Third molar development is used for dental age estimation when all the other teeth are fully mature. In most medicolegal facilities, dental age estimation is an operator-dependent procedure. During the examination of unaccompanied and undocumented minors, this procedure may lead to binary decisions around age thresholds of legal interest, namely the ages of 14, 16 and 18 years. This study aimed to test the performance of artificial intelligence to classify individuals below and above the legal age thresholds of 14, 16 and 18 years using third molar development. The sample consisted of 11,640 panoramic radiographs (9680 used for training and 1960 used for validation) of males (n = 5400) and females (n = 6240) between 6 and 22.9 years. Computer-based image annotation was performed with V7 software (V7labs, London, UK). The region of interest was the mandibular left third molar (T38) outlined with a semi-automated contour. DenseNet121 was the Convolutional Neural Network (CNN) of choice and was used with Transfer Learning. After Receiver-operating characteristic curves, the area under the curve (AUC) was 0.87 and 0.86 to classify males and females below and above the age of 14, respectively. For the age threshold of 16, the AUC values were 0.88 (males) and 0.83 (females), while for the age of 18, AUC were 0.94 (males) and 0.83 (females). Specificity rates were always between 0.80 and 0.92. Artificial intelligence was able to classify male and females below and above the legal age thresholds of 14, 16 and 18 years with high accuracy.

Factor Analysis is a popular method for modeling dependence in multivariate data. However, determining the number of factors and obtaining a sparse orientation of the loadings are still major challenges. In this paper, we propose a decision-theoretic approach that brings to light the relation between a sparse representation of the loadings and factor dimension. This relation is done through a summary from information contained in the multivariate posterior. To construct such summary, we introduce a three-step approach. In the first step, the model is fitted with a conservative factor dimension. In the second step, a series of sparse point-estimates, with a decreasing number of factors, is obtained by minimizing an expected predictive loss function. In step three, the degradation in utility in relation to the sparse loadings and factor dimensions is displayed in the posterior summary. The findings are illustrated with applications in classical data from the Factor Analysis literature. We used different prior choices and factor dimensions to demonstrate the flexibility of the proposed method.

Abstract Background Single‐session interventions have the potential to address young people's mental health needs at scale, but their effects are heterogeneous. We tested whether the mindset + supportive context hypothesis could help explain when intervention effects persist or fade over time. The hypothesis posits that interventions are more effective in environments that support the intervention message. We tested this hypothesis using the synergistic mindsets intervention, a preventative treatment for stress‐related mental health symptoms that helps students appraise stress as a potential asset in the classroom (e.g., increasing oxygenated blood flow) rather than debilitating. In an introductory college course, we examined whether intervention‐consistent messages from instructors sustained changes in appraisals over time, as well as impacts on students' predisposition to try demanding academic tasks that could enhance learning. Methods We randomly assigned 1675 students in the course to receive the synergistic mindsets intervention (or a control activity) at the beginning of the semester, and subsequently, to receive intervention‐supportive messages from their instructor (or neutral messages) four times throughout the term. We collected weekly measures of students' appraisals of stress in the course and their predisposition to take on academic challenges. Trial‐registration: OSF.io; DOI: 10.17605/osf.io/fchyn. Results A conservative Bayesian analysis indicated that receiving both the intervention and supportive messages led to the greatest increases in positive stress appraisals (0.35 SD; 1.00 posterior probability) and challenge‐seeking predisposition (2.33 percentage points; 0.94 posterior probability), averaged over the course of the semester. In addition, intervention effects grew larger throughout the semester when complemented by supportive instructor messages, whereas without these messages, intervention effects shrank somewhat over time. Conclusions This study shows, for the first time, that supportive cues in local contexts can be the difference in whether a single‐session intervention's effects fade over time or persist and even amplify.

Abstract Background Single‐session interventions have the potential to address young people's mental health needs at scale, but their effects are heterogeneous. We tested whether the mindset + supportive context hypothesis could help explain when intervention effects persist or fade over time. The hypothesis posits that interventions are more effective in environments that support the intervention message. We tested this hypothesis using the synergistic mindsets intervention, a preventative treatment for stress‐related mental health symptoms that helps students appraise stress as a potential asset in the classroom (e.g., increasing oxygenated blood flow) rather than debilitating. In an introductory college course, we examined whether intervention‐consistent messages from instructors sustained changes in appraisals over time, as well as impacts on students' predisposition to try demanding academic tasks that could enhance learning. Methods We randomly assigned 1675 students in the course to receive the synergistic mindsets intervention (or a control activity) at the beginning of the semester, and subsequently, to receive intervention‐supportive messages from their instructor (or neutral messages) four times throughout the term. We collected weekly measures of students' appraisals of stress in the course and their predisposition to take on academic challenges. Trial‐registration: OSF.io; DOI: 10.17605/osf.io/fchyn. Results A conservative Bayesian analysis indicated that receiving both the intervention and supportive messages led to the greatest increases in positive stress appraisals (0.35 SD; 1.00 posterior probability) and challenge‐seeking predisposition (2.33 percentage points; 0.94 posterior probability), averaged over the course of the semester. In addition, intervention effects grew larger throughout the semester when complemented by supportive instructor messages, whereas without these messages, intervention effects shrank somewhat over time. Conclusions This study shows, for the first time, that supportive cues in local contexts can be the difference in whether a single‐session intervention's effects fade over time or persist and even amplify.

We present a Bayesian nonparametric model for conditional distribution estimation using Bayesian additive regression trees (BART). The generative model we use is based on rejection sampling from a base model. Like other BART models, our model is flexible, has a default prior specification, and is computationally convenient. To address the distinguished role of the response in our BART model, we introduce an approach to targeted smoothing of BART models which is of independent interest. We study the proposed model theoretically and provide sufficient conditions for the posterior distribution to concentrate at close to the minimax optimal rate adaptively over smoothness classes in the high-dimensional regime in which many predictors are irrelevant. To fit our model, we propose a data augmentation algorithm which allows for existing BART samplers to be extended with minimal effort. We illustrate the performance of our methodology on simulated data and use it to study the relationship between education and body mass index using data from the medical expenditure panel survey (MEPS). Supplementary materials for this article are available online.

It is widely speculated that auditors' public forecasts of bankruptcy are, at least in part, self-fulfilling prophecies in the sense that they might actually cause bankruptcies that would not have otherwise occurred. This conjecture is hard to prove, however, because the strong association between bankruptcies and bankruptcy forecasts could simply indicate that auditors are skillful forecasters with unique access to highly predictive covariates. In this paper, we investigate the causal effect of bankruptcy forecasts on bankruptcy using nonparametric sensitivity analysis. We contrast our analysis with two alternative approaches: a linear bivariate probit model with an endogenous regressor, and a recently developed bound on risk ratios called E-values. Additionally, our machine learning approach incorporates a monotonicity constraint corresponding to the assumption that bankruptcy forecasts do not make bankruptcies less likely. Finally, a tree-based posterior summary of the treatment effect estimates allows us to explore which observable firm characteristics moderate the inducement effect.

Electronic cigarettes (e-cigarettes) provide a novel source of nicotine and may lead to experimentation by adolescents and eventually to their use of conventional cigarettes. The National Youth Tobacco Survey collected information from a nationally representative sample of students in 2016 to determine their experiences with conventional cigarettes and e-cigarettes, their intentions to use these products in the future, and their perceptions of harm and addiction associated with these products. We analyzed these data with ordered probit regression models to determine possible associations with the intention to try e-cigarettes and conventional cigarettes. This survey included 20,520 respondents, 50.6% were male, 5.9% had used conventional cigarettes, and 9.0% had used e-cigarettes. Approximately 10% thought e-cigarettes caused no harm, 3.7% thought that conventional cigarettes caused no harm, and 52.5% thought that e-cigarettes were less harmful than conventional cigarettes. Less than 1% of the students responded "Definitely yes" to the question, "Do you think you will try an e-cig?" The odds ratios for an intention to try e-cigarettes increased as the perception of harm decreased; these ratios increased from 1.0 for "A lot of harm" to 5.85 (95% CI: 3.51, 9.75) for "No harm." In 2016, the majority of students thought that e-cigarettes could cause some harm. This survey indicates that most students have not tried e-cigarettes or conventional cigarettes. The minority of students who think that e-cigarettes pose no harm and students in the ages 14 and 16 are more likely to try them.

This paper describes the creation of the Longitudinal, Intergenerational Family Electronic Micro-Database (LIFE-M), a new data resource linking vital records and decennial censuses for millions of individuals and families living in the late 19th and 20th centuries in the United States. This combination of records provides a life-course and intergenerational perspective on the evolution of health and economic outcomes. Vital records also enable the linkage of women, because they contain a crosswalk between women's birth (i.e., "maiden") and married names. We describe (1) the data sources, coverage, and linking sequence; (2) the process and supervised machine-learning methods used to link records longitudinally and across generations; and (3) the resulting linked samples, including linking rates, representativeness, and weights.

Meta-analysts often ask a yes-or-no question: Is there an intervention effect or not? This traditional, all-or-nothing thinking stands in contrast with current best practice in meta-analysis, which calls for a heterogeneity-attuned approach (i.e., focused on the extent to which effects vary across procedures, participant groups, or contexts). This heterogeneity-attuned approach allows researchers to understand where effects are weaker or stronger and reveals mechanisms. The current article builds on a rare opportunity to compare two recent meta-analyses that examined the same literature (growth mindset interventions) but used different methods and reached different conclusions. One meta-analysis used a traditional approach (Macnamara and Burgoyne, in press), which aggregated effect sizes for each study before combining them and examined moderators one-by-one by splitting the data into small subgroups. The second meta-analysis (Burnette et al., in press) modeled the variation of effects within studies-across subgroups and outcomes-and applied modern, multi-level meta-regression methods. The former concluded that growth mindset effects are biased, but the latter yielded nuanced conclusions consistent with theoretical predictions. We explain why the practices followed by the latter meta-analysis were more in line with best practices for analyzing large and heterogeneous literatures. Further, an exploratory re-analysis of the data showed that applying the modern, heterogeneity-attuned methods from Burnette et al. (in press) to the dataset employed by Macnamara and Burgoyne (in press) confirmed Burnette et al.'s conclusions; namely, that there was a meaningful, significant effect of growth mindset in focal (at-risk) groups. This article concludes that heterogeneity-attuned meta-analysis is important both for advancing theory and for avoiding the boom-or-bust cycle that plagues too much of psychological science.

Convolutional neural networks (CNN) led to important solutions in the field of Computer Vision. More recently, forensic sciences benefited from the resources of artificial intelligence, especially in procedures that normally require operator-dependent steps. Forensic tools for sexual dimorphism based on morphological dental traits are available but have limited performance. This study aimed to test the application of a machine learning setup to distinguish females and males using dentomaxillofacial features from a radiographic dataset. The sample consisted of panoramic radiographs (n = 4003) of individuals in the age interval of 6 and 22.9 years. Image annotation was performed with V7 software (V7labs, London, UK). From Scratch (FS) and Transfer Learning (TL) CNN architectures were compared, and diagnostic accuracy tests were used. TL (82%) performed better than FS (71%). The correct classifications of females and males aged ≥ 15 years were 87% and 84%, respectively. For females and males < 15 years, the correct classifications were 80% and 83%, respectively. The Area Under the Curve (AUC) from Receiver-operating Characteristic (ROC) curves showed high classification accuracy between 0.87 and 0.91. The radio-diagnostic use of CNN for sexual dimorphism showed positive outcomes and promising forensic applications to the field of dental human identification.

A growth-mindset intervention teaches the belief that intellectual abilities can be developed. Where does the intervention work best? Prior research examined school-level moderators using data from the National Study of Learning Mindsets (NSLM), which delivered a short growth-mindset intervention during the first year of high school. In the present research, we used data from the NSLM to examine moderation by teachers' mindsets and answer a new question: Can students independently implement their growth mindsets in virtually any classroom culture, or must students' growth mindsets be supported by their teacher's own growth mindsets (i.e., the mindset-plus-supportive-context hypothesis)? The present analysis (9,167 student records matched with 223 math teachers) supported the latter hypothesis. This result stood up to potentially confounding teacher factors and to a conservative Bayesian analysis. Thus, sustaining growth-mindset effects may require contextual supports that allow the proffered beliefs to take root and flourish.

Abstract Convolutional neural networks (CNN) led to important solutions in the field of Computer Vision. More recently, forensic sciences benefited from the resources of artificial intelligence, especially in procedures that normally require operator-dependent steps. Forensic tools for sexual dimorphism based on morphological dental traits are available but have limited performance. This study aimed to test the application of a machine learning setup to distinguish females and males using dentomaxillofacial features from a radiographic dataset. The sample consisted of panoramic radiographs (n = 4,003) of individuals in the age interval of 6 and 22.9 years. Image annotation was performed with V7 software (V7labs, London, UK). From Scratch (FS) and Transfer Learning (TL) CNN architectures were compared, and diagnostic accuracy tests were used. TL (82%) performed better than FS (71%). The correct classifications of females and males aged ≥ 15 years were 87% and 84%, respectively. For females and males < 15 years, the correct classifications were 80% and 83%, respectively. The Area Under the Curve (AUC) from Receiver-operating Characteristic (ROC) curves showed high classification accuracy between 0.87 and 0.91. The radio-diagnostic use of CNN for sexual dimorphism showed positive outcomes and promising forensic applications to the field of dental human identification.

We introduce Targeted Smooth Bayesian Causal Forests (tsBCF), a non-parametric Bayesian approach for estimating heterogeneous treatment effects which vary smoothly over a single covariate in the observational data setting. The tsBCF method induces smoothness by parameterizing terminal tree nodes with smooth functions and allows for separate regularization of treatment effects vs. prognostic effect of control covariates. Smoothing parameters for prognostic and treatment effects can be chosen to reflect prior knowledge or tuned in a data-dependent way. We use tsBCF to analyze a new clinical protocol for early medical abortion. Our aim is to assess the relative effectiveness of simultaneous vs. interval administration of mifepristone and misoprostol over the first nine weeks of gestation. Our analysis yields important clinical insights into how to best counsel patients seeking early medical abortion, where understanding even small differences in relative effectiveness can yield dramatic returns to public health. The model reflects our expectation that the treatment effect varies smoothly over gestation but not necessarily over other covariates. We demonstrate the performance of the tsBCF method on benchmarking experiments. Software for tsBCF is available at https://github.com/jestarling/tsbcf/ and in the Supplementary Material (Starling (2020)).

Abstract We introduce Bayesian additive regression trees (BART) for log-linear models including multinomial logistic regression and count regression with zero-inflation and overdispersion. BART has been applied to nonparametric mean regression and binary classification problems in a range of settings. However, existing applications of BART have been mostly limited to models for Gaussian “data,” either observed or latent. This is primarily because efficient MCMC algorithms are available for Gaussian likelihoods. But while many useful models are naturally cast in terms of latent Gaussian variables, many others are not—including models considered in this article. We develop new data augmentation strategies and carefully specified prior distributions for these new models. Like the original BART prior, the new prior distributions are carefully constructed and calibrated to be flexible while guarding against overfitting. Together the new priors and data augmentation schemes allow us to implement an efficient MCMC sampler outside the context of Gaussian models. The utility of these new methods is illustrated with examples and an application to a previously published dataset. Supplementary materials for this article are available online.

Flexibly modeling how an entire density changes with covariates is an important but challenging generalization of mean and quantile regression. While existing methods for density regression primarily consist of covariate-dependent discrete mixture models, we consider a continuous latent variable model in general covariate spaces, which we call DR-BART. The prior mapping the latent variable to the observed data is constructed via a novel application of Bayesian Additive Regression Trees (BART). We prove that the posterior induced by our model concentrates quickly around true generative functions that are sufficiently smooth. We also analyze the performance of DR-BART on a set of challenging simulated examples, where it outperforms various other methods for Bayesian density regression. Lastly, we apply DR-BART to two real datasets from educational testing and economics, to study student growth and predict returns to education. Our proposed sampler is efficient and allows one to take advantage of BART's flexibility in many applied settings where the entire distribution of the response is of primary interest. Furthermore, our scheme for splitting on latent variables within BART facilitates its future application to other classes of models that can be described via latent variables, such as those involving hierarchical or time series data.

To determine “what works, for whom, and under what conditions,” interventions need to be studied in diverse and heterogeneous samples. At an international scope, this degree of heterogeneity is unlikely in a single study and instead requires conducting multiple studies of the same intervention across the globe. In this paper, we provide an overview of the infrastructure required to coordinate such a series of studies, including the methodological and measurement developments that would be needed. We also discuss implications of this shared focus for the design of individual studies and the analyses of outcomes across studies. Throughout, we situate these needs in the context of developing a teacher mindset intervention that is intended to be implemented broadly in classrooms around the world.

Social stress poses a major threat to adolescent health via its effects on internalizing symptoms, such as anxiety and depression. Available interventions to help adolescents improve their stress responses, however, have not been effective in rigorous evaluation studies, or they have been difficult to administer widely. Here we show that replicable improvements in adolescent stress responses can be achieved with a short (~30-minute), scalable synergistic mindsets intervention. This intervention, which is a self-administered online training module, targets both growth mindsets (the idea that people’s intelligence can be developed in response to challenge) and stress-can-be-enhancing mindsets (the idea that people’s stress responses can be fuel for optimal performance). Its goal is to promote positive engagement with stressful events (e.g., learning from failure on a quiz or a conflict with a peer) and to encourage adolescents to use their responses to stressful events and even their bodily symptoms (e.g. racing heart, sweaty palms, butterflies in their stomach) to their advantage. In five double-blind, randomized, controlled trials (total N = 4,091 adolescents), the new synergistic mindsets intervention improved stress-related cognitions (Studies 1-2), cardiovascular reactivity (Study 3), daily internalizing symptoms and cortisol levels (Study 4), and generalized anxiety symptoms during the 2020 COVID-19 lockdowns (Study 5). Effects on downstream outcomes (in Studies 3-5) were stronger among individuals who, at baseline, held the two negative mindsets targeted by the intervention, providing evidence for the proposed mechanisms. Confidence in this conclusion comes from a conservative, Bayesian machine-learning method for detecting heterogeneity.

Nonparametric regression models have recently surged in their power and popularity, accompanying the trend of increasing dataset size and complexity. While these models have proven their predictive ability in empirical settings, they are often difficult to interpret and do not address the underlying inferential goals of the analyst or decision maker. In this article, we propose a modular two-stage approach for creating parsimonious, interpretable summaries of complex models which allow freedom in the choice of modeling technique and the inferential target. In the first stage, a flexible model is fit which is believed to be as accurate as possible. In the second stage, lower-dimensional summaries are constructed by projecting draws from the distribution onto simpler structures. These summaries naturally come with valid Bayesian uncertainty estimates. Further, since we use the data only once to move from prior to posterior, these uncertainty estimates remain valid across multiple summaries and after iteratively refining a summary. We apply our method and demonstrate its strengths across a range of simulated and real datasets. The methods we present here are implemented in an R package available at github.com/spencerwoody/possum. Supplementary materials for this article are available online.

Invited Discussion of "A Unified Framework for De-Duplication and Population Size Estimation", published in Bayesian Analysis. My discussion focuses on two main themes: Providing a more nuanced picture of the costs and benefits of joint models for record linkage and the "downstream task" (i.e. whatever we might want to do with the linked and de-duplicated files), and how we should measure performance.

When a latent shoeprint is discovered at a crime scene, forensic analysts inspect it for distinctive patterns of wear such as scratches and holes (known as accidentals) on the source shoe's sole. If its accidentals correspond to those of a suspect's shoe, the print can be used as forensic evidence to place the suspect at the crime scene. The strength of this evidence depends on the random match probability--- the chance that a shoe chosen at random would match the crime scene print's accidentals. Evaluating random match probabilities requires an accurate model for the spatial distribution of accidentals on shoe soles. A recent report by the President's Council of Advisors in Science and Technology criticized existing models in the literature, calling for new empirically validated techniques. We respond to this request with a new spatial point process model for accidental locations, developed within a hierarchical Bayesian framework. We treat the tread pattern of each shoe as a covariate, allowing us to pool information across large heterogeneous databases of shoes. Existing models ignore this information; our results show that including it leads to significantly better model fit. We demonstrate this by fitting our model to one such database.