A framework for causal inference from two-level factorial designs is proposed, which uses potential outcomes to define causal effects. The paper explores the effect of non-additivity of unit level treatment effects on Neyman's repeated sampling approach for estimation of causal effects and on Fisher's randomization tests on sharp null hypotheses in these designs. The framework allows for statistical inference from a finite population, permits definition and estimation of estimands other than ‘average factorial effects’ and leads to more flexible inference procedures than those based on ordinary least squares estimation from a linear model.
Experimental Designs for Identifying Causal Mechanisms
Experimentation is a powerful methodology that enables scientists to establish causal claims empirically. However, one important criticism is that experiments merely provide a black box view of causality and fail to identify causal mechanisms. Specifically, critics argue that, although experiments can identify average causal effects, they cannot explain the process through which such effects come about. If true, this represents a serious limitation of experimentation, especially for social and medical science research that strives to identify causal mechanisms. We consider several experimental designs that help to identify average natural indirect effects. Some of these designs require the perfect manipulation of an intermediate variable, whereas others can be used even when only imperfect manipulation is possible. We use recent social science experiments to illustrate the key ideas that underlie each of the designs proposed.
more » « less- PAR ID:
- 10401800
- Publisher / Repository:
- Oxford University Press
- Date Published:
- Journal Name:
- Journal of the Royal Statistical Society Series A: Statistics in Society
- Volume:
- 176
- Issue:
- 1
- ISSN:
- 0964-1998
- Format(s):
- Medium: X Size: p. 5-51
- Size(s):
- p. 5-51
- Sponsoring Org:
- National Science Foundation
More Like this
-
more » « less
Summary -
more » « less
Abstract Motivation Understanding causal effects is a fundamental goal of science and underpins our ability to make accurate predictions in unseen settings and conditions. While direct experimentation is the gold standard for measuring and validating causal effects, the field of causal graph theory offers a tantalizing alternative: extracting causal insights from observational data. Theoretical analysis has shown that this is indeed possible, given a large dataset and if certain conditions are met. However, biological datasets, frequently, do not meet such requirements but evaluation of causal discovery algorithms is typically performed on synthetic datasets, which they meet all requirements. Thus, real-life datasets are needed, in which the causal truth is reasonably known. In this work we first construct such a large-scale real-life dataset and then we perform on it a comprehensive benchmarking of various causal discovery methods.
Results We find that the PC algorithm is particularly accurate at estimating causal structure, including the causal direction which is critical for biological applicability. However, PC does only produces cause-effect directionality, but not estimates of causal effects. We propose PC-NOTEARS (PCnt), a hybrid solution, which includes the PC output as an additional constraint inside the NOTEARS optimization. This approach combines PC algorithm’s strengths in graph structure prediction with the NOTEARS continuous optimization to estimate causal effects accurately. PCnt achieved best aggregate performance across all structural and effect size metrics.
Availability and implementation https://github.com/zhu-yh1/PC-NOTEARS.
-
Abstract Causal effects of biodiversity on ecosystem functions can be estimated using experimental or observational designs — designs that pose a tradeoff between drawing credible causal inferences from correlations and drawing generalizable inferences. Here, we develop a design that reduces this tradeoff and revisits the question of how plant species diversity affects productivity. Our design leverages longitudinal data from 43 grasslands in 11 countries and approaches borrowed from fields outside of ecology to draw causal inferences from observational data. Contrary to many prior studies, we estimate that increases in plot-level species richness caused productivity to decline: a 10% increase in richness decreased productivity by 2.4%, 95% CI [−4.1, −0.74]. This contradiction stems from two sources. First, prior observational studies incompletely control for confounding factors. Second, most experiments plant fewer rare and non-native species than exist in nature. Although increases in native, dominant species increased productivity, increases in rare and non-native species decreased productivity, making the average effect negative in our study. By reducing the tradeoff between experimental and observational designs, our study demonstrates how observational studies can complement prior ecological experiments and inform future ones.more » « less
-
more » « less
Scientists seek to understand the causal processes that generate sustainability problems and determine effective solutions. Yet, causal inquiry in nature–society systems is hampered by conceptual and methodological challenges that arise from nature–society interdependencies and the complex dynamics they create. Here, we demonstrate how sustainability scientists can address these challenges and make more robust causal claims through better integration between empirical analyses and process- or agent-based modeling. To illustrate how these different epistemological traditions can be integrated, we present four studies of air pollution regulation, natural resource management, and the spread of COVID-19. The studies show how integration can improve empirical estimates of causal effects, inform future research designs and data collection, enhance understanding of the complex dynamics that underlie observed temporal patterns, and elucidate causal mechanisms and the contexts in which they operate. These advances in causal understanding can help sustainability scientists develop better theories of phenomena where social and ecological processes are dynamically intertwined and prior causal knowledge and data are limited. The improved causal understanding also enhances governance by helping scientists and practitioners choose among potential interventions, decide when and how the timing of an intervention matters, and anticipate unexpected outcomes. Methodological integration, however, requires skills and efforts of all involved to learn how members of the respective other tradition think and analyze nature–society systems.
-
more » « less
Abstract The mechanisms linking parental resources to their children’s opportunities are of fundamental interest to sociologists. This study is the first to provide a dynamic life course assessment of the origin–education–destination triangle using causal mediation analysis. While the age-as-equalizer hypothesis suggests that for the highly educated, direct effects of social origin are high at young ages and decrease over the life course, the age-as-stratifier hypothesis suggests that for individuals with low educational attainment, direct effects of social origin are low at young ages and increase over the life course. Findings using panel data from the German Socio-Economic Panel Study are largely consistent with these hypotheses for various measures of social origin and social destination. Overall, this study demonstrates how causal mediation analysis provides precise effect definitions that allow scholars to assess mechanisms when status transmission processes depend on educational attainment.
