CPI outlet samples from the CE: a new life for the Point-of-Purchase Survey

From 1998 to 2019, the Telephone Point-of-Purchase Survey (TPOPS) was conducted by the U.S. Census Bureau on behalf of the U.S. Bureau of Labor Statistics (BLS). The goal of TPOPS was to find out where consumers shopped for various items that make up the Consumer Price Index (CPI) market basket. BLS used the results of the survey to create a frame of establishments (called outlets) from which to draw a sample of participants in the CPI Commodities and Services (C&S) Pricing Survey.

TPOPS entered a testing phase in the fall of 1988 and was implemented in 1998, replacing its predecessor, the Continuous Point-of-Purchase Survey.¹ TPOPS was conducted quarterly through computer-assisted telephone interviewing, whereby interviewers read questions to respondents over the telephone and captured responses electronically via a collection instrument. This method reduced costs and respondent burden, while increasing the number of total responses. TPOPS used a random-digit-dialing (RDD) technique to draw, by telephone number, a sample of households for participation in the survey. The technique involved selecting area codes and telephone exchanges that represented CPI geographic areas called primary sampling units. Within an area code, unique telephone numbers were selected from banks of numbers that were at least 1 percent residential.² TPOPS started as a single-frame RDD survey, targeting only landline telephone numbers. However, as the number of cell-phone-only households increased and concerns about coverage bias grew,³ BLS began testing a cell-phone frame in 2011, implementing it in the second quarter of 2012.⁴ In 1998, households sampled by telephone number were selected to participate in TPOPS for four quarters, or 1 year, but this period was extended to eight quarters in 2015, primarily as a way to reduce costs and increase the survey’s overall response rate.⁵

By the time TPOPS was fully operational in 1998, telephone surveys had matured as data collection tools. However, over the years, TPOPS experienced a steady decline in response rates.⁶ (See chart 1.) Although this decline started after the initiation of the cell-phone frame in 2012, it cannot be explained solely by the properties of that frame.

Changes in the survey environment also played a role in the decline in response rates, with sample sizes rising and interviewers reaching households that were more reluctant to respond to TPOPS. With a static data collection budget, BLS had to make difficult tradeoffs about targeted number of survey responses, call attempts per case, and, ultimately, response rates. To improve response rates, BLS made continuous changes to its data collection methodologies for the CPI. Table 1 lists the major changes implemented over the years, excluding adjustments related to reducing respondent burden, maintaining data quality, and keeping the definitions of items surveyed in TPOPS current with the evolving marketplace. Despite these changes, BLS began testing the collection of outlet data in the Consumer Expenditure Surveys (CE) in 2016, aiming to replace TPOPS as the CPI program’s primary source of outlet frames.

Transitioning to the CE

The CE program oversees nationwide household surveys administered by the U.S. Census Bureau under contract with BLS. The surveys collect data on expenditures, income, assets and liabilities, and demographic characteristics of consumers in the United States, with the key measures being income and spending.⁷ There are two independent surveys composing the CE: the Interview Survey and the Diary Survey. Together, these two surveys cover the entire consumer market basket. The Interview Survey is administered quarterly and collects data on large, recurring expenditures made over a 3-month period, whereas the Diary Survey collects data on small, frequent expenditures that households document over two 1-week periods.⁸

CE data provide the CPI program with the total expenditures on all consumer items. These expenditures are used to estimate the relative importance of market-basket items in the calculation of the CPI. Because the CE program was already collecting total consumer expenditures for the entire CPI market basket prior to 2016, and because this collection took place in the same geographic areas in which BLS was administering its C&S Pricing Survey for the CPI, the CE could potentially be used to replace TPOPS.

In 2016, BLS assembled a team to assess the feasibility of sourcing outlet frames from the CE, an effort that involved incrementally incorporating outlet questions into the surveys and evaluating the quality of collected data. The team assessed both the benefits and drawbacks of this change.

Conceptually, consolidating outlet questions into the CE is beneficial, because the CE and TPOPS are designed to collect data on a representative sample of expenditures from households within the geographic areas covered by the CPI. Assuming no deterioration in the quality of outlet data collected in the CE, BLS could realize cost savings by streamlining survey management and data collection. Money that would have been spent on TPOPS could be used to fund projects to increase the CE sample size, boost response rates, and research alternative data collection techniques. Lastly, by consolidating surveys, BLS would not need to budget and plan to update geographic or item structures for two surveys, streamlining future transitions.

The benefits of adding outlets to the CE extend beyond survey management. The CE program collects more data than TPOPS on household characteristics, such as income. This additional household information can be used in producing research indexes for various population cohorts that are not covered by current BLS price index products or that could be better represented by products tailored to the shopping habits of cohort members. From a CE perspective, collecting more information about shopping trips (e.g., outlet name) may improve respondents’ recall of expenditures.

Transitioning to the CE for outlet information does not come without tradeoffs. From a BLS perspective, this transition means smaller sample sizes and shorter recall periods. Both tradeoffs can reduce the number of outlets reported, leading to smaller sampling-frame sizes for the C&S Pricing Survey. From a CE program perspective, the addition of outlet information means greater respondent burden.

The sample sizes in the CE Interview and Diary surveys are notably smaller than those in TPOPS. From 2006 to 2013, TPOPS averaged approximately 50,000 interviews per year, whereas the CE Interview and Diary surveys averaged approximately 21,000 and 10,200 interviews, respectively (when controlling for differences in geography). In recent years, the number of completed interviews in TPOPS has declined to an average of approximately 37,000 interviews, compared with 19,000 interviews in the CE Interview Survey and 9,000 interviews in the CE Diary Survey.

To mitigate this difference in sample sizes, the CPI program redistributed funds from the TPOPS budget to CE data collection. Most of these funds were directed to the CE Diary Survey, because its sample size is smaller than the sample size of the CE Interview Survey. This redistribution is expected to yield an approximately 50-percent increase in the number of completed CE diaries and a 10-percent increase in the number of completed CE interviews.

Further, both TPOPS and the CE Interview Survey ask respondents to recall expenses over different periods. The recall period is 3 months in the CE Interview Survey and, depending on how frequently items were purchased, 1 week to 5 years in TPOPS. The variable recall period used in TPOPS was intended to optimize the number of outlets reported for purchased items, while balancing CPI’s needs and respondent burden. For example, in TPOPS, respondents were asked to report car purchases made within the last 5 years, but a shorter recall period of 2 weeks was used for purchases of fresh fruits and vegetables. Of the 181 item categories for which expenditure data were collected in TPOPS, over half had a recall period greater than the 3-month period used in the CE Interview Survey. The recall period used in the CE Interview Survey is designed to capture quarterly expenditure estimates, which are used to produce annual expenditure estimates. In the CE Diary Survey, expenditures are reported contemporaneously for two 1-week periods per household. Only 11 of the 45 item categories whose outlets would be exclusively sourced from the CE Diary Survey had a TPOPS recall period longer than 2 weeks.

To resolve differences in the number of outlets reported—differences due to discrepancies in recall periods for item categories—BLS added a new section to the CE Interview Survey. Under this change, respondents who do not have expenditures for an item category in the standard 3-month recall period are asked to report their most recent outlet and expenditure for a variable extended recall period (between 6 months and 5 years). Table 2 summarizes how the new section is used to achieve variable recall periods (in the CE Interview Survey) that more closely match those used in TPOPS. With the exception of new and leased cars, trucks, vans, and motorcycles, item categories collected in the CE Interview Survey have outlet questions in the extended-recall questionnaire.⁹ Item categories that had a TPOPS recall period shorter than the standard 3-month recall period used in the CE Interview Survey were assigned a 6-month recall period in the extended-recall questionnaire. Otherwise, item categories were assigned a recall period equivalent to that used in TPOPS.

The extended-recall questions are designed to increase the number of outlets reported in the CE Interview Survey by increasing the rate at which households report outlets for item categories. This rate is defined as the ratio of the number of households reporting at least one outlet to the total number of households asked to report an outlet for an item category. Using data collected from the second quarter of 2019 to the first quarter of 2020, BLS found that, as expected, including the extended-recall questions in the CE Interview Survey increased the rate at which households reported outlets. Specifically, the median outlet reporting rate for an item category increased from approximately 4 percent to about 11 percent.

In addition, the inclusion of extended-recall questions led to relatively greater increases in reporting rates for less frequently purchased item categories. Chart 2 shows the distribution of outlet reporting rates for item categories in the CPI market basket, presenting results with and without the inclusion of extended-recall questions. The data are further categorized by whether an item category had a TPOPS recall period greater than or less than or equal to 90 days. As shown in the chart, with the inclusion of extended-recall questions, the median outlet reporting rate for item categories with a TPOPS recall period longer than 90 days more than tripled, rising from 3 to 11 percent, whereas the median rate for item categories with a TPOPS recall period shorter than or equal to 90 days roughly doubled, rising from 8 to 15 percent.

View Chart Data

To mitigate the added respondent burden stemming from the inclusion of outlet questions in the CE Interview Survey, BLS will continue to ask those questions only of respondents in cities where the corresponding item categories are rotated in the CPI.¹⁰ In TPOPS, this requirement reduced respondent burden, and its continuation will have the same effect by setting an upper limit to the number of questions asked of a respondent in the CE Interview Survey. To further reduce respondent burden, BLS will collect less specific location information for outlets in the CE Interview Survey than in TPOPS. In the past, TPOPS collected information on the street address, cross street, city, and state of each reported outlet. However, the CPI program determined that the benefit of collecting this detailed location information did not outweigh the cost of increased burden to CE respondents, opting instead to collect data only for an outlet’s city and state. Further, the CPI program determined that establishment location information collected in the Quarterly Census of Employment and Wages (QCEW) could provide precise address information for outlets reported in the CE.

Obtaining address information for CE outlet data

BLS matches the name and location of each reported outlet to establishments captured in QCEW data, generating a list of candidate establishments that best represent the reported outlet. The QCEW program publishes a quarterly count of employment and wages reported by employers in the United States, covering more than 95 percent of U.S. wage and salary employment.¹¹ The primary sources of QCEW data are state-level unemployment insurance programs, which provide employer name and address information. These administrative data are supplemented by two BLS surveys (the Annual Refiling Survey and the Multiple Worksite Report), which collect and verify geographic and industry information. This geographic and industry information is vital, because each establishment must be visited by BLS staff and must sell the items in the CPI market basket in order to participate in the C&S Pricing Survey.

When an address is required for an outlet reported in the CE, the CPI program links the reported outlet name and location information collected in the CE to establishments in QCEW data. In the CE Interview Survey, if a respondent reports an outlet name and indicates an in-person purchase, the respondent is asked to report the city and state of the outlet. In the CE Diary Survey, respondents are asked to report only the outlet name, and no further information about the outlet or how the purchase was made is collected. Here, unless the outlet name represents an establishment that operates entirely online, BLS staff use the CE household’s location as a proxy for the location of the outlet and assume that the purchase was made in person.

If a city and state were reported with an outlet name, BLS economists would start the matching process by searching QCEW data for establishments that match the outlet name and the reported city and state. If no establishments were found within the reported city and state, the economists would attempt to identify establishments near the reported city and state. In either case, any establishments returned from this process would be included in the candidate list of establishments that could represent the outlet in the C&S Pricing Survey. If data on an outlet’s city and state were neither reported nor collected, as would be the case for outlets captured in the CE Diary Survey, BLS economists would use the CE household’s location to determine a geographic area within which the CPI program would search the QCEW data for establishments that match the reported outlet name. Lastly, if the CPI program could not identify a representative establishment in the QCEW data, it would use data collected in the CE to locate and select a specific outlet name and address for pricing.

Timeline for CPI transition

After confirming the feasibility of the transition from TPOPS to the CE in 2017, BLS gradually added outlet questions to the CE Interview Survey. (See table 3.) In 2019, the outlet questions were fully implemented in the CE Interview Survey, and the redesigned CE Diary Survey form was first used. Meanwhile, BLS began redesigning the processing systems that create outlet-frame data for CPI outlet sample selection. These new data were first used as a sampling-frame source in early 2020, for samples that would be included in index calculations starting in November 2021. All samples used to construct the CPI and traditionally sourced from TPOPS are expected to be fully converted to data sourced from the CE by May 2025.

BLS will continue to monitor the quality of outlet data and the results of the matching algorithm for the CPI. BLS also will continue to monitor the impact of outlet questions on the quality of CE data, incorporating outlet information as a baseline requirement for the CE Interview and Diary surveys. These efforts will meet ongoing needs for unbiased and representative samples and expenditure data, allowing BLS to calculate its family of consumer price indexes.