NHTSA has several studies about the effectiveness of DRL, both pro and con. Here's a pro one:
http://www-nrd.nhtsa.dot.gov/Pubs/809760.pdf
Searches for NHTSA DRL will yield a plethora of statistical analysis on the topic.
Thank you! This is a rather interesting read, at multiple levels.
Before getting to the substance, let's take a moment to savor this glimpse into a delightfully bitter academic squabble. It is rare in a peer-reviewed paper even to mention the reviewers, other than in the Acknowledgements section to thank them for their time and helpful comments. In this paper, right up front in the abstract there's this:
"Reviewers of this paper required the inclusion of results using the odds ratio technique. The estimated the effect of DRLs are –6.3 percent, –7.9 percent, 3.8 percent, and 26 percent, respectively. None of these results were statistically significant."
In other words: we didn't want to publish these numbers, but they made us. By academic standards, that is downright testy. But it gets better in the prologue to Appendix B, where they get into the details of the math:
"This section is included at the request of the reviewers of the paper. The odds ratio is easier to understand for inexperienced analysts than the simple odds and, like the simple odds, attempts to control for a variety of factors other than the presence or absence of DRLs."
Ooooo, SMACK! They REALLY didn't like being told what to do. Having been part of this kind of process myself, I suspect what happened is a negotiation along these lines: Authors submit the paper using method A (simple odds). Reviewers say that's bogus, use method B (odds ratio). Authors say no, we really like A. Reviewers say look, use method B or we'll reject the paper. Authors say OK, how about a compromise: we'll publish both A and B. Reviewers relent, accept the paper. Final draft expresses authors' frustration. Reviewers probably piqued at the way it was presented, but technically the authors did what was asked, so they don't spike the paper.
Now, on to the actual data. Using the simple odds method, the authors conclude that DRLs reduce head-on collisions by 5%. By the usual p=0.05 standard, this result is not statistically significant; they admit this, and note the p=0.07 value. Using the odds ratio method, the authors find that DRLs actually *increase* head-on collisions by 6.3%, but this is not statistically significant either (p=0.23). I think (and will explain why if anyone actually cares, but fair warning, it would not be brief) that the reviewers were right: odds ratio is a better way to exclude exogenous variables in this case, although it's still pretty weak. By this metric, none of the results were statistically significant -- which means, in plain English, that given the available data, we cannot reject with 95% confidence the null hypothesis that DRLs make no difference. That is not surprising: small signals against a noisy background are just hard to measure.
They mentioned a number of other studies, including this rather amusing one: "Lau[17] estimates that DRLs reduce multiple vehicle crashes by 5 to 13 percent. Lau even estimates that DRLs reduce multiple vehicle nighttime crashes by 5 percent, which suggests that there may a confounding lurking variable within the data." Do tell.
