Abstract
This article suggests that contemporary philosophy of science can significantly increase our understanding of behavioral research methods. In particular, it illustrates this contention by showing how selected methodological ideas in the philosophy of science can deepen our understanding of the widely used method of exploratory factor analysis.
In his recent article, James Grice (2014) showed in detail how psychology students can be taught his novel methodology of Observation Oriented Modeling. In building the methodology and instructing students how to employ it, Grice presents a philosophy of ‘moderate realism’ to underwrite his project. This philosophy maintains that things have essences, that their natures are knowable, and that a strategy of modeling can be used to integrate knowledge about the systems under study. Grice shows in a general way how philosophy of science can make an important contribution to scientific methodology. In this short article, I show in a specific manner how the philosophy of science can be used to improve our understanding of research methods. I begin by noting the importance of contemporary philosophy of science for contributing to our understanding of scientific inquiry. I then illustrate my point by showing how selected methodological ideas in the philosophical literature can be used to deepen our understanding of the important method of exploratory factor analysis.
In recent years, philosophy of science has increasingly sought to understand science as it is practiced, and it now boasts an array of important methodological insights that are impossible to ignore when coming to grips with research methods. It can help illuminate various behavioral research methods such as exploratory data analysis, exploratory factor analysis, and tests of statistical significance. It can also help illuminate important methodological strategies such as replication, robustness, and calibration, to name but three. Haig (2014) examines these methodological matters against a backdrop of scientific realist philosophy of science. In a real sense, contemporary philosophy of science has become philosophy for science.
Philosophy of Science at Work: The Case of Exploratory Factor Analysis
Exploratory factor analysis (EFA) is a multivariate statistical method that has been widely used in psychology and many other sciences (Mulaik, 2010). It combines multiple regression and partial correlation theory in a way that facilitates the postulation of latent variables that are thought to underlie, and give rise to, patterns of correlations in new domains of observed or manifest variables. EFA is not an easy method to understand and use correctly. The paleontologist Stephen Jay Gould (1996), who was knowledgeable about the method, and employed it in some of his research on fossil organisms, described it as “a bitch of a method.”
Adding to the challenge of understanding EFA is the fact that it is a controversial method. One important aspect of the controversy is the issue of whether EFA should be considered as a method of data reduction, or a method that helps researchers explain patterns of correlated manifest variables. This issue forces factor analysts to consider the nature and interpretation of latent variables: are they to be interpreted as causal factors that explain correlations, or as summaries of those correlations?
A related issue is whether the basic form of factor analytic inference should be understood as inductive or abductive (explanatory) in nature. Expositions of EFA seldom consider its inferential nature, but when they do, the method is usually said to be inductive in character. This is not surprising, given that the origins of exploratory factor analysis can be plausibly located within 17th and 18th century empiricist philosophy of science with its inductive conception of inquiry (Mulaik, 1987). However, an inductive characterization of exploratory factor analysis is inappropriate. This is because inductive inference, being descriptive inference, cannot take the researcher from manifest effects to theoretical entities that are different in kind from those effects. However, abductive inference, which is concerned with the generation and evaluation of explanatory hypotheses, can do so. For this reason, EFA is better understood as an abductive method of theory generation (Haig, 2005; Mulaik, 2010), a characterization that coheres well with its general acceptance as a latent variable method.
Abductive inference is explanatory inference, employed, e.g., when scientists reason back from presumed effects to underlying causes. There are different forms of abductive reasoning (Thagard, 1988). EFA is a method that can facilitate the drawing of explanatory inferences that are known as existential abductions. Existential abductions enable researchers to hypothesize the existence, but not the nature, of entities previously unknown to them. The innumerable examples of existential abduction in science include the initial postulation of the presence of hidden entities such as atoms, genes, tectonic plates, and personality traits. We now know that some of these entities exist, that some of them do not, and we are unsure about the existence of others. In cases like these, the primary thrust of the initial abductive inferences is to claims about the existence of theoretical entities to explain empirical facts or phenomena. Similarly, the hypotheses obtained through the use of EFA postulate the existence of latent variables such as Spearman's g and extraversion. It remains for further research to elaborate on the first rudimentary conception of these variables and their interrelation. Sometimes this is done by constructing models of the causal mechanisms for which the latent variables are markers; we imagine something analogous to mechanisms whose nature we do know. Spearman did this by likening g to the idea of force in physics.
It is well known that EFA is a common factor analytic model in which the latent factors it postulates are referred to as common factors. Less well known is the fact that there is an important principle of scientific inference, known as the principle of the common cause (e.g., Sober, 1988), which can be used to drive the nature and shape of the existential abductive inferences involved in EFA. The principle of the common cause can be formulated concisely as follows: “Whenever two or more events are improbably, or significantly, correlated, infer one or more common causes, unless there is good reason not to.” Clearly, the principle should not be taken as a hard and fast rule, for, in many cases, proper inferences about correlated events will not be in terms of common causes. The qualifier, “unless there is good reason not to,” should be understood as an injunction to consider causal interpretations of the correlated events in addition to the common causal kind. For example, in a given research situation the correlated events might be related as direct causes, or their relationship might be mediated by a third variable in a causal sequence. Given its informal nature, the so-called principle is really a heuristic or rule of thumb, although this feature does not diminish its importance in science.
There are two features of the principle of the common cause that make it particularly suitable for use in EFA (Sober, 1988). First, it can be applied in situations where we do not know how likely it is that the correlated effects are due to a common cause. The abductive inference to common causes is a basic explanatory move, i.e., non-probabilistic, and qualitative, in nature. It is judgments about the soundness of the abductive inferences, not the assignment of probabilities, which confer initial plausibility on the factorial hypotheses spawned by EFA. Second, the principle can also be used in situations where we are essentially ignorant of the nature of the common causes. With this second feature, the principle of the common cause accommodates the fact that EFA is restricted to existential abductions.
It is important to appreciate further that the principle of the common cause does not function in isolation from other methodological constraints. Embedded in EFA, the principle helps one limit existential abductive inferences to those situations where one can reason back from correlated effects to one or more common causes. Although covariation is an important basic datum in science, not all effects are expressed as correlations and, of course, not all causes are of the common causal variety. In this way, the principle of the common cause helps to delimit the proper use of EFA.
It should also be noted here that there are other parts of philosophy of science that speak to controversial aspects of EFA. For example, the cognitive status of the dispositional explanations produced by EFA, the vexing issue of the indeterminacy of factors, and the relative importance of exploratory and confirmatory factor analysis can all be illuminated by appealing to relevant philosophical insights. Space precludes more than their mention, but Haig's articles (2005, 2014) discuss these issues in some detail.
Based on ideas in the philosophy of science literature about abduction and the principle of the common cause, EFA can be glossed as a set of multivariate procedures which help us reason in an existentially abductive manner from robust correlational data patterns to plausible explanatory proto-theories via the principle of the common cause. This understanding of EFA provides important insights into its deep structure that are not available in standard treatments of the method. I hope it is now clear that this abductive interpretation of EFA reinforces the view that it is best regarded as a latent variable method, thus distancing it from data reduction methods, such as principal components analysis, to which it is often likened.