No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
The Landscape of Causal Inference: Perspective From Citation Network Analysis
The American Statistician
Abstract
Causal inference is a fast-growing multidisciplinary field that has drawn extensive interests from statistical sciences and health and social sciences. In this paper, we gather comprehensive information on publications and citations in causal inference and provide a review of the field from the perspective of citation network analysis. We provide descriptive analyses by showing the most cited publications, the most prolific and the most cited authors, and structural properties of the citation network. Then we examine the citation network through exponential random graph models (ERGMs). We show that both technical aspects of the publications (e.g., publication length, time and quality) and social processes such as homophily (the tendency to cite publications in the same field or with shared authors), cumulative advantage, and transitivity (the tendency to cite references' references), matter for citations. We also provide specific analysis of citations among the top authors in the field and present a ranking and clustering of the authors. Overall, our paper reveals new insights into the landscape of the field of causal inference and may serve as a case study for analyzing citation networks in a multidisciplinary field and for fitting ERGMs on big networks.
... An abundance of studies exist analyzing various aspects of success in science (see, e.g., [1,2] and references therein). Most of these focus on the correlation between success and citation counts [3][4][5][6][7], productivity [8][9][10][11], collaboration [11][12][13][14], cumulative advantages also known as the Matthew effect [8,[15][16][17], and research networks [3,18,19]. The issue of gender in science and its correlation with success has also attracted a lot of attention [9,[19][20][21][22][23][24][25][26][27], as has the rise of teams in research [28][29][30][31], and other indirect factors such as the prestige of advisors, institutions, reputation of researchers, etc. [7,[32][33][34][35][36][37]. ...
... An abundance of studies exist analyzing various aspects of success in science (see, e.g., [1,2] and references therein). Most of these focus on the correlation between success and citation counts [3][4][5][6][7], productivity [8][9][10][11], collaboration [11][12][13][14], cumulative advantages also known as the Matthew effect [8,[15][16][17], and research networks [3,18,19]. The issue of gender in science and its correlation with success has also attracted a lot of attention [9,[19][20][21][22][23][24][25][26][27], as has the rise of teams in research [28][29][30][31], and other indirect factors such as the prestige of advisors, institutions, reputation of researchers, etc. [7,[32][33][34][35][36][37]. ...
... We focused on variables directly related to the production of scientific knowledge: Author productivity, impact, and collaboration. These variables have been found to be strongly correlated with researcher career trajectories in numerous other studies [3][4][5][6][7][8][9][10][11][12][13][14]. Specifically, we investigate the early productivity rate prod rate (r), early collaborator rate col rate (r) and early citation rate cit rate (r), defined for each author r as ...
We study the career lengths of researchers in theoretical high-energy physics from 1950 to 2020. Using a cohort-based analysis and bibliometric data from 30,149 authors in three physics disciplines we observe a dramatic increase in the ratio of academic dropouts over time. Where nearly all authors prior to the 1980s remained active more than a decade after their initial publication, this was the case for 50% or less by the 2010s. Consequently, the fraction of theoretical physicists who achieve full academic careers have diminished significantly in the last four decades. Additionally, we study correlations between author survivability and potential success factors, inferring that early author productivity and collaborative efforts have become increasingly strong determinants over time. A logistic regression is utilized to interpret the empirical findings through a survival model. Performance measures further indicate that early career collaboration and productivity rates have emerged as reliable predictors of author survivability during the period in which competition has intensified. Finally, we discuss the potential impacts that these developments may have on science itself.
... In this paper, apart from presenting detailed descriptive analyses of the citation networks, we also use ERGMs to study how technical features of the patents and social processes influence citation formations. Similar to the work done by An and Ding [51], we model the effects of a list of covariates we extracted from patents distinguishing between receiving and sending citations. In this study, we provide theoretical expectations on the effects of the covariates and discuss how different patent characteristics can matter for citations. ...
... This statistical methodology allows a detailed analysis of the network data and is particularly suitable for hypothesis testing. Nevertheless, as has been pointed out by An and Ding [51], studying citation networks using ERGMs poses several challenges. It is a known fact that it is difficult to fit ERGMs on large networks. ...
... For larger networks, the solution of Eq (3) cannot be found in a reasonable time. Researchers either have to study smaller sub-networks or to use very crude approximations, like for example, contrastive divergence or pseudolikelihood [51,71]. Recently a very efficient algorithm for the solution of Eq (3) was proposed [72,73]. ...
Patent Citation Analysis has been gaining considerable traction over the past few decades. In this paper, we collect extensive information on patents and citations and provide a perspective of citation network analysis of patents from a statistical viewpoint. We identify and analyze the most cited patents, the most innovative and the highly cited companies along with the structural properties of the network by providing in-depth descriptive analysis. Furthermore, we employ Exponential Random Graph Models (ERGMs) to analyze the citation networks. ERGMs enables understanding the social perspectives of a patent citation network which has not been studied earlier. We demonstrate that social properties such as homophily (the inclination to cite patents from the same country or in the same language) and transitivity (the inclination to cite references’ references) together with the technicalities of the patents (e.g., language, categories), has a significant effect on citations. We also provide an in-depth analysis of citations for sectors in patents and how it is affected by the size of the same. Overall, our paper delves into European patents with the aim of providing new insights and serves as an account for fitting ERGMs on large networks and analyzing them. ERGMs help us model network mechanisms directly, instead of acting as a proxy for unspecified dependence and relationships among the observations.
... This implies that network descriptives of patents are independent of future network additions, while they are dependent on all past citations. This implies a temporal causality (An and Ding 2018), which suggests the survival analysis approach is optimal for capturing the dynamic development of knowledge production within patent subdomains. ...
We explore a dynamic patent citation network model to explain the established link between network structure and technological improvement rate. This model, a type of survival model, posits that the *dynamic* network structure determines the *constant* improvement rate, requiring consistent structural reproduction over time. The model's hazard rate, the probability of a patent being cited, represents "knowledge production," reflecting the output of new patents given existing ones. Analyzing hydrogen technology patents, we find distinct subdomain knowledge production rates, but consistent development across subdomains. "Distribution" patents show the lowest production rate, suggesting dominant "distribution" costs in pricing. Further modeling shows Katz-centrality predicts knowledge production, outperforming subdomain classification. Lower Katz centrality in "distribution" suggests inherent organizational differences in invention. Exploitative learning (within-subdomain citations) correlates with higher patenting opportunity costs, potentially explaining slower "distribution" development, as high investment needs may incentivize monopolization over knowledge sharing.
... Finally, we investigated the sociodemographic characteristics (gender, country of origin, and the prestige of institution) of the authors of foundational papers (Section S12). We find that hidden citations capture a universal phenomenon that can emerge in any institution, regardless of their level of prestige, and we observe no statistically significant bias based on gender or country of origin in hidden citations apart from the overall biases in explicit citations that have long been observed (32)(33)(34)(35). Hence, although the methodology presented in the paper can account for additional hidden citations, it simply reflects the existing biases. ...
References, the mechanism scientists rely on to signal previous knowledge, lately have turned into widely used and misused measures of scientific impact. Yet, when a discovery becomes common knowledge, citations suffer from obliteration by incorporation. This leads to the concept of hidden citation, representing a clear textual credit to a discovery without a reference to the publication embodying it. Here, we rely on unsupervised interpretable machine learning applied to the full text of each paper to systematically identify hidden citations. We find that for influential discoveries hidden citations outnumber citation counts, emerging regardless of publishing venue and discipline. We show that the prevalence of hidden citations is not driven by citation counts, but rather by the degree of the discourse on the topic within the text of the manuscripts, indicating that the more discussed is a discovery, the less visible it is to standard bibliometric analysis. Hidden citations indicate that bibliometric measures offer a limited perspective on quantifying the true impact of a discovery, raising the need to extract knowledge from the full text of the scientific corpus.
... ERGMs are widely used for citation networks analysis. An & Ding (2018) performed the same study in the special case of publications on causal inference. They argued that some technical and social processes are underpinning citation networks. ...
The uncertainty underlying real-world phenomena has attracted attention toward statistical analysis approaches. In this regard, many problems can be modeled as networks. Thus, the statistical analysis of networked problems has received special attention from many researchers in recent years. Exponential Random Graph Models, known as ERGMs, are one of the popular statistical methods for analyzing the graphs of networked data. ERGM is a generative statistical network model whose ultimate goal is to present a subset of networks with particular characteristics as a statistical distribution. In the context of ERGMs, these graph’s characteristics are called statistics or configurations. Most of the time they are the number of repeated subgraphs across the graphs. Some examples include the number of triangles or the number of cycle of an arbitrary length. Also, any other census of the graph, as with the edge density, can be considered as one of the graph’s statistics. In this review paper, after explaining the building blocks and classic methods of ERGMs, we have reviewed their newly presented approaches and research papers. Further, we have conducted a comprehensive study on the applications of ERGMs in many research areas which to the best of our knowledge has not been done before. This review paper can be used as an introduction for scientists from various disciplines whose aim is to use ERGMs in some networked data in their field of expertise.
... Averaging (2), that is, ...
In this article, we study the hypothesis testing of the blip / net effects of treatments in a treatment sequence. We illustrate that the likelihood ratio test and the score test may suffer from the curse of dimensionality, the null paradox and the high-dimensional constraint on standard parameters under the null hypothesis. On the other hand, we construct the Wald test via a small number of point effects of treatments in single-point causal inference. We show that the Wald test can avoid these problems under the same assumptions as the Wald test for testing the point effect of treatment. The simulation study illustrates that the Wald test achieves the nominal level of type I error and a low level of type II error. A real medical example illustrates how to conduct the Wald test in practice.
... The importance of properly analyzing data and the impact of improper analyses on the validity of study results and their replicability is well documented (An & Ding, 2018;Field, 2013;Freedman, 2009;Hahn & Meeker, 1993;Heiman, 2011;Tabachnick & Fidell, 2007). The National Council of Teachers of Mathematics (NCTM, 2000) provided standards for data analysis to assist researchers that included advice to state research questions and collect appropriate data to answer them. ...
Data analysis is a significant methodological component when conducting quantitative education studies. Guidelines for conducting data analyses in quantitative education studies are common but often underemphasize four important methodological components impacting the validity of inferences: quality of constructed measures, proper handling of missing data, proper level of measurement of a dependent variable, and model checking. This paper highlights these components for novice researchers to help ensure statistical inferences are valid. We used empirical examples involving contingency tables, group comparisons, regression analysis, and multilevel modelling to illustrate these components using the Program for International Student Assessment (PISA) data. For every example, we stated a research question and provided evidence related to the quality of constructed measures since measures with weak reliability and validity evidence can bias estimates and distort inferences. The adequate strategies for handling missing data were also illustrated. The level of measurement for the dependent variable was assessed and the proper statistical technique was utilized accordingly. Model residuals were checked for normality and homogeneity of variance. Recommendations for obtaining stronger inferences and reporting related evidence were also illustrated. This work provides an important methodological resource for novice researchers conducting data analyses by promoting improved practice and stronger inferences.
... Since randomizing treatment assignments can present serious practical and ethical problems, analytic strategies to infer causal effects from observational data have proliferated across disciplines, from medical sociology and health economics, to psychology, epidemiology, and the clinical and health services sciences. Strategies used to meet challenges described in Section 2.2 have varied including those based on potential outcomes and counterfactuals [14,15], doubly robust estimation based on correctly specified missing data or complete data models [24], matching [25], propensity scores [26,27], marginal structural models [28], instrumental approaches [29], control functions [30], test for robustness of endogenous predictors [19], latent variable models [31], regression modeling [32][33][34][35], citation network analysis [36], and graphic frameworks [16,37], such as causal Directed Acyclic Graphs [DAG] [38][39][40], also known as Bayesian or Causal networks. DAGs are representations of the DGP where predictor variables, including confounders, interactive terms, and intermediate outcomes, are directed forward to impact the outcome variable of interest so that predictors are acyclic and do not cycle back on themselves. ...
In the era of big data and cloud computing, analysts need statistical models to go beyond predicting outcomes to forecasting how outcomes change when decision-makers intervene to change one or more causal factors. This paper reviews methods to estimate the causal effects of treatment choices on patient health outcomes using observational datasets. Methods are limited to those that model choice of treatment (propensity scoring) and treatment outcomes (instrumental variable, difference in differences, control function). A regression framework was developed to show how unobserved confounding covariates and heterogeneous outcomes can introduce biases to effect size estimates. In response to criticisms that outcome approaches are not systematic and subject to model misspecification error, we extend the control function approach of Lu and White by applying Best Approximating Model technology (BAM-CF). Results from simulation experiments are presented to compare biases between BAM-CF and propensity scoring in the presence of an unobserved confounder. We conclude no one strategy is ‘optimal’ for all datasets, and analyst should consider multiple approaches to assess robustness. For both observational and randomized datasets, researchers should assess how moderating covariates impact estimates of treatment effect sizes so that clinicians can understand what is best for each individual patient.
Citation behaviour is the source driver of scientific dynamics, and it is essential to understand its effect on knowledge diffusion and intellectual structure. This study explores the effect of citation behaviour on disciplinary knowledge diffusion and intellectual structure by comparing three types of citation behaviour trends, namely the high citation trend, medium citation trend, and low citation trend. The diffusion power, diffusion speed, and diffusion breadth were calculated to quantify knowledge diffusion. The properties of the global and local citation network structure were used to reflect the particular influences of citation behaviour on the scientific intellectual structure. The primary empirical results show that (a) the high citation behaviour trend could improve the knowledge diffusion speed for papers with a short citation history span. Additionally, the medium citation trend has the broadest diffusion breadth whereas the low citation behaviour trend might make the citation counts take off for papers with a long citation history span; (b) the high citation trend has a stronger influence and greater control over the intellectual structure, but this relationship is true only for papers with a short or normal citation history span. These findings could play important roles in scientific research evaluation and impact prediction.
With the wide application of statistics, it is important to identify research trends and the development of statistics. In this paper, we analyze a citation network of the top 4 statistical journals from 2001 to 2018, applying the directed spectral clustering on the ratio-of-eigenvectors (D-SCORE) method to detect the community structure of citation network. We find that statistical researchers are becoming more and more collaborative. The number of influential papers which account for the majority of citations is small. High betweenness centrality and high closeness centrality papers are concentrated in Annals of Statistics (AoS). Furthermore, we detect 4 communities and 11 sub-communities such as "High-dimensional Model", "Variable Selection", and "Covariance Matrix Analysis". Then, we compare the results of D-SCORE with three other methods and find that D-SCORE is more suitable for our citation network. Finally, we identify the dynamic nature of the communities. Our findings present trends and topological patterns of statistical papers, and the data set provides a fertile ground for future research on social networks.
Through collaboration, scientists’ human and social capital are accumulated that are considered important in the academic job market. However, little is known about whether academic past collaboration influence scientists’ mobility. To deal with treatment endogeneity, we conduct a Generalized Propensity Score analysis (GPS) and apply a novel application of the Dose-Response Function model. Using the data on 15,968 Chinese scientists from 2000 to 2012 as an illustrative case, we find that 1) the number of domestic and overseas collaborators are positively associated with scientists’ mobility and upward move, while the magnitude of the effect of overseas collaborators is far smaller than that of domestic collaborators; 2) domestic collaborators’ productivity is positively related to scientists’ move and upward move; 3) there is a stronger effect of collaborators from higher-tier universities on scientists’ upward move; 4) we do not observe a significant relationship between the recent stock of collaborators and scientists’ mobility. In addition to implications for talent policies and scientists’ career development, this study makes significant methodological contributions through introducing a new method, GPS, to address selection bias of the independent variable, i.e., scientists’ collaboration. Our results show that, with great potential to capture causality, GPS facilitates research in informetrics, scientometrics and science policy from a quantitative perspective, and enriches policy relevance of the findings.
The scientific research behavior of scholars is the core issue of scientific research. The research ideas and methods of complex networks provide a new perspective for the study of science. The scientific citation network and the scientist cooperation network are widely used to study the citation behavior of scholars and the dissemination of scientific ideas, and so far, some results have been obtained. However, due to the lack of information on the content of the article, the research based solely on the network topology has limitations and deficiencies. Combining the textual content analysis through LDA, this paper studies the distribution characteristics of content correlation between articles with citation relations and its evolution with time. It found that the distribution of citation distance has normal characteristics, but the reference distance is visible to be short. Authors have citation preferences for documents at a distance.
LARF is an R package that provides instrumental variable estimation of treatment effects when both the endogenous treatment and its instrument (i.e., the treatment inducement) are binary. The method (Abadie 2003) involves two steps. First, pseudo-weights are constructed from the probability of receiving the treatment inducement. By default LARF estimates the probability by a probit regression. It also provides semiparametric power series estimation of the probability and allows users to employ other external methods to estimate the probability. Second, the pseudo-weights are used to estimate the local average response function conditional on treatment and covariates. LARF provides both least squares and maximum likelihood estimates of the conditional treatment effects.
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.
This chapter discusses the use of directed acyclic graphs (DAGs) for causal inference in the observational social sciences. It focuses on DAGs’ main uses, discusses central principles, and gives applied examples. DAGs are visual representations of qualitative causal assumptions: They encode researchers’ beliefs about how the world works. Straightforward rules map these causal assumptions onto the associations and independencies in observable data. The two primary uses of DAGs are (1) determining the identifiability of causal effects from observed data and (2) deriving the testable implications of a causal model. Concepts covered in this chapter include identification, d-separation, confounding, endogenous selection, and overcontrol. Illustrative applications then demonstrate that conditioning on variables at any stage in a causal process can induce as well as remove bias, that confounding is a fundamentally causal rather than an associational concept, that conventional approaches to causal mediation analysis are often biased, and that causal inference in social networks inherently faces endogenous selection bias. The chapter discusses several graphical criteria for the identification of causal effects of single, time-point treatments (including the famous backdoor criterion), as well identification criteria for multiple, time-varying treatments.
Science is a complex system. Building on Latour's actor network theory, we model published science as a dynamic hypergraph and explore how this fabric provides a substrate for future scientific discovery. Using millions of abstracts from MEDLINE, we show that the network distance between biomedical things (i.e., people, methods, diseases, chemicals) is surprisingly small. We then show how science moves from questions answered in one year to problems investigated in the next through a weighted random walk model. Our analysis reveals intriguing modal dispositions in the way biomedical science evolves: methods play a bridging role and things of one type connect through things of another. This has the methodological implication that adding more node types to network models of science and other creative domains will likely lead to a superlinear increase in prediction and understanding.
Nearly a century of empirical research examines how neighborhood properties influence a host of phenomena such as crime, poverty, health, civic engagement, immigration, and economic inequality. Theoretically bundled within these neighborhood effects are institutions' and actors' social networks that are the foundation of other neighborhood-level processes such as social control, mobilization, and cultural assimilation. Yet, despite such long-standing theoretical links between neighborhoods and social networks, empirical research rarely considers or measures dimensions of geography and social network mechanisms simultaneously. The present study seeks to fill this gap by analyzing how both geography and social networks influence an important social problem in urban America: gang violence. Using detailed data on fatal and non-fatal shootings, we examine effects of geographic proximity, organizational memory, and additional group processes (e.g., reciprocity, transitivity, and status seeking) on gang violence in Chicago and Boston. Results show adjacency of gang turf and prior conflict between gangs are strong predictors of subsequent gang violence. Furthermore, important network processes, including reciprocity and status seeking, also contribute to observed patterns of gang violence. In fact, we find that these spatial and network processes mediate racial effects, suggesting the primacy of place and the group in generating gang violence.
Rankings of scholarly journals based on citation data are often met with skepticism by the scientific community. Part of the skepticism is due to disparity between the common perception of journals’ prestige and their ranking based on citation counts. A more serious concern is the inap- propriate use of journal rankings to evaluate the scientific influence of authors. This paper focuses on analysis of the table of cross-citations among a selection of Statistics journals. Data are collected from the Web of Science database published by Thomson Reuters. Our results suggest that mod- elling the exchange of citations between journals is useful to highlight the most prestigious journals, but also that journal citation data are characterized by considerable heterogeneity, which needs to be properly summarized. Inferential conclusions require care in order to avoid potential over-interpretation of insignificant differences between journal ratings. Comparison with published ratings of institutions from the UK’s Research Assessment Exercise shows strong correlation at aggregate level between assessed research quality and journal citation ‘export scores’ within the discipline of Statistics.
We have collected and cleaned two network data sets: Coauthorship and
Citation networks for statisticians. The data sets are based on all research
papers published in four of the top journals in statistics from 2003 to the
first half of 2012. We analyze the data sets from many different perspectives,
focusing on (a) productivity, patterns and trends, (b) centrality, and (c)
community structures, and present an array of interesting findings.
For (a), we find that over the 10-year period, both the average number of
papers per author and the fraction of self citations have been decreasing, but
the proportion of distant citations has been increasing. These suggest that the
statistics community has become increasingly more collaborative, competitive,
and globalized, driven by the boom of online resources and Search Engines.
For (b), we identify the most prolific, the most collaborative, and the most
highly cited authors. We also identify a handful of "hot" papers and suggest
"Variable Selection" and several other areas as the "hot" areas in statistics.
For (c), we have investigated the data sets with 6 different methods,
including D-SCORE (a new procedure we propose here). We have identified about
15 meaningful communities, including large-size communities such as
"Biostatistics", "Large-Scale Multiple Testing", "Variable Selection" as well
as small-size communities such as "Dimensional Reduction", "Objective Bayes",
"Quantile Regression", and "Theoretical Machine Learning". We find that the
community structures for the Coauthorship network and Citation network are
connected and intertwined, but are also different in significant ways.
Our findings shed light on research habits, trends, and topological patterns
of statisticians, and our data sets provide a fertile ground for future
researches on or related to social networks of statisticians.
The identification of causal peer effects (also known as social contagion or induction) from observational data in social networks is challenged by two distinct sources of bias: latent homophily and unobserved confounding. In this paper, we investigate how causal peer effects of traits and behaviors can be identified using genes (or other structurally isomorphic variables) as instrumental variables (IV) in a large set of data generating models with homophily and confounding. We use directed acyclic graphs to represent these models and employ multiple IV strategies and report three main identification results. First, using a single fixed gene (or allele) as an IV will generally fail to identify peer effects if the gene affects past values of the treatment. Second, multiple fixed genes/alleles, or, more promisingly, time-varying gene expression, can identify peer effects if we instrument exclusion violations as well as the focal treatment. Third, we show that IV identification of peer effects remains possible even under multiple complications often regarded as lethal for IV identification of intra-individual effects, such as pleiotropy on observables and unobservables, homophily on past phenotype, past and ongoing homophily on genotype, inter-phenotype peer effects, population stratification, gene expression that is endogenous to past phenotype and past gene expression, and others. We apply our identification results to estimating peer effects of body mass index (BMI) among friends and spouses in the Framingham Heart Study. Results suggest a positive causal peer effect of BMI between friends.
The term "interference" has been used to describe any setting in which one
subject's exposure may affect another subject's outcome. We use causal diagrams
to distinguish among three causal mechanisms that give rise to interference.
The first causal mechanism by which interference can operate is a direct causal
effect of one individual's treatment on another individual's outcome; we call
this direct interference. Interference by contagion is present when one
individual's outcome may affect the outcomes of other individuals with whom he
comes into contact. Then giving treatment to the first individual could have an
indirect effect on others through the treated individual's outcome. The third
pathway by which interference may operate is allocational interference.
Treatment in this case allocates individuals to groups; through interactions
within a group, individuals' characteristics may affect one another. In many
settings more than one type of interference will be present simultaneously. The
causal effects of interest differ according to which types of interference are
present, as do the conditions under which causal effects are identifiable.
Using causal diagrams for interference, we describe these differences, give
criteria for the identification of important causal effects, and discuss
applications to infectious diseases.
Has sociology become more socially integrated over the last 30 years? Recent work in the sociology of knowledge demonstrates a direct linkage between social interaction patterns and the structure of ideas, suggesting that scientific collaboration networks affect scientific practice. I test three competing models for sociological collaboration networks and find that a structurally cohesive core that has been growing steadily since the early 1960s characterizes the discipline's coauthorship network. The results show that participation in the sociology collaboration network depends on research specialty and that quantitative work is more likely to be coauthored than non-quantitative work. However, structural embeddedness within the network core given collaboration is largely unrelated to specialty area. This pattern is consistent with a loosely overlapping specialty structure that has potentially integrative implications for theoretical development in sociology.
Citation counts are central to many studies in the sociology of science. Due to cost considerations and the format of the Science Citation Index, researchers often collect citations only to a scientist's first-authored publications (so called 'straight counts'), rather than to complete counts based on all published papers. This paper examines the consequences of using straight counts to study the careers of a sample of PhD biochemists. It is found that even though these two measures are highly correlated, there is a significant chance of making substantive errors when straight counts are substituted for complete counts. The problem is found to be more severe when citations are the dependent variable rather than an independent variable. It is argued that our findings can be generalized to other fields where multiple authorship is common. Accordingly, extreme care must be used in interpreting results based on straight citation counts.
A new statistic — the modified coefficient of variation - is proposed for measuring the inequality of publication and citation counts. Its advantages include scale invariance, equal sensitivity at all levels of the distribution, allowance for random variation in publication or citation counts, and a close relationship to a probability distribution which fits observed productivity distributions. Also discussed are disciplinary differences in the distribution of productivity and the functional relationship between productivity and its determinants.
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.
Instrumental variable (IV) estimation typically requires the user to cor-rectly specify the relationship between the regressors and the outcome to obtain a consistent estimate of the effects of the treatments. This paper proposes dou-bly robust IV regression estimators that only require the user to either correctly specify the relationship between the measured confounding variables (i.e., included exogenous variables) and the outcome, or the relationship between the measured confounding variables and the IVs. We derive the asymptotic properties of the dou-bly robust IV regression estimators and investigate their finite sample properties in a simulation study. We apply our method to a study of the effect of education on earnings.
This paper uses Pierre Bourdieu’s field theory to develop tools for analyzing interdisciplinary scientific fields. Interdisciplinary
fields are scientific spaces where no single form of scientific capital has a monopoly and therefore multiple forms of scientific
capital constitute the structures and stakes of scientific competition. Scientists compete to accumulate and define forms
of scientific capital and also to set the rates of exchange between them. The paper illustrates this framework by applying
it to the interdisciplinary field of behavior genetics. Most behavior geneticists envision their participation in the field
as a means to compete for scientific capital in other fields. However, the scientific capital of behavior genetics has different
values for scientists attempting to deploy it in different neighboring fields. These values depend on situations in each field
and the ways behavior genetics mediates relationships among them. The pattern of relationships of exchange helps explain the
social hierarchy and several features of knowledge production within behavior genetics.
KeywordsBourdieu–Field theory–Scientific capital–Interdisciplinary science–Behavior genetics–Knowledge production
The instrumental variables (IV) method provides a way to estimate the causal effect of a treatment when there are unmeasured confounding variables. The method requires a valid IV, a variable that is independent of the unmeasured confounding variables and is associated with the treatment but which has no effect on the outcome beyond its effect on the treatment. An additional assumption often made is deterministic monotonicity, which says that for each subject, the level of the treatment that a subject would take is a monotonic increasing function of the level of the IV. However, deterministic monotonicity is sometimes not realistic. We introduce a stochastic monotonicity assumption, a relaxation that only requires a monotonic increasing relationship to hold across subjects between the IV and the treatments conditionally on a set of (possibly unmeasured) covariates. We show that under stochastic monotonicity, the IV method identifies a weighted average of treatment effects with greater weight on subgroups of subjects on whom the IV has a stronger effect. We provide bounds on the global average treatment effect under stochastic monotonicity and a sensitivity analysis for violations of stochastic monotonicity.We apply the methods to a study of the effect of premature babies being delivered in a high technology neonatal intensive care unit (NICU) vs. a low technology unit.
Drawing causal inference with observational studies is the central pillar of many disciplines. One sufficient condition for identifying the causal effect is that the treatment-outcome relationship is unconfounded conditional on the observed covariates. It is often believed that the more covariates we condition on, the more plausible this unconfoundedness assumption is. This belief has had a huge impact on practical causal inference, suggesting that we should adjust for all pretreatment covariates. However, when there is unmeasured confounding between the treatment and outcome, estimators adjusting for some pretreatment covariate might have greater bias than estimators without adjusting for this covariate. This kind of covariate is called a bias amplifier, and includes instrumental variables that are independent of the confounder, and affect the outcome only through the treatment. Previously, theoretical results for this phenomenon have been established only for linear models. We fill in this gap in the literature by providing a general theory, showing that this phenomenon happens under a wide class of models satisfying certain monotonicity assumptions. We further show that when the treatment follows an additive or multiplicative model conditional on the instrumental variable and the confounder, these monotonicity assumptions can be interpreted as the signs of the arrows of the causal diagrams.
In this chapter we provide an overview of the history of causal analysis in the social sciences. We review literature published from the mid-1800s to the present day, tracing the key strains of thought that lead to our current understandings of causal analysis in the social sciences. Given space limitations, we focus on three of the most important strands of causal analysis – those based on (1) constant conjunction and regularity accounts, (2) correlational and path analytic techniques, and (3) potential outcomes and counterfactual frameworks. We then return to the complexity of a Weberian approach, which contains nearly all of the elements of these three major frameworks into a single case-oriented method to causal analysis. We conclude by speculating on the future of causal analysis in the social sciences.
Most questions in social and biomedical sciences are causal in nature: what would happen to individuals, or to groups, if part of their environment were changed? In this groundbreaking text, two world-renowned experts present statistical methods for studying such questions. This book starts with the notion of potential outcomes, each corresponding to the outcome that would be realized if a subject were exposed to a particular treatment or regime. In this approach, causal effects are comparisons of such potential outcomes. The fundamental problem of causal inference is that we can only observe one of the potential outcomes for a particular subject. The authors discuss how randomized experiments allow us to assess causal effects and then turn to observational studies. They lay out the assumptions needed for causal inference and describe the leading analysis methods, including, matching, propensity-score methods, and instrumental variables. Many detailed applications are included, with special focus on practical aspects for the empirical researcher.
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.
An observational study is a nonexperimental investigation of the effects caused by a treatment. Unlike an experiment, in an observational study, the investigator does not control the assignment of treatments, with the consequence that the individuals in different treatment groups may not have been comparable prior to treatment. Analytical adjustments, such as matching, are used to remove overt bias, that is, pretreatment differences that are accurately measured and recorded. There may be pretreatment differences that were not recorded, called hidden biases, and addressing these is a central concern.
The demarcation of science from other intellectual activities-long an analytic problem for philosophers and sociologists-is here examined as a practical problem for scientists. Construction of a boundary between science and varieties of non-science is useful for scientists' pursuit of professional goals: acquisition of intellectual authority and career opportunities; denial of these resources to "pseudoscientists"; and protection of the autonomy of scientific research from political interference. "Boundary-work" describes an ideological style found in scientists' attempts to create a public image for science by contrasting it favorably to non-scientific intellectual or technical activities. Alternative sets of characteristics available for ideological attribution to science reflect ambivalences or strains within the institution: science can be made to look empirical or theoretical, pure or applied. However, selection of one or another description depends on which characteristics best achieve the demarcation in a way that justifies scientists' claims to authority or resources. Thus, "science" is no single thing: its boundaries are drawn and redrawn inflexible, historically changing and sometimes ambiguous ways.
The highly skewed distributions of productivity among scientists can be partly explained by a process of accumulative advantage. Because of feedback through recognition and resources, highly productive scientists maintain or increase their productivity, while scientists who produce very little produce even less later on. A major implication of accumulative advantage is that the distribution of productivity becomes increasingly unequal as a cohort of scientists ages. Cross-sectional survey data support this hypothesis for chemists, physicists, and mathematicians, who show strong linear increases in inequality with increasing career age. This increase is highly associated with a changing distribution of time spent on research. Another implication of accumulative advantage is also corroborated: the association among productivity, resources and esteem increases as career age increases.
Applied researchers are increasingly interested in whether and how treatment
effects vary in randomized evaluations, especially variation not explained by
observed covariates. We propose a model-free approach for testing for the
presence of such unexplained variation. To use this randomization-based
approach, we must address the fact that the average treatment effect, generally
the object of interest in randomized experiments, actually acts as a nuisance
parameter in this setting. We explore potential solutions and advocate for a
method that guarantees valid tests in finite samples despite this nuisance. We
also show how this method readily extends to testing for heterogeneity beyond a
given model, which can be useful for assessing the sufficiency of a given
scientific theory. We finally apply our method to the National Head Start
Impact Study, a large-scale randomized evaluation of a Federal preschool
program, finding that there is indeed significant unexplained treatment effect
variation.
Estimating peer effects with observational data is very difficult because of contextual confounding, peer selection, simultaneity bias, and measurement error, etc. In this paper, I show that instrumental variables (IVs) can help to address these problems in order to provide causal estimates of peer effects. Based on data collected from over 4,000 students in six middle schools in China, I use the IV methods to estimate peer effects on smoking. My design-based IV approach differs from previous ones in that it helps to construct potentially strong IVs and to directly test possible violation of exogeneity of the IVs. I show that measurement error in smoking can lead to both under- and imprecise estimations of peer effects. Based on a refined measure of smoking, I find consistent evidence for peer effects on smoking. If a student’s best friend smoked within the past 30 days, the student was about one fifth (as indicated by the OLS estimate) or 40 percentage points (as indicated by the IV estimate) more likely to smoke in the same time period. The findings are robust to a variety of robustness checks. I also show that sharing cigarettes may be a mechanism for peer effects on smoking. A 10% increase in the number of cigarettes smoked by a student’s best friend is associated with about 4% increase in the number of cigarettes smoked by the student in the same time period.
The instrumental variables (IV) method is a method for making causal
inferences about the effect of a treatment based on an observational study in
which there are unmeasured confounding variables. The method requires a valid
IV, a variable that is independent of the unmeasured confounding variables and
is associated with the treatment but which has no effect on the outcome beyond
its effect on the treatment. An additional assumption that is often made for
the IV method is deterministic monotonicity, which is an assumption that for
each subject, the level of the treatment that a subject would take if given a
level of the IV is a monotonic increasing function of the level of the IV.
Under deterministic monotonicity, the IV method identifies the average
treatment effect for the compliers (those subject who would take the treatment
if encouraged to do so by the IV and not take the treatment if not encouraged).
However, deterministic monotonicity is sometimes not realistic. We introduce a
stochastic monotonicity condition which relaxes deterministic monotonicity in
that it does not require that a monotonic increasing relationship hold within
subjects between the levels of the IV and the level of the treatment that the
subject would take if given a level of the IV, but only that a monotonic
increasing relationship hold across subjects between the IV and the treatment
in a certain manner. We show that under stochastic monotonicity, the IV method
identifies a weighted average of treatment effects with greater weight on
subgroups of subjects on whom the IV has a stronger effect. We provide bounds
on the global average treatment effect under stochastic monotonicity and a
sensitivity analysis for violations of the stochastic monotonicity assumption.
Network data arise in a wide variety of applications. Although descriptive statistics for networks abound in the literature, the science of fitting statistical models to complex network data is still in its infancy. The models considered in this article are based on exponential families; therefore, we refer to them as exponential random graph models (ERGMs). Although ERGMs are easy to postulate, maximum likelihood estimation of parameters in these models is very difficult. In this article, we first review the method of maximum likelihood estimation using Markov chain Monte Carlo in the context of fitting linear ERGMs. We then extend this methodology to the situation where the model comes from a curved exponential family. The curved exponential family methodology is applied to new specifications of ERGMs, proposed in an earlier article, having nonlinear parameters to represent structural properties of networks such as transitivity and heterogeneity of degrees. We review the difficult topic of implementing likelihood ratio tests for these models, then apply all these model-fitting and testing techniques to the estimation of linear and nonlinear parameters for a collaboration network between partners in a New England law firm.
A goal of many health studies is to determine the causal effect of a treatment or intervention on health outcomes. Often, it is not ethically or practically possible to conduct a perfectly randomized experiment, and instead, an observational study must be used. A major challenge to the validity of observational studies is the possibility of unmeasured confounding (i.e., unmeasured ways in which the treatment and control groups differ before treatment administration, which also affect the outcome). Instrumental variables analysis is a method for controlling for unmeasured confounding. This type of analysis requires the measurement of a valid instrumental variable, which is a variable that (i) is independent of the unmeasured confounding; (ii) affects the treatment; and (iii) affects the outcome only indirectly through its effect on the treatment. This tutorial discusses the types of causal effects that can be estimated by instrumental variables analysis; the assumptions needed for instrumental variables analysis to provide valid estimates of causal effects and sensitivity analysis for those assumptions; methods of estimation of causal effects using instrumental variables; and sources of instrumental variables in health studies. Copyright © 2014 John Wiley & Sons, Ltd.
The new higher order specifications for exponential random graph models introduced by Snijders et al. [Snijders, T.A.B., Pattison, P.E., Robins G.L., Handcock, M., 2006. New specifications for exponential random graph models. Sociological Methodology 36, 99–153] exhibit substantial improvements in model fit compared with the commonly used Markov random graph models. Snijders et al., however, concentrated on non-directed graphs, with only limited extensions to directed graphs. In particular, they presented a transitive closure parameter based on path shortening. In this paper, we explain the theoretical and empirical advantages in generalizing to additional closure effects. We propose three new triadic-based parameters to represent different versions of triadic closure: cyclic effects; transitivity based on shared choices of partners; and transitivity based on shared popularity. We interpret the last two effects as forms of structural homophily, where ties emerge because nodes share a form of localized structural equivalence. We show that, for some datasets, the path shortening parameter is insufficient for practical modeling, whereas the structural homophily parameters can produce useful models with distinctive interpretations. We also introduce corresponding lower order effects for multiple two-path connectivity. We show by example that the in- and out-degree distributions may be better modeled when star-based parameters are supplemented with parameters for the number of isolated nodes, sources (nodes with zero in-degrees) and sinks (nodes with zero out-degrees). Inclusion of a Markov mixed star parameter may also help model the correlation between in- and out-degrees. We select some 50 graph features to be investigated in goodness of fit diagnostics, covering a variety of important network properties including density, reciprocity, geodesic distributions, degree distributions, and various forms of closure. As empirical illustrations, we develop models for two sets of organizational network data: a trust network within a training group, and a work difficulty network within a government instrumentality.
Age dependence in organizational death rates is studied using data on three populations of organizations: national labor unions, semiconductor electronics manufacturers, and newspaper publishing companies. There is a liability of newness in each of these populations but it differs depending on whether death occurs through dissolution or by absorption through merger. Liabilities of smallness and bigness are also identified but controlling for them does not eliminate age dependence.
Productive scientists tend to hold jobs at prestigious university departments, but it is unclear whether this is because good departments hire the best scientists or because good departments encourage and facilitate research productivity. To resolve this issue, we studied the antecedents and consequences of 179 job changes by chemists, biologists, physicists, and mathematicians. Those who were upwardly mobile showed substantial increases in their rate of publication and in the rate of citation to those publications, while those who were downwardly mobile showed substantial decreases in productivity. Earlier analyses of these job changes found only a small effect of prior productivity on destination prestige. These results suggest that the effect of department affiliation on productivity is more important than the effect of productivity on departmental affiliation.
The hypothesis of cumulative advantage is widely accepted in the sociology of science, but empirical tests have been few and equivocal. One approach, originated by Allison and Stewart (1974), is to see whether inequality of productivity and recognition increases as a cohort of scientists ages. This paper extends their work by examining true cohorts of biochemists and chemists rather than synthetic cohorts. Increasing inequality is observed for counts of publications but not for counts of citations to all previous publications. It is also shown that a mathematical model of cumulative advantage does not imply increasing inequality. When the model is modified to allow for heterogeneity in the rate of cumulative advantage, however, increasing inequality is implied.
In this article we develop the theoretical properties of the propensity function, which is a generalization of the propensity score of Rosenbaum and Rubin. Methods based on the propensity score have long been used for causal inference in observational studies; they are easy to use and can effectively reduce the bias caused by nonrandom treatment assignment. Although treatment regimes need not be binary in practice, the propensity score methods are generally confined to binary treatment scenarios. Two possible exceptions have been suggested for ordinal and categorical treatments. In this article we develop theory and methods that encompass all of these techniques and widen their applicability by allowing for arbitrary treatment regimes. We illustrate our propensity function methods by applying them to two datasets; we estimate the effect of smoking on medical expenditure and the effect of schooling on wages. We also conduct simulation studies to investigate the performance of our methods.
In the Prospect Study, in 10 pairs of two primary-care practices, one practice was picked at random to receive a “depression care manager” to treat its depressed patients. Randomization inference, properly performed, reflects the assignment of practices, not patients, to treatment or control. Yet, pertinent data describe individual patients: depression outcomes, baseline covariates, compliance with treatment. The methods discussed use only (i) the random assignment of clusters to treatment or control and (ii) the hypothesis about effects being tested or inverted for confidence intervals, so they are randomization inferences in Fisher’s strict sense. There is no assumption that the covariance model generated the data, that compliers resemble noncompliers, that dependence is from additive random cluster effects, that individuals in a same cluster do not interfere with one another, or that units are sampled from a population. We contrast methods of covariance adjustment, never assuming the models are “true,” obtaining exact randomization inferences. We consider exact inference about effects proportional to doses with noncompliance and effects whose magnitude varies with the degree of improvement that would occur without treatment. A simulation examines power.
Similarity breeds connection. This principle - the homophily principle - structures network ties of every type, including marriage, friendship, work, advice, support, information transfer, exchange, comembership, and other types of relationship. The result is that people's personal networks are homogeneous with regard to many sociodemographic, behavioral, and intrapersonal characteristics. Homophily limits people's social worlds in a way that has powerful implications for the information they receive, the attitudes they form, and the interactions they experience. Homophily in race and ethnicity creates the strongest divides in our personal environments, with age, religion, education, occupation, and gender following in roughly that order. Geographic propinquity, families, organizations, and isomorphic positions in social systems all create contexts in which homophilous relations form. Ties between nonsimilar individuals also dissolve at a higher rate, which sets the stage for the formation of niches (localized positions) within social space. We argue for more research on: (a) the basic ecological processes that link organizations, associations, cultural communities, social movements, and many other social forms; (b) the impact of multiplex ties on the patterns of homophily; and (c) the dynamics of network change over time through which networks and other social entities co-evolve.
We use a trading metaphor to study knowledge transfer in the sciences as well
as the social sciences. The metaphor comprises four dimensions: (a) Discipline
Self-dependence, (b) Knowledge Exports/Imports, (c) Scientific Trading
Dynamics, and (d) Scientific Trading Impact. This framework is applied to a
dataset of 221 Web of Science subject categories. We find that: (i) the
Scientific Trading Impact and Dynamics of Materials Science And Transportation
Science have increased; (ii) Biomedical Disciplines, Physics, And Mathematics
are significant knowledge exporters, as is Statistics & Probability; (iii) in
the social sciences, Economics, Business, Psychology, Management, And Sociology
are important knowledge exporters; (iv) Discipline Self-dependence is
associated with specialized domains which have ties to professional practice
(e.g., Law, Ophthalmology, Dentistry, Oral Surgery & Medicine, Psychology,
Psychoanalysis, Veterinary Sciences, And Nursing).
The statistical modeling of social network data is difficult due to the complex dependence structure of the tie variables. Statistical exponential families of distributions provide a flexible way to model such dependence. They enable the statistical characteristics of the network to be encapsulated within an exponential family random graph (ERG) model. For a long time, however, likelihood-based estimation was only feasible for ERG models assuming dyad independence. For more realistic and complex models inference has been based on the pseudo-likelihood. Recent advances in computational methods have made likelihood-based inference practical, and comparison of the different estimators possible.In this paper, we present methodology to enable estimators of ERG model parameters to be compared. We use this methodology to compare the bias, standard errors, coverage rates and efficiency of maximum likelihood and maximum pseudo-likelihood estimators. We also propose an improved pseudo-likelihood estimation method aimed at reducing bias. The comparison is performed using simulated social network data based on two versions of an empirically realistic network model, the first representing Lazega's law firm data and the second a modified version with increased transitivity. The framework considers estimation of both the natural and the mean-value parameters.The results clearly show the superiority of the likelihood-based estimators over those based on pseudo-likelihood, with the bias-reduced pseudo-likelihood out-performing the general pseudo-likelihood. The use of the mean value parameterization provides insight into the differences between the estimators and when these differences will matter in practice.
This article provides an introductory summary to the formulation and application of exponentialrandomgraphmodels for socialnetworks. The possible ties among nodes of a network are regarded as random variables, and assumptions about dependencies among these random tie variables determine the general form of the exponentialrandomgraphmodel for the network. Examples of different dependence assumptions and their associated models are given, including Bernoulli, dyad-independent and Markov randomgraphmodels. The incorporation of actor attributes in social selection models is also reviewed. Newer, more complex dependence assumptions are briefly outlined. Estimation procedures are discussed, including new methods for Monte Carlo maximum likelihood estimation. We foreshadow the discussion taken up in other papers in this special edition: that the homogeneous Markov randomgraphmodels of Frank and Strauss [Frank, O., Strauss, D., 1986. Markov graphs. Journal of the American Statistical Association 81, 832–842] are not appropriate for many observed networks, whereas the new model specifications of Snijders et al. [Snijders, T.A.B., Pattison, P., Robins, G.L., Handock, M. New specifications for exponentialrandomgraphmodels. Sociological Methodology, in press] offer substantial improvement.
Social behavior over short time scales is frequently understood in terms of actions, which can be thought of as discrete events in which one individual emits a behavior directed at one or more other entities in his or her environment (possibly including himself or herself). Here, we introduce a highly flexible framework for modeling actions within social settings, which permits likelihood-based inference for behavioral mechanisms with complex dependence. Examples are given for the parameterization of base activity levels, recency, persistence, preferential attachment, transitive/cyclic interaction, and participation shifts within the relational event framework. Parameter estimation is discussed both for data in which an exact history of events is available, and for data in which only event sequences are known. The utility of the framework is illustrated via an application to dynamic modeling of responder radio communications during the early hours of the World Trade Center disaster.
The small-world phenomenon formalized in this article as the coinci-dence of high local clustering and short global separation, is shown to be a general feature of sparse, decentralized networks that are neither completely ordered nor completely random. Networks of this kind have received little attention, yet they appear to be widespread in the social and natural sciences, as is indicated here by three dis-tinct examples. Furthermore, small admixtures of randomness to an otherwise ordered network can have a dramatic impact on its dy-namical, as well as structural, properties—a feature illustrated by a simple model of disease transmission.
Propensity score matching estimators (Rosenbaum and Rubin, 1983) are widely used in evaluation research to estimate average treatment effects. In this article, we derive the large sample distribution of propensity score matching estimators. Our derivations take into account that the propensity score is itself estimated in a first step, prior to matching. We prove that first step estimation of the propensity score affects the large sample distribution of propensity score matching estimators. Moreover, we derive an adjustment to the large sample variance of propensity score matching estimators that corrects for first step estimation of the propensity score. In spite of the great popularity of propensity score matching estimators, these results were previously unavailable in the literature.
We attempt to clarify, and show how to avoid, several fallacies of causal inference in experimental and observational studies. These fallacies concern hypothesis tests for covariate balance between the treated and control groups, and the consequences of using randomization, blocking before randomization, and matching after treatment assignment to achieve balance. Applied researchers in a wide range of scientific disciplines seem to fall prey to one or more of these fallacies. To clarify these points, we derive a new three-part decomposition of the potential estimation errors in making causal inferences. We then show how this decomposition can help scholars from different experimental and observational research traditions better understand each other's inferential problems and attempted solutions. We illustrate with a discussion of the misleading conclusions researchers produce when using hypothesis tests to check for balance in experiments and observational studies.
Despite their popularity, conventional propensity score estimators (PSEs) do not take into account uncertainties in propensity scores. This paper develops Bayesian propensity score estimators (BPSEs) to model the joint likelihood of both propensity score and outcome in one step, which naturally incorporates such uncertainties into causal inference. Simulations show that PSEs using estimated propensity scores tend to overestimate variations in the estimates of treatment effects—that is, too often they provide larger than necessary standard errors and lead to overly conservative inference—whereas BPSEs provide correct standard errors for the estimates of treatment effects and valid inference. Compared with other variance adjustment methods, BPSEs are guaranteed to provide positive standard errors, more reliable in small samples, can be readily employed to draw inference on individual treatment effects, etc. To illustrate the proposed methods, BPSEs are applied to evaluating a job training program. Accompanying software is available on the author's website.
An instrument manipulates a treatment that it does not entirely control, but the instrument affects the outcome only indirectly through its manipulation of the treatment. The idealized prototype is the randomized encouragement design, in which subjects are randomly assigned to receive either encouragement to accept the treatment or no such encouragement, but not all subjects comply by doing what they are encouraged to do, and the situation is such that only the treatment itself, not disregarded encouragement alone, can affect the outcome. An instrument is weak if it has only a slight impact on acceptance of the treatment, that is, if most people disregard encouragement to accept the treatment. Typical applications of instrumental variables are not ideal; encouragement is not randomized, although it may be assigned in a far less biased manner than the treatment itself. Using the concept of design sensitivity, we study the sensitivity of instrumental variable analyses to departures from the ideal of random assignment of encouragement, with particular reference to the strength of the instrument. With these issues in mind, we reanalyze a clever study by Angrist and Krueger concerning the effects of military service during World War II on subsequent earnings, in which birth cohorts of very similar but not identical age were differently “encouraged” to serve in the war. A striking feature of this example is that those who served earned more, but the effect of service on earnings appears to be negative; that is, the instrumental variables analysis reverses the sign of the naive comparison. For expository purposes, this example has the convenient feature of enabling, by selecting different birth cohorts, the creation of instruments of varied strength, from extremely weak to fairly strong, although separated by the same time interval and thus perhaps similarly biased. No matter how large the sample size becomes, even if the effect under study is quite large, studies with weak instruments are extremely sensitive to tiny biases, whereas studies with stronger instruments can be insensitive to moderate biases.
Science is a social field of forces, struggles, and relationships that is defined at every moment by the relations of power among the protagonists. Scientific choices are guided by taken-for-granted assumptions, interactive with practices, as to what constitutes real and important problems, valid methods, and authentic knowledge. Such choices also are shaped by the social capital controlled by various positions and stances within the field. This complex and dynamic representation thus simultaneously rejects both the absolutist-idealist conception of the immanent development of science and the historicist relativism of those who consider science as purely a conventional social construct. The strategies used in science are at once social and intellectual; for example, strategies that are founded on implicit agreement with the established scientific order are thereby in affinity with the positions of power within the field itself. In established scientific fields of high autonomy, revolutions no longer are necessarily at the same time political ruptures but rather are generated within the field themselves: the field becomes the site of a permanent revolution. Under certain conditions, then, strategies used in struggles for symbolic power transcend themselves as they are subjected to the crisscrossing censorship that represents the constitutive reason of the field. The necessary and sufficient condition for this critical correction is a social organization such that each participant can realize specific interest only by mobilizing all the scientific resources available for overcoming the obstacles shared by all his or her competitors. Thus, the type of analysis here illustrated does not lead to reductive bias or sociologism that would undermine its own foundations. Rather it points to a comprehensive and reflexive objectivism that opens up a liberating collective self-analysis.
In observational studies with exposures or treatments that vary over time, standard approaches for adjustment of confounding are biased when there exist time-dependent confounders that are also affected by previous treatment. This paper introduces marginal structural models, a new class of causal models that allow for improved adjustment of confounding in those situations. The parameters of a marginal structural model can be consistently estimated using a new class of estimators, the inverse-probability-of-treatment weighted estimators.