AN EVALUATION OF THE 1988 CURRENT POINT-OF-PURCHASE CATI FEASIBILITY TEST Clyde Tucker and Robert Casady, Bureau of Labor Statistics James Lepkowski, The University of Michigan Clyde Tucker, BLS, 600 E Street, N.W., Room 5217, Washington, D.C., 20212 KEY WORDS: mode effects, length effects, survey costs An important assumption underlying this research is that I. INTRODUCTION In 1988, the Bureau of Labor Statistics (BLS), in conjunction with the Bureau of the Census, conducted a test of the feasibility of transforming the Current Point-of- Purchase Survey (CPP) from a personal interview to a telephone interview. The Bureau of the Census conducts this survey for BLS as part of the Consumer Price Index (CPI) program. Its purpose is to develop and maintain a timely list of retail, wholesale, and service establishments at which people shop for specified consumer items. This list serves as a sampling frame for BLS to update and maintain the sample of outlets it uses in pricing goods and services for the CPI. The survey is conducted on a five-year rotation schedule in which one-fifth of the 88 CPI primary sampling units (PSUs) are surveyed yearly in April, May, and June. Each year approximately 5000 consumer units or CUs (similar to household) are asked about their expenditures in almost 150 commodity categories with varying recall periods. Conducting the CPP survey with CATI can offer several potential advantages. After the initial development costs, the cost for contacting a household by telephone will be substantially less than in person. Furthermore, supervisors can exercise more quality control over the interviewing in a centralized CATI facility. Using telephone survey methodology, the survey can be conducted continuously in all PSUs. This ultimately will reduce the cost and complexity of outlet sampling by eliminating the need for hiring and training a new staff in each PSU every five years and by transforming processing and sampling into routine tasks. A telephone survey also will provide greater flexibility with respect to adding new commodities to the CPI or changing PSUs, thus, making it more timely. The complexity of administration should be less for a telephone survey because there will be fewer layers of bureaucracy between the data collectors and those using the data. Finally, the burden for any single respondent will be less because each respondent will be asked about expenditures for only a portion of the commodity categories covered in the personal interview. This will necessitate an increase in sample size which hopefully will be more than offset by other cost savings. Along with the above advantages come some disadvantages as well. About 7 percent of all households do not have telephones and, therefore, will not be covered by the survey. Response rates in telephone surveys are usually substantially lower than those in personal interviews. In addition to the loss of data due to undercoverage and nonresponse, the quality of the data collected by telephone may not be as good as that collected in personal interviews. the coverage bias and increased nonresponse which would result from abandoning the personal interview mode ultimately will have very little effect on the estimates of price change in the CPI. Outlets are sampled at a rate proportional to estimated sales within a commodity category. Any bias resulting from the incorrect estimates of sales in the category will lead only to inefficiences in estimating price change and not to a bias in price change so long as each store's estimated measure of size has an expected value which is greater than zero. II. PREVIOUS RESEARCH As more and more survey organizations have considered switching to telephone surveys, interest in the effects of the mode of interview on survey results has intensified. This interest has produced a considerable volume of research over the last 25 years. These comparative studies, as Paul Biemer (1988) has pointed out, actually assess the differences between two data collection systems and do not simply estimate a mode effect. Most of the research has focused on differences in the amount of nonresponse (Groves and Kahn 1979; Drew et al. 1988), the demographic characteristics of the samples (Mulry-Liggan 1983; Thornberry and Massey 1983; Kormendi 1988), and the estimates for the variables of substantive interest (Hochstim 1967; Rogers 1976; Schuman et al. 1985). Some studies have provided details of differences in the costs of the two methodologies (Biemer 1983; Groves and Lepkowski 1986; Sirken and Casady 1988; Groves 1989). 111. STUDY DESIGN The CPP feasibility test was conducted in September through November, 1988 in four of the PSUs from the spring 1988 CPP sample. The nineteen PSUs in the spring survey were classified into four groups according to size, and one PSU from each group was purposefully chosen for the test. The four PSUs selected were Chicago, New Orleans, Tucson, and the urbanized part of Halifax County, North Carolina. Each respondent in the personal-interview survey is asked about purchases in 143 of the 166 commodity categories because there are two versions of the questionnaire administered to half samples. Most of the categories (120) are on both versions, but 23 categories are unique to each version. For the feasibility test, all 166 commodity categories were divided first into eight versions, each containing approximately 20 categories. To gain some understanding about respondent burden in a telephone CPP survey, however, the eight versions were paired to produce four forty-category questionnaires that were also a part of the test. In order to ensure comparability, the forty-category telephone surveys and the two personal questionnaires were split into the eight 508