An interesting discussion recently emerged (here and here) of a study comparing two agricultural experiments (involving different seeds of cowpeas to farmers in Tanzania) -- one blinded, one unblinded. As Andrew Gelman (Columbia) notes: "Bulte et al. find much different results in the two experiments and
attribute the difference to expectation effects (when people know
they’re receiving an experiment they behave differently); Ozler is
skeptical and attributes the different outcomes to various practical
differences in implementation of the two experiments."
Andrew Gelman (Columbia) passes along (and seeks input on) this intereting post from the long (and triple) jump coach at Boise State University, Jeff Petersmeyer (who also coached horizontal jumpers at the 2012 Olympics). Jeff wonders how to better deploy statistical modeling into the service of predicting jumpers' potential. Evidently, a "Moneyball" perspective is migrating into the world of track & field.
A paper now circulating in SSRN (and forthcoming in Judicature) explores the impact of Twombly and Iqbal on dismissal rates and does so with an interesting methodological twist. A New Look: Dismissal Rates in Federal Civil Cases, by Scott Dodson (Hastings), contributes to the growing empirical literature by coding at the individual claim--rather than the case--level. The abstract follows.
"In the wake of Twombly and Iqbal, a number of studies have been conducted to
determine the decisions' effects on dismissal practice in federal civil cases.
However, those studies have tended to code whole cases rather than claims --
leading to the ambiguous coding category of “mixed” dismissals and to problems
in characterizing the nature of the dispute -- and have failed to distinguish
between legal sufficiency and factual sufficiency, potentially masking important
detail about the effects of the pleadings changes.
This paper begins to fill in that detail. I compiled an original dataset of
district court opinions and coded each claim -- rather than whole case --
subject to an adjudicated Rule 12(b)(6) motion. For each claim, I also
determined whether the court resolved the motion on grounds of legal or factual
sufficiency. This methodology opened an unprecedented level of granularity in
The data reveal statistically significant increases in the dismissal rate
overall and in a number of subsets of claims. I also find an increase in the
relative prevalence and efficacy of factual-insufficiency arguments for
dismissal. Perhaps surprisingly, I find a decrease in the relative prevalence
and efficacy of legal-insufficiency arguments for dismissal. These data and
insights on the rationales of dismissals are new to the literature and suggest
that Twombly and Iqbal are affecting both movant strategy and judicial
Cornell colleagues, Ted Eisenberg and Marty Wells, empirically analyze leading ranking metrics for refereed law journals in their recent paper, Ranking Law Journals and the Limits of Journal Citation Reports. Their analysis of ranking outcomes emphasizes a pre-occupation with ordinal ranking and database bias. The abstract follows:
schools, scholars, and journals emphasize ordinal rank. Journal rankings
published by Journal Citation Reports (JCR) are widely used to assess research
quality, which influences important decisions by academic departments,
universities, and countries. We study refereed law journal rankings by JCR,
Washington and Lee Law Library (W&L), and the Australian Research Council
(ARC). Both JCR’s and W&L’s multiple measures of journals can be represented
by a single latent factor. Yet JCR’s rankings are uncorrelated with W&L’s.
The differences appear to be attributable to underrepresentation of law journals
in JCR’s database. We illustrate the effects of database bias on rankings
through case studies of three elite journals, the Journal of Law &
Economics, Supreme Court Review, and the American Law & Economics Review.
Cluster analysis is a supplement to ordinal ranking and we report the results of
a cluster analysis of law journals. The ARC does organize journals into four
large groups and provides generally reasonable rankings of journals. But
anomalies exist that could be avoided by checking the ARC groups against
citation-based measures. Entities that rank should use their data to provide
meaningful clusters rather than providing only ordinal ranks."
Dan Ho (Stanford) has put together an outstanding Program for this year's conference. Event's begin later this morning with Ted Eisenberg's (Cornell) 10-hour empirical training workshop. Insofar as the Workshop proved quite popular (registration closed quickly due to participant demand), workshops will be planned for future CELS. In addition, Ted is presenting a similar workshop at AALS in January.
A nice pair of posts from Andrew Gelman's (Columbia--Poli Sci, Statistics) blog (here) illustrates and briefly discusses the virtues of visually conveying uncertainty in regression results. The initial post illustrates two competing visual approaches. A follow-up post includes (or links to) helpful coding.
The Society for Empirical Legal Studies Executive Director, Dawn Chutkow, passed along the following that might interest those planning to attend the CELS at Stanford in November.
Ted Eisenberg (Cornell) will
conduct an Empirical Training Workshop on November 8 - 9, in connection with CELS
2012 at Stanford Law School. Enrollment is limited. The workshop begins the day
before the conference. A brief description and a link for more
Empirical Training Workshop is intended for professors and students who seek an
introduction to the statistical and programming skills needed to conduct
quantitative empirical legal research. Professor Eisenberg will guide
participants through an intensive 10-hour course on statistical analysis in the
legal context. Pre-registration and a small fee are required.
In 2005, the National Research Council (NRC) evaluated the “More Guns, Less Crime” hypothesis using county-level crime data for the period 1977-2000. 17 of the 18 NRC panel members essentially concluded that the existing research was inconclusive on whether "right-to-carry" laws increased or decreased crime.
"We evaluate the NRC evidence, and improve and expand on the report’s county data analysis by analyzing an additional six years of county data as well as state panel data for the period 1977-2006. We also present evidence using both a more plausible version of the Lott and Mustard specification, as well as our own preferred specification (which, unlike the Lott and Mustard model used in the NRC report, does control for rates of incarceration and police). While we have considerable sympathy with the NRC’s majority view about the difficulty of drawing conclusions from simple panel data models, we disagree with the NRC report’s judgment that cluster adjustments to correct for serial correlation are not needed. Our randomization tests show that without such adjustments the Type 1 error soars to 44-75 percent. In addition, the conclusion of the dissenting panel member that RTC laws reduce murder has no statistical support.
Our paper highlights some important questions to consider when using panel data methods to resolve questions of law and policy effectiveness. Although we agree with the NRC’s cautious conclusion regarding the effects of RTC laws, we buttress this conclusion by showing how sensitive the estimated impact of RTC laws is to different data periods, the use of state versus county data, particular specifications, and the decision to control for state trends. Overall, the most consistent, albeit not uniform, finding to emerge from both the state and county panel data models conducted over the entire 1977-2006 period with and without state trends and using three different specifications is that aggravated assault rises when RTC laws are adopted."
Interesting post (and related discussion) over at PrawfsBlawg involving Martin Pritikin's (Whittier) initial "jump" onto the ELS "bandwagon" and into ELS scholarship. Notable to me is Martin's willingness to undertake the "heroic" task of data creation (rather than secondary analysis of existing data sets). It is also hearting to hear reports from participants about the efficacy of the growing number of empirical legal studies workshops. Finally, the comments' helpfulness evidences how the ELS community stands ready to assist those willing to engage.
The folks over at The Stata Blog recently polled readers (obviously, a non-random selection of Stata users) on their favorite Stata command. While some may find the results (here) themselves interesting, others might find unfamiliar commands that could prove useful.
While not particularly legal per se, results from a test of professional violinists' ability to identify music from a Stradivarius as opposed to other and newer, expensive violins, originally published in The Strad (Feb. 2007), were featured in a recent NPR segment (here). (The NPR segment includes two audio clips for anyone interested and desiring to test their own musical acumen.)
Notably (and certainly within the ELS Blog sweet-spot), the researchers employed a double-blind test. "Researchers gathered professional violinists in a hotel room in Indianapolis. They had six violins — two Strads, a Guarneri and three modern instruments. Everybody wore dark goggles so they couldn't see which violin was which." Ironically, "the only statistically obvious trend in the choices was that one of the Stradivarius violins was the least favorite, and one of the modern instruments was slightly favored."
On the Stata listserv I recently stumbled across a resource (here) for anyone looking for a helpful "how-to" explanation of Stata's margins command and the resulting adjusted predictions and marginal effects output. Credit goes to Richard Williams, a sociologist at Notre Dame for sharing his slides.
Over the years I have repeatedly emphasized to my students in empirical methods classes of the need to "get underneath" the data and results, particularly for secondary analyses. By that I mean researchers invariably benefit from on-the-ground insights into and outside perspectives on what their data (and research design) actually capture and results suggest. Kyle Graham's helpful post over at Concurring Opinion illustrates how this general point can work when evaluating a possible empirical project.
While wanting to avoid igniting something of a statistical holy war, I did want to point to an argument from the good folks at Stata for using poisson over log-linear regression (and negative binomial regression) in some cases. The post (here) quite helpfully includes suggested code as well as Stata output illustrations.