Reliability in Comparative and Ethnographic Observations: The Example of High Inference Father-Child Interaction Measures

The theory of reliability and reliability estimates, nearly a century old, has rarely been employed in anthropology, both for lack of familiarity and related problems of computation. This theory is reviewed and considered for use with anthropological data. A set of procedures is provided which combine existing methods to solve the practical problems in use of the theory to assess the reliabilities of composite measurement scales combining multiple measures, of individual independent measurements of a single concept, and of individual cases scored on the composite scales. These procedures are also embodied in a computer program, the results of which are explicated. While domain sampling assumptions are the only requirements of assessing reliability of composite scales, strict assumptions and validation procedures are discussed for the assessment of individual variable reliabilities. An illustration using reliability theory is drawn from cross-cultural studies for "high inference" measures from four different studies of father child interaction. Validity issues are illustrated both in terms of tests for measurement bias and construct validity for the hypothesized relation between the father child bond and beliefs in high gods. Reliability theory is equally applicable in comparative and ethnographic case studies. It offers research practices and theoretical understandings that are capable of integrating and mediating discourse between many of the splintered schools of thought in anthropology, and healing some of the rifts between them. Essentialist biases are discussed as one reason why the theory is not more, widely employed.