Clark Glymour

We describe a unification of old and recent ideas for formulating graphical models to explain time series data, including Granger causality, semi-automated search procedures for graphical causal models, modeling of contemporaneous influences in times series, and heuristic generalized additive model corrections to linear models. We illustrate the procedures by finding a structure of exogenous variables and mediating variables among time series of remote geospatial indices of ocean surface temperatures and pressures. The analysis agrees with known exogenous drivers of the indices, not assumed in the analysis. Automated search applied to the residuals after regressing each series on its lags and the lags of its Granger causes yields a graphical model of “comtemporaneous” causal relations identical with the qualitative graphical relations from the time series. A similar analysis produces reasonable results when applied to candidate climate indices obtained by a clustering method from sea surface temperature and sea level pressure data.

The Bell Curve aims to establish a set of causal claims. I argue that the methodology of The Bell Curve is typical of much of contemporary social science and is intrinsically defective. I claim better methods are available for causal inference from observational data, but that those methods would yield no causal conclusions from the data used in the formal analyses in The Bell Curve. Against the laissez-faire social policies advocated in the book, I claim that when combined with common sense and other information, the informal data mustered in The Bell Curve support a range of "liberal" social policies

Twenty years ago, Nancy Cartwright wrote a perceptive essay in which she clearly distinguished causal relations from associations, introduced philosophers to Simpson’s paradox, articulated the difficulties for reductive probabilistic analyses of causation that flow from these observations, and connected causal relations with strategies of action (Cartwright 1979). Five years later, without appreciating her essay, I and my (then) students began to develop formal representations of causal and probabilistic relations, which, subsequently informed by the work of computer scientists and statisticians, led eventually to a practical theory of causal inference and prediction, a theory incorporating some of the sensibilities Cartwright had voiced (Glymour et al. 1987; Spirtes et al. 1993). That theory, and ideas related to it, have become a subfield of computer science with contributions far deeper than mine from many sources, and its inferential and predictive techniques have been successfully applied in biology, economics, educational research, geology and space physics.

Linear structural equation models (SEMs) are widely used in sociology, econometrics, biology, and other sciences. A SEM (without free parameters) has two parts: a probability distribution (in the Normal case specified by a set of linear structural equations and a covariance matrix among the “error” or “disturbance” terms), and an associated path diagram corresponding to the causal relations among variables specified by the structural equations and the correlations among the error terms. It is often thought that the path diagram is nothing more than a heuristic device for illustrating the assumptions of the model. However, in this paper, we will show how path diagrams can be used to solve a number of important problems in structural equation modelling.

Recent literature in philosophy of science has addressed purported notions of explanatory virtues—‘explanatory power’, ‘unification’, and ‘coherence’. In each case, a probabilistic relation between a theory and data is said to measure the power of an explanation, or degree of unification, or degree of coherence. This essay argues that the measures do not capture cases that are paradigms of scientific explanation, that the available psychological evidence indicates that the measures do not capture judgements of explanatory power, and, finally, that the measures do not provide useful methods for selecting hypotheses. 1. Introduction2. Some Proposed Measures of Explanatory Virtues3. Descriptive Inadequacy3.1 Excellent but false explanations3.2 Causal explanation4. Psychological Inadequacy5. Finding the Truth6. Conclusion

Of the many proposals for inferring genetic regulatory structure from microarray measurements of mRNA transcript hybridization, several aim to estimate regulatory structure from the associations of gene expression levels measured in repeated samples. The repeated samples may be from a single experimental condition, or from several distinct experimental conditions; they may be “equilibrium” measurements or time series; the associations may be estimated by correlation coefficients or by conditional frequencies (for discretized measurements) or by some other statistic. This paper describes two elementary statistical difficulties for all such procedures, no matter whether based on Bayesian updating, conditional independence testing, or other machine learning procedures such as simulated annealing or neural net pruning. One difficulty obtains if large numbers of cells are aggregated in a measurement of expression levels from a common population of cells; the other obtains if small numbers of cells are aggregated or if samples are separately aggregated over different populations of cells.

After reviewing theoretical reasons for doubting that machine learning methods can accurately infer gene regulatory networks from microarray data, we test 10 algorithms on simulated data from the sea urchin network, and on microarray data for yeast compared with recent experimental determinations of the regulatory network in the same yeast species. Our results agree with the theoretical arguments: most algorithms are at chance for determining the existence of a regulatory connection between gene pairs, and the algorithms that perform better than chance are nonetheless so errorprone as to be of little practical use in these applications.

By combining experimental interventions with search procedures for graphical causal models we show that under familiar assumptions, with perfect data, N - 1 experiments suffice to determine the causal relations among N > 2 variables when each experiment randomizes at most one variable. We show the same bound holds for adaptive learners, but does not hold for N > 4 when each experiment can simultaneously randomize more than one variable. This bound provides a type of ideal for the measure of success of heuristic approached in active learning methods of casual discovery, which currently use less informative measures

S CIENTISTS often claim that an experiment or observation tests certain hypotheses within a complex theory but not others. Relativity theorists, for example, are unanimous in the judgment that measurements of the gravitational red shift do not test the field equations of general relativity; psychoanalysts sometimes complain that experimental tests of Freudian theory are at best tests of rather peripheral hypotheses; astronomers do not regard observations of the positions of a single planet as a test of Kepler's third law, even though those observations may test Kepler's first and second laws. Observations are regarded as relevant to some hypotheses in a theory but not relevant to others in that same theory. There is another kind of scientific judgment that may or may not be related to such judgments of relevance: determinations of the accuracy of the predictions of some theories are not held to provide tests of those theories, or, at least, positive results are not held to support or confirm the theories in question. There are, for example, special relativistic theories of gravity that predict the same phenomena as does general relativity, yet the theories are regarded as..