One potential problem as ichin4 points out (http://slashdot.org/comments.pl?sid=191520&cid=15 740518) is that the dependent variable here is percent return to shareholders. One can arguably make the case that absolute return is more important when determining compensation. Whatever one thinks, if the study were worth anything at all it would at least provide an analysis of both (perhaps arguing that one is better) to allow the readers to make some sort of informed judgements.
Of course, the above issue may not have been an issue at all if the authors had used reasonable statistical techniques. You can read the actual study yourself at http://www.dolmatconnell.com/resources/2006DCPTech 100Study.pdf. The upshot is that this "study" isn't much more than a bunch of charts and graphs. The most sophisticated statistical measure it uses is the arithmetic mean. While I don't expect such a study to use non-linear models, fixed effects or other such techniques (though it would be nice), I would at least expect to see a simple regression (better yet a multiple regression -- the number of potential lurking variables here is enormous) and a p-value. Is there really a statistical relationship between shareholder return (however defined) and executive compensation? You can't tell from this study. The only thing this study accomplishes is getting DolmatConnell & Partners some cheap publicity, perhaps duping some (perhaps overcompensated) CEO to hire their services.
