Response and Sequencing Errors in Surveys: A Discrete Contagious Regression Analysis

• Published in: Journal of the American Statistical Association Vol. 88; no. 423; pp. 775 - 781
• Main Author: Hill, Daniel H.
• Format: Journal Article
• Language: English
• Published: Alexandria, VA Taylor & Francis Group 01.09.1993; American Statistical Association
• Subjects: Aggregation; Applications and Case Studies; At risk population; Contagious processes; Degrees of freedom; Error rates; Exact sciences and technology; Interviews; Mathematics; Probability and statistics; Proxy reporting; Regression analysis; Reliability; Response variance; Sampling theory, sample surveys; Sciences and techniques of general use; Sequencing; Statistical variance; Statistics; Surveys; Truncation; Regression model; Level analysis; Reliability; Growth of error; Sample survey; Measurement error; Interview
• ISSN: 0162-1459; 1537-274X
• DOI: 10.1080/01621459.1993.10476338

Skip sequencing in surveys reduces costs but also acts as a mechanism of contagion in transmitting error from one survey item to subsequent items. The error process is modeled as a contagious, stochastic one in which both the initiating and induced errors follow the Poisson distribution. The resulting compound distribution, first used by Thomas in 1949, is then used as the basis of a discrete contagious regression model in which characteristics of respondents, subject individuals, and interviewers are allowed to affect the intensity of both initiating and induced errors. The model is applied to data from the 1984 Survey of Income and Program Participation (SIPP) Reinterview Program. The results suggest that induced-sequencing errors are at least as important a source of losses in reliability as are initiating-response or spontaneous-procedural errors. The relative importance of induced-sequencing errors is directly proportional to the sequence length. Because the relationships between individual characteristics and the various types of error differ significantly, conclusions based on narrow definitions of response errors do not necessarily generalize when sequencing errors are included. The regression estimates suggest that older and less educated respondents provide less reliable data. Older female interviewers, however, are found to obtain more reliable data.