Thinking about the Future: The Methodology of Projective Inference

• In Event: Rethinking Standard Statistical Practice

Abstract

This paper provides an in-depth methodological framework for thinking about a form of inference that is often neglected but that is essential for thinking about future political systems. Projective inference evaluates the achievability of plans or projections about the future. It builds upon but it is different from descriptive inference, such as the analysis of censuses or surveys, that portrays the world as it was and as it is, interpretive approaches, such as ethnographic studies or discourse analysis, that explore people’s past and current understandings of the world in which they live, or causal inference, such as econometric modeling or experiments, that establishes prior and contemporaneous causes of social outcomes. Through projection, humans contrive projects that put together diverse elements to achieve important purposes and values. Examples include the Constitution for a fledgling democratic republic in 1789, a Transcontinental Railroad in the 1860s, and a California Global Warming Solutions Act (AB 32) in 2006.

Projections must, of course, start from current circumstances captured by descriptive and interpretive inferences, and they must rely upon the opportunities for change identified by causal and interpretive understandings. But then they can and should go beyond descriptive, interpretive, and causal inference to make an active effort to “envision and shape the future” by considering what values we want the future to embody, the paths to those futures—including new and inventive routes—and the likelihood that such a future can be produced. Projective inferences must not only consider the causal relationships, probabilities, and outcomes that might determine the future, but also the values and decision rules that lead to actions that could affect the future through the changes they make to those relationships, probabilities and outcomes.

Projective inferences usually must contend with levels of uncertainty about the world that cannot be easily captured by statistical methods, and they must deal with complex questions about prospective preferences that will shape the future. In doing this, projections can be thought of as “futures analysis” or “foresight thinking” which enjoyed a short period of popularity but not much academic attention in the 1960s and 1970s. As a result, foresight thinking and projective inference have not become very well-developed fields of inquiry, and they lack a strong intellectual foundation. Although projection can be at least partly accomplished with time-series forecasting or simulations such as Integrated Assessment Models for climate change that are considered respectable and well-tested techniques among methodologists, a complete discussion requires a much broader perspective that challenges methodologists to think in new ways.

This paper considers five approaches to projective inference: statistical forecasting and modeling, technological forecasting, scenario analysis, configurative analysis, and robust decision-making (RDM) under deep uncertainty. It argues that these approaches ask different questions about the future: The first two (statistical and technological forecasting) ask what is likely? The second two (scenario and configurative analysis) ask what is possible and best? The last one (RDM) asks what must be avoided or can be adventitiously exploited? Correspondingly, beyond considering the core elements of relationships and outcomes that define a projection problem, they consider different additional elements. Statistical and technological forecasting focus on determining probabilities of relationships and outcomes. Scenario and configurative approaches focus on the values of outcomes. And robust decision-making approaches focus on robust actions and decision-rules that will ensure reasonable, if not optimal, outcomes. These different questions and emphases lead to different methods and procedures – all of which constitute legitimate forms of projective inference based upon modern understandings of scientific method as reviewed in detail in the last part of the paper.

The history of political methodology since the 1950s can be written as a succession of new methodological concerns. First there was the behavioral revolution and a concern with descriptive inference. Then there was a focus on causal inference through observational studies using multivariate statistical methods and experimental methods using careful designs and statistical methods. And in parallel there has been a concern with the methods of interpretive inference such as qualitative research, discourse analysis, and the interpretation of meanings. Projective inference poses a new set of issues encompassing all of these methods but going beyond them by presenting new problems of characterizing uncertainty and comprehending values. This paper challenges political scientists to take on these problems.

Author

• Henry E. Brady, University of California, Berkeley