Suchergebnisse

Many survey researchers believe that the quality of answers to attitude questions may be improved by using no-opinion filters to remove those responses that are not based on true opinions. The authors investigate whether reliability, one aspect of data quality, is greater for filtered questions. Split-ballot experiments from two telephone surveys of the Akron, Ohio metropolitan area are used to estimate the effect of filtered questions on the reliability of items from Srole's anomia scale, Rosenberg's self-esteem scale, and a scale of attitudes toward lawyers. There is no evidence that filtered questions improve reliability, and in the case of self-esteem, reliability may be reduced on the filtered form.

A method suggested by the psychol'ts D. T. Campbell & D. W. Fiske ('Convergent and Discriminant Validation by the Multitrait-Multimethod Matrix,' Psychological Bulletin, 1959, 56, 81-105) for the study of test-method reliability is applied to a nat'l sample of PO data (N=1,504 adults) re public policy questions. This method requires computation of a larger matrix of is between responses to diff items under varying methods of application. Reliability is defined as 'the agreement between 2 efforts to measure the same trait through maximally similar methods.' The 15 policy areas studied show considerable variation in item-method reliability, but items measuring policy views on open housing, Sch integration, & full employment show item-method reliability that compares favorably with psychol'al tests. The advantages of this 'heterotrait matrix approach' are confirmed by the analysis. In sample surveys card sort, closed & open items, or direct observation can be taken as alternatives for studying the same traits. Pol'al sci'ts could then confront the problem of the validity of their measures rather than simply asserting the fact. 3 Tables. Modified HA.

In interpreting income as a variable in a study it may be necessary to include the income picture covering several years. On the basis of recall, considerable error of non-random sort is likely. If direction or magnitude of change, for changes of excess of 4% is important, error may be introduced by recall. There is a relationship (r = .58) between magnitude of income change (diff between current reports) and magnitude of unreliability (diff between first current report and recall report). L. P. Chall.