No description available.

No description available.

<p>Graphs are the natural framework of many of today’s highest impact computing applications: from online social networking, to Web search, to product recommendations, to chemistry, to bioinformatics, to knowledge bases, to mobile ad-hoc networking. To develop successful applications in these domains, we often need representation learning methods ---models mapping nodes, edges, subgraphs or entire graphs to some meaningful vector space. Such models are studied in the machine learning subfield of graph representation learning (GRL). Previous GRL research has focused on learning node or entire graph representations through associational tasks. In this work I study higher-order (k>1-node) representations of graphs in the context of both associational and counterfactual tasks.<br> </p>

No description available.

Most research questions and theory in quantitative peace and conflict research are fundamentally causal. However, a large gap exists in the extant literature between research question and research methodology. Not only does most existing methodology fail to achieve what most quantitative peace scholars attempt, but many researchers do not appear to be aware of these limitations. In this dissertation, I outline five key shortcomings within this literature that, left unaddressed, create results that are not informative of the questions quantitative peace researchers are interested in. This dissertation demonstrates solutions addressing these shortcomings with two applied chapters, conducting causal research designs on a study examining the economic impact of United Nations peacekeeping operations and the effect of human rights treaties on repression, respectively. I find that conventionally-established results in the literature change dramatically when exposed to methodological changes informed by the causal inference literature.

This dissertation investigates novel statistical approaches for causal effect estimation in observational settings, where controlled experimentation is infeasible and confounding is the main hurdle in estimating causal effect. As such, deconfounding constructs the main subject of this dissertation, that is (i) to restore the covariate balance between treatment groups and (ii) to attenuate spurious correlations in training data to derive valid causal conclusions that generalize. By incorporating ideas from representation learning, adversarial matching, generative causal estimation, and invariant risk modeling, this dissertation establishes a causal framework that balances the covariate distribution in latent representation space to yield individualized estimations, and further contributes novel perspectives on causal effect estimation based on invariance principles. The dissertation begins with a systematic review and examination of classical propensity score based balancing schemes for population-level causal effect estimation, presented in Chapter 2. Three causal estimands that target different foci in the population are considered: average treatment effect on the whole population (ATE), average treatment effect on the treated population (ATT), and average treatment effect on the overlap population (ATO). The procedure is demonstrated in a naturalistic driving study (NDS) to evaluate the causal effect of cellphone distraction on crash risk. While highlighting the importance of adopting causal perspectives in analyzing risk factors, discussions on the limitations in balance efficiency, robustness against high-dimensional data and complex interactions, and the need for individualization are provided to motivate subsequent developments. Chapter 3 presents a novel generative Bayesian causal estimation framework named Balancing Variational Neural Inference of Causal Effects (BV-NICE). Via appealing to the Robinson factorization and a latent Bayesian model, a novel variational bound on likelihood is derived, explicitly characterized by the causal effect and propensity score. Notably, by treating observed variables as noisy proxies of unmeasurable latent confounders, the variational posterior approximation is re-purposed as a stochastic feature encoder that fully acknowledges representation uncertainties. To resolve the imbalance in representations, BV-NICE enforces KL-regularization on the respective representation marginals using Fenchel mini-max learning, justified by a new generalization bound on the counterfactual prediction accuracy. The robustness and effectiveness of this framework are demonstrated through an extensive set of tests against competing solutions on semi-synthetic and real-world datasets. In recognition of the reliability issue when extending causal conclusions beyond training distributions, Chapter 4 argues ascertaining causal stability is the key and introduces a novel procedure called Risk Invariant Causal Estimation (RICE). By carefully re-examining the relationship between statistical invariance and causality, RICE cleverly leverages the observed data disparities to enable the identification of stable causal effects. Concretely, the causal inference objective is reformulated under the framework of invariant risk modeling (IRM), where a population-optimality penalty is enforced to filter out un-generalizable effects across heterogeneous populations. Importantly, RICE allows settings where counterfactual reasoning with unobserved confounding or biased sampling designs become feasible. The effectiveness of this new proposal is verified with respect to a variety of study designs on real and synthetic data. In summary, this dissertation presents a flexible causal inference framework that acknowledges the representation uncertainties and data heterogeneities. It enjoys three merits: improved balance to complex covariate interactions, enhanced robustness to unobservable latent confounders, and better generalizability to novel populations. / Doctor of Philosophy / Reasoning cause and effect is the innate ability of a human. While the drive to understand cause and effect is instinct, the rigorous reasoning process is usually trained through the observation of countless trials and failures. In this dissertation, we embark on a journey to explore various principles and novel statistical approaches for causal inference in observational studies. Throughout the dissertation, we focus on the causal effect estimation which answers questions like ``what if" and ``what could have happened". The causal effect of a treatment is measured by comparing the outcomes corresponding to different treatment levels of the same unit, e.g. ``what if the unit is treated instead of not treated?". The challenge lies in the fact that i) a unit only receives one treatment at a time and therefore it is impossible to directly compare outcomes of different treatment levels; ii) comparing the outcomes across different units may involve bias due to confounding as the treatment assignment potentially follows a systematic mechanism. Therefore, deconfounding constructs the main hurdle in estimating causal effects. This dissertation presents two parallel principles of deconfounding: i) balancing, i.e., comparing difference under similar conditions; ii) contrasting, i.e., extracting invariance under heterogeneous conditions. Chapter 2 and Chapter 3 explore causal effect through balancing, with the former systematically reviews a classical propensity score weighting approach in a conventional data setting and the latter presents a novel generative Bayesian framework named Balancing Variational Neural Inference of Causal Effects(BV-NICE) for high-dimensional, complex, and noisy observational data. It incorporates the advance deep learning techniques of representation learning, adversarial learning, and variational inference. The robustness and effectiveness of the proposed framework are demonstrated through an extensive set of experiments. Chapter 4 extracts causal effect through contrasting, emphasizing that ascertaining stability is the key of causality. A novel causal effect estimating procedure called Risk Invariant Causal Estimation(RICE) is proposed that leverages the observed data disparities to enable the identification of stable causal effects. The improved generalizability of RICE is demonstrated through synthetic data with different structures, compared with state-of-art models. In summary, this dissertation presents a flexible causal inference framework that acknowledges the data uncertainties and heterogeneities. By promoting two different aspects of causal principles and integrating advance deep learning techniques, the proposed framework shows improved balance for complex covariate interactions, enhanced robustness for unobservable latent confounders, and better generalizability for novel populations.

Suicide is the tenth leading cause of death among the general U.S. population and the second leading cause of death among those under the age of 45. Veterans are at a particularly high risk, representing 13.7% of the suicides among adult Americans in 2019, despite accounting for only 7% of the total population. The high rates have led the Veterans Health Administration (VHA) and others to develop predictive tools to help identify at-risk patients and facilitate targeting for suicide prevention. Suicide prediction is a notoriously complex challenge. Current theories suggest that suicidal behavior is a result of complex interactions between psychological, clinical, biological, social, and environmental factors. Despite decades of effort in suicide research, predictive abilities for suicide have remained at near-chance levels for the past 50 years. As a result, many organizations, including the VHA, have shifted to advanced statistical methods, such as machine learning, based predictive models. These models have improved suicide prevention efforts by leveraging individual-level electronic health records to detect patterns and identify individuals at highest risk of suicide, thereby enabling the delivery of additional mental health resources to those flagged by the model. Although researchers have made important advances in recent years, we have limited knowledge about how facility-level factors, such as variables related to access and capacity, may affect suicide-related events and aid suicide prediction and prevention. Understanding these factors is important because research demonstrates that clinic operations factors can have direct effects on suicide outcomes and are more easily changed by policymakers and facility managers relative to biological or social factors. The overarching aim of this dissertation was to improve risk prediction and suicide outcomes in the Veteran population by investigating the impact of clinic operations characteristics on suicide outcomes through the use of two methodologies: machine learning and causal inference. Its three specific aims were the following: (1) to investigate the performance of risk-prediction models after adding facility-level predictors of suicide-risk to commonly used machine learning algorithms, (2) and to explore the potential bias in machine learning based suicide risk prediction by stratifying the models by age, sex, race, and ethnicity. The third aim (3) used an instrumental variables approach to explore the causal relationship between virtual care utilization and individual-level suicide related events. Findings from Aim 1 were mixed and showed that adding facility attributes to suicide risk prediction models, specifically logistic regression and elastic net models, could accurately identify a larger number of individuals at greatest risk of suicide, depending on the specification of the model. The analysis from Aim 2 uncovered notable differences in the sensitivity of these models within various subgroups, with enhanced benefits observed for Black, Non-Hispanic, male, and younger populations. Aim 3 results highlighted that a rise in the proportion of virtual mental health visits compared to all visits significantly reduces suicide-related incidents, suggesting that the implementation of virtual mental health services could lower the incidence of suicide outcomes. The findings of this dissertation underscore the value of integrating clinic characteristics into suicide prevention efforts, offering a nuanced approach to improving predictive accuracy and mitigating biases in machine-learning models through the incorporation of facility-level factors. Furthermore, the application of causal inference methods provides critical policy-relevant insights, helping to answer fundamental 'why' questions that underpin suicide-related outcomes. Overall, these findings advocate for a broadening of perspectives from individual-level factors to include facility-level predictors, thereby enhancing the scope and effectiveness of suicide prevention efforts. / 2026-08-28T00:00:00Z

Public bus transit operates within an intricate network of routes and stops, where delays are common and can propagate throughout the transit system, affecting systemreliability, passenger satisfaction, and operational efficiency. Existing research on bus delay propagation has primarily focused on route-level delays correlation-basedanalysis, lacking a comprehensive understanding of underlying causal mechanisms of bus delay propagation from a network-level perspective. To enhance our understanding of bus delay propagation within urban transit systems, this study aims to develop a new approach that captures the causal relationshipsbetween stop delays, integrating their temporal and spatial characteristics. Utilizing a causal discovery algorithm for time series data, the thesis infers causal relationshipsfrom bus operation time series data. It then analyze the resulting causal graphs based on complex network measurement indicators. A case study using GTFS data of Stockholm, Sweden, was conducted. The results reveal that stops with a high degree of connections significantly influence delay propagation, with the network exhibiting a community structure that includes both high-degree and low-degree stops. Stops are classified based on their levels into four distinct delay propagation patterns. Critical stops are identified as either delay aggravation or absorption stops, based on their Momentary Conditional Independence (MCI) values. A new metric was constructed, underscoring the importance of considering delays across the entire network rather than isolating analysis to individual routes. The comparison with traditional correlation-based analysis highlights instances of low correlation among stops with high causality and high correlations without underlying causality, emphasizing the deeper insight from the causal approach

For the last 25 years, research on the effects of community colleges on baccalaureate degree attainment has concluded that community colleges drastically reduce the likelihood of attaining a bachelor's degree compared to the effects of four-year institutions on this likelihood. The thesis of this dissertation is that community colleges have been misjudged as institutions that tend to perpetuate social and economic stratification; what previous studies on the topic have found is based on systematic differences in the student populations. Community college students are consistently more at risk of failing academically than four-year students. Then, the positive impact that four-year colleges have on their students compared to the impact of two-year colleges is to a great extent due to the fact that four-year students tend to have more resources and means to handle college requirements than two-year students. The main challenges to analyze two- and four-year sector effects relies on identifying community college students who resemble four-year college students and then compare their outcomes. This dissertation expands on previous research that has only looked at the effect of community colleges on students' educational outcomes by including labor market outcomes. The analyses conducted in this study primarily relied on propensity score matching (PSM) and the Heckman two-stage estimation procedures to reduce bias in the analysis by accounting for non-random selection into the treatment. In addition, the analytic samples were disaggregated by gender and ethnicity. To estimate the effects of interest, a nationally representative sample that is longitudinal and panel in nature was used: The National Education Longitudinal Study of 1988 (NELS:88).Results revealed that neither the two- nor the four-year sectors were able to help students with very low probabilities of graduation from a four-year college. A new financial aid approach that bridges merit-based and aid-based perspectives is proposed. Community colleges, by welcoming a greater proportion of first-time, full-time undergraduate students, many of whom are underrepresented in higher education, and by helping their students to perform similarly than four-year college students in the outcomes analyzed, are conceptualized as engines for mobility helping surpass economic and social stratification of opportunities in American society.

Estimating causal effects in clinical trials is often complicated by treatment noncompliance and missing outcomes. In time-to-event studies, estimation is further complicated by censoring. Censoring is a type of missing outcome, the mechanism of which may be non-ignorable. While new estimates have recently been proposed to account for noncompliance and missing outcomes, few studies have specifically considered time-to-event outcomes, where even the intention-to-treat (ITT) estimator is potentially biased for estimating causal effects of assigned treatment. In this thesis, we develop a series of parametric potential-outcome (PPO) survival models, for the analysis of randomised controlled trials (RCT) with time-to-event outcomes and noncompliance. Both ignorable and non-ignorable censoring mechanisms are considered. We approach model-fitting from a likelihood-based perspective, using the EM algorithm to locate maximum likelihood estimators. We are not aware of any previous work that addresses these complications jointly. In addition, we give new formulations for the average causal effect (ACE) and the complier average causal effect (CACE) to suit survival analysis. To illustrate the likelihood-based method proposed in this thesis, the HIP breast cancer trial data \citep{Baker98, Shapiro88} were re-analysed using specific PPO-survival models, the Weibull and log-normal based PPO-survival models, which assume that the failure time and censored time distributions both follow Weibull or log-normal distributions. Furthermore, an extended PPO-survival model is also derived in this thesis, which permits investigation into the impact of causal effect after accommodating certain pre-treatment covariates. This is an important contribution to the potential outcomes, survival and RCT literature. For comparison, the Frangakis-Rubin (F-R) model \citep{Frangakis99} is also applied to the HIP breast cancer trial data. To date, the F-R model has not yet been applied to any time-to-event data in the literature.