Jeff Gill

Penalized regression, standard errors, and Bayesian lassos

Penalized regression methods for simultaneous variable selection and coe-cient estimation, especially those based on the lasso of Tibshirani (1996), have received a great deal of attention in recent years, mostly through frequen- tist models. Properties such as consistency have been studied, and are achieved by difierent lasso variations. Here we look at a fully Bayesian formulation of the problem, which is ∞exible enough to encompass most versions of the lasso that have been previously considered. The advantages of the hierarchical Bayesian for- mulations are many. In addition to the usual ease-of-interpretation of hierarchical models, the Bayesian formulation produces valid standard errors (which can be problematic for the frequentist lasso), and is based on a geometrically ergodic Markov chain. We compare the performance of the Bayesian lassos to their fre- quentist counterparts using simulations, data sets that previous lasso papers have used, and a di-cult modeling problem for predicting the collapse of governments around the world. In terms of prediction mean squared error, the Bayesian lasso performance is similar to and, in some cases, better than, the frequentist lasso.

The Effects of Turnout on Partisan Outcomes in U.S. Presidential Elections 1960–2000

It is commonly believed by pundits and political elites that higher turnout favors Democratic candidates, but the extant research is inconsistent in finding this effect. The purpose of this article is to provide scholars with a methodology for assessing the likely effects of turnout on an election outcome using simulations based on survey data. By varying simulated turnout rates for five U.S. elections from 1960 to 2000, we observe that Democratic advantages from higher turnout (and Republican advantages from lower turnout) have steadily ebbed since 1960, corresponding to the erosion of class cleavages in U.S. elections.

Effect of implementation of a paediatric neurocritical care programme on outcomes after severe traumatic brain injury: a retrospective cohort study

BACKGROUND Outcomes after traumatic brain injury are worsened by secondary insults; modern intensive-care units address such challenges through use of best-practice pathways. Organisation of intensive-care units has an important role in pathway effectiveness. We aimed to assess the effect of a paediatric neurocritical care programme (PNCP) on outcomes for children with severe traumatic brain injury. METHODS We undertook a retrospective cohort study of 123 paediatric patients with severe traumatic brain injury (Glasgow coma scale scores ≤8, without gunshot or abusive head trauma, cardiac arrest, or Glasgow coma scale scores of 3 with fixed and dilated pupils) admitted to the paediatric intensive-care unit of the St Louis Childrens Hospital (St Louis, MO, USA) between July 15, 1999, and Jan 15, 2012. The primary outcome was rate of categorised hospital discharge disposition before and after implementation of a PNCP on Sept 17, 2005. We developed an ordered probit statistical model to assess adjusted outcome as a function of initial injury severity. We assessed care-team behaviour by comparing timing of invasive neuromonitoring and scored intensity of therapies targeting intracranial hypertension. FINDINGS Characteristics of treated patients (aged 3-219 months) were much the same between treatment periods. Before PNCP implementation, 33 (52%) of 63 patients had unfavourable disposition at hospital discharge (death or admission to an inpatient facility) and 30 (48%) had a favourable disposition (home with or without treatment); after PNCP implementation, 20 (33%) of 60 patients had unfavourable disposition and 40 (67%) had favourable disposition (p=0·01). Seven (11%) patients died before PNCP implementation compared with two (3%) deaths after implementation. The probit model indicated that outcome improved across the spectrum of Glasgow coma scale scores after resuscitation (p=0·02); this improvement progressed with increasing injury severity. Kaplan-Meier analysis suggested that neuromonitoring was started earlier and maintained longer after implementation of the PNCP (p=0·03). Therapeutic intensity scores were increased for the first 3 days of treatment after PNCP implementation (p=0·0298 for day 1, p=0·0292 for day 2, and p=0·0471 for day 3). The probit model suggested that increasing age (p=0·03), paediatric risk of mortality III scores (p=0·0003), and injury severity scores (p=0·02) were reliably associated with increased probability of unfavourable outcomes whereas white race (p=0·01), use of intracranial pressure monitoring (p=0·001), and increasing Glasgow coma scale scores after resuscitation (p=0·04) were associated with increased probability of favourable outcomes. INTERPRETATION Outcomes for children with traumatic brain injury can be improved by altering the care system in a way that stably implements a cooperative programme of accepted best practice. FUNDING St Louis Childrens Hospital and the Sean Glanvill Foundations.

ESTIMATION IN DIRICHLET RANDOM EFFECTS MODELS

We develop a new Gibbs sampler for a linear mixed model with a Dirich- let process random effect term, which is easily extended to a generalized linear mixed model with a probit link function. Our Gibbs sampler exploits the properties of the multinomial and Dirichlet distributions, and is shown to be an improvement, in terms of operator norm and efficiency, over other commonly used MCMC algorithms. We also investigate methods for the estimation of the precision parameter of the Dirichlet process, finding that maximum likelihood may not be desirable, but a posterior mode is a reasonable approach. Examples are given to show how these models perform on real data. Our results complement both the theoretical basis of the Dirichlet process nonparametric prior and the computational work that has been done to date.

The 2011–2016 Transdisciplinary Research on Energetics and Cancer (TREC) Initiative: Rationale and Design

PurposeRecognition of the complex, multidimensional relationship between excess adiposity and cancer control outcomes has motivated the scientific community to seek new research models and paradigms.MethodsThe National Cancer Institute developed an innovative concept to establish a center grant mechanism in nutrition, energetics, and physical activity, referred to as the Transdisciplinary Research on Energetics and Cancer (TREC) Initiative. This paper gives an overview of the 2011–2016 TREC Collaborative Network and the 15 research projects being conducted at the centers.ResultsFour academic institutions were awarded TREC center grants in 2011: Harvard University, University of California San Diego, University of Pennsylvania, and Washington University in St. Louis. The Fred Hutchinson Cancer Research Center is the Coordination Center. The TREC research portfolio includes three animal studies, three cohort studies, four randomized clinical trials, one cross-sectional study, and two modeling studies. Disciplines represented by TREC investigators include basic science, endocrinology, epidemiology, biostatistics, behavior, medicine, nutrition, physical activity, genetics, engineering, health economics, and computer science. Approximately 41,000 participants will be involved in these studies, including children, healthy adults, and breast and prostate cancer survivors. Outcomes include biomarkers of cancer risk, changes in weight and physical activity, persistent adverse treatment effects (e.g., lymphedema, urinary and sexual function), and breast and prostate cancer mortality.ConclusionThe NIH Science of Team Science group will evaluate the value added by this collaborative science. However, the most important outcome will be whether this transdisciplinary initiative improves the health of Americans at risk of cancer as well as cancer survivors.

We Have to Be Discrete About This: A Non-Parametric Imputation Technique for Missing Categorical Data

Missing values are a frequent problem in empirical political science research. Surprisingly, there has been little attention to the match between the measurement of the missing values and the correcting algorithms used. While multiple imputation is a vast improvement over the deletion of cases with missing values, it is often ill suited for imputing highly non-granular discrete data. We develop a simple technique for imputing missing values in such situations, which is a variant of hot deck imputation, drawing from the conditional distribution of the variable with missing values to preserve the discrete measure of the variable. This method is tested against existing techniques using Monte Carlo analysis and then applied to real data on democratisation and modernisation theory. We provide software for our imputation technique in a free and easy-to-use package for the R statistical environment.

Whose Variance Is It Anyway? Interpreting Empirical Models with State-Level Data

Researchers commonly apply inferential statistical procedures to population data from the 50 U.S. states as if they were estimating population parameters from sample statistics. This method is incorrect because with population data there is no need to make inferences about quantities that are already known. Instead, authors should simply provide evidence that their specified model provides a good fit to the data. Summary measures of variance as well as the full engine of Bayesian statistics perform this function. This research note demonstrates why the current practice of making inferences from population data with the null hypothesis significance test is wrong, provides some specific examples of problems in the literature, and gives prescriptive advice about correctly assessing and conveying empirical model results.

What to Do When Your Hessian is Not Invertible Alternatives to Model Respecification in Nonlinear Estimation

What should a researcher do when statistical analysis software terminates before completion with a message that the Hessian is not invertible? The standard textbook advice is to respecify the model, but this is another way of saying that the researcher should change the question being asked. Obviously, however, computer programs should not be in the business of deciding what questions are worthy of study. Although noninvertable Hessians are sometimes signals of poorly posed questions, nonsensical models, or inappropriate estimators, they also frequently occur when information about the quantities of interest exists in the data through the likelihood function. The authors explain the problem in some detail and lay out two preliminary proposals for ways of dealing with noninvertable Hessians without changing the question asked.

Nonparametric Priors for Ordinal Bayesian Social Science Models: Specification and Estimation

A generalized linear mixed model, ordered probit, is used to estimate levels of stress in presidential political appointees as a means of understanding their surprisingly short tenures. A Bayesian approach is developed, where the random effects are modeled with a Dirichlet process mixture prior, allowing for useful incorporation of prior information, but retaining some vagueness in the form of the prior. Applications of Bayesian models in the social sciences are typically done with “uninformative” priors, although some use of informed versions exists. There has been disagreement over this, and our approach may be a step in the direction of satisfying both camps. We give a detailed description of the data, show how to implement the model, and describe some interesting conclusions. The model utilizing a nonparametric prior fits better and reveals more information in the data than standard approaches.

Ralph's Pretty-Good Grocery versus Ralph's Super Market: Separating Excellent Agencies from the Good Ones

What makes a public agency perform at a high level? Some agencies are doing extremely well in their environment and it may be because they are lucky enough to have access to plentiful resources, excellent management, and a supportive public. Unfortunately, cases such as these provide little prescriptive evidence for public managers looking to improve their own agencys performance. We apply a new quantitative technique (SWAT) to educational outcome data for 534 school districts in Texas and identify those districts doing extremely well given their fixed and often limited inputs. This approach is useful because the truly superior agencies are those that do more with less, and public managers who lead their organizations to high performance levels despite limited resources provide potential solutions to others.