-
IEEE Transactions on Pattern Analysis... Feb 2023Explainability is crucial for probing graph neural networks (GNNs), answering questions like "Why the GNN model makes a certain prediction?". Feature attribution is a...
Explainability is crucial for probing graph neural networks (GNNs), answering questions like "Why the GNN model makes a certain prediction?". Feature attribution is a prevalent technique of highlighting the explanatory subgraph in the input graph, which plausibly leads the GNN model to make its prediction. Various attribution methods have been proposed to exploit gradient-like or attention scores as the attributions of edges, then select the salient edges with top attribution scores as the explanation. However, most of these works make an untenable assumption - the selected edges are linearly independent - thus leaving the dependencies among edges largely unexplored, especially their coalition effect. We demonstrate unambiguous drawbacks of this assumption - making the explanatory subgraph unfaithful and verbose. To address this challenge, we propose a reinforcement learning agent, Reinforced Causal Explainer (RC-Explainer). It frames the explanation task as a sequential decision process - an explanatory subgraph is successively constructed by adding a salient edge to connect the previously selected subgraph. Technically, its policy network predicts the action of edge addition, and gets a reward that quantifies the action's causal effect on the prediction. Such reward accounts for the dependency of the newly-added edge and the previously-added edges, thus reflecting whether they collaborate together and form a coalition to pursue better explanations. It is trained via policy gradient to optimize the reward stream of edge sequences. As such, RC-Explainer is able to generate faithful and concise explanations, and has a better generalization power to unseen graphs. When explaining different GNNs on three graph classification datasets, RC-Explainer achieves better or comparable performance to state-of-the-art approaches w.r.t. two quantitative metrics: predictive accuracy, contrastivity, and safely passes sanity checks and visual inspections. Codes and datasets are available at https://github.com/xiangwang1223/reinforced_causal_explainer.
PubMed: 35471869
DOI: 10.1109/TPAMI.2022.3170302 -
Journal of Family Psychology : JFP :... Mar 2023Attributional accuracy focuses on the extent to which one person accurately judges the reasons that another person acts the way they do. Research has shown that...
Attributional accuracy focuses on the extent to which one person accurately judges the reasons that another person acts the way they do. Research has shown that relationship quality, individual factors, and the overall context of a discussion, all play a role in the accuracy of attributions within adult relationships. However, little research has examined these patterns for parents and adolescents. Within the parent-adolescent literature, research on informant discrepancies has found agreement between family members' reports can highlight the overall functioning of the parent-child relationship but has not focused on perceptions of motives. This study assessed mothers' and adolescents' attributional accuracy during conflictual discussions and the extent to which such accuracy was associated with their relationship quality and individual perspective-taking abilities. One hundred twenty-three mother ( = 43) and adolescent ( = 14, 54% female, 52% White) dyads participated in a discussion about an issue commonly causing adolescent guilt and rated their own and their partner's motives during the discussion. They also self-reported on their relationship and perspective-taking abilities. Results showed that mothers and teens reported overall more positive discussion motives, when their relationship was better. There was also moderate agreement between mothers' and adolescents' attributed motives and their partners' self-reported motives. Adolescents' accuracy regarding both positive and negative maternal motives improved with better perspective-taking. Mothers' accuracy improved with better relationship quality, but only regarding negative adolescent motives. This study highlights individual factors may be more relevant for adolescent attributional accuracy whereas relational factors may be more relevant for mothers' accuracy. (PsycInfo Database Record (c) 2023 APA, all rights reserved).
Topics: Adult; Humans; Female; Adolescent; Male; Mothers; Mother-Child Relations; Parent-Child Relations; Guilt; Self Report
PubMed: 36480371
DOI: 10.1037/fam0001051 -
Journal of Substance Use and Addiction... Oct 2023Discrimination is associated with poor mental health and substance use among Black Americans, but research is needed on mediators and moderators of these relationships....
INTRODUCTION
Discrimination is associated with poor mental health and substance use among Black Americans, but research is needed on mediators and moderators of these relationships. This study tested whether: 1) discrimination is associated with current alcohol, tobacco (cigarette or e-cigarette), and cannabis use among US Black emerging adults; 2) psychological distress (PD) and positive well-being (PW) are mediators of discrimination-substance use relationships; and 3) these relationships are moderated by sex and attributions to discrimination (racial vs. nonracial).
METHODS
Using data from a 2017 US nationally representative survey, we conducted bivariate and multiple-group moderated mediation analyses among 1118 Black American adults aged 18-28. The study assessed discrimination and attribution to discrimination using the Everyday Discrimination scale, past 30-day PD with the Kessler-6 scale, and past 30-day PW with the Mental Health Continuum Short Form. We utilized probit regression for all structural equation models and adjusted final models for age.
RESULTS
Discrimination was positively associated with past 30-day cannabis and tobacco use directly and indirectly through PD in the overall model. Among males who reported race as the sole/main attribution to discrimination, discrimination was positively associated with alcohol, cannabis, and tobacco use through PD. Among females who reported race as the sole/main attribution to discrimination, discrimination was positively associated with cannabis use through PD. Discrimination was positively associated with tobacco use among those who reported nonracial attributions to discrimination and with alcohol use among those whose attribution was not assessed. Discrimination was positively associated with PD among those who reported race as a secondary attribution to discrimination.
CONCLUSIONS
Discrimination specifically attributed to race may contribute to greater PD and in turn alcohol, cannabis, and tobacco use among Black emerging adults, especially males. Future substance use prevention and treatment efforts targeted to Black American emerging adults may benefit from addressing racial discrimination and PD.
Topics: Adult; Female; Humans; Male; Black People; Electronic Nicotine Delivery Systems; Racism; Substance-Related Disorders; Black or African American; Adolescent; Young Adult; Psychological Distress
PubMed: 37230392
DOI: 10.1016/j.josat.2023.209080 -
Bioethics May 2024In biomedical ethics, there is widespread acceptance of moral realism, the view that moral claims express a proposition and that at least some of these propositions are...
In biomedical ethics, there is widespread acceptance of moral realism, the view that moral claims express a proposition and that at least some of these propositions are true. Biomedical ethics is also in the business of attributing moral obligations, such as "S should do X." The problem, as we argue, is that against the background of moral realism, most of these attributions are erroneous or inaccurate. The typical obligation attribution issued by a biomedical ethicist fails to truly capture the person's actual obligations. We offer a novel argument for rife error in obligation attribution. The argument starts with the idea of an epistemic burden. Epistemic burdens are all of those epistemic obstacles one must surmount in order to achieve some aim. Epistemic burdens shape decision-making such that given two otherwise equal options, a person will choose the option that has the lesser of epistemic burdens. Epistemic burdens determine one's potential obligations and, conversely, their non-obligations. The problem for biomedical ethics is that ethicists have little to no access to others' epistemic burdens. Given this lack of access and the fact that epistemic burdens determine potential obligations, biomedical ethicists often can only attribute accurate obligations out of luck. This suggests that the practice of attributing obligations in biomedical ethics is rife with error. To resolve this widespread error, we argue that this practice should be abolished from the discourse of biomedical ethics.
Topics: Humans; Morals; Dissent and Disputes; Moral Obligations; Ethicists; Bioethics
PubMed: 38367255
DOI: 10.1111/bioe.13275 -
Cognitive Science Oct 2021Young children, like adults, understand that human agents can flexibly choose different actions in different contexts, and they evaluate these agents based on such...
Young children, like adults, understand that human agents can flexibly choose different actions in different contexts, and they evaluate these agents based on such choices. However, little is known about children's tendencies to attribute the capacity to choose to robots, despite increased contact with robotic agents. In this paper, we compare 5- to 7-year-old children's and adults' attributions of free choice to a robot and to a human child by using a series of tasks measuring agency attribution, action prediction, and choice attribution. In morally neutral scenarios, children ascribed similar levels of free choice to the robot and the human, while adults were more likely to ascribe free choice to the human. For morally relevant scenarios, however, both age groups considered the robot's actions to be more constrained than the human's actions. These findings demonstrate that children and adults hold a nuanced understanding of free choice that is sensitive to both the agent type and constraints within a given scenario.
Topics: Adult; Child; Child, Preschool; Humans; Robotics; Social Perception
PubMed: 34606132
DOI: 10.1111/cogs.13043 -
Journal of Epidemiology Aug 2023Identifying which exposures cause disease and quantifying their impacts is essential in promoting and monitoring public health. When multiple exposures are involved,...
BACKGROUND
Identifying which exposures cause disease and quantifying their impacts is essential in promoting and monitoring public health. When multiple exposures are involved, measuring individual contributions becomes challenging.
METHODS
The authors propose a disease attribution method based on aggregate data or summary statistics of individual-level data, possibly from multiple data sources.
RESULTS
Using the proposed method, the burden of disease is apportioned to the independent and interaction effects of each of its major risk factors and all the other factors as a whole. This scheme guarantees that 100% is the total share of the burden.
CONCLUSION
The calculation is simple and straightforward; therefore, it is recommended for use in studies on disease burden.
Topics: Humans; Disease Attributes; Cost of Illness; Public Health; Japan; Causality
PubMed: 35283399
DOI: 10.2188/jea.JE20210084 -
Indian Journal of Psychological Medicine May 2022The COVID-19 pandemic and its associated psychological distress led people to engage in attributing several health-related behaviors and consequences at the community...
BACKGROUND
The COVID-19 pandemic and its associated psychological distress led people to engage in attributing several health-related behaviors and consequences at the community and international levels. A scoping review was conducted to explore the existing literature on the use of attribution theory in understanding the psychological phenomena underlying health-related behavior and consequences during the pandemic.
METHODS
We conducted the literature review using Arksey and O'Malley's methodological framework for scoping review. Studies were identified through a comprehensive search of the following six databases: MEDLINE through PubMed, ProQuest, JSTOR, Scopus, ScienceDirect, and Google Scholar. All databases were searched for entries in English from September 2019 to September 2021 to correspond to the advent of the pandemic.
RESULTS
Several elements influence attributions and the influences of the attributions on people's responses to information and the consequences of attributions in influencing people's responses to information and behavior changes in the context of the COVID-19 pandemic. The importance of attribution errors leading to stigmatization and responsibility framing, both crucial for implementing pandemic control measures and enhancing psychological well-being, were also highlighted.
CONCLUSION
More research is needed in this field to inform people-centered policies and pandemic preparedness plans to mitigate the potentially devastating psychosocial consequence of the pandemic or other public health emergencies.
PubMed: 35656422
DOI: 10.1177/02537176221091675 -
Brain Sciences Oct 2020The aim was to investigate behavioral reactions and event-related potential (ERP) responses in healthy participants under conditions of personalized attribution of...
The aim was to investigate behavioral reactions and event-related potential (ERP) responses in healthy participants under conditions of personalized attribution of emotional appraisal vocabulary to one-self or to other people. One hundred and fifty emotionally neutral, positive and negative words describing people's traits were used. Subjects were asked to attribute each word to four types of people: one-self, loved, unpleasant and neutral person. The reaction time during adjectives attribution to one-self and a loved person was shorter than during adjectives attribution to neutral and unpleasant people. Self-related adjectives induced higher amplitudes of the N400 ERP peak in the medial cortical areas in comparison with adjectives related to other people. The amplitude of P300 and P600 depended on the emotional valence of assessments, but not on the personalized attribution. The interaction between the attribution effect and the effect of emotional valence of assessments was observed for the N400 peak in the left temporal area. The maximal amplitude of N400 was revealed under self-attributing of emotionally positive adjectives. Our results supported the hypothesis that the emotional valence of assessments and the processing of information about one-self or others were related to the brain processes that differ from each other in a cortical localization or time dynamics.
PubMed: 33120879
DOI: 10.3390/brainsci10110782 -
The Gerontologist Jun 2023This study examined the relationship between number of attributed reasons for everyday discrimination and all-cause mortality risk, developed latent classes of...
BACKGROUND AND OBJECTIVES
This study examined the relationship between number of attributed reasons for everyday discrimination and all-cause mortality risk, developed latent classes of discrimination attribution, and assessed whether these latent classes were related to all-cause mortality risk among U.S. older Black women.
RESEARCH DESIGN AND METHOD
Participants were from the 2006 and 2008 waves of the Health and Retirement Study (N = 1,133; 335 deaths). Vital status was collected through the National Death Index through 2013 and key informant reports through 2019. Latent class analyses were conducted on discrimination attributions. Weighted Cox proportional hazards model was used to predict all-cause mortality. Analyses controlled for demographic characteristics, socioeconomic status, and health.
RESULTS
Reporting greater attributions for everyday discrimination was associated with higher mortality risk (hazard ratio [HR] = 1.117; 95% confidence interval [CI]: 1.038-1.202; p < .01), controlling for demographic characteristics, socioeconomic status, and health as well as health behaviors. A 4-class solution of the latent class analysis specified the following attribution classes: No/Low Attribution; Ancestry/Gender/Race/Age; Age/Physical Disability; High on All Attributions. When compared to the No/Low Attribution class, membership in the High on All Attributions class was associated with greater mortality risk (HR = 2.809; CI: 1.458-5.412; p < .01).
DISCUSSION AND IMPLICATIONS
Findings underscore the importance of everyday discrimination experiences from multiple sources in shaping all-cause mortality risk among older Black women. Accordingly, this study problematizes the homogenization of Black women in aging research and suggests the need for health interventions that consider Black women's multiplicity of social statuses.
Topics: Female; Humans; Black or African American; Latent Class Analysis; Social Class; Mortality
PubMed: 35678164
DOI: 10.1093/geront/gnac080 -
International Journal of Behavioral... Jun 2021Illness beliefs are significant contributors to health outcomes. Beliefs about the cause of physical symptoms are considered particularly important among those with...
BACKGROUND
Illness beliefs are significant contributors to health outcomes. Beliefs about the cause of physical symptoms are considered particularly important among those with medically unexplained symptoms and illnesses (MUS); yet little is known about causal beliefs among those with the most severe MUS (i.e., Veterans). The goal of the current study was to examine Veteran's causal attributions of their physical symptoms.
METHOD
A total of 91 combat Veterans with MUS were surveyed using a mixed-methods design about the cause of their physical symptoms, physical symptom severity, and PTSD symptoms. Causal attributions of physical symptoms were analyzed through thematic response analysis and grouped into categories. Chi-square analysis was used to assess the distribution of causal attribution types across Veterans with varying physical symptom severity and PTSD symptom severity.
RESULTS
Veterans with MUS reported an average of 7.9 physical symptoms, and attributed the cause of their symptoms to seven different categories ("Do not Know," "Stress/Mental Health," "Deployment/Environment," "Functional/Symptom," "Medically Explained," "Medically Unexplained Syndrome," and "Lifestyle"). Exploratory chi-square analysis revealed significant differences in causal attributions across physical symptom severity and severity of PTSD symptoms. Veterans with more severe PTSD and Veterans with more severe physical symptoms were more likely to attribute their MUS to stress/mental health or to a medically unexplained syndrome compared with those with low/no PTSD symptoms and physical symptom severity. Veterans with minimal PTSD and Veterans with minimal physical symptom severity were more likely to attribute the cause of their MUS to lifestyle choices (e.g., exercise/diet) compared with those with high PTSD and physical symptom severity.
CONCLUSION
Veterans with MUS endorse multiple, varied causal attributions for their physical symptoms, suggesting more complex causal beliefs than typically assumed. This has important implications for patient-provider communication and development of concordance around MUS treatment.
PubMed: 32691396
DOI: 10.1007/s12529-020-09918-0