MAIN POINTS

Many concepts employed in the social sciences are complex and difficult to measure with single indicators. Indexes and scales are composite measures constructed through the combination of two or more items or indicators. Indexes usually involve adding together measurements from single indicators; scales are more carefully constructed in an attempt to ensure that only one dimension is measured.

The combination of two or more items or indicators yields an index. Four major problems are involved in constructing indexes: definition of purpose for which the index is being compiled, selection and collection of sources of data, selection of the base comparison, and selection of methods of aggregation and weighting. Attitude indexes, also referred to as arbitrary scales, involve a battery of questions that are selected on an a priori basis on a zero-to-four or a one-to-five answer scale that is arbitrarily scored.

Likert scaling is a method designed to measure people's attitudes. Six steps can be distinguished in the construction of a Likert scale: 1) compiling a list of possible scale items, 2) administering these items to a random sample of respondents, 3) computing a total score for each respondent, 4) determining the contribution or discriminative power (DP) of items, 5) selecting the scale items, and 6) testing reliability.

Guttman scaling is designed to incorporate an empirical test of the unidimensionality of a set of items as an integral part of the scale construction process. Guttman scales are unidimensional as well as cumulative. In practice, a perfect Guttman scale is rarely obtainable; consequently, Guttman developed a criterion for evaluating the unidimensional and cumulative assumptions: the coefficient of reproducibility (CR), which measures the degree of conformity to a perfect scalable pattern. There are two major steps in constructing a Guttman scale: 1) selecting scale items, and 2) calculating the coefficient of reproducibility.

Factor analysis is a statistical technique for classifying a large number of interrelated variables into a smaller number of dimensions or factors. It is a useful method for the construction of multiple item scales, where each scale represents a dimension of a more abstract construct. First, bivariate correlations are computed, and then these correlations are placed in a matrix format. The correlation matrix is used as the input data in the factor analysis procedure. Finally, a composite scale is developed for each factor