Small-Area Analyses Using Public American Community Survey Data: A Tree-Based Spatial Microsimulation Technique

Quantitative sociologists and social policymakers are increasingly interested in local context. Some city-specific studies have developed new primary data collection efforts to analyze inequality at the neighborhood level, but methods from spatial microsimulation have yet to be broadly used in sociology to take better advantage of existing public data sets. The American Community Survey (ACS) is the largest household survey in the United States and indispensable for detailed analysis of specific places and populations. The authors propose a technique, tree-based spatial microsimulation, to produce “small-area” (census-tract) estimates of any person- or household-level phenomenon that can be derived from ACS microdata variables. The approach is straightforward and computationally efficient, based only on publicly available data, and it provides more reliable estimates than do prevailing methods of microsimulation. The authors demonstrate the technique’s capabilities by producing tract-level estimates, stratified by race/ethnicity, of (1) the proportion of people in the census-tract population who have children and work in an essential occupation and (2) the proportion of people in the census-tract population living below the federal poverty threshold and in a household that spends greater than 50 percent of monthly income on rent or owner costs. These examples are relevant to understanding the sociospatial inequalities dramatized by the coronavirus disease 2019 pandemic. The authors discuss potential extensions of the technique to derive small-area estimates of variables observed in surveys other than the ACS.

What Goes Up Might Not Come Down: Modeling Directional Asymmetry with Large-N, Large-T Data

Modeling asymmetric relationships is an emerging subject of interest among sociologists. York and Light advanced a method to estimate asymmetric models with panel data, which was further developed by Allison. However, little attention has been given to the large- N, large- T case, wherein autoregression, slope heterogeneity, and cross-sectional dependence are important issues to consider. The authors fill this gap by conducting Monte Carlo experiments comparing the bias and power of the fixed-effects estimator to a set of heterogeneous panel estimators. The authors find that dynamic misspecification can produce substantial biases in the coefficients. Furthermore, even when the dynamics are correctly specified, the fixed-effects estimator will produce inconsistent and unstable estimates of the long-run effects in the presence of slope heterogeneity. The authors demonstrate these findings by testing for directional asymmetry in the economic development–CO2 emissions relationship, a key question in macro sociology, using data for 66 countries from 1971 to 2015. The authors conclude with a set of methodological recommendations on modeling directional asymmetry.

Validating Sequence Analysis Typologies Using Parametric Bootstrap

In this article, the author proposes a methodology for the validation of sequence analysis typologies on the basis of parametric bootstraps following the framework proposed by Hennig and Lin (2015). The method works by comparing the cluster quality of an observed typology with the quality obtained by clustering similar but nonclustered data. The author proposes several models to test the different structuring aspects of the sequences important in life-course research, namely, sequencing, timing, and duration. This strategy allows identifying the key structural aspects captured by the observed typology. The usefulness of the proposed methodology is illustrated through an analysis of professional and coresidence trajectories in Switzerland. The proposed methodology is available in the WeightedCluster R library.

A General Panel Model for Unobserved Time Heterogeneity with Application to the Politics of Mass Incarceration

Panel data analysis is common in the social sciences. Fixed effects models are a favorite among sociologists because they control for unobserved heterogeneity (unexplained variation) among cross-sectional units, but estimates are biased when there is unobserved heterogeneity in the underlying time trends. Two-way fixed effects models adjust for unobserved time heterogeneity but are inefficient, cannot include unit-invariant variables, and eliminate common trends: the portion of variance in a time-varying variable that is invariant across cross-sectional units. This article introduces a general panel model that can include unit-invariant variables, corrects for unobserved time heterogeneity, and provides the effect of common trends while also allowing for unobserved unit heterogeneity, time-varying coefficients, and time-invariant variables. One-way and two-way fixed effects models are shown to be restrictive forms of this general model. Other restrictive forms are also derived that offer all the usual advantages of one-way and two-way fixed effects models but account for unobserved time heterogeneity. The author uses the models to examine the increase in state incarceration rates between 1970 and 2015.

Can You Really Study an Army on the Internet? Comparing How Status Tasks Perform in the Laboratory and Online Settings

Laboratory experiments have a long history within sociology, with their ability to test causality and their utility for directly observing behavior providing key advantages. One influential social psychological field, status characteristics and expectation states theory, has almost exclusively used laboratory experiments to test the theory. Unfortunately, laboratory experiments are resource intensive, requiring a research pool, laboratory space, and considerable amounts of time. For these and other reasons, social scientists are increasingly exploring the possibility of moving experiments from the lab to an online platform. Despite the advantages of the online setting, the transition from the lab is challenging, especially when studying behavior. In this project, we develop methods to translate the traditional status characteristics experimental setting from the laboratory to online. We conducted parallel laboratory and online behavioral experiments using three tasks from the status literature, comparing each task’s ability to differentiate on the basis of status distinctions. The tasks produce equivalent results in the online and laboratory environment; however, not all tasks are equally sensitive to status differences. Finally, we provide more general guidance on how to move vital aspects of laboratory studies, such as debriefing, suspicion checks, and scope condition checks, to the online setting.

Using Social Networks to Supplement RDD Telephone Surveys to Oversample Hard-to-Reach Populations: A New RDD+RDS Approach

Random digit dialing (RDD) telephone sampling, although experiencing declining response rates, remains one of the most accurate and cost-effective data collection methods for generating national population-based estimates. Such methods, however, are inefficient when sampling hard-to-reach populations because the costs of recruiting sufficient sample sizes to produce reliable estimates tend to be cost prohibitive. The authors implemented a novel respondent-driven sampling (RDS) approach to oversample cigarette smokers and lesbian, gay, bisexual, and transgender (LGBT) people. The new methodology selects RDS referrals or seeds from a probability-based RDD sampling frame and treats the social networks as clusters in the weighting and analysis, thus eliminating the intricate assumptions of RDS. The authors refer to this approach as RDD+RDS. In 2016 and 2017, a telephone survey was conducted on tobacco-related topics with a national sample of 4,208 U.S. adults, as well as 756 referral-based respondents. The RDD+RDS estimates were comparable with stand-alone RDD estimates, suggesting that the addition of RDS responses from social networks improved the precision of the estimates without introducing significant bias. The authors also conducted an experiment to determine whether the number of recruits would vary on the basis of how the RDS recruitment question specified the recruitment population (closeness of relationship, time since last contact, and LGBT vs. tobacco user), and significant differences were found in the number of referrals provided on the basis of question wording. The RDD+RDS sampling approach, as an adaptation of standard RDD methodology, is a practical tool for survey methodologists that provides an efficient strategy for oversampling rare or elusive populations.

Uncovering Sociological Effect Heterogeneity Using Tree-Based Machine Learning

Individuals do not respond uniformly to treatments, such as events or interventions. Sociologists routinely partition samples into subgroups to explore how the effects of treatments vary by selected covariates, such as race and gender, on the basis of theoretical priors. Data-driven discoveries are also routine, yet the analyses by which sociologists typically go about them are often problematic and seldom move us beyond our biases to explore new meaningful subgroups. Emerging machine learning methods based on decision trees allow researchers to explore sources of variation that they may not have previously considered or envisaged. In this article, the authors use tree-based machine learning, that is, causal trees, to recursively partition the sample to uncover sources of effect heterogeneity. Assessing a central topic in social inequality, college effects on wages, the authors compare what is learned from covariate and propensity score–based partitioning approaches with recursive partitioning based on causal trees. Decision trees, although superseded by forests for estimation, can be used to uncover subpopulations responsive to treatments. Using observational data, the authors expand on the existing causal tree literature by applying leaf-specific effect estimation strategies to adjust for observed confounding, including inverse propensity weighting, nearest neighbor matching, and doubly robust causal forests. We also assess localized balance metrics and sensitivity analyses to address the possibility of differential imbalance and unobserved confounding. The authors encourage researchers to follow similar data exploration practices in their work on variation in sociological effects and offer a straightforward framework by which to do so.

Reconsidering the Reference Category

Social scientists often present modeling results from categorical explanatory variables, such as gender, race, and marital status, as coefficients representing contrasts to a “reference” group. Although choosing the reference category may seem arbitrary, the authors argue that it is an intrinsically meaningful act that affects the interpretability of results. Reference category selection foregrounds some contrasts over others. Also, selecting a culturally dominant group as the reference can subtly reify the notion that dominant groups are the most “normal.” The authors find that three of four recently published tables in Demography and American Sociological Review that include race or gender explanatory variables use dominant groups (i.e., male or white) as the reference group. Furthermore, the tables rarely state what the reference is: only half of tables with race variables and one-fifth of tables with gender variables explicitly specify the reference category; the rest leave it up to the reader to check the methods section or simply guess. As an alternative to this apparently standard practice, the authors suggest guidelines for intentionally and responsibly choosing a reference category. The authors then discuss alternative ways to convey results from categorical explanatory variables that avoid the problems of reference categories entirely.

Multigroup Segregation Analyses with Covariates

The author introduces methods for the decomposition analysis of multigroup segregation measured by the index of dissimilarity, the squared coefficient of variation, and Theil’s entropy measure. Using a new causal framework, the author takes a unified approach to the decomposition analysis by specifying conditions that must be satisfied to decompose segregation into unexplained and explained components. Here, the unexplained component represents the direct effects of the group variable on the conditional probability of acquiring a social position—such as a residential district in an analysis of residential segregation or an occupation in an analysis of occupational segregation—and the explained component represents indirect effects of the group variable on the outcome through covariates. The major merit of this approach is its ability to control individual-level covariates for the decomposition analysis of segregation. Two methods, one for semiparametric outcome models with the identity link function and the other for semiparametric outcome models with the multinomial logit link function, are introduced in this unified framework. The application of these methods focuses on occupational segregation among racial/ethnic groups. Father’s occupation, subject’s educational attainment, and the region of interview are included as covariates, using data from the General Social Surveys.

Uncertain Choices: The Heterogeneous Multinomial Logit Model

In this article, a modeling strategy is proposed that accounts for heterogeneity in nominal responses that is typically ignored when using common multinomial logit models. Heterogeneity can arise from unobserved variance heterogeneity, but it may also represent uncertainty in choosing from alternatives or, more generally, result from varying coefficients determined by effect modifiers. It is demonstrated that the bias in parameter estimation in multinomial logit models can be substantial if heterogeneity is present but ignored. The modeling strategy avoids biased estimates and allows researchers to investigate which variables determine uncertainty in choice behavior. Several applications demonstrate the usefulness of the model.