Tighter Confidence Intervals Research Articles

Time-varying parameter energy demand functions: Benchmarking state-space methods against rolling-regressions

Time-varying parameters and elasticities are an appealing extension to constant parameter energy demand functions. In a recent study Altinay and Yalta (2016) use a modified rolling-regression method to approximate time-varying elasticities of demand for natural gas in Istanbul. In a related literature the state-space econometric framework has been used to directly/formally estimate such time-varying effects in energy studies. Through a Monte Carlo simulation exercise, we compare and contrast these two methods and provide evidence that rolling regressions fail to obtain ‘accurate’ estimates (and hence economic implications) of time-varying coefficients in around 80% of our replications for small samples and 40% of replications in large samples. Conversely state-space models are ‘accurate’ 60% of the time in small samples, and 90% of the time in larger samples. We further argue that rolling regressions can lead to unsatisfactory policy recommendations more often than might be considered acceptable, by generating ‘over-confident’ estimates of the wrong elasticity value (i.e. ‘inaccurate’ coefficient estimates with tight confidence intervals that never include the true coefficient). Various robustness checks confirm the invariance of our conclusions to: missing variables; serially dependent errors; a mixture of stationary and non-stationary variables; and choices regarding window size. Flexible least squares and structural time series models are also considered for completeness.

Effects of class size reduction in upper grades: Evidence from Seoul, Korea

We study the effects of class size reduction (CSR) in upper grades (10–12) on college entrance exam scores, using merged Korean administrative data. This data set provides a rare opportunity to investigate the effects of high school class sizes on student outcomes. Seoul has maintained strong restrictions on educational choices to achieve the ideal of equal educational opportunities. The identification mainly comes from the large-scale CSR policy implemented in 2002. By using the within-school between-cohort variation in class size intended by the CSR policy, we find that the effects of high school class sizes on test scores are small with tight confidence intervals. We confirm this result by using the additional within-school variation in class size arising from the major division, especially the implicit rule of class allocation between majors. Further, the distributional effects are no different from the average effects either.

Validation by Monte Carlo sampling of experimental observation functionals

SummaryWe present and analyze a validation procedure for a given state estimate u⋆ of the true field utrue based on Monte Carlo sampling of experimental observation functionals. Our method provides, given a set of N possibly noisy local experimental observation functionals over the spatial domain Ω, confidence intervals for the L2(Ω) error in state and the error in L2(Ω) outputs. For L2(Ω) outputs, our approach also provides a confidence interval for the output itself, which can be used to improve the initial output estimate based on u⋆. Our approach implicitly takes advantage of variance reduction, through the proximity of u⋆ to utrue, to provide tight confidence intervals even for modest values of N. We present results for a synthetic model problem to illustrate the elements of the methodology and confirm the numerical properties suggested by the theory. Finally, we consider an experimental thermal patch configuration to demonstrate the applicability of our approach to real physical systems.

Inferences in panel data with interactive effects using large covariance matrices

We consider efficient estimation of panel data models with interactive effects, which relies on a high-dimensional inverse covariance matrix estimator. By using a consistent estimator of the error covariance matrix, we can take into account both cross-sectional correlations and heteroskedasticity. In the presence of cross-sectional correlations, the proposed estimator eliminates the cross-sectional correlation bias, and is more efficient than the existing methods. The rate of convergence is also improved. In addition, we find that when the statistical inference involves estimating a high-dimensional inverse covariance matrix, the minimax convergence rate on large covariance estimations is not sufficient for inferences. To address this issue, a new “doubly weighted convergence” result is developed. The proposed method is applied to the US divorce rate data. We find that our more efficient estimator identifies the significant effects of divorce-law reforms on the divorce rate, and provides tighter confidence intervals than existing methods. This provides a confirmation for the empirical findings of Wolfers (2006) under more general unobserved heterogeneity.

Genetically low vitamin D concentrations and myopic refractive error: a Mendelian randomization study.

Myopia prevalence has increased in the past 20 years, with many studies linking the increase to reduced time spent outdoors. A number of recent observational studies have shown an inverse association between vitamin D [25(OH)D] serum levels and myopia. However, in such studies it is difficult to separate the effects of time outdoors and vitamin D levels. In this work we use Mendelian randomization (MR) to assess if genetically determined 25(OH)D levels contribute to the degree of myopia. We performed MR using results from a meta-analysis of refractive error (RE) genome-wide association study (GWAS) that included 37 382 and 8 376 adult participants of European and Asian ancestry, respectively, published by the Consortium for Refractive Error And Myopia (CREAM). We used single nucleotide polymorphisms (SNPs) in the DHCR7, CYP2R1, GC and CYP24A1 genes with known effects on 25(OH)D concentration as instrumental variables (IV). We estimated the effect of 25(OH)D on myopia level using a Wald-type ratio estimator based on the effect estimates from the CREAM GWAS. Using the combined effect attributed to the four SNPs, the estimate for the effect of 25(OH)D on refractive error was -0.02 [95% confidence interval (CI) -0.09, 0.04] dioptres (D) per 10 nmol/l increase in 25(OH)D concentration in Caucasians and 0.01 (95% CI -0.17, 0.19) D per 10 nmol/l increase in Asians. The tight confidence intervals on our estimates suggest the true contribution of vitamin D levels to degree of myopia is very small and indistinguishable from zero. Previous findings from observational studies linking vitamin D levels to myopia were likely attributable to the effects of confounding by time spent outdoors.

Estimation of flux ratios without uptake or release data: Application to serine and methionine metabolism

Model-based metabolic flux analysis (MFA) using isotope-labeled substrates has provided great insight into intracellular metabolic activities across a host of organisms. One challenge with applying MFA in mammalian systems, however, is the need for absolute quantification of nutrient uptake, biomass composition, and byproduct release fluxes. Such measurements are often not feasible in complex culture systems or in vivo. One way to address this issue is to estimate flux ratios, the fractional contribution of a flux to a metabolite pool, which are independent of absolute measurements and yet informative for cellular metabolism. Prior work has focused on “local” estimation of a handful of flux ratios for specific metabolites and reactions. Here, we perform systematic, model-based estimation of all flux ratios in a metabolic network using isotope labeling data, in the absence of uptake/release data. In a series of examples, we investigate what flux ratios can be well estimated with reasonably tight confidence intervals, and contrast this with confidence intervals on normalized fluxes. We find that flux ratios can provide useful information on the metabolic state, and is complementary to normalized fluxes: for certain metabolic reactions, only flux ratios can be well estimated, while for others normalized fluxes can be obtained. Simulation studies of a large human metabolic network model suggest that estimation of flux ratios is technically feasible for complex networks, but additional studies on data from actual isotopomer labeling experiments are needed to validate these results. Finally, we experimentally study serine and methionine metabolism in cancer cells using flux ratios. We find that, in these cells, the methionine cycle is truncated with little remethylation from homocysteine, and polyamine synthesis in the absence of methionine salvage leads to loss of 5-methylthioadenosine, suggesting a new mode of overflow metabolism in cancer cells. This work highlights the potential for flux ratio analysis in the absence of absolute quantification, which we anticipate will be important for both in vitro and in vivo studies of cancer metabolism.

Treatment Effect Estimation with Data-Driven Variable Decomposition

One fundamental problem in causal inference is the treatment effect estimation in observational studies when variables are confounded. Control for confounding effect is generally handled by propensity score. But it treats all observed variables as confounders and ignores the adjustment variables, which have no influence on treatment but are predictive of the outcome. Recently, it has been demonstrated that the adjustment variables are effective in reducing the variance of the estimated treatment effect. However, how to automatically separate the confounders and adjustment variables in observational studies is still an open problem, especially in the scenarios of high dimensional variables, which are common in big data era. In this paper, we propose a Data-Driven Variable Decomposition (D$^2$VD) algorithm, which can 1) automatically separate confounders and adjustment variables with a data driven approach, and 2) simultaneously estimate treatment effect in observational studies with high dimensional variables. Under standard assumptions, we show experimentally that the proposed D$^2$VD algorithm can automatically separate the variables precisely, and estimate treatment effect more accurately and with tighter confidence intervals than the state-of-the-art methods on both synthetic data and real online advertising dataset.

Concentration inequalities in the infinite urn scheme for occupancy counts and the missing mass, with applications

An infinite urn scheme is defined by a probability mass function $(p_j)_{j\geq1}$ over positive integers. A random allocation consists of a sample of $N$ independent drawings according to this probability distribution where $N$ may be deterministic or Poisson-distributed. This paper is concerned with occupancy counts, that is with the number of symbols with $r$ or at least $r$ occurrences in the sample, and with the missing mass that is the total probability of all symbols that do not occur in the sample. Without any further assumption on the sampling distribution, these random quantities are shown to satisfy Bernstein-type concentration inequalities. The variance factors in these concentration inequalities are shown to be tight if the sampling distribution satisfies a regular variation property. This regular variation property reads as follows. Let the number of symbols with probability larger than $x$ be $\vec{\nu}(x)=|\{j:p_j\geq x\}|$. In a regularly varying urn scheme, $\vec{\nu}$ satisfies $\lim_{\tau \rightarrow0}\vec{\nu}(\tau x)/\vec{\nu}(\tau)=x^{-\alpha}$ for $\alpha\in[0,1]$ and the variance of the number of distinct symbols in a sample tends to infinity as the sample size tends to infinity. Among other applications, these concentration inequalities allow us to derive tight confidence intervals for the Good--Turing estimator of the missing mass.

Inter-rater reliability of distal ureteral diameter ratio compared to grade of VUR

Management of vesicoureteral reflux (VUR) remains controversial, and reflux grade constitutes an important prognostic factor. Recent work has suggested that distal ureteral diameter ratio (UDR) is a predictive factor relative to clinical outcome independent of grade. Previous studies have noted significant inter-rater variability with grading of VUR. The present study compared inter-rater reliability of reflux grade and UDR in children with primary VUR. Four pediatric urologists independently reviewed, in a blinded fashion, voiding cystourethrograms. For each renal unit, grade was assigned according to the standardized international scale. The UDR was calculated by dividing the largest ureteral diameter within the false pelvis by the distance between L1-L3 vertebral bodies. Correlation within each rater was determined using Pearson's correlation coefficient. Reliability of VUR grade and UDR were calculated using two-way ANOVA model inter-rater agreement. Four independent raters reliably measured VUR grade (ICC=0.87, 95% CI=0.78-0.93) and UDR (ICC=0.95, 95% CI=0.92-0.97). While UDR and grade were equally reliable measures, UDR had a tighter confidence interval. For each rater, grade and UDR were well correlated (r=0.73-0.84; P<0.0001). For higher-grade reflux, grade was more variable than UDR (Summary Figure). Using empirical thresholds, theincreased variability of grade compared with UDR maylead to significant differences in clinical decision-making among physicians (P=0.022). Known discordance with grading reflux emphasizes the need for a more objective VUR measurement, as clinicians and parents often opt for clinical intervention based on both clinical course and the likelihood of spontaneous resolution. While ICC for UDR and grade were not significantly different, the confidence intervals for grade were wider due to greater variability among grade measurements. This suggests that using UDR measurements may lead to more accurate characterization of VUR and ultimately more consistent clinical decision-making across providers. Ureteral diameter ratio has good inter-rater reliability among pediatric urologists, with less clinically relevant variability than VUR grade. Ureteral diameter ratio is a more objective and reliable measure than grade, and may be more useful in clinical decision-making.

Inflation-targeting and real interest rate parity: A bias correction approach

This paper investigates whether inflation-targeting influences real interest rate parity (RIP) by a bias correction approach under cross-sectional dependence. The recursive mean adjustment (RMA) method proposed by So and Shin (1999) and Shin and So (2001) is employed to correct the downward bias in the panel unit root tests and in the half-life estimates of real interest rate differentials for traded and non-traded goods. The empirical findings differ depending on whether we apply the RMA. More importantly, the empirical results show that as more homogeneous economies become involved in terms of inflation-targeting regime, stronger mean reversion and much a tighter confidence interval are present. Thus, inflation-targeting plays an important role in providing favorable evidence for long-run RIP.

Damage Caps and the Labor Supply of Physicians: Evidence from the Third Reform Wave

Nine states adopted caps on non-economic damages during the third medical malpractice reform wave from 2002–05, joining twenty-two other states with caps on non-economic or total damages. We study the effects of these reforms on physician supply. Across a variety of difference-in-differences (DiD), triple differences, and synthetic control methods, in both state- and county-level regressions, we find, with tight confidence intervals, no evidence that cap adoption leads to an increase in total patient care physicians, or in specialties that face high liability risk (with a possible exception for plastic surgeons), or in rural physicians.

Abstract 1830: Answering the challenge of cancer genomic testing

Abstract Cancer genomes are a hodgepodge of mutations. Single nucleotide somatic variants, complex structural rearrangements, whole and partial gene deletions and amplifications, splice variants and expression changes—these are the potential mutation types in cancer. Tumor heterogeneity is common, tumor samples can be solid or liquid biopsies, and they are often contaminated with adjacent normal tissue. Cancer samples are treated with formalin for archival purposes causing additional genomic variations. All of these factors combine to make genomic analysis in cancer far more complicated than that of normal germline tissue. Recent efforts to define gold standard human genomes lack many features characteristic of cancer genomes. A single cancer sample may contain a few distinct variant types, but typically not enough to confidently assess the accuracy of cancer variant detection algorithms. The Horizon Quantitative Multiplex, an engineered FFPE cell line mix with variants at allele frequencies (AFs) from 1.0 to 41.5%, and the Acrometrix Oncology Hotspot Control with &amp;gt;500 SNVs and small indels in one sample are available standards. In the Horizon cell line, we detected 18/18 SNVs and 4/4 indels. Meanwhile, we found that the artificial nature of the Acrometrix sample made it incompatible with next generation sequencing. While samples of this type have limited utility, there is a clear need for samples containing the full spectrum of mutation types and frequences. Since the commercially available standards are of limited utility, we aimed to create a cancer standard set that consists of many cancer cell lines in combination with some real primary cancer samples. We started with a collection of 28 cancer cell lines including 817 known SNVs, 62 small indels, 21 deletions, 23 amplifications, and 14 gene fusions. We simulated tumor heterogeneity by mixing the cell lines at various ratios, generating variant allele frequencies down to 1% and emulated reduced purity by mixing cell lines with paired normal samples at ratios down to 10%. This gave us our platform accuracy metrics for tumor-only somatic analysis. These experiments required sequencing over 200 samples and mixes. Using augmented exome and cancer gene panel (targeting &amp;gt;1,500 cancer genes), we were able to interrogate substantial quantities of variants. We detected 16136/16146 SNVs at 5% AF, 639/646 indels at 10% AF, 29/30 CNAs at 20% purity, and 14/14 gene fusions using the panel, for example. These high variant counts gave us tighter confidence intervals and accuracy metrics for detection of all major cancer variant types. After validating our variant calling approaches via this set, we applied our algorithms to primary tumor samples both formalin treated and fresh frozen, and also tested them both with and without a paired normal tissue sample. These gave us the ability to understand the fundamental effects of formalin fixation, and the benefits of including the paired normal sample, which will be covered in this presentation. Citation Format: Michael J. Clark. Answering the challenge of cancer genomic testing. [abstract]. In: Proceedings of the 107th Annual Meeting of the American Association for Cancer Research; 2016 Apr 16-20; New Orleans, LA. Philadelphia (PA): AACR; Cancer Res 2016;76(14 Suppl):Abstract nr 1830.

Within-group relatedness is correlated with colony-level social structure and reproductive sharing in a social fish.

In group-living species, the degree of relatedness among group members often governs the extent of reproductive sharing, cooperation and conflict within a group. Kinship among group members can be shaped by the presence and location of neighbouring groups, as these provide dispersal or mating opportunities that can dilute kinship among current group members. Here, we assessed how within-group relatedness varies with the density and position of neighbouring social groups in Neolamprologus pulcher, a colonial and group-living cichlid fish. We used restriction site-associated DNA sequencing (RADseq) methods to generate thousands of polymorphic SNPs. Relative to microsatellite data, RADseq data provided much tighter confidence intervals around our relatedness estimates. These data allowed us to document novel patterns of relatedness in relation to colony-level social structure. First, the density of neighbouring groups was negatively correlated with relatedness between subordinates and dominant females within a group, but no such patterns were observed between subordinates and dominant males. Second, subordinates at the colony edge were less related to dominant males in their group than subordinates in the colony centre, suggesting a shorter breeding tenure for dominant males at the colony edge. Finally, subordinates who were closely related to their same-sex dominant were more likely to reproduce, supporting some restraint models of reproductive skew. Collectively, these results demonstrate that within-group relatedness is influenced by the broader social context, and variation between groups in the degree of relatedness between dominants and subordinates can be explained by both patterns of reproductive sharing and the nature of the social landscape.

Does Medical Malpractice Reform Increase Physician Supply? Evidence from the Third Reform Wave

Nine states adopted caps on non-economic damages during the third medical malpractice reform wave from 2002-2005, joining twenty-two other states with caps on non-economic or total damages. We study the effects of these reforms on physician supply. Across a variety of difference-in-differences (DiD), triple differences, and synthetic control methods, in both state- and county-level regressions, we find, with tight confidence intervals, no evidence that cap adoption leads to an increase in total patient care physicians, increases in specialties that face high liability risk (with a possible exception for plastic surgeons), nor increases in in rural physicians.The online appendix is available from SSRN at http://ssrn.com/abstract=2761901.

MP55-03 INTER-RATER RELIABILITY OF DISTAL URETERAL DIAMETER RATIO COMPARED TO GRADE FOR VESICOURETERAL REFLUX

You have accessJournal of UrologyPediatrics: Urinary Tract Infection and Vesicoureteral Reflux1 Apr 2016MP55-03 INTER-RATER RELIABILITY OF DISTAL URETERAL DIAMETER RATIO COMPARED TO GRADE FOR VESICOURETERAL REFLUX Amanda Swanton, Angela M. Arlen, Siobhan E. Alexander, Kathleen Kieran, Douglas W. Storm, and Christopher S. Cooper Amanda SwantonAmanda Swanton More articles by this author , Angela M. ArlenAngela M. Arlen More articles by this author , Siobhan E. AlexanderSiobhan E. Alexander More articles by this author , Kathleen KieranKathleen Kieran More articles by this author , Douglas W. StormDouglas W. Storm More articles by this author , and Christopher S. CooperChristopher S. Cooper More articles by this author View All Author Informationhttps://doi.org/10.1016/j.juro.2016.02.594AboutPDF ToolsAdd to favoritesDownload CitationsTrack CitationsPermissionsReprints ShareFacebookTwitterLinked InEmail INTRODUCTION AND OBJECTIVES Management of vesicoureteral reflux (VUR) in children remains controversial, and is dependent upon individual risk for further infections, potential for renal scarring, and likelihood of spontaneous VUR resolution. VUR grade constitutes an important prognostic factor; however studies have demonstrated poor inter-rater reliability. Recent work suggests that distal ureteral diameter ratio (UDR) may be a more predictive alternative for assessing clinical outcomes than grade of reflux. We aimed to characterize the inter-rater reliability of VUR grade and UDR in children with VUR. METHODS Voiding cystourethrograms (VCUG) of 20 pediatric patients (31 VUR-affected kidneys) were independently reviewed by four pediatric urologists in a blinded fashion. For each renal unit, grade was assigned according to the standardized international scale. UDR was calculated by dividing the largest ureteral diameter within the false pelvis by the distance between the L1-L3 vertebral bodies. The mean grade and mean UDR was calculated for each affected kidney. Correlation within each rater was determined using a Pearson’s correlation coefficient. Reliability of VUR grade and UDR was calculated using intraclass correlation coefficients (ICC) using a two-way ANOVA model interrater agreement. RESULTS VUR grade (ICC = 0.87, 95% CI = 0.78-0.93) and UDR (ICC = 0.95, 95% CI = 0.92-0.97) were reliably measured by four independent raters. While UDR and grade were equally reliable measures, UDR had a tighter confidence interval. For each rater, grade and UDR were well-correlated (r = 0.73-0.84; p <0.0001). In the upper ranges of measurements, grade was more variable than UDR [Figure]. Using an empirical threshold, the increased variability with grade may lead to significantly more differences in clinical decision-making among physicians (p = 0.022). CONCLUSIONS UDR has good inter-rater reliability among pediatric urologists. There was significantly more clinically relevant variability with grade than with UDR. Our study demonstrates that UDR is a more objective and reliable measure than grade, and may be a useful adjunct in clinical decision making and categorizing VUR. © 2016FiguresReferencesRelatedDetails Volume 195Issue 4SApril 2016Page: e735-e736 Advertisement Copyright & Permissions© 2016MetricsAuthor Information Amanda Swanton More articles by this author Angela M. Arlen More articles by this author Siobhan E. Alexander More articles by this author Kathleen Kieran More articles by this author Douglas W. Storm More articles by this author Christopher S. Cooper More articles by this author Expand All Advertisement Advertisement PDF downloadLoading ...

MP55-04 REVISITING THE ROLE OF GENDER ON EARLY DIAGNOSED PRIMARY VESICOURETERAL REFLUX IN INFANTS WITH PRENATAL HYDRONEPHROSIS

characterize the inter-rater reliability of VUR grade and UDR in children with VUR. METHODS: Voiding cystourethrograms (VCUG) of 20 pediatric patients (31 VUR-affected kidneys) were independently reviewed by four pediatric urologists in a blinded fashion. For each renal unit, grade was assigned according to the standardized international scale. UDR was calculated by dividing the largest ureteral diameter within the false pelvis by the distance between the L1-L3 vertebral bodies. The mean grade and mean UDR was calculated for each affected kidney. Correlation within each rater was determined using a Pearson’s correlation coefficient. Reliability of VUR grade and UDR was calculated using intraclass correlation coefficients (ICC) using a two-way ANOVA model interrater agreement. RESULTS: VUR grade (ICC 1⁄4 0.87, 95% CI 1⁄4 0.78-0.93) and UDR (ICC 1⁄4 0.95, 95% CI 1⁄4 0.92-0.97) were reliably measured by four independent raters. While UDR and grade were equally reliable measures, UDR had a tighter confidence interval. For each rater, grade and UDR were well-correlated (r 1⁄4 0.73-0.84; p <0.0001). In the upper ranges of measurements, grade was more variable than UDR [Figure]. Using an empirical threshold, the increased variability with grade may lead to significantly more differences in clinical decision-making among physicians (p 1⁄4 0.022). CONCLUSIONS: UDR has good inter-rater reliability among pediatric urologists. There was significantly more clinically relevant variability with grade than with UDR. Our study demonstrates that UDR is a more objective and reliable measure than grade, and may be a useful adjunct in clinical decision making and categorizing VUR.

In response: challenges for statistical evaluation of ecotoxicological experiments--an industry perspective.

In response: challenges for statistical evaluation of ecotoxicological experiments--an industry perspective.

On the lognormality of historical magnetic storm intensity statistics: Implications for extreme‐event probabilities

AbstractAn examination is made of the hypothesis that the statistics of magnetic storm maximum intensities are the realization of a lognormal stochastic process. Weighted least squares and maximum likelihood methods are used to fit lognormal functions to −Dst storm time maxima for years 1957–2012; bootstrap analysis is used to established confidence limits on forecasts. Both methods provide fits that are reasonably consistent with the data; both methods also provide fits that are superior to those that can be made with a power‐law function. In general, the maximum likelihood method provides forecasts having tighter confidence intervals than those provided by weighted least squares. From extrapolation of maximum likelihood fits: a magnetic storm with intensity exceeding that of the 1859 Carrington event, −Dst ≥ 850 nT, occurs about 1.13 times per century and a wide 95% confidence interval of [0.42, 2.41] times per century; a 100 year magnetic storm is identified as having a −Dst ≥ 880 nT (greater than Carrington) but a wide 95% confidence interval of [490, 1187] nT.

Robust human body model injury prediction in simulated side impact crashes

This study developed a parametric methodology to robustly predict occupant injuries sustained in real-world crashes using a finite element (FE) human body model (HBM). One hundred and twenty near-side impact motor vehicle crashes were simulated over a range of parameters using a Toyota RAV4 (bullet vehicle), Ford Taurus (struck vehicle) FE models and a validated human body model (HBM) Total HUman Model for Safety (THUMS). Three bullet vehicle crash parameters (speed, location and angle) and two occupant parameters (seat position and age) were varied using a Latin hypercube design of Experiments. Four injury metrics (head injury criterion, half deflection, thoracic trauma index and pelvic force) were used to calculate injury risk. Rib fracture prediction and lung strain metrics were also analysed. As hypothesized, bullet speed had the greatest effect on each injury measure. Injury risk was reduced when bullet location was further from the B-pillar or when the bullet angle was more oblique. Age had strong correlation to rib fractures frequency and lung strain severity. The injuries from a real-world crash were predicted using two different methods by (1) subsampling the injury predictors from the 12 best crush profile matching simulations and (2) using regression models. Both injury prediction methods successfully predicted the case occupant's low risk for pelvic injury, high risk for thoracic injury, rib fractures and high lung strains with tight confidence intervals. This parametric methodology was successfully used to explore crash parameter interactions and to robustly predict real-world injuries.

Robust Inference on Average Treatment Effects with Possibly More Covariates than Observations

This paper concerns robust inference on average treatment effects following model selection. In the selection on observables framework, we show how to construct confidence intervals based on a doubly-robust estimator that are robust to model selection errors and prove that they are valid uniformly over a large class of treatment effect models. The class allows for multivalued treatments with heterogeneous effects (in observables), general heteroskedasticity, and selection amongst (possibly) more covariates than observations. Our estimator attains the semiparametric efficiency bound under appropriate conditions. Precise conditions are given for any model selector to yield these results, and we show how to combine data-driven selection with economic theory. For implementation, we give a specific proposal for selection based on the group lasso and derive new technical results for high-dimensional, sparse multinomial logistic regression. A simulation study shows our estimator performs very well in finite samples over a wide range of models. Revisiting the National Supported Work demonstration data, our method yields accurate estimates and tight confidence intervals.