倾向值匹配与因果推论:方法论述评

社会科学对于因果关系的关注使得倾向值匹配(propensity score matching)的方法受到越来越多的重视。本文从以下几个方面对这一方法进行述评。首先,本文系统梳理了倾向值匹配方法的历史、发展及其对于社会学调查研究的意义。其次,通过一个具体实例,本文展示了倾向值匹配的具体实施过程。第三,从密尔对因果关系的论述以及统计学的反事实框架(counterfactual framework)出发,本文阐释了倾向值匹配如何通过"控制"混淆变量从而满足非混淆假设(unconfoundedness)来帮助研究者提出因果性结论。第四,本文将倾向值匹配与回归中断设计和工具变量进行了比较。最后,在简要分析倾向值匹配的局限性之后,本文讨论了倾向值匹配的最新发展趋势。     

Family Structure and Children’s Education Outcome: Evidence from Uruguay

As the developed world has experienced a shift away from the traditional two-biological parent family, scholars have sought to understand how children are faring in non-traditional homes. Debate has arisen over assertions that children from non-traditional families do less well in school. Concerns about selection issues as well as a paucity of cross-cultural evidence, have led some scholars to question the influence of family structure on educational attainment. Using data from the 2006 Uruguayan household survey, we evaluated the relationship of family structure with children’s education using two different methods to deal with selection problems, an instrumental variables approach and propensity score matching. Both approaches yield evidence that growing up in non-traditional family structures seems to be negative related with the schooling of Uruguayan boys, with more muted results for girls. Interestingly, Uruguay is a developing country with two peculiarities, that is, a culture that experienced fairly rapid modernization in terms of institutions—including family transition—especially compared with other South American nations, and meanwhile an intriguingly high level of school drop-out, unusually high for Uruguay’s overall level of development.     

The Effect of Social Information on Giving from Lapsed Donors: Evidence from a Field Experiment

Using data from an experiment carried out by a large nonprofit organization, this paper finds that lapsed donors who received a solicitation letter referencing a relatively high donation made by another donor (high social information) were more generous in giving, but overall less likely to make a donation, relative to the baseline (low social information) group. After using the propensity score matching to correct for pretreatment differences in the two experimental groups, the estimated effect of high social information on the average donation amount is an increase of $14.95 (45 %). However, high social information is also found to reduce the probability a lapsed donor will give by 4.1 %. Thus, high social information can have potentially offsetting effects when applied to lapsed donors. Nonprofits should consider this trade-off when employing social information fundraising techniques to solicit donations from lapsed donors.     

户籍制度改革对流动人口收入的影响研究

以往关于户籍制度对收入的影响的诸多研究基本上遵循的是工资差异分解模型的思路。然而,将外来人口和本地市民的工资收入差异进行分解,无法克服样本选择性与遗失关键变量带来的偏误。本文基于中国家庭收入调查数据,采用新近发展的倾向得分匹配方法来控制样本选择偏误,并采用自抽样法(Bootstrap)法进行统计推断,以克服小样本偏误,试图回答户籍限制对流动人口收入的影响。研究结果发现:(1)户籍制度对流动人口收入影响非常显著,克服选择性偏误之后的负向效应要高于普通回归结果和代理变量回归结果;(2)户籍限制对倾向得分较低的流动人口的收入效应是正向的,对倾向得分较高(能力较高)的流动人口产生影响为负;(3)这一不利影响随着市场化进程推进变得更显著。     

Presentation matters: how mode effects in item non-response depend on the presentation of response options

Comparability of data across modes is an important issue in survey research. In this paper we discuss item non-response to attitudinal questions in telephone and web surveys. We present results from a survey experiment conducted in Italy and in Spain that compares different presentations of response options in an online setting with a benchmark telephone survey. In line with earlier studies we find that (A), the share of non substantial answers in the online survey depends on how the response option is presented. Comparing different presentations in an online survey to the standard approach of telephone surveys using propensity score matching, we find that (B), the share and pattern of non substantial answers, is most similar across the two modes in the online survey when the it is captured in an instruction on each screen for the online survey. Our findings are of particular relevance for the design of multiple mode or mix-mode surveys of attitudinal questions using online and telephone modes.     

Parents’ Union Dissolution and Adolescents’ School Performance: Comparing Methodological Approaches

We use data from the National Longitudinal Study of Adolescent Health and the Adolescent Health and Academic Achievement Study to estimate how parents’ union dissolution influences changes in adolescents’ mathematics course work gains, overall grade point average, and course failure rates during a window of approximately 1 year (N =2,629). A primary purpose of this study is demonstrating the utility of propensity score matching techniques for studying topics such as ours that pose methodological challenges such as dealing with endogeneity and selection bias. We compare propensity score matching techniques to ordinary least squares (OLS) regression methods to show and discuss comparability of results obtained using these different procedures. Findings suggest that associations between parents’ union dissolution and achievement may be causal, regardless of method used.     

Die Sch?tzung von kausalen Effekten: überlegungen zu Methoden der Kausalanalyse anhand von Kontexteffekten in der Schule

This article discusses the problem of causal inference based on contextual peer effects in education. For this purpose, I first introduce the counterfactual approach to causality and the fundamental problem of causal inference. Subsequently, experiments and a number of statistical methods are discussed as possible approaches to estimate causal effects. In particular, conditioning on observables using common linear regression models as well as matching procedures (such as propensity score matching), fixed-effects models, the difference-in-difference approach, instrumental variables, and the regression discontinuity design are discussed. The introduction of these methods is neither focused on the mathematical background nor on the estimation procedure but rather on the general logic and the assumptions connected with a causal interpretation of the estimated effects. The paper closes with an analysis of compositional peer effects in German elementary schools to illustrate the adequate treatment of selection processes.     

Attitudes and Values of Scottish Farmers: “Yeoman” and “Entrepreneur” as Factors,Not Distinct Types1

Abstract We investigate the validity of the classification of farmers into two or more distinct types on the basis of attitudes, objectives, and values. Factor analysis of data from a survey of farm pluriactivity in Scotland (N=472) shows that attributes ascribed to “yeoman” and “entrepreneur” types emerge as separate factors rather than as a single bipolar factor. No clustering of the factor scores corresponding to distinct types is found. We also present correlations of the factor scores with attitudes to on- and off-farm pluriactivity and with observables such as age and off-farm employment. The desirability of analyzing survey data by methods which do not make assumptions about typology is discussed.     

Parenting with a Disability and Child Mental Health: A Propensity Score Analysis

Little research has examined the impact of having a parent with a disability on child mental health. This study used data from 7,116 families of children who participated in the 2006 National Health Interview Survey and propensity score matching (PSM) to investigate this question. Prior to PSM, differences in Child Mental Health Brief scores were significant with a moderate effect size. After PSM, differences in Child Mental Health Brief scores were still significant, but the effect size was small. Results indicate that much of the mental health risk for children of parents with disabilities is connected to related issues.     

Retiree Health Insurance and Disengagement from a Career Job

Over the past two decades the prevalence of partial/phased retirements has increased dramatically, redefining retirement and the way in which retirement benefits are evaluated. Specifically the effect of retirement benefits on the transition away from a state of career employment has become the primary issue of interest. This study uses data obtained from the Health and Retirement Study (HRS) and the Rand HRS files, to examine the relationship between access to retiree health insurance (RHI) and the decision to leave one’s career job. We employ a Cox Proportional Hazard Model to estimate how RHI affects the probability that an individual disengages from their career job, given they have not yet done so. Results indicate that those with access to RHI are 21% more likely to leave their career employer in all time periods than similar individuals without RHI. Several robustness tests including stratified estimation and propensity score matching are performed and no evidence of bias is detected.     

The Impact of Housing Instability on Child Maltreatment: A Causal Investigation

More than a half million children are confirmed as victims of maltreatment by the child welfare system each year. Children from unstably housed families are over-represented in child maltreatment reports, and a growing body of evidence links housing problems to maltreatment and Child Protective Services (CPS) investigation. The present study applies two propensity score analysis approaches—greedy matching and propensity score weighting—to data from the Fragile Families and Child Well-being Study to move toward a causal explanation of child maltreatment behaviors among mothers in low-income households. Utilizing two separate methods to correct for overt selection bias, the present study finds that housing instability leads to a small increase in maltreatment behaviors, yet this small positive net impact on child maltreatment does not fully explain the over-representation of unstably housed families in the child welfare system. Families experiencing housing problems likely have a range of needs that require earlier, targeted intervention to mitigate consequences of poverty, domestic violence, and maternal depression. Child welfare services should invest resources in housing assistance programs in-house as well as through partnerships with local public housing authorities to stabilize families, reduce housing-related strain on caregivers, and promote family preservation.     

Effect of Marriage and Spousal Criminality on Recidivism

The authors analyzed whether the effect of marriage on recidivism varied by spousal criminality. For this purpose, they used propensity score matching and full population data from Statistics Denmark on all unmarried and previously convicted men from birth cohorts 1965–1985 (N = 102,839). The results showed that marriage reduced recidivism compared to nonmarriage only when the spouse had no criminal record. Similarly, marriage to a nonconvicted spouse reduced recidivism significantly more than marriage to a convicted spouse. These findings not only underline how important marriage is for social integration but also stress the heterogeneous nature of the protective effects of marriage.     

Economic opportunities and threats in contentious environmental politics: A view from the European South

Aiming to contribute to the discussion of how opportunities and threats are put into motion by focusing on the understudied economic factors, this article presents four illustrative accounts from the post-dictatorial experience of environmental contentious politics in Greece, Spain and Portugal, selectively. First, it documents that the incorporation of professional environmental organizations into dominant sustainable development practices in Greece is tied to an economic opportunity spiral that merges from contemporary economic liberalization policies and practices. Secondly, on the basis of protest-event analysis it shows how economic liberalization in Greece acts as a constraint for environmental protests, but an opportunity for anti-privatization ones. Third, via protest-case analysis it documents how economic dependency performs as a spiral of opportunities and constraints for groups tied to different interests in Crete. For example, tourism interest groups appear to opt for environmental protest more often when these interests are threatened by non-tourism-related environmentally damaging activities. Finally, using protest-case analysis the article illustrates how in Southern European local environmental conflict, spirals of economic opportunities and constraints portrayed in state responses to protesters, play a significant role in shaping the dynamics of local environmental contentious politics.     

Crisis nursery effects on child placement after foster care

The results of a study of the relationship between receiving crisis nursery services and the placement outcomes for young children leaving the child welfare system in Illinois are reported in this paper. The placement outcomes for children leaving foster care whose families received crisis nursery support prior to the children's placement in foster care is compared to the placement outcomes for children whose families received only foster care services. The children in two samples were identified by matching crisis nursery children's data from FY 2006 with children's data in the Illinois Child Abuse and Neglect Tracking System and Children Youth and Services Information System databases. After children served by crisis nursery and foster care services were identified, a comparison group of children with like-characteristics whose families received only foster care services was identified using propensity score matching. The children were followed until their out-of-home placement was terminated or until June 30, 2009. The placement outcomes and the length of stay were compared for the two groups. Using logistical regression analysis the results showed that children whose families received crisis nursery services prior to foster care placement were twice as likely to be reunited with their biological families (birth or extended family members) when compared to children whose families received only foster care services. The difference in the length-of-stay in foster care was not statistically significant when the two groups were compared. This preliminary study using administrative data shows that receiving crisis nursery services may have positive effects on the children's ultimate placement outcome after foster care. Additional research is needed to further explore the relationship between placement outcome and crisis nursery services.     

Effects of Homelessness and Child Maltreatment on the Likelihood of Engaging in Property and Violent Crime During Adulthood

The current study employs data from the U.S. National Longitudinal Study of Adolescent Health (N = 10,818; 56% female; 21% African American) to test how experiencing homelessness through emerging adulthood and child maltreatment predict adult property crime and violent crime. Unlike many published studies on homelessness, we used propensity score matching to correct for selection bias between homeless and nonhomeless individuals. Logistic regression models were run to predict violent and property crime in adulthood, controlling for child maltreatment. Participants who experienced homelessness by age 26 were 1.6 times more likely to commit violent crime in adulthood and almost 30% more likely to commit property crime. Those who were victims of child maltreatment were 15 times more likely to commit property crime, but no more likely to commit violent crime. Findings show that comparing statistically equivalent groups, homelessness through emerging adulthood significantly predicted adult criminality while child maltreatment showed more variable results. Greater prevention efforts aimed at children and adolescents at high risk of experiencing homelessness, as well as more intensive outreach services to homeless youth, may moderate exposure and reduce reliance on criminal survival behaviors.     

The role of child and parent savings in promoting expectations for university education among middle school students in Ghana: A propensity score analysis

The volume of research on high educational expectations among young people is increasing. Yet, there are still research gaps in what we know about the key predictors of young people's educational expectations, especially in the Sub-Saharan African context. This study aims to complement existing research on asset effects by testing two related propositions about how child and parent savings relate independently to children's expectations for university education. We use propensity score analysis with data from a random sample of junior high school students in Ghana. Findings reveal that parent savings are positively associated with higher probability of children having higher educational expectation. We also find that parents' savings have stronger effect on children's educational expectations compared to children's savings. We discuss implications for early exposure of young people to savings opportunities and its potential to ensure educational progression and optimum psychological and economic benefits to students.     

EXPERT OPINION AND PRODUCT QUALITY: EVIDENCE FROM NEW YORK CITY RESTAURANTS

We analyze whether consumers' quality perception and/or producer investment of New York City restaurants, measured by Zagat scores, responds to newly appearing expert opinion, measured by Michelin scores. Answering this question is of general economic interest as it applies to all markets with information asymmetries. Employing a difference‐in‐differences approach as well as a propensity score matching approach we find significant Michelin treatment effects on food and décor quality. Based on these changes, we find a Michelin‐induced price increase of approximately 30% per Michelin star. To examine whether the improved food and nonfood quality is based on restaurant investments or is merely imagined, we analyze nonfood investments by referring to Wine Spectator wine list awards. Our analysis suggests that Michelin‐reviewed restaurants are significantly more likely to invest in their wine list than others. As a result, Michelin reviewed restaurants are more likely to improve food and nonfood (esp. décor) quality leading to significant price increases. However, while restaurants that increase prices only due to décor and service improvements are more likely to go out of business, food improvements appear to secure a restaurant's survival. (JEL D11, L15, L66)     

Variation in the Relationship Between Education and Marriage: Marriage Market Mismatch?

Educational expansion has led to greater diversity in the social backgrounds of college students. We ask how schooling interacts with this diversity to influence marriage formation among men and women. Relying on data from the 1979 National Longitudinal Survey of Youth (N = 3208), we use a propensity score approach to group men and women into social strata and multilevel event history models to test differences in the effects of college attendance across strata. We find a statistically significant, positive trend in the effects of college attendance across strata, with the largest effects of college on first marriage among the more advantaged and the smallest-indeed, negative-effects among the least advantaged men and women. These findings appear consistent with a mismatch in the marriage market between individuals' education and their social backgrounds.     

Differences in Relationship Quality Measurement and Covariate Influence Between Individuals in First and Second Marriages: A Propensity Score Analysis

Few studies have examined relationship quality construct measurement and covariate influence differences between first married and remarried individuals. We used multiple-sample confirmatory factor analysis and structural regression invariance testing to assess relationship quality measurement and covariate influence differences between these two groups. We account for selection bias on 11 observed selection variables by comparing outcomes before and after propensity score greedy matching procedures. Using a sample of 2,030 first married and 488 remarried individuals from the Relationship Evaluation Survey, prematching results indicate some significant differences between groups. Postmatching results, however, indicate that all measurement and structural parameters can be constrained to be equal between groups without worsening model fit. Implications, limitations, and future directions for research are discussed.     

Factors Related to Successful Outcomes Among Preschool Children Born to Low‐Income Adolescent Mothers

The purpose of this study was to describe how the experiences and circumstances of the most successful children born to low‐income adolescent mothers differed from the experiences and circumstances of the least successful children over the first 54 months of their lives. Success was defined as scoring in the top quartile for this sample on the Peabody Picture Vocabulary Test‐Revised (PPVT‐R) when the children were 54 months old and about to make the transition to kindergarten. The experiences of the 22 children with the highest scores on the PPVT‐R were contrasted with the experiences of the 22 children who had the lowest scores on the PPVT‐R for this sample. The two groups differed markedly on measures of caregiving and home environment assessed when the children were 12, 24, 36, and 54 months old. In addition, mothers of the most successful children achieved more years of education, were more likely to be employed, had fewer children on average, tended to live in more desirable neighborhoods, and were more likely to be living with a male partner. Case studies are also presented on four families (two from the most successful group and two from the least successful group) to further illustrate how the experiences and circumstances of children in the two groups differed.