When running comparisons between groups of stocks, it is important to avoid falling into the trap known as 'survivorship bias'. The best description I've read about survivorship bias is in the article "Two Common Errors in Empirical Financial Research" by Burton G. Malkiel (introduction here). Incidentally, Malkiel is also the author of "A Random Walk Down Wallstreet", an oft-referenced book for proponents of EMH.
Basically, survivorship bias occurs because there is a tendency for failed companies not to be included in performance research because these companies no longer exist. For example, if you looked at the S&P 500 today, and looked at the performance of these companies in the past several years, the returns you calculated would overstate the actual returns of the index, because you would only be including companies that were successful enough to remain on the S&P 500 today. Survivorship bias has been one of the principles used to dispute many results that have shown low P/E outperformance.
In our research, we have had to be careful to avoid survivorship bias in our results. As such, we have had to flag companies that have not survived, and attempt to determine what did happen to these companies (bankruptcy? buyout? two events which both result in removal from the index, but with very different implications from a results point of view).