Statistical latent-variable models make unmeasured concepts cause their indicators.

Evidence

In factor analytic, latent variable, LISREL approaches the basiclevel factors—latent variables—cause the indicators. … Blalock’s 1979 American Sociological Association presidential address illustrates nicely the statistical view of the concept-indicator problem: it is almost all about cause running from unmeasured variables to indicators. We can use Bollen’s analysis of democracy as an example of this way of thinking about concepts. Figure 2.4 shows a typical latent variable model with its unmeasured latent variables, its indicators, and error terms. Bollen and Grandjean (1981) explicitly examine the multidimensionality of the democracy concept. Formally, the indicators are the effect of the latent causes: Here X1 and X2 are unmeasured basic-level concepts which produce indicators xi (this notation is unconventional in the latent variable context, but keeps with my notation for three-level concepts). — Gary Goertz, Social Science Concepts, 392

Categorization

• DEFINE: Clarifies the LISREL/factor-analytic direction of causality: latent basic-level concepts are treated as causes, indicators as effects.
• XREF: Connects to measurement-theory work on concept-indicator relations, e.g. the three-level framework elaborated elsewhere in this book.