Reader Comments

Post a new comment on this article

Flawed analysis of a flawed analysis

Posted by droodman on 11 May 2012 at 12:21 GMT

The present paper is right to raise and examine the issue of outliers, which may be generated in part by bad data. But the analysis is not credible. The authors seem to favor an alternative regression model, but never justify it, and in particular never confront the realities that a) the original authors did it too; b) that it is *less* conservative than the original authors' favored approach; and c) is in fact known to be biased. What is reported as sensitivity to removal of outliers is actually in large part "sensitivity" to switching to this alternative, biased estimator. This can be seen easily in the p values Table 4, which thus contradict several statements in text.

Separately, I have blogged about why I think the original study raises serious doubts too, for reasons not addressed here:

No competing interests declared.