Erowid References Database
|
Lyvers M, Hasking P.
“Have Halpern et al. (2004) detected 'residual neuropsychological effects' of MDMA? Not likely”.
Drug Alcohol Depend. 2004 Aug 16;75(2):149-52; discussion 1.
|
|
Abstract
The preliminary study by Halpern et al. (2004) is a commendable attempt to isolate correlates of 'Ecstasy' use from some of the many confounds that have plagued previous work on this controversial issue. However, Halpern et al. go far beyond their data by concluding that the few significant differences they found, out of a great number of comparisons conducted on a small sample of subjects, actually represent 'residual neuropsychological effects of MDMA.' Indeed, as their findings fail to establish a cause–effect relationship between heavy MDMA use and neurocognitive impairment, their use of the term 'residual effect' is very misleading.
There are several serious problems with the Halpern et al. (2004) report. First, their analyses were statistically unsound.
They obtained scores on 39 separate measures from only 39 subjects (even fewer in the case of the RSAT measures) and then subjected those scores to 117 between-group comparisons. Two basic rules of inferential statistics were thus violated: (1) There were as many measures as there were subjects, a situation likely to lead to artefactual or meaningless results (Tabachnik and Fidell, 2001, p. 117). Of the nine comparisons out of 117 that were termed 'significant' (most of which disappeared after additionally adjusting for family of origin), most compared 11 heavy users to 16 non-users; the comparisons on the two RSAT measures involved only nine heavy users. (2) There was no correction for the inflated probability of making a Type I error because of the large number (117) of between-group comparisons. If an uncorrected alpha of P = 0.05 is used, six comparisons should show up as 'significant' because of chance alone (0.05 × 117). The simplest alpha correction would be 0.05/117 which yields a critical alpha level of P = 0.00004, but this significance level was not reached by any of their so-called 'significant' findings. The authors' failure to adjust for Type I error seems incongruous with their previous studies (e.g., Pope et al., 2001, 2003) in which similar analyses were conducted. In these previous studies the authors set the alpha level at 0.01 'to provide some correction for the multiple comparisons' (Pope et al., 2003, p. 305). In short, the data analysis looks somewhat like a fishing expedition that yielded very little (if anything) that can be justified on statistical grounds. To their credit, Halpern et al. did acknowledge the alpha inflation problem in their paper, but at the same time did nothing to correct it (i.e., by adjusting the alpha level), and instead make general conclusions about the differences between heavy and non-user groups 'on a range of measures'.
|
| # |
| Submit Comment |
|
[
Cite HTML ]
