Posted by Larry Hoover on May 11, 2009, at 20:56:40
In reply to Re: Antidepressants Hardly Help ( Time Magazine), posted by chumbawumba on May 11, 2009, at 19:10:06
> The phrase "I have profound contempt for this latest work." just makes me think you're pro-medication? Just a wild guess. Nothing wrong with that, it just suggests a bias.
Well, here's another in a lengthy thread on the subject.
https://www.dr-bob.org/babble/20080420/msgs/824891.htmlI have contempt for "scientists" who resort to deception in furtherance of an argument that has no factual basis. His paper was political, not scientific.
For example, the HAM-D is not an interval scale, it is ordinal. It is inappropriate to submit HAM-D scores to the statistical methods employed by Kirsch.
From another post I made, I quoted from other scientists' criticisms of the paper:
(Me)Well, let's see what his peers have said, shall we? From the reviews appended to the original article, and BMJ:
"In conclusion, the paper of Kirsch and his colleagues presents nothing that was not previously known, but it does introduce empirically unsupported conclusions and erroneous interpretation that are potentially misleading."
(Me) Oh, I said the same things in my critique.
"Among other things, these applications have revealed that the misuse of ordinal scaled data can produce erroneous data and drive inaccurate conclusions. Consequently, concerns must be raised over the accuracy of the results of the meta-regression performed by Kirsch et al, given they have undertaken sophisticated mathematical operations on data which do not support such activities. Moreover, it is worth noting that even the calculation of a mean, a standard deviation, and a change score are invalid on ordinal data, given that these all assume equal interval scaling."
(Me) Translation: The statistical methods applied during the meta-analysis (of the ordinal Hamilton Depression Scale scores) are not meaningful. Ergo, any conclusions therefrom suffer from the same limitation.
"In each case the null hypothesis that the Kirsch et al estimator is unbiased has been tested and overwhelmingly rejected."
(Me)Re-analysis of Kirsch's methods demonstrate that his methodology negatively biased the outcomes.
(Me)And, even if one accepts the premise that these data are analyzable via this methodolgy, a recalculation under more rigorous procedures provides this outcome:
"If the weighted mean difference is used (an equally, or more valid approach given that all studies utilised the same outcome measure, namely the HRSD) effect sizes expressed in HRSD scores are larger than reported in this study (2.8 vs 1.8), and paroxetine and venlafaxine reach the NICE criteria for 'clinical significance' (HRSD change > 3)."
(Me)Aside, I had estimated the effect size plotted on Table 4 at about d=3, so I feel validated that my common-sensical critical-thinking test of Kirsch's stats is supported mathematically.
Regards,
Lar
poster:Larry Hoover
thread:895119
URL: http://www.dr-bob.org/babble/20090505/msgs/895244.html