Lagged Structures and Cross—Sectional Methods

Abstract

Causal structures involving a dynamic relation with time lags present difficulties if approached with conventional multivariate cross-sectional techniques. It is shown that customary tests for causal structure, e. g. partial correlations, break down or yield ambiguous results. Some further implications of lagged structure, and an extension of the idea to variability in space, are discussed in the paper.

An Investigation of the Relationship Between Work and Family Size Decisions over Time

“…For simplicity, but without loss of generality, population values of variances (uii) and covariances (uij) can be used to solve for the focal parameters in Figure It can be seen, when comparing b62 to bs3 and b35 to b36, respectively, that the recursive and nonrecursive representations of causality yield quite different results for the key structural parameters. The differences persist even if one were to assume that (a) time intervals are short between cause and effect, (b) variables are stationary over time, and (c) an averaging of variables over time achieves equilibrium conditions (c.f., Carlsson, 1972;James et al, 1982). For all three assumptions, key variances andlor covariances, and hence the focal causal parameters, diverge between nonrecursive and recursive models.…”

Linear Equations and Nonlinear Estimation

“…Heise (1975: 228ff. ) demonstrates the bias that can be introduced when using unequilibrated cases in analysis, and Carlsson (1972) shows how cross-sectional estimates may be greatly misleading as indicators of system parameters in dynamic systems. Two related strategies should be considered when this specification assumption is problematic.…”

On the concept of complexity and its relationship to the methodology of policy-oriented research

“…Since there are several excellent accounts of this topic (Pelz and Andrews, 1964;Pelz and Lew, 1970;Heise, 1970;Carlsson, 1972); we omit further discussion on causal lags here, concentrating upon the assumptions of the nature of systems underlying linear models. The effects on causal inferences of erroneous estimation of causal lags depend on the processes involved; therefore, they are of different magnitude in different contexts.…”