No full-text available
To read the full-text of this research,
you can request a copy directly from the author.
Causal Inference in Sociological Studies
Abstract
Acknowledgments:The authors would like to thank Melissa Hardy, David Harding, and Felix Elwert for comments,on an earlier draft of this paper. 1
... Importantly, the motivations for these experimental studies are grounded in concerns related to selection and post-treatment xx confounding which are endemic to observational studies of racism and place-based relations, and these must be addressed to construct the closest counterfactuals to our relational theories (all three empirical chapters consider these methodological issues) (Gangl 2010;Morgan and Winship 2014;Oakes et al. 2015;Sen and Wasow 2016;Sharkey and Faber 2014). The response to charges of "vagueness" and "description" leveled at sociological work based in counterfactual regression models should not be to cede the grounds of "real" causal inference to the interventionist frame and only use words such as "links" and "drivers" (Hernán 2018;Winship and Morgan 1999;Winship and Sobel 2004), but rather to sharpen our estimands so that they most accurately reflect theoretically plausible assumptions about time-varying confounding and multiple mediationwithout sacrificing theoretical generalizability for the sake of internal validity and precision. In other words, embracing that the goal of counterfactual models is most often causal understanding, which is more or less justified given the underlying social theory, study design, and counterfactual assumptions. ...
... In other words, embracing that the goal of counterfactual models is most often causal understanding, which is more or less justified given the underlying social theory, study design, and counterfactual assumptions. As long argued by sociological methodologists such as Christopher Winship, Michael Sobel, and others (Gangl 2010;Morgan and Winship 2014;Winship and Morgan 1999;Winship and Sobel 2004), the use of regression models with many control variables has increased exponentially in applied quantitative sociology, with coefficients of the target independent variable interpreted as demonstrating a "link" to the outcome. These scholars contend that the widening semantic divide between "associational" studies and "causal inference" when applying these models to observational data is used to sidestep critical inferential issues (especially in longitudinal settings) related to study design, confounding, and mediation. ...
... These scholars contend that the widening semantic divide between "associational" studies and "causal inference" when applying these models to observational data is used to sidestep critical inferential issues (especially in longitudinal settings) related to study design, confounding, and mediation. Ceding of the "causal inference" space in quantitative xxi sociology to only (quasi-)experimental designsand thereby ceding normative authority to fields grounded in positivist rather than relational theoretical frameworksis a disservice to the many important ways in which holistic, etiologic counterfactual models speak to causal explanation in social theory and policy (Elwert and Winship 2014;Itzigsohn and Brown 2020;Mackie 1974;Matthay et al. 2020;Moffitt 2005;Muntaner 2013;Pearl 2014;Schwartz, Prins, et al. 2016;Schwartz, Gatto, and Campbell 2017;Sharkey and Elwert 2011;Winship and Sobel 2004;Wodtke et al. 2011). ...
In this dissertation, I examine how we quantify the dynamic, cumulative effects of relational social exposures with longitudinal survey data. In Chapter 1, I demonstrate a new mediation framework for describing what are often conceptualized problematically as “neighborhood effects.” Findings from this study clarify the reciprocal, life-course process through which neighborhood is implicated in the early production of social inequality. In Chapter 2, I extend this mediation framework to respond to theoretical critiques of how variables for race are used in common regression frameworks in attempts to study structural racism. I demonstrate an alternative counterfactual approach to explain how multiple racialized systems dynamically shape health over time, examining racial inequities in cardio-metabolic risk. I decompose the observed disparity into three types of effects: a controlled direct effect (“unobserved racism”), proportions attributable to interaction (“racial discrimination”), and pure indirect effects (“emergent discrimination”). I discuss the limitations of counterfactual approaches while highlighting how they can be combined with critical theories to quantify how interlocking systems produce racial health inequities. In Chapter 3, I use this framework to examine the Black-white wealth gap in the United States. Descriptive and qualitative analyses have identified many mechanisms underlying wealth correlations across successive generations, but few studies have quantified the relative contributions of these interconnected and racialized systems of reproduction to the total gap we observe today. I define a wealth gap in 2015–17 between the grandchildren of those racialized as Black and the grandchildren of those racialized as white in 1968–70. I use a fully interacted counterfactual mediation framework to decompose this disparity into the historical, racialized contributions of 1) effects of home values in 1968–70 on home values in successive generations and 2) effects via educational attainment in successive generations. Findings from this study contribute to our understanding of the dynamic, racialized process of multigenerational place-based wealth accumulation and support the importance of historically contingent social policy centered on reparative justice.
... The extent to which the results indicate causal relationships is subject to interpretation, depending for instance on the assessment of confounding and selection effects (see e.g. Gangl, 2010;Winship and Sobel, 2010;Keele, 2015). I combined all applicable World Value Surveys (WVS, 1981(WVS, -2014 and Arab Barometer surveys (AB, 2006(AB, -2014 collected between 2001 and 2014 and representing 15 MENA countries: Algeria, Bahrain, Egypt, Iraq, Jordan, Kuwait, Lebanon, Libya, Morocco, Palestine, Saudi Arabia, Sudan, Tunisia, and Yemen. ...
... I set out to address this issue and advance existing knowledge by theorizing and testing the context-dependency of commonly discussed relationships between generalized trust on the one hand and socio-economic and religious factors on the other. In line with the dominant approach in the existing literature (see Winship and Sobel, 2010;Bauer and Freitag, 2018), I applied large-scale regression analyses to 47 surveys from the AB and WVS. ...
... Second, more advanced data or matching methods could help flesh out the mechanism at work and filter out potential identification effects (Gangl, 2010). Third, experiments could be designed to circumvent selection effects and test the spillover effect of political trust (see Winship and Sobel, 2010;Keele, 2015). Last, biological factors might be cofounders (and even suppressors) here too (see Cawvey et al., 2018;Uslaner, 2018), but such information is not available, at least not cross-nationally. ...
Our knowledge of social trust's drivers in the MENA region is limited and there are good reasons to expect that theories based on Western countries cannot be copied to the MENA one-to-one. Arguing for a broader and at the same time context-sensitive comparative approach, I translate the 'societal winners', social capital, and religious beliefs mechanisms explaining trust to the MENA context. Moreover, I acknowledge intraregional diversity and test how the impact of these factors also differs among MENA countries. Empirically, I synchronize 47 surveys from 15 MENA countries, which provides the broadest and most systematic assessment of trust in the MENA to date. The results show that the societal-winner mechanism does not hold: employed, higher education and wealthier citizens are not more trusting. However, higher-educated citizens distrust other citizens more, particularly in the strongest autocracies. Religiosity seems pivotal too. Among others, service-attending citizens are more trusting, mainly where regimes regulate religious affairs. Overall, this study provides insight into what shapes generalized social trust in the Middle East and North Africa and it underscores that at a comparative level we need to consider interregional and intra-regional forms of context-dependency were we to formulate a broadly applicable theoretical framework of trust's drivers.
... A fourth issue which has not been considered in the literature as far as we are aware is whether it makes sense to talk about APC models as causal models. From a counterfactual perspective (for reviews, see Morgan 1999, Winship andSobel 2004), the assertion that Age, Period, or Cohort have causal effects is highly problematic. Holland (1986), as well as others, has argued that only manipulable variables can have a causal effect (for further discussion see Winship and Sobel 2004). ...
... From a counterfactual perspective (for reviews, see Morgan 1999, Winship andSobel 2004), the assertion that Age, Period, or Cohort have causal effects is highly problematic. Holland (1986), as well as others, has argued that only manipulable variables can have a causal effect (for further discussion see Winship and Sobel 2004). In other words, only variables for which it is possible to potentially change an individual's value on the causal (treatment) variable should be the subject of counterfactual causal analysis. ...
... Conversely, although night always follows the day, we do not believe that day causes the night because of the absence of a causal mechanism by which day might cause night, and our belief that both are related to a separate causal factor, the rotation of the earth. (For a more detailed discussion of the different concepts of causality, see Sobel 1995, Brady 2003, and Winship and Sobel 2004 Assuming we can specify the mechanisms through which Age, Period, or Cohort affect an outcome, does it make sense to talk about the mechanisms involved as having causal effects? ...
Robert Mare, Stephen Morgan and members of Harvard’s Applied Statistics Colloquium for comments
... They are simply unavoidable, especially as one keeps counterfactuality in mind. Winship and Sobel (2004) contributed a précis of the history of uses of "causality" in a range of developments that began with structural equation modeling and continued through the 1990s, developments in which both authors were major principals. Their discussions are usefully descriptive in the manner of a review, but the greater usefulness comes from their highlighting of impacts by chief critics of the laxity with which many social scientists engaged in purportedly "causal analysis" with little or no attention to assumptions being made. ...
... The more difficult question has to do with theorization-not practices of commentary on "the greats" but rather Cartwright's concern as expressed in her question, Where are the new causal theories of substantive processes? Winship and Sobel (2004) limited their review of early "philosophical literature" mainly to Hume, which is fitting inasmuch as he got the articulation of doubt as a skepticism on track. It was Kant who left the idea of a science of human affairs as so eminently doubtable as to be a non-starter. ...
A short primer on inference (observational, mensural, causal, and statistical), requiring very little mathematics.
... Time-invariant covariates include whether the child is female (= 1), maternal race (black = 1, nonblack = 0), maternal ethnicity (Hispanic/Latina = 1, non-Hispanic/Latina = 0), maternal age, maternal education at time 1 (less than high school, high school, some college compared with college or more; timeinvariant because education did not change much over the study period), and the mother's propensity to be included in the analytic sample. Logistic regressions predicting inclusion in the analytic sample based on certain characteristics from the first wave the NLSY79 produce predicted probabilities or propensities to be included in the analytic sample (Morgan and Sørensen 1999;Winship and Sobel 2004). These predicted probabilities, when included as control variables, account for the unequal propensities to be included in the analytic sample, thus modeling the missing data mechanism and correcting the correlation of the error term with the covariates due to these missing data (Morgan and Sørensen 1999;Willson et al. 2007;Winship and Sobel 2004). ...
... Logistic regressions predicting inclusion in the analytic sample based on certain characteristics from the first wave the NLSY79 produce predicted probabilities or propensities to be included in the analytic sample (Morgan and Sørensen 1999;Winship and Sobel 2004). These predicted probabilities, when included as control variables, account for the unequal propensities to be included in the analytic sample, thus modeling the missing data mechanism and correcting the correlation of the error term with the covariates due to these missing data (Morgan and Sørensen 1999;Willson et al. 2007;Winship and Sobel 2004). To the extent that the inclusion model was correctly specified, it appears that selection bias does not change the results once the time-varying and time-invariant covariates are included, as there were neither substantive differences nor differences in statistical significance between the results in the models reported here and the models without the inclusion adjustment (results available upon request). ...
While many studies use parental socioeconomic status and health to predict children's health, this study examines the interplay over time between child and maternal health across childhood and adolescence. Using data from women in the National Longitudinal Study of Youth 1979 cohort and their children (N = 2,225), autoregressive cross-lagged models demonstrate a reciprocal relationship between child activity limitations and maternal health limitations in direct effects of child activity limitations on maternal health limitations two years later and vice versa-net of a range of health-relevant time-varying and time-invariant covariates. Furthermore, there are indirect effects of child activity limitations on subsequent maternal health limitations and indirect effects of maternal health limitations on subsequent child activity limitations via intervening health statuses. This study examines how the interplay between child and maternal health unfolds over time and describes how these interdependent statuses jointly experience health disadvantages.
... The fourth factor is the intensity of reading news. Watching television is unrelated to geographical knowledge, but reading international news in newspapers can influence geographical knowledge (Winship, 2014). The fifth factor is the intensity of social media access. ...
One of the abilities that students must master in the 21st century is literacy skills, one of which is geographic literacy. Based on previous research, students' geographic literacy, in general, still needs to improve. This research aims to determine and analyze the geographic literacy abilities of State High School (SMA) students in Jambi City. This research uses a survey method. The data processing results show that the geographic literacy skills of 190 State High School students in Jambi City are in the medium category with a percentage of 51% or 97 students. This means the students can understand the essence of interaction, interconnection, and the contextual implications of spatial interaction phenomena in life. Daily for some instances only. Meanwhile, the number of students in the high geographical literacy category is 43 or 23%, so it can be interpreted that students can understand the essence of interaction, interconnection and the implications of spatial interaction phenomena contextually in everyday life. Geographic literacy skills in the low category are 50 students or 26%, which means that students still need to understand the essence of interaction, interconnection and the implications of spatial interaction phenomena contextually in everyday life.
... First, the present findings are based on self-reported data. Another limitation relates to the issue of causation, a limitation facing many non-experimental studies (Winship & Sobel, 2004). There is need to examine the use and usefulness of science homework management strategies across rural and urban settings at the high school level, as the role of educational aspirations in science homework behaviour may be more pronounced at this level. ...
This study was an attempt to test if students differed in their self-concept, attitude and their perception of the usefulness of Physics and Chemistry in terms of their type of school and location. To test this hypothesis, t-test was used to compare differences in the mean scores in terms of type of school and location. Pearson's Product Moment Correlation coefficient was used to determine the relationships between independent and dependent variable. Four hundred and forty six (446) students constituted the sample. The attitude towards Physics and Chemistry and the perception of the type of school showed there was no significant difference between single-sex and co- educational school. There was no significant difference between single- sex and co- educational school (mixed) students in their self- concept in Physics and Chemistry. There was no significant difference between single- sex and co- educational school (mixed) students in their perception of the usefulness of Physics and Chemistry. Results also show that urban students have better perception of the usefulness of Physics and Chemistry than the rural students. The difference in perception of the usefulness of Chemistry is statistically significant but the difference in perception of the usefulnessofPhysicsisnotstatisticallysignificant. Bothruralandurbanstudentshad positive attitude and self-concept for Physics and Chemistry.
... These methods include DID models relying on propensity score matching (PSM) estimations, which are applied to create a control group that is fully comparable, based on observables, with the treatment group (Caliendo and Kopeinig 2008). The propensity score is a balancing score including a function of the observed covariates, which displays a conditional probability of the assignment to the treatment (Gangl 2010;Gangl and DiPrete 2004;Morgan and Winship 2015;Rosenbaum and Rubin 1983;Winship and Sobel 2001). PSM matches all treatment and control cases with (nearly) the same propensity score as a kind of "virtual twins" (Foster et al. 2011) for the calculation of the average treatment effect. ...
The chapter asks about possible causal effects of migration on subjective well-being (SWB) measured by self-reported overall life satisfaction. By combining the emigration sample of the German Emigration and Remigration Panel Study (GERPS) with a quasi-counterfactual sample of internationally non-mobile Germans provided by the Socio-Economic Panel Study (SOEP) the difference-in-difference analyses show that emigration is actually accompanied by an increase in SWB. Based on propensity score matching procedures and compared to non-mobile German stayers, German first-time emigrants show a significant increase in SWB shortly after arrival in their host country. For most emigrants, migration pays off not only economically via increasing incomes but also with regard to an increase in life satisfaction. However, the underlying analysis has certain limitations and we therefore discuss the significance of the presented evidence and consequences and challenges for future research.
... Sociologists of sociology have found that causal arguments are widespread and important in the sociological community. They're important to sociologists' styles of thought and self-understandings, even if not everyone is happy with this, misgivings have been voiced, and "causalism" has been criticized (Abbott 1998(Abbott , 2016Hirschman 2016;Winship and Sobel 2004). ...
I argue that what-makes-it-possible questions are a distinct and important kind of sociological research question. What is social phenomenon P made possible or enabled by? Results won’t be about P’s causes and causal relationships, but about its enablers and enabling relationships. I examine the character of what-makes-it-possible questions and claims, how they can be empirically investigated, and what they’re good for. If I’m right, they provide a unique perspective on social phenomena, they show how the social world doesn’t come ready-made, and they open up new avenues for research.
... Second, if we also make a common assumption that covariates are measured pre-treatment [24], then only the structures in columns of A, B, C, and D will be possible. Furthermore, conditional independence relations would be sufficient to distinguish among each of these graphical models. ...
... The processes of employees' self-selection and selection by others, such as employers, determine the assignment of the treatment as well as the probabilities of participating in such programmes systematically. This selectivity bias usually leads to biased estimates since each of the effects of the programme are not identical to the differences of the result variables for participants and non-participants (Winship and Sobel 2004). ...
Human capital theory and the life-course perspective are used to investigate how economic modernisation, as well as developments in the labour market after the West German “economic miracle”, impacted employers’ supply of further education and training on the job, and employees’ increased participation in these arrangements. Additionally—controlling for the aforementioned structural change and economic cycles—it is analysed whether participation in further training minimises employees’ risk of dismissal and heightens their commitment to a company. The hypotheses are tested using longitudinal data and time series—allowing the analysis of employees’ participation in further education and training on the job, and the careers of West Germans born between 1956 and 1978 for the 1972–2008 periods—by procedures of event history analysis and episode splitting in a dynamic multi-level design. Systematic period and cohort effects of structural change in the economy and labour markets on companies’ supply of, and employees’ participation in, continued vocational training on the job have been revealed. Participation in further training reduces employees’ risk of dismissal, as well as their mobility between companies. Participants’ adaptation to structural change via job-related further training is correlated with increased employment security, professional flexibility, and commitment to the employer.
... j odvozená pozice na trhu práce? Získané znalosti či postoje utvrzené v rámci uzavřeného sociálního okruhu? Jedná se o zlomkový odraz celkové třídní pozice? Zatímco samotné užití regresního modelu zodpovězení těchto otázek vyžaduje, řada vícerozměrných analýz si je ani neklade a zůstává u prostého konstatování "kontroly" vlivu vzdělání apod. [srov.Sobel, Winship 2009]. 6 K této otázce téžAbbott [2001: 122]: "Naše opovržení (deskripcí) je neupřímné, jelikož snadnost počítačového zpracování udělala z regrese samotné deskriptivní metodu. V mo- ...
Cultural capital is an important part of the conceptual apparatus of research on inequalities and social reproduction. The putative transformation of cultural hierarchies in contemporary society, however, opens up the question of whether it still makes any sense to speak of 'legitimate taste' and eventually of what the nature this legitimate taste might be. This article examines what constitutes legitimate culture in the context of university-level study. It focuses on the differentiation of taste and the way in which the space of cultural consumption is structured by academic disciplines and university faculties. The article draws on data from a questionnaire survey of first- and second-year students at Charles University in 2017 (n = 5127) and conducts a multiple correspondence analysis. It shows that the first dimension of the cultural space of students can be interpreted as the axis of overall cultural capital without any specific differentiation. The third dimension of cultural space, by contrast, convincingly captures a cleavage between traditional cultural capital and a new form of cultural capital. The amount of cultural capital accumulated depends on the kind of academic disciplines studied, but another significant structuring element of cultural capital is the environment of individual university faculties itself.
... 'Similar' in this context means similar with respect to the chance to receive the treatment. Other predictors of the outcome that do not influence treatment assignment do not need to be taken into account (Gangl 2015;Winship & Sobel 2004). Furthermore, mediators should not be taken into account since they lead to post-treatment bias (i.e., an underestimation of the direct causal effect) (Rosenbaum 1984). ...
Typically, associations between being unemployed and policy attitudes are explained with reference to economic self‐interest considerations of the unemployed. Preferences for labour market policies (LMP) and egalitarian preferences are the prime example and the focus of this study. Its aim is to challenge this causal self‐interest argument: self‐interest consistent associations of unemployment with policy preferences are neither necessarily driven by self‐interest nor necessarily causal. To that end, this article first confronts the self‐interest argument with a broader perspective on attitudes. Given that predispositions (e.g., value orientations) are stable and influence more specific policy attitudes, it is at least questionable whether people change their policy attitudes simply because they get laid off. Second, the article derives a non‐causal argument behind associations between unemployment and policy attitudes, arguing that these might be spurious associations driven by individuals’ socioeconomic background. After all, the entire socioeconomic background of a person is simultaneously related to both the risk of getting unemployed (‘selection into unemployment’) and distinct political socialisation experiences from early childhood onwards. Third, this article uses methods inspired by a counterfactual account on causality to test the non‐causal claims. Analyses are carried out using the fourth wave of the European Social Survey and applying entropy balancing to control for selection bias. In only two of the 31 analysed countries do unemployment effects on egalitarian orientations remain significant after controlling for selection bias. The same holds for effects on active LMP attitudes with the exception of six countries. Attitudes towards passive LMP are to some degree an exception since effects remain in a third of the countries. Robustness checks and Bayes factor replications showing evidence for the absence of unemployment effects support the general impression from these initial analyses. After discussing this article's results and limitations, its broader implications are considered. On the one hand, the article offers a new perspective on the conceptualisation and measurement of unemployment risk. On the other hand, its theoretical argument, as well as its treatment of the resulting selection bias, can be broadly applied. Thus, this article can contribute to many other research questions regarding the (ir)relevance of individual life events for political attitudes and political behaviour.
... Kahneman Tversky Slovic (1982) Lau Redlawsk(2001, 953) Holland (1986,959) (looking for the cause of an effect) (studying the effect of a cause) 2008,3 (causal effect) (Rosenbaum 2002;Rubin 1974) Morgan Winship(2007, 280) 13 (instrumental variable) Morgan Winship(2007, 193-200) 14 CMC (ignorability) (Winship and Sobel 2004) 2012 13 Holland(1986, 947) (fundamental problem of causal inference) Rosenbaum Rubin(1983) i ...
News and many public comments indicate that the “92 consensus” was the crucial issue to affect the result of Taiwan 2012 presidential election.This paper aims to study the effect of the 92 consensus on voters’ choices in 2012. This paper reviews the core assumptions, boundary of application and analytical methods of the “issue voting” theory. Moreover, it focus on studying the impacts of the 92 consensus that are presumed to be endogenously correlated with party identification. Based on the approach of “studying the effects of a cause”, as well as using the “2012 Taiwan’s Election and Democratization Study” dataset (TEDS2012-T and TEDS2012),
this paper applies “propensity score matching” (PSM) method to investigate the issue effect of the “92 consensus” on voting choices during 2012 election. The results from the data analysis demonstrate that positions supportive of the “92 consensus” account for about twenty percent of supporting rates to
pro-Ma voters in the period of the electoral campaign; meanwhile, positions oppositional to the “92 consensus” would contribute about thirteen percent of supporting rates for pro-Tsai voters. After the election the influential
probability of the “92 consensus” was dramatically downsized to ten percent of supporting rates to pro-Ma voters; however, the percentage for pro-Tsai voters was slightly reduced to twelve percent. These findings provide more valid and credible estimates toward the influential probability of the “92 consensus” issue during the 2012 elections. Moreover, the statistical findings over various time-points also verify the successful transformation of the “92 consensus” to be identified as a salient issue across pro-Ma and pro-Tsai voters. It indeed achieved substantial influences toward the processes and result of 2012 presidential election.
... Although selfreport data are commonly used in self-regulation research (Pintrich, 2004;Zimmerman, 2008), and while students have certain advantages as observers of their own homework behavior (e.g., some aspects of homework behavior are not easily observable by parents or trained observers), there is a need to replicate our findings in future investigation that includes other behavioral measures (i.e., in addition to prior standardized mathematics achievement used in our present study as one objective measure). Another limitation concerns the issue of causation, a challenge that often faces nonexperimental research (e.g., Winship & Sobel, 2004). Informed by related theoretical frameworks (Boekaerts & Corno, 2005;Eccles & Wigfield, 2002;Pintrich & Zusho, 2002), in the present investigation we sought to control for potential variables that are known to influence homework management. ...
The authors examined self-regulation of mathematics homework behavior (i.e., mathematics homework management). The participants consisted of 796 eighth-grade students (46 classes) in China. Multilevel results showed that mathematics homework management was positively associated with value belief at the class and individual level. At the individual level, students' management in mathematics homework was positively related to affective attitude, expectancy belief, learning-oriented reasons, homework interest, parent education, teacher feedback, adult-oriented reasons, and value belief. Meanwhile, students' management in mathematics homework was negatively related to time spent on television. Our findings were discussed in the context of related theoretical frameworks (e.g., self-regulation and expectancy value) as well as previous findings pertaining to homework.
... Another limitation concerns the issue of causation, a challenge that typically faces many nonexperimental studies (Winship & Sobel, 2004). Informed by theoretical frameworks pertaining to regulation of motivation (e.g., Eccles & Wigfield, 2002;Pintrich, 2004;Pintrich & Zusho, 2002), the current investigation attempts to control for potential confounding variables. ...
As many students face the enduring challenge of maintaining their motivation to complete homework assignments, there is a critical need to pay close attention to homework motivation management (i.e., students’ efforts to sustain or enhance their motivation in order to complete homework assignments that might be boring or difficult). Yet, in spite of research showing that homework motivation has a powerful influence on homework performance and academic achievement, there have been few attempts to systematically investigate models of factors that influence homework motivation management.
... A second problem regarding causality is often left undiscussed in the counterfactual literature (for an exception, Winship and Sobel 2001). This is the problem of causal direction. ...
Extant research uses regression analysis with macro-level data to study the effect of immigration on crime in the contemporary USA. These studies have found mostly null or negative associations between the two variables. I point out three problems with these analyses. (i) Some studies use negative binomial regression inappropriately, in effect measuring determinants of frequencies rather than rates; (ii) all existing estimates arguably suffer from severe undercontrol, overcontrol, or both and cannot establish the direction of causality (if any) between immigration and crime; (iii) all studies present estimates that give equal weight to areas of differing population size. Taken together, these limitations render the research inconclusive. I show how to avoid these problems in a regression analysis of the effect of recent immigration on homicide rates in a sample of 91 US cities in 2000. Estimates point to a negative effect of immigration on homicide, but are not statistically significant.
... Over the past decades, empirical research on causal relationships using propensity score methods has been increasing in the social sciences Pischke, 2009, 2014;Caliendo and Kopeinig, 2008;Harding, 2003;Imai and van Dyk, 2004;Imbens and Rubin, 2015;Imbens and Wooldridge, 2009;Morgan, 2001;Morgan and Harding, 2006;Morgan and Winship, 2007;Normand et al., 2001;Sekhon, 2011;Smith, 1997;Sobel, 1995Sobel, , 1996Sobel, , 2000Winship and Morgan, 1999;Winship and Sobel, 2004). One of the great promises of propensity scores is that through procedures like matching or weighting, scholars are enabled to construct a dataset in which treated and untreated cases, for specific propensity score values, have similar odds of receiving a treatment. ...
This article reviews and comments on three major expansions of propensity score methods in recent decades. First, how to use generalized propensity scores to tackle multi-categorical or continuous treatment variables is shown in procedures of propensity score regression adjustment and propensity score weighting. Second, the counterfactual framework of causal inference in the analysis of mediation mechanisms is reviewed and the decomposition of the causal relationship between variables into causal direct effects and causal indirect effects is illustrated. Third, the heterogeneous treatment effect across the distribution of propensity score values is discussed in the framework of the stratification-multilevel model. For each methodological breakthrough, this article comments on potential issues which deserve serious attention in the practical application of these methods.
... Although much care was taken to control for possible confounding variables and alternative explanations, other predictor variables might have had an effect on homework distraction had I included them. Unfortunately, it is difficult to address the issue of causality in nonexperimental research in general (Winship & Sobel, 2004), and with homework research in particular, because homework is influenced by more factors than any other instructional activity (H. Cooper, 2001). ...
Background: Students continue to struggle with homework distraction well into the secondary school years. Recently, the concern over homework distraction has been growing, as new electronic media have offered diverse and nearly ubiquitous forms of diversion to students while they are doing homework. It is surprising to note, however, that a systematic examination of a broad spectrum of factors that contribute to homework distraction is noticeably absent from much contemporary literature. Thus, there is a critical need to examine a range of variables that may influence homework distraction and, consequently, what implications might be drawn from this line of research to help students better handle homework distraction.
Purpose: The aim of the present study is to propose and test empirical models of variables posited to predict homework distraction at the secondary school level, with the models informed by (a) relevant theoretical approaches (e.g., volitional control) and (b) findings from homework research that alluded to a number of factors that may influence homework distraction.
Research Design: The study reported here used cross-sectional survey data.
Participants: The participants were 1,800 students from 97 classes in the southeastern United States: 969 eighth graders from 52 classes, and 831 eleventh graders from 45 classes.
Results: Results from the multilevel analyses revealed that most of the variance in homework distraction occurred at the student level, with grade level as the only significant predictor at the class level. Findings further revealed that at the student level, the variation in homework distraction was influenced by gender, self-reported grades, the context of doing homework athome, and student attitudes toward homework.
... Another related limitation relates to the issue of causation, a limitation facing virtually all nonexperimental research (Winship & Sobel, 2004). Although much care was taken to control for possible confounding variables (informed by research and theorizing on the regulation of motivation), other predictor variables might have had an effect on groupwork motivation management had they been included (e.g., the quality of online class or collaborative activities as perceived by students). ...
Background: Online learning is becoming a global phenomenon and has a steadily growing influence on how learning is delivered at universities worldwide. Motivation of students, however, has become one of the most serious problems in one important aspect of online learning—online collaborative groupwork or online group homework. It is surprising to note that few empirical studies have focused on how to enhance and sustain student motivation to work together in online learning environments.
Purpose: The propose of the present study is to propose and test empirical models of variables posited to predict students’ motivation management in online groupwork, with the models informed by (a) research and theorizing on regulation of motivation and (b) findings from online groupwork that alluded to a number of factors that may influence motivation management in online learning environments.
Research Design: The study reported here used cross-sectional survey data.
Participants: The participants were 150 graduate students from 46 online groups in the southeastern United States.
Results: Results from the multilevel analyses revealed that most of the variance in groupwork motivation management occurred at the student level, with online groupwork interest as the only significant predictor at the group level. At the student level, the variation in groupwork motivation management was positively related to student initiative, includingarranging the environment, managing study time, and help seeking. In addition, groupwork motivation management was positively related to feedback from the instructor and peers.
Conclusion: As most of the variance in online groupwork motivation management occurred at the student level rather than at the group level, online groupwork motivation management was largely a function of individual student characteristics and experiences. The present study further suggests that feedback and student initiative (arranging the environment, managing study time, and help seeking) play an important role in online groupwork
motivation management. Consequently, it would be beneficial to promote feedback among the instructor and group members in the online groupwork process. In addition, it would be beneficial to encourage students to take more initiative in online groupwork settings to better manage their motivation.
... Let subscript i represent the ith member in U. We further denote y 1 i as the ith member's potential outcome if treated (i.e., when d i ¼ 1), and y 0 i as the ith member's potential outcome if untreated (i.e., when d i ¼ 0). The framework for counterfactual reasoning in causal inference (Heckman 2005;Holland 1986;Manski 1995;Morgan and Winship 2007;Rubin 1974;Sobel 2000;Winship and Sobel 2004) states that we should conceptualize a treatment effect as the difference in potential outcomes associated with different treatment states for the same member in U: ...
Since the seminal introduction of the propensity score (PS) by Rosenbaum and Rubin, PS-based methods have been widely used for drawing causal inferences in the behavioral and social sciences. However, the PS approach depends on the ignorability assumption: there are no unobserved confoun-ders once observed covariates are taken into account. For situations where this assumption may be violated, Heckman and his associates have recently developed a novel approach based on marginal treatment effects (MTEs). In this article, we (1) explicate the consequences for PS-based methods when aspects of the ignorability assumption are violated, (2) compare PS-based methods and MTE-based methods by making a close examination of their identification assumptions and estimation performances, (3) apply these two approaches in estimating the economic return to college using data from the National Longitudinal Survey of Youth (NLSY) of 1979 and discuss their discrepancies in results. When there is a sorting gain but no systematic baseline difference between treated and untreated units given observed covariates,
... Although the present study included two age groups (eighth-and 11th-grade students), the findings were based on a cross-sectional survey, rather than repeated measures at different time points. Another related limitation relates to the issue of causation, a limitation facing virtually all nonexperimental research (Winship & Sobel, 2004). Although much care was taken to control for possible confounding variables (informed by research and theorizing on self-regulation), other predictor variables might have had an effect on homework management had they been included. ...
The authors examined empirical models of variables posited to predict homework management at the secondary school level. The participants were 866 eighth-grade students from 61 classes and 745 eleventh-grade students from 46 classes. Most of the variance in homework management occurred at the student level, with affective attitude and homework interest appearing as 2 significant predictors at the class level. At the student level, homework management was positively associated with learning-oriented reasons, affective attitude, self-reported grade, family homework help, homework interest, teacher feedback, and adult-oriented reasons. On the other hand, homework management was negatively associated with time spent watching television. In addition, Black girls, compared with Black boys, were more likely to manage their homework assignments.
... 2. The causality this paper refers to is defined in the counterfactual framework (see, e.g., Holland 1986; Morgan and Winship 2007;Sobel 1996;Winship and Morgan 1999;Winship and Sobel 2004). ...
Although there has been a large body of empirical literature assessing the role of social capital in the labor market, it is still unclear whether the relation-ship between measures of social capital and individual outcomes is causal due to the presence of endogeneity. This article offers a systematic review of the progress made by social scientists in alleviating the endogeneity bias and improving causal inferences on the effects of social capital in the labor market. The full typology of social capital is examined and four most salient sources of endogeneity that limit casual inference are addressed.
... Selection bias occurs when data is collected in ways which systematically distort it -where some factor determining the outcome of interest is also at play in determining whether those affected by it will be included in the final sample or not (Heckman 1979;Shadish, Cook and Campbell 2002;Rosenbaum 2002;Winship and Sobel 2004;Rosenbaum 2005;Nichols 2007). The mechanisms of (reasons for, causes of ) consent, then, are of interest where they may include factors that have substantial associations with the outcome of interest -for example where young people most at risk of something don't want to talk about it and hence are less likely to consent to do so. ...
I first began talking with children when I was a child. Their thoughts informed me and helped to trigger my imagination as my learning and development grew and took the directions that it did. As an early childhood educator, I continued to talk to children, about how they perceived the world and what they wanted and needed to learn. Fortunately, in my experiences as an early childhood teacher, curricular decision-making has been open enough to allow me, operating quietly in one classroom, to accommodate children’s voices. I am pleased that the Queensland Early Years Curriculum is written as a play-based, collaborative one that encourages children’s input into their learning. This is based on the Reggio Emilia and new sociology of childhood’s conceptualisation of the agentic child.
The agentic child is capable and competent, learning and growing through interaction with others (Corsaro 1997). Within this construct, childhood has social standing of its own; children are positioned as ’being‘ rather than ’becoming‘ (James, Jenkins and Prout 1998). Gandini (1993, in QSA, 2006) states: ‘[c]hildren are strong, rich and capable. All children have preparedness, potential, curiosity and interest in constructing their learning, negotiating with everything their environment brings to them’ (p. 10). Adults — such as teachers and parents — become co-learners who negotiate, challenge and guide while sharing power with children (Woodrow 1999). Research or any other relationship between adults and children is with children rather than about them. Power is negotiated between the researcher and child participants in data collection (Fasoli 2001). Children’s voices are given serious consideration (Sorin 2003).
... Addressing the effects of UI reforms on occupational trajectories raises also a statistical challenge for future research, which is related with the estimation of policy effects when the treatment and control groups differ in their observable and unobservable characteristics. Using DD-matching algorithms may offer a way to deal with these problems by providing a nonparametric estimate of the causal effects (Heckman, Ichimura andTodd 1997, 1998;Imbens 2004;Winship and Sobel 2004;Dias et al. 2008). DD-matching algorithms correct for the problem of differences among the control and treatment groups by defining the causal effect that is to be estimated as the (average) difference between observed outcomes among those affected and the weighted average of observed outcomes among those not affected groups. ...
... First, the present findings are based on self-reported data. Another limitation relates to the issue of causation, a limitation facing many non-experimental studies (Winship & Sobel, 2004). Other predictor variables (e.g., adult monitoring and perception of instrumentality of academic tasks) might have an effect on homework management strategies had they been included. ...
The aim of this study was to examine whether student achievement and school location may influence a range of homework management strate-gies. The participants were 633 rural and urban students in Grade 8. These homework management strategies include: (a) setting an appropriate work environment, (b) managing time, (c) handling distraction, (d) monitoring mo-tivation, and (e) controlling negative emotion. Compared with low-achieving students, high-achieving students reported more frequently working to man-age their workspace, budget time, handle distraction, monitor motivation, and control emotion while doing homework. Urban middle school students, com-pared with their rural counterparts, reported being more self-motivated during homework.
... Another limitation relates to the issue of causation, an issue that faces virtually all non-experimental research (Winship and Sobel 2004). Although much care was taken to control for possible confounding variables, other predictor variables such as quality of homework and students' employment status might have had an effect on HMM had they been included. ...
This study examines models of variables posited to predict students’ homework motivation management (HMM), based on survey data from 866 8th graders (61 classes) and 745 11th graders (46 classes) in the south-eastern USA. Most of the variance in HMM occurred at the student level, with parent education as the only significant predictor at the class level. At the student level, HMM was positively associated with family help, peer-oriented reasons, learning-oriented reasons, homework interest, arranging the environment and managing time. Girls (compared with boys) and Blacks (compared with Whites) were more likely to manage homework motivation. The gender difference in HMM was more pronounced among Blacks (compared with Whites) and among 8th graders (compared with 11th graders).
... How do we adjust for these factors in examining the relative likelihood that a woman serving in the military will have a birth? Matching is a powerful, nonparametric alternative to the regression model that isolates the effects of a variable-in our case, military service-to the domain of covariates where the treatment is most prevalent (Smith, 1997;Winship & Morgan, 1999;Winship & Sobel, 2004). We examined fertility differences between military women and civilian women who were, on average, alike across a range of factors that also impinge on fertility. ...
Although female employment is associated with lower levels of completed fertility in the civilian world, we find family formation rates among U.S. military women to be comparatively high. We compare enlisted women with civilian women using the National Longitudinal Survey of Youth (N = 3,547), the only data set to measure simultaneously the nuptiality and fertility of both populations. Using propensity score matching, we show that the fertility effect derives primarily from early marriage in the military, a surprisingly ‘‘family-friendly’’ institution. This shows that specific organizational and economic incentives in a working environment may offset the more widespread contemporary social and economic factors that otherwise depress marriage and fertility.
... Second, the findings were based on a cross-sectional survey, rather than repeated measures at different time points. Another related limitation relates to the issue of causation, a limitation facing virtually all nonexperimental research (Winship & Sobel, 2004). Although much care was taken to control for possible confounding variables (informed by research and theorizing on time management), other predictor variables might have had an effect on groupwork time management had they been included. ...
The aim of this study is to propose and empirically examine multilevel models of students' time management in online collaborative groupwork. Student- and group-level predictors of groupwork time management were analyzed in a survey of 204 graduate students from 61 groups in the Southeast of U.S. Results from the multilevel analyses revealed that most of the variance in time management occurred at the student level, with peer-oriented reasons being the only significant predictor at the group level. Results further revealed that groupwork time management was positively related to feedback and students' efforts in arranging the study environment.
... Future research is needed to continue to unpack the specific underlying causal mechanisms. Smith 1997;Sobel 1995;Winship and Sobel 2004). The rationale is that an unbiased estimate of the treatment effect may be obtained on observational data by conditioning on the "propensity score"-the probability for an individual being assigned to the "treatment" (Rosenbaum and Rubin 1983). ...
... First, the present findings are based on self-reported data. Another limitation relates to the issue of causation, a limitation facing many non-experimental studies (Winship & Sobel, 2004). There is need to examine the use and usefulness of science homework management strategies across rural and urban settings at the high school level, as the role of educational aspirations in science homework behaviour may be more pronounced at this level. ...
This study was an attempt to test if students differed in their self-concept, attitude and
their perception of the usefulness of Physics and Chemistry in terms of their type of
school and location. To test this hypothesis, t-test was used to compare differences in the
mean scores in terms of type of school and location. Pearson's Product Moment
Correlation coefficient was used to determine the relationships between independent
and dependent variable. Four hundred and forty six (446) students constituted the
sample. The attitude towards Physics and Chemistry and the perception of the type of
school showed there was no significant difference between single-sex and coeducational
school. There was no significant difference between single- sex and coeducational
school (mixed) students in their self- concept in Physics and Chemistry.
There was no significant difference between single- sex and co- educational school
(mixed) students in their perception of the usefulness of Physics and Chemistry.
Results also show that urban students have better perception of the usefulness of
Physics and Chemistry than the rural students. The difference in perception of the
usefulness of Chemistry is statistically significant but the difference in perception of the
usefulness of Physics is not statistically significant. Both rural and urban students had
positive attitude and self-concept for Physics and Chemistry.
... 7 I draw here from Henry Brady's four theories of causality: neo-Humean regularity theory, manipulation theory, counterfactual theory, and mechanisms and capacities Brady (2002). For further reading on causation in the social sciences see Gerring (2005), Holland (1986), Marini and Singer (1988), McKim and Turner (1997), Pearl (2000), Ringer (1989), Sobel (1995), Thompson (2003), Waldner (2002), Wendt (1998), Winship and Sobel, (2004). ...
... The discussion of the counterfactual model and propensity score matching in this paper follows the discussion presented inMorgan and Winship (2007). See alsoWinship and Morgan (1999),Winship and Sobel (2004), andMorgan and Harding (2006). ...
Abstract Private cram schooling is prevalent in Taiwan. Students go to ,cram schools or seek private tutoring after regular school hours in order ,to receive extra learning or gain a competitive,edge. Since cram schooling is believed to have,positive effects on learning achievement, which in turn will affectstratification process, the present paper attempts to answer ,the following questions: (1) What factors influence students’ participation in math ,cramming? ,(2) Does cram schooling for math work? (3) If it works, how big is the average effect? (4) What kinds of student benefit most and least from math cramming?, Using data gathered by Taiwan Education Panel Study (TEPS) in 2001 and 2003, the present research employs,the method,ofpropensity score matching to estimate the average,treatment effect of the ,9 graders ,who ,participated in math ,cramming programs. The present research finds that family backgrounds,and the previous math performance,would influence the chances of students’ participation in cram schooling
... Kahneman Tversky Slovic (1982) Lau Redlawsk(2001, 953) Holland (1986,959) (looking for the cause of an effect) (studying the effect of a cause) 2008,3 (causal effect) (Rosenbaum 2002;Rubin 1974) Morgan Winship(2007, 280) 13 (instrumental variable) Morgan Winship(2007, 193-200) 14 CMC (ignorability) (Winship and Sobel 2004) 2012 13 Holland(1986, 947) (fundamental problem of causal inference) Rosenbaum Rubin(1983) i ...
Abstract will be provided by author.
... Another limitation relates to the issue of causation, a limitation facing virtually all nonexperimental research (Winship & Sobel, 2004). Although much care was taken to control for the possible confounding variables (informed by interest theories and theoretical models of homework), other predictor variables might have had an effect on homework interest had they Xu been included. ...
This aim of this study was to test empirical models of variables posited to predict homework interest at the secondary school level. Student- and class-level predictors of homework interest were analyzed in a survey of 1,046 8th graders from 63 classes and of 849 11th graders from 48 classes. Most of the variance in homework interest occurred at the student level, with grade level appearing as the only significant predictor at the class level. At the student level, the variation in homework interest was positively associated with affective attitude toward homework, motivational orientation toward homework, student initiative in monitoring homework motivation, teacher feedback, and self-reported grade. Girls reported statistically significant higher scores in homework interest than did boys.
China provides a critical case for how social connections affect political trust because of the unique guanxi networks and the symbiosis between the state and non-governmental associations. Based on survey data from urban China, findings from both standard models and auxiliary IV analysis propose that: (1) informal social ties impair political trust since guanxi networking offers provisional publics for the transmission of political stimuli; (2) formal social ties foster political trust because associations and the state coexist in a state-dominant symbiotic sphere that brings about political trust assimilations by helping associations realize their agendas and interests.
This chapter provides the nationally representative evidence on the relationship between religion and subjective well-being for the case of China. Research on Western societies tends to find a positive association between being religious and level of well-being. This chapter hypothesizes to find a positive association between religion and well-being in China too, but argues social capital, for which strong evidence is often found in Western societies, is unlikely to be an important mechanism because religion in China is generally non-congregational. Instead, the private and subjective dimension of religion matters for well-being in China by helping adherents have an improved sense of social status relative to the non-religious in the context of rapid social change and growing inequality.KeywordsSocial capitalSubjective social statusSubjective well-beingReligionReligiosity
Using data from twenty-two provinces in China, this chapter assesses whether a rural-to-urban migrant’s wage in China is positively influenced by the number of fellow villagers who migrated to urban areas. Heckman’s two-stage method is used to correct for sample selection. Natural disaster in the village of origin is used as an instrumental variable to deal with other endogeneity biases. The results presented here verify significant effects of origin-based networks, and show that network effects are underestimated without taking the endogeneity into account.KeywordsSocial capitalRural-to-urban migrantsJob searchWagesAcquaintance networkSample selection
This chapter returns to the hotly debated concept of social capital. Scholars have mostly emphasized its structuralism as the attribute of embedded resources, neglecting its constructivism, which inevitably led to a missing link in dialectically understanding social capital as a practical concept. This chapter uses the specific case of interactions between General Gengyao Nian (Nian) and Emperor Yongzheng (Yong) in the Qing Dynasty of China to further explain the concept. Moreover, Pierre Bourdieu’s important terms “habitus” and “field” are introduced to address the paradoxes splitting the historical reality and theoretic expatiation. Primitive explanatory models are proposed to probe the genesis, attributes, and measurement of social capital.
This chapter systematically reviews the endogeneity problem and model identification issues in existing studies on the labour market role of social capital, following three research lines focusing on using contacts, used social capital and accessed social capital. This chapter shows that identifying causality in social capital research calls for improving model specification and data collection, exploiting exogenous variables, clearly specifying working assumptions and replication studies.KeywordsCausal effectsSocial capitalLabour marketModel identificationLiterature review
Since endogeneity problem is pervasive in identifying the causal effects of social capital, this chapter discusses four major sources of estimation bias derived from endogeneity, which are: omitted variable bias, self-selection bias, sample-selection bias, and simultaneity bias. Focusing on these identification challenges, this chapter also explains how to apply advanced approaches to deal with them according to examples in labour economics and sociological literature.KeywordsEndogeneitySocial capitalOmitted variable biasSelf-selection biasSample-selection biasSimultaneity bias
The present investigation linked grade, gender, and maths achievement to homework management strategies using data from 305 Chinese students in grades 7, 8, and 9. These strategies included arranging the environment, managing time, handling distraction, monitoring motivation, and controlling potentially interfering emotion. A three-way MANOVA examined the effects of grade, gender, and maths achievement on homework management strategies. Grade or gender was not related to homework management strategies. Meanwhile, high-achieving students (compared with low-achieving students) were more likely to arrange the environment, manage time, handle distraction, monitor motivation, and control negative emotion.
The literature on causality is wide-ranging, difficult at times, and controversial.1 Rather than delving into all of the nuances of this material, this chapter and the next have a much more modest goal. Through the analysis of examples, these chapters aim to sensitize the reader to various conceptions of causality discussed by social and statistical scientists. With this overview in mind, the reader of this book will be better able to critique assertions about causal effects in the subsequent chapters and in other research reports. By reading the cited material, the reader can gain a more detailed understanding of causality.
Drawing on the contextual, evaluative, and summarizing studies of this book, this chapter explicates 11 uses for multilevel models and defines relevant vocabulary, concepts, and notational conventions. Because multilevel models are composed of both fixed and random components, statisticians refer to them as mixed models (Littell et al. 2006). Because multilevel models focus on data at different hierarchical levels, educational researchers refer to them as hierarchical models (Raudenbush and Bryk 2002). In the contextual analyses of data at one point in time, the level-1 response variable and its covariates are conceptualized as being grouped (or contained) within the level-2 units. In the analyses of data at several points in time, the level-1 response variable is an observation at a time point on an entity, and the repeated observations on that entity are said to be grouped or contained by that entity. The entity (e.g., a person, an organization, a country) is the level-2 unit. Ideally, multilevel models assess change on disaggregated data at several points in time (e.g., the scores on the repeated assessments of individual students who are grouped into classrooms). When the chapters of this book model aggregated data, it is because the disaggregated data are not available.1 By applying special cases of generalized linear mixed models—the Poisson and logit—some chapters model response variables that are not normally distributed. By applying multilevel models, all of the core chapters address the clustering of level-1 units when they are contained within level-2 units.
The complexity of actual cause and effect relationships in social life can lead quickly to confused thinking and muddled discussions. Helpful here are distinctions that allow one to speak about some causes as different from others. Our chapter describes several distinctions among causes that we find especially useful for social science. First, taking a broad view of what “causes” are, we discuss some issues concerning whether causes are manipulable or preventable. Then, we consider the distinction between proximal and distal causes, connecting these to concepts of mediation and indirect effects. Next, we propose ways that concepts related to the distinction between necessary and sufficient causes in case-oriented research may be also useful for quantitative research on large samples. Afterward, we discuss criteria for characterizing one cause as more important than another. Finally, we describe ultimate and fundamental causes, which do not concern the relationship between an explanatory variable and outcome so much as the causes of properties of the systems in which more concrete causal relationships exist.
The past decade has seen an explosion in the availability and use of biomarkers data as a result of innovative discoveries and recent development of new biological and molecular techniques. Biomarkers are essential for at least four key purposes in biomedical research and public health practice: they are used for disease detection, diagnosis, prognosis, to identify patients who are most likely to benefit from selected therapies, and to guide clinical decision making. Determining the predictive and diagnostic value of biomarkers, singly and in combination, is essential to their being used effectively, and this has spurred the development of new statistical methodologies to assess the relationship between biomarkers and clinical outcomes. In this paper, we review both standard and novel statistical methods used for biomarker selection. We focus on techniques that could be readily applied to HIV infection research. Particular attention is given to deriving variable importance measures based on doubly robust methods such as targeted maximum likelihood estimation. We conclude by providing an example of the application of three novel techniques to a dataset from the Hormonal Contraception and HIV Genital Shedding and Disease Progression Study (GS Study) to select, among many candidate biomarkers, the best subset that is significantly associated with a CD4 cell decline to less than 350 cells/mm 3.
This paper is an empirical assessment of the average effect of tax consultants on non-business taxpayers’ tax burden. The author uses German income tax return data to identify tax returns filed by a tax consultant. In contrast to previous research, propensity score matching is used to construct an appropriate control group to eliminate problems arising from self-selection. The effect of tax consultants is found to be twofold and depends on the income level. On the one hand, tax consultants reduce the tax burden of clients on a high income. Possible reasons for this are the progressive income tax rate as well as income-dependent deductions, e.g., donations, which offer tax consultants greater scope to make use of their superior tax knowledge. However, this effect is reversed for clients on comparatively low incomes. This finding is in line with the interpretation that, besides the already identified disburdening impact, a tax consultant can also act as a preceding enforcement instance that mitigates clients’ excessively aggressive filing strategies.
While most work on causation in ethnography addresses the normative question of what ethnographers should do, this article addresses the empirical question of what ethnographers actually do. Specifically, it investigates whether ethnographic articles make causal arguments and how these arguments are made. The authors draw on a content analysis of 48 ethnographic articles sampled from four groups of sociological journals: contemporary generalist journals, contemporary specialist journals, mid-20th-century generalist journals—all in the United States—and contemporary generalist journals in Mexico. They find that ethnographies in U.S. contemporary generalist journals are most likely to advance strong and central causal claims and to use logical and rhetorical devices comparable to those used in quantitative articles. They also find that most Mexican ethnographic articles undertake a different kind of project, which they call “shedding light” on social phenomena. In addition to offering one methodological and one substantive suggestion to account for these findings, the authors highlight their implications for the sociology of social science.
Fixed-term contracts have become very relevant in the transition from school to work. Using data from the German Socio-Economic Panel (GSOEP) for the period 1984–2006, this article analyses differences in the timing of receiving a fixed-term contract or a permanent contract throughout the duration of first-job search and whether fixed-term contracts are associated with lower initial wages. Competing risk duration models reveal that school leavers initially receive more often permanent contracts but that a certain proportion also accepts temporary jobs. For transitions to both fixed-term and permanent contracts, we find that the longer the time spent searching for a job, the lower the transition probability to a job. The negative duration dependence effect is even more pronounced for fixed-term employment. Comparing labour market entrants with similar individual and job characteristics using propensity score matching techniques show that entrants earn significantly less in fixed-term jobs compared to permanent employment. Wage penalties are slightly larger for those who actually choose fixed-term contracts in their first job, whereas randomly allocated job entrants would suffer lower wage losses. Especially graduates from tertiary education suffer high initial wage losses in fixed-term contracts.
In accordance to Boudon's structural-individualistic action model, it is the aim of this paper to investigate the casual effects of individual abilities, resources of the parental home, socially selective educational transitions as well as teaching and learning conditions in schools on the development of reading literacy and its dispersion in respect of social origin. It is assumed that the socially selective transition into the secondary school tracks contributes to the general deterioration of the mean reading literacy when controlling for social origin and individual achievement. The socially selective transition from primary to secondary school contributes to the increase of social inequality of reading literacy existing already since the start of the pupils’ schooling. Since it is not possible to isolate these causes with comparative-static cross-sectional data like PIRLS or PISA empirically, we use a design of generating quasi-longitudinal data with three time references, namely reading literacy at the point of enrolment into elementary school, at the age of 9–10 years, and at the age of 15 years by the pair wise matching of “synthetic twins” with identical status criteria. The empirical analysis of the German data of PISA 2000 and PIRLS 2001 confirm the effects of the individual abilities depending on social origin, the social selectivity of transition into the secondary schools, the sorting and selection performances of the school system, the allocation of children into different learning contexts, and the schooling of both the individual development of reading competences and the social inequality of reading literacy.
Can education increase an individual’s IQ? This has been one of the most incendiary and controversial questions in the social sciences in the past few decades. The greatest firestorm occurred after the publication of Arthur Jensen’s 1969 article in The Harvard Education Review, “How Much Can We Boost IQ and Scholastic Achievement?”¹ The controversy was further fueled by Richard Herrnstein’s 1971 Atlantic Monthly article, “IQ.”² Then, after smoldering for two decades, the passion and acrimony reignited with publication in 1994 of The Bell Curve by Richard Herrnstein and Charles Murray.
Inference of latent feature models in the Bayesian nonparametric setting is generally difficult, especially in high dimensional settings, because it usually requires proposing features from some prior distribution. In special cases, where the integration is tractable, we could sample feature assignments according to a predictive likelihood. However, this still may not be efficient in high dimensions. We present a novel method to accelerate the mixing of latent variable model inference by proposing feature locations from the data, as opposed to the prior. This sampling method is efficient for proper mixing of the Markov chain Monte Carlo sampler, computationally attractive because this method can be performed in parallel, and is theoretically guaranteed to converge to the posterior distribution as its limiting distribution.
The recent literature on evaluating manpower training programs demonstrates that alternative nonexperimental estimators of the same program produce a array of estimates of program impact. These findings have led to the call for experiments to be used to perform credible program evaluations. Missing in all of the recent pessimistic analyses of nonexperimental methods is any systematic discussion of how to choose among competing estimators. This paper explores the value of simple specification tests in selecting an appropriate nonexperimental estimator. A reanalysis of the National Supported Work Demonstration Data previously analyzed by proponents of social experiments reveals that a simple testing procedure eliminates the range of nonexperimental estimators that are at variance with the experimental estimates of program impact.
The propensity score is the conditional probability of assignment to a particular treatment given a vector of observed covariates. Both large and small sample theory show that adjustment for the scalar propensity score is sufficient to remove bias due to all observed covariates. Applications include: (i) matched sampling on the univariate propensity score, which is a generalization of discriminant matching, (ii) multivariate adjustment by subclassification on the propensity score where the same subclasses are used to estimate treatment effects for all outcome variables and in all subpopulations, and (iii) visual representation of multivariate covariance adjustment by a two- dimensional plot.
Reduction of class size to increase academic achievement is a policy option that is currently of great interest. Although the results of small-scale randomized experiments and some interpretations of large-scale econometric studies point to positive effects of small classes, the evidence has been seen by some scholars as ambiguous. Project STAR in Tennessee, a 4-year, large-scale randomized experiment on the effects of class size, provided persuasive evidence that small classes had immediate effects on academic achievement. However, it was not clear whether these effects would persist over time as the children returned to classes of regular size or would fade, as have the effects of most other early education interventions. This article reports analyses of a 5-year follow-up of the students in that experiment. The analyses described here suggest that class size effects persist for at least 5 years and remain large enough to be important for educational policy. Thus, small classes in early grades appear to have lasting benefits.
David Card and Alan B. Krueger have already made national news with their pathbreaking research on the minimum wage. Here they present a powerful new challenge to the conventional view that higher minimum wages reduce jobs for low-wage workers. In a work that has important implications for public policy as well as for the direction of economic research, the authors put standard economic theory to the test, using data from a series of recent episodes, including the 1992 increase in New Jersey's minimum wage, the 1988 rise in California's minimum wage, and the 1990-91 increases in the federal minimum wage. In each case they present a battery of evidence showing that increases in the minimum wage lead to increases in pay, but no loss in jobs. A distinctive feature of Card and Krueger's research is the use of empirical methods borrowed from the natural sciences, including comparisons between the "treatment" and "control" groups formed when the minimum wage rises for some workers but not for others. In addition, the authors critically reexamine the previous literature on the minimum wage and find that it, too, lacks support for the claim that a higher minimum wage cuts jobs. Finally, the effects of the minimum wage on family earnings, poverty outcomes, and the stock market valuation of low-wage employers are documented. Overall, this book calls into question the standard model of the labor market that has dominated economists' thinking on the minimum wage. In addition, it will shift the terms of the debate on the minimum wage in Washington and in state legislatures throughout the country.
Matched sampling is a standard technique in the evaluation of treatments in observational studies. Matching on estimated propensity scores comprises an important class of procedures when there are numerous matching variables. Recent theoretical work (Rubin, D. B., and Thomas, N., 1992a, reprinted in this volume as Chapter 15) on affinely invariant matching methods with ellipsoidal distributions provides a general framework for evaluating the operating characteristics of such methods. Moreover, Rubin and Thomas (1992b, reprinted in this volume as Chapter 16) uses this framework to derive several analytic approximations under normality for the distribution of the first two moments of the matching variables in samples obtained by matching on estimated linear propensity scores. Here we provide a bridge between these theoretical approximations and actual practice. First, we complete and refine the nomal-based analytic approximations, thereby making it possible to apply these results to practice. Second, we perform Monte Carlo evaluations of the analytic results under normal and nonnormal ellipsoidal distributions, which confirm the accuracy of the analytic approximations, and demonstrate the predictable ways in which the approximations deviate from simulation results when normal assumptions are violated within the ellipsoidal family. Third, we apply the analytic approximations to real data with clearly nonellipsoidal distributions, and show that the thoretical expressions, although derived under artificial distributional conditions, produce useful guidance for practice. Our results delineate the wide range of settings in which matching on estimated linear propensity scores performs well, thereby providing useful information for the design of matching studies.
We outline a framework for causal inference in settings where assignment to a binary treatment is ignorable, but compliance with the assignment is not perfect so that the receipt of treatment is nonignorable. To address the problems associated with comparing subjects by the ignorable assignment - an "intention-to-treat analysis" - we make use of instrumental variables, which have long been used by economists in the context of regression models with constant treatment effects. We show that the instrumental variables (IV) estimand can be embedded within the Rubin Causal Model (RCM) and that under some simple and easily interpretable assumptions, the IV estimand is the average causal effect for a subgroup of units, the compliers. Without these assumptions, the IV estimand is simply the ratio of intention-to-treat causal estimands with no interpretation as an average causal effect. The advantages of embedding the IV approach in the RCM are that it clarifies the nature of critical assumptions needed for a causal interpretation, and moreover allows us to consider sensitivity of the results to deviations from key assumptions in a straightforward manner. We apply our analysis to estimate the effect of veteran status in the Vietnam era on mortality, using the lottery number that assigned priority for the draft as an instrument, and we use our results to investigate the sensitivity of the conclusions to critical assumptions.
Two-stage least squares (TSLS) is widely used in econometrics to estimate parameters in systems of linear simultaneous equations and to solve problems of omitted-variables bias in single-equation estimation. We show here that TSLS can also be used to estimate the average causal effect of variable treatments such as drug dosage, hours of exam preparation, cigarette smoking, and years of schooling. The average causal effect in which we are interested is a conditional expectation of the difference between the outcomes of the treated and what these outcomes would have been in the absence of treatment. Given mild regularity assumptions, the probability limit of TSLS is a weighted average of per-unit average causal effects along the length of an appropriately defined causal response function. The weighting function is illustrated in an empirical example based on the relationship between schooling and earnings.
The abstract for this document is available on CSA Illumina.To view the Abstract, click the Abstract button above the document title.
Rubin's model for causal inference in experiments and observational studies is enlarged to analyze the problem of "causes causing causes" and is compared to path analysis and recursive structural equations models. A special quasi-experimental design, the encouragement design, is used to give concreteness to the discussion by focusing on the simplest problem that involves both direct and indirect causation. It is shown that Rubin's model extends easily to this situation and specifies conditions under which the parameters of path analysis and recursive structural equations models have causal interpretations.
In a recent paper in Mind
1 Prof. Arthur W. Burks has made an interesting proposal for the construction of a logic of causal propositions. Such a logic is greatly needed, for however completely the word ‘cause’ has been eliminated from epistemology, it is still very much a part of the working vocabulary of most empirical scientists. Prof. Burks begins, wisely, by avoiding the Humean controversy, and asks instead whether it is possible to give a clear and consistent explication within a system of logic of the term ‘cause’ as it is used in common speech. While I am wholly in accord with Prof. Burks’ suggestion that such an explication is possible and useful, my own investigations into the question have led me to a definitional proposal that is somewhat different from his.
Written by one of the preeminent researchers in the field, this book provides a comprehensive exposition of modern analysis of causation. It shows how causality has grown from a nebulous concept into a mathematical theory with significant applications in the fields of statistics, artificial intelligence, economics, philosophy, cognitive science, and the health and social sciences. Judea Pearl presents and unifies the probabilistic, manipulative, counterfactual, and structural approaches to causation and devises simple mathematical tools for studying the relationships between causal connections and statistical associations. The book will open the way for including causal analysis in the standard curricula of statistics, artificial intelligence, business, epidemiology, social sciences, and economics. Students in these fields will find natural models, simple inferential procedures, and precise mathematical definitions of causal concepts that traditional texts have evaded or made unduly complicated. The first edition of Causality has led to a paradigmatic change in the way that causality is treated in statistics, philosophy, computer science, social science, and economics. Cited in more than 5,000 scientific publications, it continues to liberate scientists from the traditional molds of statistical thinking. In this revised edition, Judea Pearl elucidates thorny issues, answers readers’ questions, and offers a panoramic view of recent advances in this field of research. Causality will be of interests to students and professionals in a wide variety of fields. Anyone who wishes to elucidate meaningful relationships from data, predict effects of actions and policies, assess explanations of reported events, or form theories of causal understanding and causal speech will find this book stimulating and invaluable.
Problems involving causal inference have dogged at the heels of statistics since its earliest days. Correlation does not imply causation, and yet causal conclusions drawn from a carefully designed experiment are often valid. What can a statistical model say about causation? This question is addressed by using a particular model for causal inference (Holland and Rubin 1983; Rubin 1974) to critique the discussions of other writers on causation and causal inference. These include selected philosophers, medical researchers, statisticians, econometricians, and proponents of causal modeling.
Adjustments for bias in observational studies are not always confined to variables that were measured prior to treatment. Estimators that adjust for a concomitant variable that has been affected by the treatment are generally biased. The bias may be written as the sum of two easily interpreted components: one component is present only in observational studies; the other is common to both observational studies and randomized experiments. The first component of bias will be zero when the affected posttreatment concomitant variable is, in a certain sense, a surrogate for an unobserved pretreatment variable. The second component of bias can often be addressed by an appropriate sensitivity analysis.
Two-stage least squares (TSLS) is widely used in econometrics to estimate parameters in systems of linear simultaneous equations and to solve problems of omitted-variables bias in single-equation estimation. We show here that TSLS can also be used to estimate the average causal effect of variable treatments such as drug dosage, hours of exam preparation, cigarette smoking, and years of schooling. The average causal effect in which we are interested is a conditional expectation of the difference between the outcomes of the treated and what these outcomes would have been in the absence of treatment. Given mild regularity assumptions, the probability limit of TSLS is a weighted average of per-unit average causal effects along the length of an appropriately defined causal response function. The weighting function is illustrated in an empirical example based on the relationship between schooling and earnings.
Typically, using data from a nonexperimental study, quantitative sociologists model one or more outcomes as a function of independent variables, interpreting the resulting parameter estimates as effects. This article compares the usual approach to causal inference in sociology with an alternative approach that builds explicitly on a counterfactual account of causation. The comparison is relevant because sociologists typically interpret (incorrectly) parameter estimates obtained using the first approach as supporting causal statements that are counterfactual. To make matters concrete, the author reconsiders an attainment model of Featherman and Hauser, who are interested in comparing the effects of family background on achievement, by sex and time. This analysis suggests the coefficients in the regressions of respondent's education and occupational status on background should not be interpreted as effects. However, because a child's sex is determined without regard for his or her subsequent achievements, sex may be viewed as randomly assigned, justifying treating sex as the cause and the background variables as covariates. Unfortunately, the way in which the data were collected preclude such a treatment.
This chapter is concerned with a particular formalization that has proved useful in empirical work, hence the juxtaposition of causality and inference. It also bears close relation to notions of strictly exogenous and predetermined variables, which have considerable operational significance in statistical inference, and to the concepts of causal orderings and reliability which are important in model construction in econometrics and engineering. Causality is defined in terms of predictability; it cannot be an acceptable definition of causation for most philosophers of science. The chapter focuses on the operational usefulness of the definition in the construction, estimation, and application of econometric models. It considers the logical relationships among Wiener–Granger causality, Simon's definition of causal ordering, the engineer's criterion of reliability, and the concept of structure. It also discusses that unidirectional causality from X to Y is not equivalent to the assertion that X is predetermined in a particular behavioral relationship whose parameters are to be estimated. It further focuses on parameterization problems, processes that are nonautoregressive or have deterministic components or are nonstationary, and inference about many variables.
Matching to control for covariates in the estimation of treatment effects is not common in sociology, where multivariate data are most often analyzed using multiple regression and its generalizations. Matching can be a useful way to estimate these effects, especially when the treatment condition is comparatively rare in a population, and controls are numerous but mostly unlike the treatment cases. Matching on numerous covariates is abetted by the estimation of propensity scores, or functions of the probability that cases are treatments rather than controls. This procedure is illustrated in the estimation of the effects of an organizational innovation on Medicare mortality within hospitals; the data set is very large, but innovative hospitals few, and many of the remaining hospitals are quite unlike the hospitals constituting the treatment subsample. Results are based on a variance-components model that is extended to consider the effects of an additional covariate. They show effects of the organizational innovation comparable to those estimated via multiple regression models but with substantially reduced standard errors.
Using data from a two-stage probability sample of U.S. high school students, an attempt is made to estimate the effect that dropping out has on cognitive achievement test scores. Each sampled dropout from a school is matched by a multivariate procedure to a student who remained in the same school. The matched pair differences are then adjusted using analysis of covariance. The possibility that important covariates have been omitted from the analysis is addressed through tests of ignorable treatment assignment and through sensitivity analyses.
The recent literature on evaluating manpower training programs demonstrates that alternative nonexperimental estimators of the same program produce an array of estimates of program impact. These findings have led to the call for experiments to be used to perform credible program evaluations. Missing in all of the recent pessimistic analyses of nonexperimental methods is any systematic discussion of how to choose among competing estimators. This article explores the value of simple specification tests in selecting an appropriate nonexperimental estimator. A reanalysis of the National Supported Work Demonstration data previously analyzed by proponents of social experiments reveals that a simple testing procedure eliminates the range of nonexperimental estimators at variance with the experimental estimates of program impact.
Social scientists never have access to true experimental data of the type sometimes available to laboratory scientists.1 Our inability to use laboratory methods to independently vary treatments to eliminate or isolate spurious channels of causation places a fundamental limitation on the possibility of objective knowledge in the social sciences. In place of laboratory experimental variation, social scientists use subjective thought experiments. Assumptions replace data. In the jargon of modern econometrics, minimal identifying assumptions are invoked.
This paper is about the logic of interpreting recursive causal theories in sociology. We review the distinction between associations and effects and discuss the decomposition of effects into direct and indirect components. We then describe a general method for decomposing effects into their components by the systematic application of ordinary least squares regression. The method involves successive computation of reduced-form equations, beginning with an equation containing only exogenous variables, then computing equations which add intervening variables in sequence from cause to effect. This generates all the information required to decompose effects into their various direct and indirect parts. This method is a substitute for the often more cumbersome computation of indirect effects from the structural coefficients (direct effects) of the causal model. Finally, we present a way of summarizing this information in tabular form and illustrate the procedures using an empirical example.
When assignment to treatment group is made solely on the basis of the value of a covariate, X, effort should be concentrated on estimating the conditional expectations of the dependent variable Y given X in the treatment and control groups. One then averages the difference between these conditional expectations over the distribution of X in the relevant population. There is no need for concern about "other" sources of bias, e.g., unreliability of X, unmeasured background variables. If the conditional expectations are parallel and linear, the proper regression adjustment is the simple covariance adjustment. However, since the quality of the resulting estimates may be sensitive to the adequacy of the underlying model, it is wise to search for nonparallelism and nonlinearity in these conditional expectations. Blocking on the values of X is also appropriate, although the quality of the resulting estimates may be sensitive to the coarseness of the blocking employed. In order for these techniques to be useful in practice, there must be either substantial overlap in the distribution of X in the treatment groups or strong prior information.
In an observational study, detecting hidden bias involves checking that treatment effects appear where they should, and not elsewhere. For instance, treated and control groups are often compared with respect to outcomes the treatment should not affect. This paper uses such a test for bias to obtain a confidence set for an unobserved covariate. The impact of this unobserved covariate is indicated by a sensitivity analysis with the covariate confined to the confidence set. In this way, a test for bias may indicate either the presence and magnitude of a hidden bias or else a reduction in the sensitivity to bias.
If treatment assignment is strongly ignorable, then adjustment for observed covariates is sufficient to produce consistent estimates of treatment effects in observational studies. A general approach to testing this critical assumption is developed and applied to a study of the effects of nuclear fallout on the risk of childhood leukemia. R.A. Fisher's advice on the interpretation of observational studies was “Make your theories elaborate”; formally, make causal theories sufficiently detailed that, under the theory, strongly ignorable assignment has testable consequences.
The human propensity to think in causal terms is well known (Young 1978), and the manner in which judgments about causation are made in everyday life has been studied extensively by psychologists (Einhorn and Hogarth 1986; White 1990). No doubt this propensity contributes, for better or worse, to the persistence of causal language in scientific discourse, despite some influential attempts (for example, Russell 1913) to banish such talk to the prescientific era.