Factor Analysis and the Construct Indicator Relationship

Publisher Website
Google Scholar
Abstract
The use of factor analysis to validate theoretical constructs and to underpin multiple indicator measurement strategies ignores two methodological principles: a necessary condition for the operationalization of a theoretical construct is not the same as a sufficient condition, and a correlation does not imply a particular model of causation. It is pointed out that factor analysis of the traditional (exploratory) kind used as a basis for construct validation breaches these principles; it does not provide sufficient evidence of construct validity, and it is incapable of dealing with any structure of the relations between variables other than in terms of a set of underlying causes (common factors). Alternative causal interpretations can be evaluated only through the confirmatory methods of structural modeling. Using this approach in the analysis of Richmond Basic Skills Test data, the value of alternatives to the classical factor model is clearly demonstrated. Some implications for measurement strategy and theory construction in social science are considered.