte Carlo iterations. A surprisingly large number of reports do not contain a justifiable definition of robustness and many studies are conducted with an insullicient number of iterations to achieve satisfactory statistical conclusion validity. The implication is that we do not follow our own advice regarding the management of Type I and Type 11 errors when conducting Monte Carlo experiments. This paper reports a straightforward application of a well-known procedure for the purpose of objectively determining the exact number of iterations necessary to confidently detect departures from robustness in Monte Carlo results. A table of the number of iterations necessary to detect departures from a series of nominal Type I error rates is included.