Search

Prior research has suggested that children living in a disadvantaged neighborhood have lower achievement test scores, but these studies typically have not estimated causal effects that account for neighborhood choice. Recent studies used propensity score methods to account for the endogeneity of neighborhood exposures, comparing disadvantaged and nondisadvantaged neighborhoods. We develop an alternative propensity function approach in which cumulative neighborhood effects are modeled as a continuous treatment variable. This approach offers several advantages. We use our approach to examine the cumulative effects of neighborhood disadvantage on reading and math test scores in Los Angeles. Our substantive results indicate that recency of exposure to disadvantaged neighborhoods may be more important than average exposure for children's test scores. We conclude that studies of child development should consider both average cumulative neighborhood exposure and the timing of this exposure.

The first part of the dissertation focusses on spatial and temporal modeling of point processes on linear networks. Point processes on/near linear networks can simply be defined as point events occurring on or near line segment network structures embedded in a certain space. A separable modeling framework is presented that fits a formation and a dissolution model of point processes on linear networks over time. Two major applications of the separable temporal model are spider web building activity in brick mortar lines and wildfire ignition origins near road networks.The second part of the dissertation focusses on analyses of large energy databases, specifically the Energy Atlas database. The main motivation of this part is to explore and understand the issues of balancing necessary data resolution while maintaining consumer privacy.The issue of data resolution and its importance are explored by first tackling a specific policy objective. This is achieved by applying a longitudinal quantile regression model to parcel-level monthly energy consumption in the Westwood neighborhood; the model results aid in fulfilling efficiency goals outlined in the California Senate Bill 350. Then the issue of record privacy is explored through a review of current privacy methods, implementation, data ownership, and concluded with avenues of future research.

Respondent-driven sampling (RDS) is a method for sampling from a target population by leveraging social connections. RDS is invaluable to the study of hard-to-reach populations. However, RDS is costly and can be infeasible. RDS is infeasible when RDS point estimators have small effective sample sizes (large design effects) or when RDS interval estimators have large confidence intervals relative to estimates obtained in previous studies or poor coverage. As a result, researchers need tools to assess whether or not estimation of certain characteristics of interest for specific populations is feasible in advance. In this paper, we develop a simulation-based framework for using pilot data-in the form of a convenience sample of aggregated, egocentric data and estimates of subpopulation sizes within the target population-to assess whether or not RDS is feasible for estimating characteristics of a target population. in doing so, we assume that more is known about egos than alters in the pilot data, which is often the case with aggregated, egocentric data in practice. We build on existing methods for estimating the structure of social networks from aggregated, egocentric sample data and estimates of subpopulation sizes within the target population. We apply this framework to assess the feasibility of estimating the proportion male, proportion bisexual, proportion depressed and proportion infected with HIV/AIDS within three spatially distinct target populations of older lesbian, gay and bisexual adults using pilot data from the caring and Aging with Pride Study and the Gallup Daily Tracking Survey. We conclude that using an RDS sample of 300 subjects is infeasible for estimating the proportion male, but feasible for estimating the proportion bisexual, proportion depressed and proportion infected with HIV/AIDS in all three target populations.

Permanent supportive housing programs, which provide high-need homeless individuals with long-term housing and supportive services, are thought to be crucial for addressing chronic homelessness. However, many individuals who enroll into permanent supportive housing programs exit within a short period of time, often to unsuitable destinations. This paper utilizes a random survival forest model to predict the outcomes of permanent supportive housing programs in Los Angeles County.The model demonstrates moderate success out-of-sample, with a concordance of 75% between expected risk of exit and observed length of stay. The identification of negative outcomes is similarly successful, with an AUC of 0.7. Organization-level covariates are found to be the most important predictors. Other important factors include age, previous homeless experience, and variables related to client income and benefits. On the other hand, most demographic variables, client health, and client disabilities are found to play relatively small roles in predicting outcomes.

Cognitive social structures (CSS) is an area in social network research that has enduring importance but lacks flexible models. In this paper we consider statistical models for CSS systems where we observe a three-dimensional binary array of relational ties characterized by the "sender" of the relation, the "receiver" of the relation, and the "perceiver" of the relation from the "sender" to the "receiver". Such systems have been represented as networks by Krackhardt. Durante, Dunson and Vogelstein proposed a flexible Bayesian nonparametric approach to model the population distribution of network valued data, in which the joint distribution of the edges probabilities is defined through a mixture model that reduces dimensionality and incorporates information within each mixture component based on latent space models. Inspired by this work, we modify the model to characterize cognitive social structures by adding a parameter for cognitive error. As a case study, we apply our model to Krackhardt's network data of 21 managers in a high-tech machine manufacturing firm. The results show distinct effects of cognitive error and illustrate that our model is capable of characterizing cognitive social structures. The results also motivate future improvement on transitivity and triangle relations.

Current literature debates the desirability of urban revitalization and gentrification for cities. While municipal revitalization of central cities --- and the influx of well-educated and higher income individuals typically associated with it --- can revive moribund and marginalized communities, both processes are associated with heightening inequality and adverse outcomes for the poor and for racial or ethnic minorities. This thesis explores one such negative outcome: the extent to which the gentrification is associated with discriminatory policing practices. Specifically, it tests whether gentrification is associated with more order maintenance policing (OMP) --- a policing strategy that targets low-level, highly discretionary offenses such as disorderly behavior and disturbing the peace, and which historically has targeted the poor and minority groups. Pushing the limitations of the extant literature, this thesis tests whether the relationship between gentrification and OMP ought to be understood as a fundamentally spatial process, and considers a suite of spatial models for the relationship between gentrification and policing as the consequence of endogenous and exogenous spatial interactions, as well as test for omitted spatial variables. Data from both New York City and Los Angeles are collected and analyzed. Results and implications are discussed.

Nowadays online social media becomes the key factor for people to share and exchange information to the community, those real-time social media platforms such as Facebook, Twitter, and Instagram connect people to the others all over the world. With these platforms, people are able to share their thoughts and feelings to thousands of people, or even more. In this thesis, I used data from Twitter to analysis people's opinion about the India government's announcement in 2016 for scraping the bank notes of 500 and 1,000 rupee. And try to figure out people's emotions with their words in their tweets. Thus, the people's thoughts could be collected in a more efficient way.

With a rise in the amount of network data comes increased need for flexible and interpretable network models. Exponential-family random graph models (ERGM) are widely used to analyze small- to medium-sized networks, but suffer from model degeneracy which detracts from their application. In Part I of this dissertation we address this problem by developing novel statistics for ERGM. We focus on the modeling of transitivity in networks as it is a key feature of many real-world networks, but most attempts to account for it within ERGM have induced model degeneracy. The statistics we propose combine the strategies of the transformed statistics proposed by Horvat et al. (2015) and the regularized statistics proposed by Fellows (2012b). They include statistics to capture transitivity, clustering, and a new class of moment statistics to improve goodness of fit. We characterize our newly introduced statistics along with those of Horvat et al. (2015) and Fellows (2012a) using recent theoretical developments regarding ERGM degeneracy. We also compare them theoretically and in practice to the geometrically weighted statistics of Snijders et al. (2006) that are currently the most commonly used to model transitivity in ERGM.In Part II of this dissertation we develop models to rate and rank items based on network data, and demonstrate many advantageous properties of these models. The impetus for this work came from research on ranking statistics journals by Varin et al. (2016). They present a quasi-Stigler model that is a great improvement over the commonly used but statistically indefensible Impact Factor, especially in the quantification of ratings uncertainty. However, the quasi-Stigler model does not fully leverage the network structure of the data and underestimates uncertainty. In addition to applying latent space models to the network rating problem, we identify a fast computational method for fitting the models. We also develop a new latent network model that leverages the

A spatial point process is a random pattern of points on a space $A \subseteq \mathbb{R}^d$. Typically $A$ will be a $d$-dimensional box. Point processes on a plane have been well-studied. However, not much work has been done when it comes to modeling points on $\mathcal{S}^{d-1}\subset \mathbb{R}^d$. There is some work in recent years focusing on extending exploratory tools on $\mathbb{R}^d$ to $\mathcal{S}^{d-1}$, such as the widely used Ripley's $\mathcal{K}$ function. In this dissertation, we propose a more general framework for modeling point processes on $\mathcal{S}^2$. The work is motivated by the need for generative models to understand the mechanisms behind the observed crater distribution on Venus. We start from a background introduction on Venusian craters. Then after an exploratory look at the data, we propose a suite of Exponential Family models, motivated by the Von Mises-Fisher distribution and its generalization. The model framework covers both Poisson-type models and more sophisticated interaction models. It also easily extends to modeling marked point process. For Poisson-type models, we develop likelihood-based inference and an MCMC algorithm to implement it, which is called MCMC-MLE. We compare this method to other procedures including generalized linear model fitting and contrastive divergence. The MCMC-MLE method extends easily to handle inference for interaction models. We also develop a pseudo-likelihood method (MPLE) and demonstrate that MPLE is not as accurate as MCMC-MLE. In addition, we discuss model fit diagnostics and model goodness-of-fit. We also address a few practical issues with the model, including the computational complexity, model degeneracy and sensitivity. Finally, we step away from point process models and explore the widely used presence-only model in Ecology. While this model provides a different angle to approach the problem, it has a few notable defects. The major contributions to spatial point process analysis are, 1)

This dissertation contains two distinct parts. The first part presents a theory for the observation process in surveys of human populations. Focus resides on the effect of survey instrument structure on the observed responses. If the effect of instrument structure on a survey question can be estimate and certain exchangeability conditions hold, sampling from the set of all possible survey instrument structures provides a way to assess the observed sample. This part then introduces probability distributions that aggregate over the of possible instrument structures thereby defining a way to view data without an instrument structure. Methods to incorporate instrument structure are demonstrated in several examples including a survey of a large organization. This dataset requires new techniques to account for dependence between units in the population. The second part introduces a method to estimate latent networks of interacting nodes when a group-level response variable is available. In large populations, the number of possible ties between nodes grows large. The models named CoordiNet and TriadNet place a prior over the dyads included in the model for the group-level response. Doing so induces parsimony in the estimated social structures. This method is applied to data from professional sports.

The study of hard-to-reach populations presents significant challenges. Typically, a sampling frame is not available, and population members are difficult to identify or recruit from broader sampling frames. This is especially true of populations at high risk for HIV/AIDS. Respondent-driven sampling (RDS) is often used in such settings with the primary goal of estimating the prevalence of infection. In such populations, the number of people at risk for infection and the number of people infected are of fundamental importance. This article presents a case-study of the estimation of the size of the hard-to-reach population based on data collected through RDS. We study two populations of female sex workers and men-who-have-sex-with-men in El Salvador. The approach is Bayesian and we consider different forms of prior information, including using the UNAIDS population size guidelines for this region. We show that the method is able to quantify the amount of information on population size available in RDS samples. As separate validation, we compare our results to those estimated by extrapolating from a capture-recapture study of El Salvadorian cities. The results of our case-study are largely comparable to those of the capture-recapture study when they differ from the UNAIDS guidelines. Our method is widely applicable to data from RDS studies and we provide a software package to facilitate this.

Parties are the main vehicles of representation in modern, democratic societies. Party systems, that is the number and the size of all the parties within a country, can vary greatly across countries. There is an ongoing debate in the political science literature about the appropriate way to reduce the dimensionality of the cross-country party system data for comparative purposes. This thesis reviews that literature and offers a new solution: Principal Component Analysis to find the most important information in the data matrix. I use data from 17 advanced democracies from 1970-2013. I conduct analyses using various related methods (Principal Component Analysis, Principal Component Analysis on the Residuals, kernel Principal Component Analysis, Non-Linear Principal Component Analysis, Principal Component Analysis on log-ratio transformed variables and Principal component Analysis on non-centered variables). I find that the most important differences across countries are: “the size of the biggest two parties”, “competition between the two biggest parties”, “existence of a third party” and “balanced multipartism.” I argue that most of the current political science literature uses summary measures that are only correlated with the first of those four dimensions. I suggest a strategy for incorporating a measure of the second dimension that relies on indices of opposition structure.

Respondent-driven sampling (RDS) is commonly used to sample from hard-to-reach populations, such as female sex workers or people who inject drugs, since traditional methods are unable to efficiently survey members due to the typically highly stigmatized nature of the population. Estimating the size of these populations is often desirable so that organizations such as the Centers for Disease Control and Prevention (CDC) can provide the proper amount of aid. However, due to the nature of RDS, traditional methods of population size estimation do not provide good estimates. Therefore, in order to effectively utilize the data we can collect from these populations, we must develop new population size methods designed specifically for RDS data.In this dissertation, I first explore some of the assumption in RDS using a data set of female sex workers in Kampala, Uganda. This is an exploratory look at the recruitment based on geographical location, with a focus on determining whether the RDS is able to actually reach all areas of the map instead of getting stuck in a certain region, as well as looking at the effects of age on recruitment. Then, I introduce a new method of estimating population size that uses concepts from capture-recapture methods while modeling the RDS as a successive sampling process. This extends a current method of population size estimation based on RDS called Successive Sampling for Population Size Estimation (SS-PSE) to include more than one sample, incorporating the information from a capture-recapture design. I develop the Bayesian framework for the model including posterior sampling with a Markov chain Monte Carlo algorithm. Then, using simulation studies, I compare my method with various existing methods in the literature, as well as assessing the frequentist properties of my method.

This paper provides the first rigorous examination of the network determinants of modern bilateral environmental cooperation. Using a new relational data set of all bilateral environmental agreements (BEAs) signed from 2006-16, I characterize the topology of the modern BEA network and generate hypotheses about the data-generating process that gives rise to it. These hypotheses are tested using both exponential-family random graph models (ERGMs) and additive and multiplicative effects (AME) models for relational data. I find that the network formed by modern BEA ties is characterized by influential and highly active “brokers,” that economic development is a major predictor of bilateral environmental collaboration while regime type has little explanatory power, and that countries are highly motivated to pursue environmental cooperation across geographical regions. I argue that network methods can both complement and greatly enrich the work of political scientists who study international environmental cooperation using other methodologies.

UCLA develops and uses a learning management system to provide an online environment for students to access and interact with course content. The data collected by the learning management system is a direct measure of student activity, and provides information that augments known information about the student, such as assignment grades and demographics. This paper assesses UCLA's learning management system data for its usefulness in creating an early warning system that will advise instructors and students of whether a student is likely to receive a failing grade. The data are used in two analyses: an exploratory analysis of how students use the learning management system, and a predictive model to forecast end-of-term grades based on partial-term information. Recommendations on how to generalize the results across the UCLA undergraduate population are drawn from the findings of the analyses.

We develop a class of exponential-family point processes based on a latent social space to model the coevolution of social structure and behavior over time. Temporal dynamics are modeled as a discrete Markov process specified through individual transition distributions for each actor in the system at a given time. We prove that these distributions have an analytic closed form under certain conditions and use the result to develop likelihood-based inference. We provide a computational framework to enable both simulation and inference in practice. Finally, we demonstrate the value of these models by analyzing alcohol and drug use over time in the context of adolescent friendship networks.

This paper provides a review of statistical models in finance for portfolio optimization and portfolio performance evaluation. Based on the assumptions of modern portfolio theory, we discuss five portfolio optimizing models. We then classify portfolio performance evaluation measures into four generalized categories, including the most common performance/risk ratios, the incremental return, the preference-based measures, and the market timing measures. Under each category we review the typical measures with their advantages and drawbacks, and discuss approaches to refine on the drawbacks.In the empirical study section that follows we build five portfolios based on the portfolio optimizing models. Eleven performance evaluation measures are applied to the portfolios, and are compared according to their effectiveness.

Respondent-driven sampling (RDS) is a network sampling methodology used worldwide to sample key populations at high risk for HIV/AIDS who often practice stigmatized/illegal behaviors and are not typically reachable by conventional sampling techniques. In RDS, study participants recruit their peers to enroll, resulting in a sampling mechanism that is unknown to researchers. Current estimators for RDS data require many assumptions about the sampling process, including that recruiters choose people from their network uniformly at random to participate in the study. However, this is likely not true in practice. We believe that people recruit based on observable covariates, such as age, frequency of interaction, geography, socioeconomic status, or social capital.To model preferential recruitment, I develop a sequential two-sided rational-choice framework, referred to as the RCPR model. At each wave of recruitment, each recruiter has a utility for selecting each peer, and symmetrically each peer has a utility for being recruited by each recruiter. Each person also has utilities for selecting themself (not recruiting or not participating). People in the network behave in a way that maximizes their utility given the constraints of the network and the restrictions on recruitment. Although a person's utility is not observed, it can be modeled as a linear combination of observable nodal or dyadic covariates plus unobserved pair-specific heterogeneities. This framework allows generative probabilistic network models to be created for the RDS recruitment process. The models can incorporate observable characteristics of the population and have interpretable parameters. It greatly increases the sophistication of the modeling of the RDS sampling mechanism. Inference can be made about the preference coefficients by maximizing the likelihood of the observed recruitment chain given the observed covariates. As the likelihood is computationally intractable, I develop a Bayesian framework

Respondent-Driven Sampling is a widely-used method for sampling hard-to-reach human populations by link-tracing over their social networks. Inference from such data requires specialized techniques because the sampling process is both partially beyond the control of the researcher, and partially implicitly defined. Therefore, it is not generally possible to directly compute the sampling weights for traditional design-based inference, and likelihood inference requires modeling the complex sampling process. As an alternative, we introduce a model-assisted approach, resulting in a design-based estimator leveraging a working network model. We derive a new class of estimators for population means and a corresponding bootstrap standard error estimator. We demonstrate improved performance compared to existing estimators, including adjustment for an initial convenience sample. We also apply the method and an extension to the estimation of HIV prevalence in a high-risk population.

We propose a log-linear model to assess the consistency of ego reports of dyadic outcomes. We do so specifically in the context where males and females report on shared events, and we demonstrate how inconsistencies can be assessed by using a log-linear model that estimates separate mixing totals for each set of reports. This modelling approach immediately allows us to determine where inconsistencies in reports occur. To demonstrate how our method can be easily implemented for survey data, we apply it to both the 1992 National Health and Social Life Survey and the 2002 National Survey of Family Growth. Our analysis identifies inconsistencies in male and female reports of concurrent partnerships and the number of biological children.

Dependent phenomena, such as relational, spatial and temporal phenomena, tend to be characterized by local dependence in the sense that units which are close in a well-defined sense are dependent. In contrast with spatial and temporal phenomena, though, relational phenomena tend to lack a natural neighbourhood structure in the sense that it is unknown which units are close and thus dependent. Owing to the challenge of characterizing local dependence and constructing random graph models with local dependence, many conventional exponential family random graph models induce strong dependence and are not amenable to statistical inference. We take first steps to characterize local dependence in random graph models, inspired by the notion of finite neighbourhoods in spatial statistics and M-dependence in time series, and we show that local dependence endows random graph models with desirable properties which make them amenable to statistical inference. We show that random graph models with local dependence satisfy a natural domain consistency condition which every model should satisfy, but conventional exponential family random graph models do not satisfy. In addition, we establish a central limit theorem for random graph models with local dependence, which suggests that random graph models with local dependence are amenable to statistical inference. We discuss how random graph models with local dependence can be constructed by exploiting either observed or unobserved neighbourhood structure. In the absence of observed neighbourhood structure, we take a Bayesian view and express the uncertainty about the neighbourhood structure by specifying a prior on a set of suitable neighbourhood structures. We present simulation results and applications to two real world networks with 'ground truth'.

BackgroundRespondent-driven sampling is used worldwide to estimate the population prevalence of characteristics, such as HIV/AIDS and associated risk factors in hard-to-reach populations. Estimating the total size of these populations is of great interest to national and international organizations; however, reliable measures of population size often do not exist.MethodsSuccessive sampling-population size estimation (SS-PSE) along with network size imputation allows population size estimates to be made without relying on separate studies or additional data (as in network scale-up, multiplier, and capture-recapture methods), which may be biased.ResultsTen population size estimates were calculated for people who inject drugs, female sex workers, men who have sex with other men, and migrants from sub-Saharan Africa in six different cities in Morocco. SS-PSE estimates fell within or very close to the likely values provided by experts and the estimates from previous studies using other methods.ConclusionsSS-PSE is an effective method for estimating the size of hard-to-reach populations that leverages important information within respondent-driven sampling studies. The addition of a network size imputation method helps to smooth network sizes allowing for more accurate results. However, caution should be used particularly when there is reason to believe that clustered subgroups may exist within the population of interest or when the sample size is small in relation to the population.

Realistic stochastic models for the co-evolution of social relations and individual behavior over time have broad applicability in social science. The stochastic actor oriented model and temporal exponential-family random graph model have proven useful in modeling longitudinal social networks with nodal covariates while latent space approaches to network analysis offer unique insights into social phenomena. We borrow ideas from these frameworks to construct a spatial temporal exponential-family of point processes (STEPP) to jointly model the co-evolution of social relations and individual behavior in discrete time. We develop likelihood-based inference of STEPP parameters for spatial temporal data as well as latent space inference for longitudinal social networks. We utilize the general STEPP framework to construct a virtual laboratory for simulating social systems and interventions. This virtual laboratory provides a novel simulation tool for developing and assessing potential strategies for influencing behavior in particular communities. We apply these methods to a study of risky behavior in adolescent friendship networks to model and simulate social processes associated with drug and alcohol use in middle school students.

In recent years, there has been a proliferation of preferential trade agreements (PTAs). Through these treaties, countries agree to reduce trade barriers and open their economies to one another. Besides facilitating cooperation between member countries, PTAs also create exclusions that may harm non-members. The scholarship on trade agreements has focused on two questions: (1) when do countries join PTAs? and (2) how do PTAs impact trade cooperation? There is little consensus on answers due, at least in part, to methodological obstacles. The indirect effects of PTAs---on non-member countries---are important parts of the puzzle of whether they increase trade cooperation. Using network analysis, this paper shows that PTAs proliferate especially between small countries and active trade partners. The apparent impact of these treaties on trade is positive, once sufficient time-spans are included. By explicitly modeling indirect effects in a network, this paper reduces measurement bias and generates more accurate estimates.