continued to unfold, statistical agencies around the world have responded with a sense of urgency to prepare for a world when most transactions leave electronic records that will become the principal source of price and quantity data.

Recognizing these realities, in opening remarks to the panel, BLS stated its objective to “convert a significant proportion of the CPI market basket from traditional collection to non-traditional sources and collection modes, including harnessing large-scale data, by 2024.” To assist in this process, the agency commissioned the Committee on National Statistics of the National Academies of Sciences, Engineering, and Medicine to assemble a panel of experts to provide guidance on how the CPI might be improved by accelerating the transition to an approach that blends multiple data sources.¹ The panel also was asked to consider ways to improve the measurement of traditionally difficult-to-measure sectors such as medical care and housing and, finally, to assess the policy and research value of developing price indexes to track differential inflation rates experienced by population subgroups such as those defined by income level. This summary highlights a subset of the panel’s recommendations for modernizing BLS’s CPI program.

THE POTENTIAL OF ALTERNATIVE DATA FOR CONSTRUCTION OF ELEMENTARY INDEXES

Point-of-sale data, household-generated home scanner data, and data scraped from the web are the primary “nontraditional” data sources that have been successfully exploited for price measurement. These data are generated for a wide range of items, often in near real time, and provide key information, including the price, source outlet, quantity, and characteristics of the item.

A key motivation driving data modernization is the potential to improve timeliness, relative to survey alternatives, in detecting what consumers are buying and from where. At no time has this need for updated data collection methods been exposed more starkly than during the COVID-19 pandemic. Statistical agencies that systematically use transactions data in their CPI programs—such as the Australian Bureau of Statistics (ABS), where less than 2 percent of the price quotes (by expenditure weight) are collected by field staff in-stores—were able to provide up-to-date information about prices being paid and about shifts in consumers’ buying patterns even as access to outlets became highly limited during lockdowns.

___________________

¹ The charge to the panel is reproduced in full in Chapter 1.

available, but where cost or accuracy benefits may emerge from pursuing alternative sources. BLS should embark on a broad-based strategy of accelerating and significantly enhancing the use of transactions data and other alternative data sources in CPI compilation. Embracing alternative data sources now, and moving forward aggressively with research for their integration, will ensure that the accuracy and timeliness of the CPI will not be compromised in the future. The data modernization strategy will involve:

• Identifying promising alternative data sources and then prioritizing the work needed to evaluate and incorporate these data into the items/strata where they can be applied;
• Continuing development of a robust research agenda that supports incorporation of alternative data and associated new methodologies more broadly beyond just price quote replacement;
• Continuing research assessing the quality of new types of data;
• Developing staff expertise that includes more data scientists and other specialists;
• Creating a cross-agency strategy for data acquisition with the possibility of joint contracts across statistical agencies;
• Carrying out a strong communication strategy to inform stakeholders of plans and implementation details.

Testing of indexes constructed from alternative data sources and methodologies will be a key part of data modernization. Before BLS incorporates alternative data for specific item categories into the official CPI, it will be important to have a significant overlap period (perhaps as long as two years) during which parallel indexes based on new data sources can be tested and compared against their traditionally constructed counterparts.

BLS has the opportunity to build upon the experiences of statistical agencies that have expedited incorporation of nonsurvey data into their CPIs. These agencies have navigated key challenges such as managing risks associated with the use of privately collected data and assessing the capacity of alternative data sources to track product quality changes. Where rapid quality changes are common, such as for high-tech items, combining datasets that include information about product characteristics expands the opportunities for improving price measurement.

One analytic task where immediate progress can be made is the development of elementary indexes constructed from web-scraped data. Beginning in March 2020, due to initial COVID-19 shutdowns, BLS had to improvise as monthly in-person collection of price data from retailers and businesses by field staff came to a halt. For an extended period, price checkers were limited to filling up virtual carts online to check prices. This

process, brought on as a stopgap measure in the face of the immediate crisis, only mimicked the in-store price checking activity. Converting opportunities for permanently automating web-scraping of price data should be a high priority for the CPI. In evaluating the usefulness of web-scaped data for elementary index estimation, food, electronics, and apparel should be priority categories. Data for these categories are readily available with a large share of transactions already online, and work by other statistical agencies and private-sector organizations have demonstrated feasibility. In the short term, BLS could consider obtaining web-scraped data from outside vendors but ultimately BLS should develop automated web-scraping methodologies within the agency. BLS should also continue and expand work with large companies to understand how they record data, and then collaborate on building programming interfaces that can be run behind company firewalls to provide the statistics needed by BLS.

HIGHER-LEVEL AGGREGATION AND SHIFTING CONSUMER BEHAVIOR

With the availability of near real-time information from private-sector data sources, coupled with an ever-increasingly dynamic economy, the collection and dissemination of timely data has become a basic expectation of statistical agencies. While the composition of what consumers buy is always evolving, the shifts were especially dramatic during the pandemic. Spending on travel, food away from home, and clothing worn outside the home declined dramatically, while demand for computer and communications equipment needed for remote work and home deliveries surged. Improving the ability to track such shifts is an essential goal.

Revising Consumer Expenditure Survey Weights More Frequently

The primary method used by most statistical offices to determine the composition of households’ expenditures is to ask them directly. In the United States, this process is carried out using the Consumer Expenditure Survey (CE), which has for decades been the statistical system’s most comprehensive source of data on households’ income and expenditures. And—because a nationally representative survey conducted by a government statistical agency is needed for benchmarking estimates of consumer expenditures in a way that links buying patterns to households—a version of the CE will continue to have value for the foreseeable future.

However, if the relevance of the CPI is to be maintained, changes in the way market basket shares (weights) are estimated must be a high priority. To improve the timeliness of the CPI and the accuracy with which it captures changing buying patterns, BLS must (1) update upper-level

weights—which currently, on average, lag 36 months behind actual expenditures in a given period—more frequently and rapidly, and (2) improve the accuracy of weights applied to specific items that the CE measures poorly and for which alternative data likely are more accurate.

Ideally, the expenditure data used to calculate CPI weights would come from a single 12-month period ending no more than six months prior to their introduction. For example, new CPI weights introduced in January 2022 would reflect expenditure patterns from July 2020 to June 2021. This production schedule may take time to achieve so, as an interim step in mitigating the timeliness problem, weights should be updated annually using two-year rolling averages of the CE data. Under this setup, the rolling weights would still lag real-time market realities, but not by as much as they do in the current two-year cycle.

Broadening Sources of Data Used for Estimating Expenditure Weights

In the not-to-distant future, detailed price and quantity information likely will be available from a range of nonsurvey sources for almost all products, which will allow much more frequent (and possibly more accurate) updating of CPI weights. With supplementing and complementing the CE data in mind, BLS should invest in collecting comprehensive data for individual spending using electronic means of payments such as credit/debit cards or electronic payment processors (e.g., PayPal or Stripe). Initially, these new data could be applied to the chained Consumer Price Index for All Urban Consumers (CPI-U) or to a new experimental index. Later, after an adequate period of study, expenditure pattern estimates used to construct CPI weights should be derived as a blend of data on spending from (1) the CE, (2) timely private sources, and (3) the national accounts. Given limitations of electronic transaction data to link information on prices and quantities to specific households—and, in turn, to population subgroups—research into use of alternative data sources for estimating upper-level weights should initially be directed toward production of the national-level CPI.

The national accounts—specifically, Personal Consumption Expenditure (PCE) data—from the Bureau of Economic Analysis can also be used, albeit retrospectively, to produce a more accurately weighted experimental CPI. One advantage of PCE data is that they are benchmarked to a census of retail establishments (conducted every five years) and a variety of other merchant-based sources, so they reflect a more comprehensive accounting of transactions by consumers. A first step might involve creating an experimental CPI that uses PCE-generated weights at the upper (243 item) level but that is otherwise no different from the CPI. Another option for blending PCE and CE data that BLS should test for the purpose of updating

upper-level expenditure weights is to continue using the CE as the benchmark for most CPI categories, then integrating PCE data to adjust the acknowledged weakest categories of the CE. The experiences of statistical agencies in several other countries that already use this strategy can be drawn upon to expedite this line of research. For example, Statistics Canada (2021) published a special edition price index using credit and debit card data supplied by the Bank of Canada and alternative weights based on national accounts data “to account for pandemic related expenditure shifts at more detailed levels of geography and CPI components.”

Longer-Term Planning

The strategy outlined above presumes that BLS will adopt a data approach for establishing upper-level expenditure weights that blends data from national accounts and private companies. A longer-run alternative to the data purchase model would be for BLS to set up an in-house operation for collecting the needed data. Since point-of-sale scanner data are not perfectly suited for use in the CPI (for reasons discussed in Chapter 2), the project should focus on collecting scanner data directly from households. BLS should begin exploring development of a household-based scanner recording program that would capture prices, quantities, and item characteristics of purchases made by surveyed respondents. In addition to its value for estimating item strata weights, this method of obtaining spending information would be useful for construction of elementary aggregates. BLS should consider “leapfrogging” traditional methodologies of handheld scanners that require large initial investments and look to modern approaches using a custom mobile phone app. Technologies are changing fast, so the most durable solutions may be based on flexible, mostly software-based approaches.

MODERNIZING MEASUREMENT OF DIFFICULT-TO-MEASURE EXPENDITURE CATEGORIES

A small subset of goods and services account for a large share of expenditures by households, and therefore have an outsized impact on the CPI. If new data sources can be harnessed for improving price measurement in these categories, the return on investments by BLS may be especially large.

Housing

Housing services represent by far the largest component of most consumers’ cost of living. Owner-occupied housing is particularly important, both because it accounts for about three-fourths of the shelter category and

because statistical agencies have not fully converged on a standard measurement approach. Prominent reviews of the CPI program have endorsed BLS’s rental equivalence method—which essentially estimates the amount that would be required for a homeowner to rent a home with all the characteristics of that household’s owned home—because, in many situations, it corresponds closely to a cost-of-living concept for measuring housing services. BLS should continue using rental equivalence as the primary approach to estimating the price of housing services for owner-occupied units.

The CPI methodology has traditionally relied on survey data to provide information on rent changes and housing expenditure shares. However, a number of new data sources for rents have emerged in recent decades, resulting from the expansion of large institutional landlords and property management companies. Supplementing the CPI Housing survey with data from such sources could help improve the accuracy of imputations of rent changes to the owner-occupied stock. BLS should seek to identify new data sources that would allow for improved coverage of single-family homes and of areas where houses are predominantly owner-occupied. New data sources could also improve the CPI’s ability to reflect rapid changes in rent growth by allowing rent for a given housing unit to be measured in consecutive months, allowing changes to be assessed over short periods of time. By contrast, the CPI Housing Survey only samples each housing unit every six months, which is problematic in times when rents change rapidly.

In addition to their value in measuring rent changes, alternative information sources—including property tax records and American Community Survey data—could be used in conjunction with the CPI Housing Survey. Specifically, BLS should consider strategies for estimating expenditure shares for owner-occupied housing that would make use of the rich housing characteristics information often available in property tax data.

In tracking the price of shelter in the United States, geographic detail is especially important because so much variation exists across and within regions. With this in mind, BLS should publish additional detail on the housing components of the CPI, such as indexes by structure type and for a larger number of metropolitan areas than the roughly 20 areas for which they are currently published. Broadening the geographic scope of the CPI could be facilitated by de-linking the housing sample from the samples of other CPI items.

Imputing rent for owner-occupied homes works best when there is a high degree of overlap—in terms of geography and housing quality—between the market of homes for sale and the rental market. When overlap is more limited, a user cost approach might be helpful to improve estimates of the price of housing services. Thus, although this panel is of the view that the rental equivalence approach should continue to be the primary method used in the CPI, research conducted on data at the micro

level would be valuable for testing where the rental equivalence method is performing well and where it is not (such as for pricing housing services associated with higher-end properties). As part of its research program, BLS should compare rental equivalence estimates to user cost estimates for individual properties. Research on alternative methods for housing could lay the groundwork for eventually publishing housing indexes using different methodologies.

Medical Care

Medical care is another large, growing, and rapidly changing consumer expenditure category. Even though the scope of the CPI covers only the share of the sector directly paid for by consumers (and not the share of spending by businesses and government), medical care is an important component of the CPI. Within medical care, health insurance is the largest expenditure made by consumers. For this reason, a pressing methodological decision facing BLS is whether to continue pricing insurance using an indirect method or to migrate to a direct pricing approach. The direct approach involves estimating total health insurance premium prices; the indirect approach involves pricing health insurance using prices of medical care blended with information about retained earnings of insurance providers. The underlying logic of the indirect approach is to separate out the prices of medical care goods and services from the portion of insurance premiums retained by insurance companies to cover administrative costs and profits. Although perhaps second best conceptually, the indirect method represents a pragmatic approach that at least partially reflects quality change and care utilization. The indirect method has practical advantages and therefore should, in the short to medium run, continue to be the method for pricing health insurance in the CPI.

The above recommendation notwithstanding, declining response rates in the medical care components of the commodities and services surveys are making reliance on the indirect method increasingly difficult. For this reason, and for its attractiveness conceptually, BLS should investigate historical differences between the indirect and direct approach doing a true apples-to-apples comparison. A “whole health insurance price deflator” that is a weighted average between the CPI’s current health insurance deflator and the various deflators for the medical services financed by insurance should be calculated and compared to the deflator used in the direct approach. If this research reveals that the two approaches do not differ greatly historically, BLS could revisit its reliance on the indirect method.

One source of concern with the current indirect approach is the volatility exhibited in the insurance services component of the total health insurance price index. For example, a particularly heavy flu season leading

to high utilization of health care services will reduce retained earnings and push down that component of the index. For this reason, among others, a number of potential improvements to the indirect approach should be considered. To better capture what consumers actually pay for insurance, and which does not depend on utilization rates, BLS should explore using a rolling average of retained earnings per unit of health services (where retained earnings equal premiums, less medical expenses), rather than an annual value. This approach will mean that actual changes in the cost of health insurance—stemming from changes in regulation, market structure, or technology—will appear more slowly in the CPI; this limitation could be an acceptable tradeoff if the problem of excess volatility now built into the index can be mitigated.

Also, there is currently a long lag between the time that prices of health insurance change and their incorporation into the CPI. Shortening this lag would improve the accuracy and timeliness of the index. For the purpose of tracking changes in health insurance prices, BLS should consider switching from using annual data on profits net of premiums to quarterly data.

For health insurance, developing methods and identifying data sources for adjusting prices to reflect quality change in insurance policies over time will continue be a top priority. As part of this work, BLS should continue evaluating how to accelerate incorporation of claims data to improve the coverage, detail, and timeliness of price and quantity information in the medical care component of the CPI. Such data may also be useful for research on broader questions, such as about productivity in the medical care sector. A pilot program is currently under way at BLS indicating promising uses of claims data including for the construction of experimental disease-based price indexes.

Supplemental Subgroup Price Indexes

Price indexes and other economic statistics tailored to describe and track the experiences of specific population subgroups are a growing research and policy need. The rationale for producing price indexes for population subgroups is clear for purposes such as adjusting Social Security benefits and marginal tax rates or for specifying transfer payments for which only certain groups are eligible.

One factor that can lead to differential inflation rates—and, as it turns out, the easiest one to measure—is that people purchase different baskets of goods and services. In general, research using a simple reweighting approach to reflect different consumption patterns has tended to detect only minimal differences in inflation rates faced by different groups. To fully portray differential inflation, subgroup price indexes must also account for different prices paid for similar items. Recent research based on diverse

data sources has revealed clear patterns of differential price inflation, in particular across income groups. Crucially, this research suggests that the greatest source of heterogeneity in households’ inflation rates is variation in prices paid for the same types of goods—not from variation in broadly defined consumption bundles. Research and policy making stand to benefit a great deal if the underlying trends in price inflation faced by different population groups can be more accurately measured and, in turn, better understood. Because of the urgency of issues related to income and wealth inequality, social welfare, and poverty, developing price indexes for population subgroups along the income distribution should be a high priority for BLS. Identifying data sources that would ultimately allow production of price indexes by income quintile or, if possible, decile is a key part of this work. The potential return from investments in developing income-defined subgroup price indexes is further enhanced by ongoing work at the Bureau of Economic Analysis to produce prototype statistics on the distribution of personal income across households.

As with other aspects of CPI modernization discussed in this report, long-term promise for creative initiatives for subgroup indexes comes with the increased availability of microdata containing information on prices that households actually pay and on details about the items purchased. However, electronic transactions data as currently generated do not cover all consumer expenditures. For this reason, the next generation of empirical studies on inflation inequality will need to draw on additional, alternative data sources—perhaps most importantly, a household-based scanner recording program that captures prices, quantities, and item characteristics of purchases made by surveyed respondents.

The above-described research linking individuals to their purchases strongly suggests the need for approaches that blend multiple data sources—encompassing survey data that cover the full consumption basket, including item categories for which electronic transaction data are still incomplete, and commercial data sources that allow deep analyses of prices paid and product detail—in a way that accounts for the full range of consumer expenditures.

ORGANIZATIONAL CONSIDERATIONS

Given that it has performed reliably for decades, the survey-based methodology underlying the CPI is commonly assumed to be the gold standard for estimating price changes. However, as with other economic statistics rooted in the application of a 20th century survey-centric system, the resulting estimates have been affected by falling survey response rates and increasing costs. Accordingly, the data collection model of statistical agencies is shifting. To fully capitalize on emerging data opportunities to

improve the quality and timeliness of the CPI, a paradigm shift at BLS will be required that lessens reliance on older survey-based approaches.

In addition to the methodological challenges—such as coverage, representativeness, and scope of variables present—practical considerations hinder development of a mixed data infrastructure that includes public, private, survey, and nonsurvey data. Legal constraints, privacy concerns, and high data acquisition costs—acutely present for the U.S. case—have slowed the incorporation of commercial and even government administrative data sources into social and economic statistics (NASEM, 2020). Modernization of the data system is further complicated by the decentralized statistical system of the United States. Especially within such a system, there are compelling reasons for agencies to pursue integrated data collection and production processes. More extensive collaboration between the Census Bureau, BLS, and Bureau of Economic Analysis (BEA)—along with other statistical agencies that collect key economic data, such as the Department of Agriculture—is needed to advance the acquisition and use of alternative data sources in the production of economic statistics. More specifically, such coordination will allow the statistical system to negotiate common, unified, comprehensive contracts with companies that collect applicable data.

To navigate the above-described complexities, and to establish authority and accountability within BLS, the agency should build data modernization into its organizational structure. BLS should designate a single, high-level person within the agency, preferably at the deputy commissioner level, whose job is to lead data transformation efforts. Having this responsibility explicitly designated would facilitate a focused, coordinated effort and would ensure accountability. This person also could be the visible point person in collaborations with other statistical agencies. A key objective is to avoid duplicative efforts that likely would arise if data transformation proceeded in a more decentralized way.

In addition to facilitating administration of workflows, such formalized institutional arrangements would signal a commitment by senior leadership to expand the use of alternative data sources for statistical purposes. The data transformation lead would also be part of the team to develop communication strategies to work with Congress to seek the necessary resources to implement changes and highlight the value of the task to user communities.

BLS should continue to look externally for data modernization models as well. With price measurement in particular, ample opportunities exist to learn from and adopt innovative approaches pioneered by statistical offices internationally. BLS should enhance its contacts and collaborations with CPI staff in statistical agencies beyond the U.S. system. Other countries have made significant progress on data transformation—specifically in methods blending scanner and web-scraped data with survey sources—and

CPI staff would benefit from more fully investigating successes and failures experienced during these efforts. Beyond national statistical offices, some of the most innovative work on price measurement—often involving electronic transaction data or web-scraped price data—has taken place in academic settings, so continued collaboration with these experts is likewise encouraged.

The kind of data modernization envisioned in this report will require upfront investments in data acquisition, updating of CPI program production procedures and IT systems, and staff training. In the future, the CPI staff skills will need to shift at least partially away from those needed to analyze structured, survey-based price information and toward those needed to process unstructured price data. In addition to hiring staff with data science skills, BLS should strive to develop this talent inhouse by supporting and rewarding staff who pursue training and educational opportunities to develop the technical expertise that will facilitate data transformation efforts in coming years.

Additionally, because confidence in and understanding by data users of official statistics is critical, successful modernization of the CPI will require that BLS provide clear and consistent communication to stakeholders about the re-design on an ongoing basis. This includes advance notice of changes in an easy to find location on its website, detail about alternative data sources incorporated, transparency around experimental indexes, and updates on the timelines of projects as they evolve.

This page intentionally left blank.