The Landscape of Causal Inference: Perspective From Citation Network Analysis

ABSTRACT

Causal inference is a fast-growing multidisciplinary field that has drawn extensive interests from statistical sciences and health and social sciences. In this article, 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 article 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. Supplementary materials for this article are available online.

KEYWORDS:

• Big network
• Causal inference
• Citation network
• ERGM

Supplementary Material

The supplementary materials contain additional tables and figures.

Acknowledgments

The authors would like to thank Dr. Jian Xu for the help with data collection. The authors are grateful to Professor Nicole Lazar and anonymous reviewers for the valuable comments on earlier drafts of this article.

Notes

1 To assess the quality of this name measure, we extract a random sample of 300 records. We find 34 common east Asian names, among which 24 (71%) are correctly coded as ones while the other 10 are incorrectly coded as zeros. Among the 266 non-East Asian names, 246 (92.4%) are correctly coded as zeros. In general, the measurement error tends to bias the estimated effect of this variable toward zero.

2 We use the Times Higher Education World University Ranking, because of its comprehensive coverage of universities in the world. The data are available at http://www.timeshighereducation.co.uk/world-university-rankings/2014-15/world-ranking.

3 To provide sensitivity analysis, we also fit the first ERGM on the core network. That the results do not significantly differ from those based on the full network indicates that the results from the second ERGM based on the core network may be (reasonably) generalizable to the full network.

4 When there are no endogenous network formation processes in the model, like in the first ERGM, PMLE is equivalent to MCMLE. Note that if MCMLE describes the true model, PMLE may under-estimate endogenous network formation processes and provide more conservative inferences (i.e., wider confidence intervals) on covariate effects (van Duijn, Gile, and Handcock 2009).

5 Figure A3 provides some assessment of the goodness of fit of Model 3 (Goodreau et al. 2008). The model seems to have captured structural features of the network well, as there are only a few places at which the observed statistics lie out of the 95% confidence intervals of the statistics in simulated networks.