Estimating causal effects of treatments in randomized and nonrandomized studies.

By Donald B. Rubin. 1974

A discussion of matching, randomization, random sampling, and other methods of controlling extraneous variation is presented. The objective is to specify the benefits of randomization in estimating causal effects of treatments. The basic conclusion is that randomization should be employed whenever possible but that the use of carefully controlled nonrandomized data to estimate causal effects is a reasonable and necessary procedure in many cases. Recent psychological and educational literature has included extensive criticism of the use of nonrandomized studies to estimate causal effects of treatments (e.g., Campbell & Erlebacher, 1970). The implication in much of this literature is that only properly randomized experiments can lead to useful estimates of causal effects. If taken as applying to all fields of study, this position is untenable. Since the extensive use of randomized experiments is limited to the last half century,8 and in fact is not used in much scientific investigation today,4 one is led to the conclusion that most scientific "truths" have been established without using randomized experiments. In addition, most of us successfully determine the causal effects of many of our everyday actions, even interpersonal behaviors, without the benefit of randomization. Even if the position that causal effects of treatments can only be well established from randomized experiments is taken as applying only to the social sciences in which

Subfields: Optimization & Training, Causal Inference & Discovery

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Optimization & Training

In Causal Inference & Discovery

Citations in the corpus

Concrete Problems in AI Safety (2016, background): “The first is counterfactual reasoning [106, 129, 117, 30], where one asks “what would have happened if the world were different in a certain way”? In some sense, distributional shift can be thought of as a particular type of counterfactual, and so understanding counterfactual reasoning is likely to ”
Causal Reasoning for Algorithmic Fairness (2018, background): “Another commonly used term is potential outcomes [34], a terminology reflecting that strictly speaking such outcomes are not truly counterfactual until an exposure effectively takes place.”
Invariant Risk Minimization (2019, positive): “Rubin’s ignorability [44] plays the same role.”
Counterfactual Normalization: Proactively Addressing Dataset Shift and Improving Reliability Using Causal Mechanisms (2018, methodology): “The proposed method involves the estimation of coun-terfactuals, which can be formalized using the Neyman-Rubin potential outcomes framework (Neyman, 1923; Rubin, 1974).”
Multi-Level Cause-Effect Systems (2015, methodology): “Standard approaches to causation using graphical models (Spirtes et al., 2000; Pearl, 2009) or potential outcomes (Rubin, 1974) presuppose this step—these methods focus on discovering the causal relations among a given set of well-defined causal variables.”
Metalearners for estimating heterogeneous treatment effects using machine learning (2017, methodology): “Framework and Definitions We use the Neyman–Rubin potential outcome framework (11, 12) and assume a superpopulation or distribution P from which a realization of N independent random variables is given as the training data.”
The Econometrics of Randomized Experiments (2016, methodology): “In part of the discussion in this chapter, therefore, we view the sample at hand as the full population of interest, following the approaches taken by Fisher (1925, 1935), Neyman (1935), and subsequently by Rubin (1974, 1978, 2007).”
Approximate residual balancing: debiased inference of average treatment effects in high dimensions (2016, background): “Our goal is to estimate average treatment effects in the potential outcomes framework, or Rubin causal model (Rubin, 1974; Imbens and Rubin, 2015).”
Opportunities and obstacles for deep learning in biology and medicine (2017, background): “Causal inference is often framed in terms of counterfactual question [267].”
Learning When-to-Treat Policies (2019, methodology): “We formulate causal effects in terms of potential outcomes (Neyman, 1923; Robins, 1986; Rubin, 1974).”
Learning Representations for Counterfactual Inference (2016, methodology): “The model described above is known as the Rubin-Neyman causal model (Rubin, 1974; 2011).”
Direct and Indirect Effects (2001, methodology): “The subscripted notation Yx is borrowed from the potential-outcome framework of Rubin (1974).”
Causal inference in statistics: An overview (2009, methodology): “…equation models (SEM) used in economics and social science (Goldberger, 1973; Duncan, 1975), the potentialoutcome framework of Neyman (1923) and Rubin (1974), and the graphical models developed for probabilistic reasoning and causal analysis (Pearl, 1988; Lauritzen, 1996; Spirtes et al., 2000;…”
Comment: Demystifying Double Robustness: A Comparison of Alternative Strategies for Estimating a Population Mean from Incomplete Data. (2008, background): “The notion of potential outcomes was introduced by Neyman (1923) for randomized experiments and by Rubin (1974a) for nonrandomized studies.”
Demystifying Double Robustness: A Comparison of Alternative Strategies for Estimating a Population Mean from Incomplete Data (2007, methodology): “The notion of potential outcomes was introduced by Neyman (1923) for randomized experiments and by Rubin (1974a) for nonrandomized studies.”