Not every credible interval is credible: Evaluating robustness in the presence of contamination in Bayesian data analysis

As Bayesian methods become more popular among behavioral scientists, they will inevitably be applied in situations that violate the assumptions underpinning typical models used to guide statistical inference. With this in mind, it is important to know something about how robust Bayesian methods are to the violation of those assumptions. In this paper, we focus on the problem of contaminated data (such as data with outliers or conflicts present), with specific application to the problem of estimating a credible interval for the population mean. We evaluate five Bayesian methods for constructing a credible interval, using toy examples to illustrate the qualitative behavior of different approaches in the presence of contaminants, and an extensive simulation study to quantify the robustness of each method. We find that the “default” normal model used in most Bayesian data analyses is not robust, and that approaches based on the Bayesian bootstrap are only robust in limited circumstances. A simple parametric model based on Tukey’s “contaminated normal model” and a model based on the t-distribution were markedly more robust. However, the contaminated normal model had the added benefit of estimating which data points were discounted as outliers and which were not.     

Cognition at risk: Gestalt/feature-intensive processing, attention deficit, and substance abuse

Attention deficit hyperactive disorder (ADHD) is associated with elevated levels of substance abuse, but the cognitive linkages involved have been little explored. The present study builds upon earlier work showing that more gestalt, as opposed to feature-intensive, styles of processing are conductive to substance abuse, especially in those with ADHD symptoms. The work reported here indicates that a protective factor against substance abuse may lie in relatively efficient heuristic processing, especially synthetic or inferential heuristic processing, but further demonstrates that those with ADHD symptoms tend to have poorer abilities in this regard, even at the subclinical, nondiagnosed level. Results are discussed in terms of the gestalt/featureintensive processing theory of cognition.     

Attention problems and executive functions in children with spina bifida and hydrocephalus.

This study addressed the incidence of Attention-Deficit/Hyperactivity Disorder (ADHD) subtypes in children with spina bifida meningomyelocele and shunted hydrocephalus (SBH) as well as differences in executive functions among these subtypes. Parent rating scales revealed that 31% of the group with SBH could be identified with AD/HD, mostly the Inattentive type (23%). The group with SBH differed from normal controls on cognitive measures of executive functions, but subtype differences were not significant. Multivariate tests showed that children with SBH were rated with greater difficulties on the Behavior Rating Inventory of Executive Function (BRIEF) compared to controls; those with SBH and any subtype of ADHD differed from those with SBH and no ADHD; and those with ADHD (Combined Type) differed significantly from those with ADHD (Predominantly Inattentive Type). Subtype differences on univariate tests in the latter comparison were significant on the BRIEF Inhibit scale, showing more disinhibition in those with SBH and ADHD (Combined Type), but no significant differences were apparent on the BRIEF Sustain, Shift, and Initiate scales. The results show that the incidence of ADHD in children with SBH exceeds the population rate, is represented by problems with inattention rather than with impulsivity and hyperactivity; and that as with non-brain injured individuals, subtype differences in cognitive function remain to more clearly delineated.     

From beliefs to behavior: the mediating role of hostile response selection in predicting aggression

This study evaluated the complementary roles that aggressive normative beliefs and hostile response selections play in predicting adolescents' aggressive behavior. Self‐, peer‐, and teacher‐reported data were collected from 2003 young adolescents from a variety of ethnic backgrounds (Black, Latino, Asian, White) during the fall of their sixth grade year to test the hypothesis that adolescents' hostile response selections mediate the association between their normative beliefs and aggressive reputations among their peers and teachers. Structural equation modeling analyses supported our hypothesis, and the model was found to be consistent across both gender and ethnicity. The results suggest that even though gender and ethnic differences in average levels of aggressive cognitions and behavior exist, the general process linking cognitions and behaviors is the same for all young adolescents. Aggr. Behav. 00:1–20, 2005. © 2005 Wiley‐Liss, Inc.     

Toward a hierarchical model of criminal thinking: evidence from item response theory and confirmatory factor analysis

Item response theory (IRT) methods were applied to items from the 80-item Psychological Inventory of Criminal Thinking Styles (PICTS; G. D. Walters, 1995) to determine how well they measure the latent trait of criminal thinking in a group of 2,872 male medium security prison inmates. Preliminary analyses revealed that the 64 PICTS thinking style items, 32 PICTS proactive criminal thinking items, and 24 PICTS reactive criminal thinking items were sufficiently unidimensional to meet the local independence requirements of IRT. The PICTS was fitted to a 2-parameter logistic-graded response IRT model, the results of which showed that the 8 items measuring denial of harm (Sentimentality) displayed weak discrimination (a < 0.5), whereas most of the proactive and reactive items displayed moderate to good discrimination (a > 1.0). Information function analysis revealed that all 3 components of a hierarchical model of criminal thinking--PICTS total scale, PICTS proactive factor, and PICTS reactive factor--displayed greater precision at higher rather than lower levels of the trait dimension. The study findings indicate that items from the PICTS Sentimentality scale do a poor job of measuring general criminal thinking, whereas items from the other 7 PICTS thinking style scales provide their most precise estimates at the upper end of the trait dimension.     

Differential role of the hippocampus in response-outcome and context-outcome learning: evidence from selective satiation procedures

Instrumental performance in rats with hippocampal lesions is insensitive to the degradation of action-outcome contingencies, but sensitive to the effects of selective devaluation by satiation. One interpretation of this dissociation is that damage to the hippocampus impairs the formation of context-outcome associations upon which the effect of contingency degradation, but not selective satiation, relies. Here, we provide a direct assessment of this interpretation, and showed that conditioned responding to contexts did not show sensitivity to selective satiation (Experiment 1), and confirmed that instrumental performance was sensitive to selective satiation (Experiment 2) following hippocampal cell loss.     

Deletion of the GluA1 AMPA receptor subunit impairs recency-dependent object recognition memory

Deletion of the GluA1 AMPA receptor subunit impairs short-term spatial recognition memory. It has been suggested that short-term recognition depends upon memory caused by the recent presentation of a stimulus that is independent of contextual-retrieval processes. The aim of the present set of experiments was to test whether the role of GluA1 extends to nonspatial recognition memory. Wild-type and GluA1 knockout mice were tested on the standard object recognition task and a context-independent recognition task that required recency-dependent memory. In a first set of experiments it was found that GluA1 deletion failed to impair performance on either of the object recognition or recency-dependent tasks. However, GluA1 knockout mice displayed increased levels of exploration of the objects in both the sample and test phases compared to controls. In contrast, when the time that GluA1 knockout mice spent exploring the objects was yoked to control mice during the sample phase, it was found that GluA1 deletion now impaired performance on both the object recognition and the recency-dependent tasks. GluA1 deletion failed to impair performance on a context-dependent recognition task regardless of whether object exposure in knockout mice was yoked to controls or not. These results demonstrate that GluA1 is necessary for nonspatial as well as spatial recognition memory and plays an important role in recency-dependent memory processes.     

Adding to the Mix: Integrating ELSI into a National Nanoscale Science and Technology Center

This paper describes issues associated with integrating the study of Ethical, Legal and Social Issues (ELSI) into ongoing scientific and technical research and describes an approach adopted by the authors for their own work with the center for nanophase materials sciences (CNMS) at the Oak Ridge national laboratory (ORNL). Four key questions are considered: (a) What is ELSI and how should it identify and address topics of interest for the CNMS? (b) What advantages accrue to incorporating ELSI into the CNMS? (c) How should the integration of ELSI into the CNMS take place? (d) How should one judge the effectiveness of the activity? We conclude that ELSI research is not a monolithic body of knowledge, but should be adapted to the question at hand. Our approach focuses on junctures in the R&;D continuum at which key decisions occur, avoids topics of a purely ethical nature or advocacy, and seeks to gather data in ways that permit testing the validity of generalization. Integrating ELSI into the CNMS allows dealing with topics firmly grounded in science, offers concrete examples of potential downstream applications and provides access to the scientists using the CNMS and their insights and observations. As well, integration provides the opportunity for R&;D managers to benefit from ELSI insights and the potential to modify R&;D agendas. Successful integration is dependent on the particular ELSI question set that drives the project. In this case questions sought to identify key choices, information of value to scientists, institutional attributes, key attributes of the CNMS culture, and alternatives for communicating results. The opportunity to consult with scientists on ELSI implications is offered, but not promoted. Finally, ELSI effectiveness is judged by observing the use to which research products are put within the CNMS, ORNL, and the community of external scholars.