National Academies Press: OpenBook

A Consumer Food Data System for 2030 and Beyond (2020)

Chapter: 2 ERS's Current Consumer Food and Nutrition Data Infrastructure

« Previous: 1 Introduction
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

2

ERS’s Current Consumer Food and Nutrition Data Infrastructure

The U.S. Department of Agriculture’s (USDA’s) Economic Research Service (ERS) is responsible for collecting information and conducting research on a broad range of policy-rich domains. One such domain is represented by the Consumer Food Data System (CFDS)—defined by the agency as its “portfolio of data resources that measure, from the perspective of a consumer, food and nutrition conditions and the factors that affect those conditions” (Larimore et al., 2018, p. 1). This portfolio of data resources is used in economic analysis by researchers within and outside ERS along with related data resources from other sources. The value of CFDS is enhanced by ERS collaborations, both within USDA (with sister agencies Food and Nutrition Services [FNS], Agricultural Research Service [ARS], Center for Nutrition Policy and Promotion [CNPP], and others) and outside USDA (with National Center for Health Statistics [NCHS], the Census Bureau, Bureau of Labor Statistics [BLS], National Cancer Institute, and others).

Figure 2.1 illustrates the data inputs to the CFDS and the data outputs that result. Lines between inputs and outputs illustrate that ERS combines multiple inputs to provide public outputs. One input to CFDS includes the administrative data from the Supplemental Nutrition Assistance Program (SNAP) and the Special Supplemental Nutrition Program for Women, Infants, and Children (WIC), two large USDA food programs. The Next Generation Data Platform was initiated by ERS in collaboration with FNS and the Census Bureau to add state-level administrative data from SNAP and WIC to the Census Bureau’s Data Linkage Infrastructure, available to researchers only in a Federal Statistical Research Data Center (FSRDC), a secure data center (see Box 2.1). Other inputs to CFDS include

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Image
FIGURE 2.1 Overview of the Consumer Food and Nutrition Data System.
NOTES: ARS = Agricultural Research Service; ATUS = American Time Use Survey; ERS = Economic Research Service; FARA = Food Access Research Atlas; FEA = Food Environment Atlas; FoodAPS = National Household Food Acquisition and Purchase Survey; FSRDC = Federal Statistical Research Data Center; IRI = IRI Worldwide, a vendor of proprietary data; NHANES = National Health and Nutrition Examination Survey; Nielsen = a vendor of proprietary data; NORC = National Opinion Research Center at the University of Chicago, home of the NORC Data Enclave; NPD = NPD Group, a vendor of proprietary data; SNAP = Supplemental Nutrition Assistance Program; WIC = Special Supplemental Nutrition Program for Women, Infants, and Children.
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

data from probability sample surveys (both data from stand-alone surveys, such as Food Acquisition and Purchase Survey (FoodAPS), and data resulting from the addition of ERS modules to surveys conducted by other agencies); proprietary/commercial data (purchased from vendors); and combined data sources, such as the nutrient or food composition (crosswalk-linkage) databases (collaborative efforts among ARS, ERS, and others).

Information developed through linkage is generally more useful than the sum of its parts. Linking databases has great potential to increase their value, and it is a key ERS approach to producing public outputs. Record linkage involves finding the same entity among two or more data sources, which enables linkages over time or across programs. This is extremely valuable and, when successful, it results in an expanded database, especially if there is significant overlap between individuals in at least one of the two databases. However, record linkage can also be time-consuming and expensive to conduct, and it can be prone to error.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

Record matching projects generally require access to a secure data center because of their reliance on Personally Identifiable Information (refer to Box 2.1). As noted above, the most important project undertaken by ERS in this category is its work in support of the Next Generation Data Platform, which makes state-level SNAP and WIC administrative data available for research at an FSRDC. Research projects linking the SNAP data to data from the American Community Survey have also been undertaken by ERS researchers at an FSRDC (Newman and Scherpf, 2013). FoodAPS also involved linkage with SNAP files to verify self-reported SNAP status.

Other linkage projects offer valuable insights, including those employing probabilistic matching and semantic matching. Crosswalk databases link items collected on a survey (such as quantity and type of food consumed) with important attributes (such as nutrients included and their quantities.) The nutrient databases (also called food composition databases), produced by ARS in collaboration with ERS and others, are examples of these crosswalk databases. Further development in nutrient databases are described in Poti and colleagues (2017), and Carlson and colleagues (2019) describe an additional crosswalk between the ARS data and scanner data. The geographic databases from ERS, which include the Food Access Research Atlas

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

(FARA) and Food Environment Atlas (FEA), are also crosswalk databases. These make use of respondent geography, such as county of residence, to link to attributes about that county, such as the percentage of households that are low-income (from the American Community Survey), or average distance from housing units to the closest food store.

ERS uses each of the input data resources in conjunction with the others to prepare public outputs. Public outputs typically include tables or spreadsheets, including crosswalk databases, graphs, maps, and public-use microdata files available on the ERS (or ARS) Website. These data products are checked by the agency to ensure that respondent confidentiality has been protected. The Office of Management and Budget’s (2005) Statistical Policy Working Paper 22 describes statistical disclosure limitation techniques that are used for this purpose,1 but there is ongoing research to develop improved methods, especially at the Census Bureau. ERS also provides confidential FoodAPS microdata files that can be accessed by the public for approved projects at the NORC Data Enclave, a secure data center where respondent confidentiality is protected (refer to Box 2.1).

Most of the data sources described here already play a central role in the current data infrastructure; others are recent innovations that do not yet have a central role but present new opportunities. An example of the latter is the Next Generation Data Platform, which is not widely known among the research community outside of government. It provides state-level administrative data on participation and benefits from SNAP and WIC, as well as on programs sponsored by other agencies, such as the Temporary Assistance for Needy Families (TANF) program and Medicare. Thanks to the Next Generation Data Platform, SNAP and WIC administrative data can be linked to Census Bureau survey data for approved projects. These data are available only to users for approved projects in a secure FSRDC. The data also incorporate Protected Identification Keys (PIKs), which support authorized users in linking individual records across these sources.

Policy and research questions drive ERS’s data investment choices to maintain and advance the CFDS. Broadly speaking, CFDS products, in conjunction with data from other sources, are intended to serve descriptive and monitoring purposes and to provide inputs into causal research. Supporting causal research places greater demands on the data infrastructure than the purely descriptive function places on it. Core topics in such causal analyses may be assessed or reassessed over time. They include understanding the effects of the food environment on diet and health, understanding links between the diet and health of consumers, identifying the extent to which diets are out of balance with dietary guidelines, and measuring the effectiveness of USDA’s food and nutrition assistance programs in improving outcomes.

___________________

1 See https://www.hhs.gov/sites/default/files/spwp22.pdf.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

Many of the questions in CFDS are intended to have a strong geospatial dimension. Data showing geographic variation in outcomes over time can be used to estimate causal impacts. For example, data that contain outcome measures as well as identifications of the county of birth or residence of program recipients at early ages can be used to gauge how people’s exposure to different SNAP policies—that is, to the rollout over time of SNAP policies in different states or counties—affects those outcomes. Another example where fine geographic information is important is in studying how the food environment interacts with locations where program benefits are redeemed, such as grocery stores (e.g., Allcott et al., 2019).

Administrative and commercial/proprietary data have proved to be useful for revealing such geographic granularity. The Store Tracking and Redemption System (STARS)2 from FNS (1989–2017), TDLinx from Nielsen (2004-2017), and ReCount from the NPD Group (1998–2017) have all been used to assess characteristics of the food retail environment, such as the locations and characteristics of food retailers and restaurants.3 Descriptive information on the ways food acquisition and consumption vary based on context is important, but it is also critical to have data on outcomes across locations and time periods that reflect responses to policies over time. Such data can be used to measure how program changes and other changes in the food environment affect food acquisition, food consumption, nutrition, and health.

In the remainder of this chapter, we provide detail on ERS efforts to improve its use of alternative data sources, including surveys, administrative, proprietary/commercial, and combined data, to improve and expand its products. We also point out the collaborators who have facilitated this work. Section 2.1 describes ERS survey data initiatives and summarizes some of the other key federal surveys of importance to economic analysis of the food environment.

One key ERS initiative is FoodAPS, which is currently a stand-alone survey. This is described in the first subsection of 2.1. Another ERS survey-related innovation is the development of modules that are added to surveys conducted by other agencies. Included in that category are the Food Security Module, added to 11 surveys, the Flexible Consumer Behavior Survey, which was added to the National Health and Nutrition Examination Survey (NHANES), and the Eating and Health Module, added to the BLS American Time Use Survey. These are described in the second subsection of 2.1.

___________________

2 STARS includes information about authorized SNAP retailers.

3 See, for example, Taylor and Villas-Boas (2016a, 2016b), which examines the food store choices of low-income households; and Smith and colleagues (2016), which uses FoodAPS data to examine the “SNAP benefit cycle,” in which SNAP participants exhibit higher food consumption shortly after receiving their benefits, followed by lower consumption toward the end of the benefit month.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

Section 2.2 describes the administrative data for SNAP and WIC and summarizes ERS initiatives using those data, including the Next Generation Data Platform. Section 2.3 describes ERS use of proprietary data, including both food acquisition databases and store and restaurant location databases as well as the innovative products that have resulted. Included are two subsections, one describing the advantages of using proprietary data and the second describing the disadvantages. Finally, Section 2.4 describes the ARS Nutrient databases. Figure 2.1 shows these as both an input and an output because they are continually updated and expanded.

2.1. SURVEY DATA SOURCES

Data from probability sample surveys have traditionally been strong in providing representative measures of the population, but in recent years this strength has been challenged by increasing difficulties with participation rates. That in turn makes it important to note the comparatively high respondent burden and low response rates for some surveys. Survey data are also comparatively expensive to conduct on a per-observation basis. Nevertheless, survey data can provide insights into household- and person-level variables about outcomes, information that is frequently missing in administrative data.

Table 2.1 lists and summarizes the national probability sample surveys that the panel thinks have been most important to the analysis of consumer food and nutrition conditions over the past decade or more. Two of these are repeated cross-sectional surveys, two are panel surveys, and one is a longitudinal survey.4 All collect household-level detail, demographic information, and some self-reported program participation. Of these, NHANES is the only survey to collect detailed information about food consumption on the What We Eat in America Module, sponsored by ARS. NHANES also includes extensive self-reported demographic and health-related information as well as results from a physical examination and biomarker specimens from a qualified medical practitioner. There is a long history of food consumption data both on NHANES and on the Continuing Survey of Food Intakes by Individuals (CSFII) collected by ARS in the 1980s and 1990s until 2001, when ARS and NCHS merged their respective food-related collections into NHANES.5

___________________

4 Repeated cross-sectional surveys are conducted regularly, but with a new random sample selected each time. Panel surveys include at least some of the same sampled units in subsequent iterations of the survey to better capture changes over time. Longitudinal surveys collect information from only the same sampled units over time.

5 For an overview of USDA and HHS food consumption surveys, 1936–1998, see https://www.cdc.gov/nchs/tutorials/Dietary/SurveyOrientation/DietaryDataOverview/Info1.htm.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

TABLE 2.1 Summary of Federal Surveys Containing Consumer Food-Related Data

Survey Name National Health and Nutrition Examination Survey (NHANES) Panel Study of Income Dynamics (PSID)
Source NCHS and ARS, conducted by a contractor University of Michigan
Goal To assess health and nutritional status of adults and children. To assess the dynamic and interactive aspects of family economics, demography, and health.
Sample Annual cross-sectional probability sample of 5,000 (achieved) households and individuals. Oversamples persons 60 and over, African Americans, and Hispanics. Current version since 2001. Longitudinal. In 1968, a nationally representative sample of 5,000 families. Oversampled low-income. Genealogical design. In 2017 sample consisted of 10,000 families. Data collected biannually.
What It Collects Demographic, socioeconomic, dietary, and health. The examination consists of medical, dental, and physiological measurements, as well as laboratory tests. Includes What We Eat in America Module, Food Security Module, and Flexible Consumer Behavior Survey. Demographics, employment, income, wealth, expenditures, health, marriage, childbearing, child development, philanthropy, education, etc. Data on food at home, food away from home, total amount of Food Stamps. Food Security Module included in some interviews.
Downside Food-intake recall method undercounts consumption. No panel data. No data on food prices or expenditures; food acquired without reimbursement. No detail on food at home and away from home. No data on food prices, expenditures, consumption, or food acquired without reimbursement.
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Consumer Expenditure Survey (CEX) Survey of Income and Program Participation (SIPP) National Health Interview Survey (NHIS)
BLS, conducted by the Census Bureau Census Bureau NCHS, collected by Census Bureau
To learn how Americans spend their money. To assess income and program participation. To assess health conditions.
Annual cross-sectional probability sample of 6,900 (achieved) households. Began annual collection in 1979. Has a panel component with quarterly interviews. A series of almost quadrennial national probability-sample household panel surveys beginning in 1983. Quarterly interviews until 2014, then annual. Initial 2014 panel sample of 53,000 households. Annual cross-sectional probability sample of expected 35,000 households (in 2019). Began in 1957. Oversamples persons 60 and over, African Americans, and Hispanics.
Expenditures, demographics, and income. Two separate surveys: the Interview Survey and the Diary Survey. The quarterly Interview Survey collects data on large and recurring expenditures with 3-month recall (rent and utilities); and the Diary Survey collects data on small, frequently purchased items, including most food and clothing. Demographic characteristics, labor force participation, cash and noncash income and assets, costs for medical, shelter, child care, dependent care, and other. Occasionally includes the Food Security Module and other topical modules. Monthly event history for 4-month recall (reference period) prior to 2014. Annual reference period since. Incidence of acute and chronic conditions, injury, physician visits, hospitalizations, and related topics using a stable core and changing modules on current health topics. Since 2011 includes adult food security module.
Limited breakdown of spending for food at home. No data on food consumption, quantities purchased, or prices. Annual recall method likely to be subject to undercount. No data about food expenditures, consumption, or prices. No panel data. No data on food expenditures, consumption, or prices.
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

Other surveys important to the analysis of the food environment include the Panel Survey of Income Dynamics (PSID), conducted by the University of Michigan with a variety of sponsors, mostly federal, including the National Science Foundation, National Institute on Aging, and National Institute of Child Health and Human Development, as well as ERS; the Consumer Expenditure Survey (CEX) sponsored by the Bureau of Labor Statistics and conducted by the Census Bureau; the Survey of Income and Program Participation (SIPP), sponsored and conducted by the Census Bureau; and the National Health Interview Survey (NHIS), sponsored by the National Center for Health Statistics and collected by the Census Bureau. However, even with this substantial history of data collection, major information gaps about food and nutrition for the U.S. population remain. The last row in Table 2.1 notes the inadequacies of each survey for purposes of monitoring food and nutrition conditions.

There have also been a large number of one-time surveys connected with particular studies, such as the large and ambitious National Food Stamp Program Survey conducted by FNS in 1996,6 the Healthy Incentives Pilot7 conducted by FNS in 2011–2012, and the Summer Electronic Benefit Transfer for Children study conducted by FNS in 2011–2014.8 These surveys have yielded important insights about food security, nutrition outcomes, and poverty at national, subnational and household levels that have a wide range of policy research applications. Some of these one-off surveys have even been part of randomized controlled trials, thereby extending causal understanding of the way policy changes affect outcomes in peoples’ lives. As an example, the Healthy Eating Pilot showed how subsidizing healthy purchases with SNAP can affect outcomes.

ERS initiatives in probability sample surveys include the 2012 FoodAPS and the addition of modules to surveys conducted by other agencies. These initiatives are described in the two sections below.

The National Household Food Acquisition and Purchase Survey (FoodAPS)9

With guidance from a National Academy of Sciences study convened by the Committee on National Statistics (NRC, 2005), ERS in partnership with FNS launched FoodAPS to close several key data gaps that had been hampering policy research. FoodAPS was intended to capture information

___________________

6 See https://www.ers.usda.gov/topics/food-nutrition-assistance/food-assistance-data-collaborative-research-programs/national-data-sets/#NFSPS.

7 See https://www.fns.usda.gov/snap/hip.

8 See https://clinicaltrials.gov/ct2/show/NCT02877147.

9 Information in this section is drawn from Larimore and colleagues (2018) and presentations made by ERS staff and others at the workshops described in Appendixes A, B, and C.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

about food acquisitions from all sources (food purchases for consumption at home, food purchases for consumption away from home, and food acquired without monetary payment, by source) and to capture information about respondents’ local food environment, such as distance to the nearest grocery store (by type). Another key gap was filled by using administrative data to identify actual SNAP participants.

FoodAPS was collected under the Confidential Information Protection and Statistical Efficiency Act (CIPSEA) of 2002. CIPSEA requires that the collected data be used strictly for statistical purposes and promises respondents high levels of data protection against disclosure of confidential information.10

FoodAPS was conducted between April 2012 and January 2013. ERS has been planning a second version of the survey, FoodAPS-2, which is described later in this subsection. FoodAPS was conducted with interviews spread over a few months, making it difficult to leverage changes in policy or the food environment to understand in a causal fashion how such changes impact food choices, food security, nutrition, or nutrition-related health outcomes

FoodAPS used a nationally representative probability sample of 4,826 households. Four target populations were of particular interest: SNAP-participating households, non-SNAP households with incomes below 100 percent of the federal poverty guideline (and therefore SNAP-eligible), non-SNAP households with incomes between 100 and 185 percent of the federal poverty guideline, and non-SNAP households with incomes above 185 percent of the federal poverty guideline (Page et al., 2019).

FoodAPS oversampled SNAP participants and other low-income households because a primary goal of the survey was to understand the food acquisition behaviors of these groups. The achieved sample included 1,581 SNAP recipient households identified from a list of then-current SNAP participants and 1,197 other low-income households. Together these two household categories made up more than half of the total sample.11 See Box 2.2 for an overview of the FoodAPS sample design. This important use of persons known to be on SNAP as a sampling frame enabled FoodAPS to identify and recruit a sufficient number of actual SNAP recipients.

___________________

10 See Implementation Guidance for Title V of the E-Government Act, Confidential Information Protection and Statistical Efficiency Act of 2002 (CIPSEA) at https://obamawhitehouse.archives.gov/sites/default/files/omb/assets/omb/inforeg/proposed_cispea_guidance.pdf.

11 More precisely, FoodAPS used lists of recent SNAP participants to facilitate finding them for recruitment into the survey. Not all states provided these lists, however. Because of household mobility, changes in program participation status over time, and the absence of lists in some states, FoodAPS relied on a combination of self-reports, verification (when possible) with matching to updated state files, and the presence of observed EBT transactions in FNS’s ALERT (Anti-Fraud Locator EBT Retailer Transactions) file (again, when possible) to identify the 1,581 SNAP households.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

Households participating in the survey were asked to report each “food event”—a food purchase or acquisition—for a 7-day period for all household members. Additionally, FoodAPS collected data on factors that may affect household purchases and food demand, including household income, food assistance program participation, size of the household, food security, health status, food allergies and intolerances, and diet and nutrition knowledge. The survey also collected household information on major nonfood expenditures, such as rent or mortgage, public transportation, and health insurance premiums and other health-related expenses.

To provide information about health-outcome variables, FoodAPS included variables needed to calculate healthy eating index (HEI) scores and body mass index (BMI) scores (the later to capture incidence of obesity), and the Food Security Module (described in the next section). Other key covariates included self-reported SNAP and WIC participation, two administrative measures of SNAP participation, gender, race, marital status, household size, income, education, age, work, and rural tract. See Courtemanche, Denteh, and Tchernis (2019), Meyer and Mittag (2019), and Kang and Moffit (2019) for an assessment of these measures.

The inclusion of information about food acquired without monetary payment is a distinctive feature of FoodAPS. Such foods are an important food source for many families, especially low-income families. The survey collected information on foods acquired from food banks, food pantries, relatives, friends, and home gardens, as well as children’s receipt of a USDA school meal (whether purchased for full or reduced price or received for free). Notwithstanding the value of this approach, FoodAPS respondents may underreport their acquisition of such food due to stigmas associated with it. The survey also captured the geographic location of food events and the distance from the household to food retailers and restaurants.

FoodAPS was pioneering in linking survey data to auxiliary data from a range of sources to reduce respondent burden and enhance capacity for data analysis. It made use of proprietary scanner data (discussed in section 2.3) to create food item descriptions and item weights, and it used SNAP administrative records (discussed in section 2.2) to create the sampling frame and allow for data quality checks on self-reported SNAP use.12 Thirteen data sources were used to enhance the FoodAPS geography component—specifically, to fill in details about the local food environment, such as location and density of retailers, measures of access to these retailers, local food prices, and area demographics. USDA food nutrient databases (see section 2.4) were used to add micro- and macro-nutrient content and food pattern equivalents to the micro record generated by FoodAPS.

___________________

12 The survey contractor matched survey records to state program files to obtain a limited amount of information, such as participation status, date and dollar amount of last issuance, and details (date, place, amount) of EBT transactions by survey households.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

Due to its rich content, FoodAPS has generated a substantial body of research on nutrition assistance, dietary quality of food spending, geographic access to retailers, food acquisition away from home, food security, and food prices.13 Data from the survey have been used in research on the nutritional quality of food purchases and acquisitions, the economics of local food retail access, evaluation of nutrition assistance programs, and other topics (Larimore et al., 2018; Page et al., 2019; Wilde and Ismail, 2018; Kirlin and Denbaly, 2017). Important descriptive and monitoring work has come out of FoodAPS, along with some work on causal questions.

FoodAPS provides data that other existing data sources do not offer. For example, NHANES identifies food intake quantities and health-related information, but it does not capture item-level food purchases or prices, and information on participation in nutrition assistance programs is self-reported by respondents. SIPP provides self-reported food assistance program participation data, with the possibility of self-reported data about Supplemental Security Income, Social Security Disability Insurance, and Old Age, Survivors and Disability Insurance, but it does not provide good data on food intake, quantities acquired, or spending. CEX collects disaggregated food spending information, but it does not have information on quantities acquired, prices, or intake. And proprietary food retail consumer panels do not include foods purchased from restaurants or food acquired without monetary payment from food pantries (Page et al., 2019).

FoodAPS combines administrative data with survey data to generate more reliable—although not perfect—estimates of program participation. Courtemanche, Denteh, and Tchernis (2019), Meyer and Mittag (2019), and Kang and Moffitt (2019) have found inconsistencies in the quality of FoodAPS appended program data across states. In a presentation to the panel for this study, Colleen Heflin laid out reasons for such inconsistencies.14 For example, some states may maintain monthly data while others do not; and disbursement dates or caseload information may or may not be available (the papers cited above found that two states did not report disbursement dates and five states did not provide caseload data).

Nonetheless, while the measures of SNAP based on administrative records are imperfect, the findings of Courtemanche, Denteh, and Tchernis (2019) and Kang and Moffitt (2019) suggest that they are satisfactory in the sense that whatever errors exist do not seem to meaningfully affect overall conclusions (although Meyer and Mittag [2019] are more nuanced in their conclusions). Having three different measures of SNAP participation,

___________________

13 A list of publications using FoodAPS data can be found at https://www.ers.usda.gov/data-products/foodaps-national-household-food-acquisition-and-purchase-survey/research-projects-and-publication.

14Appendix C, Meeting 3, includes a summary of this presentation.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

two of which are administrative, allows for a combined measure that is surely superior to the misreported rates of self-reported program participation found in most U.S. surveys. The aforementioned research found that the biggest challenge in using FoodAPS was data missing because only 22 of 27 states agreed to provide their administrative data for use in the project, and only 20 provided their data in time to be used. To carry out research on outcomes associated with multiple program use, these papers argued for the integration of administrative data on participation in other programs, such as WIC and Medicaid, in addition to SNAP. Since Medicaid is administered by the U.S. Department of Health and Human Services (HHS), doing so would impose important additional costs and coordination issues.

USDA also invested in understanding the strengths and weaknesses of FoodAPS from the perspective of researchers and data users (Wilde and Ismail, 2018). Data users noted several strengths, compared to other data sources, such as inclusion of administrative data on participation in SNAP in participating states, oversampling of SNAP participants, completeness of food acquisition sources including both food consumed at home and food consumed away from home, and linkage of purchase events to specific retailer locations. (Unfortunately, however, distances provided for the last type of information were based on distances between retail location and home rather than distance related to the actual shopping trip). Data users also had favorable views of FoodAPS documentation and support provided by USDA.

Limitations of FoodAPS identified by researchers and data users included the long wait for the initial data release, rounds of updates to files as data cleaning continued after initial release, some missing item-level data, and (less frequently) inconsistent classification of some retail chains and implausible values for some variables. The NORC Data Enclave15 facilitated the use of confidential data, but this involved financial costs for researchers. Moreover, a USDA confidentiality review was required before downloading output or uploading user-provided inputs (such as user-written codes), and the response time for this type of review was variable, with longer response times and greater difficulty for reviews that were in some way atypical or nonstandard.

Data users also had suggestions for improvements in data about WIC participants, food pantry use, and other topics. It was noted that the reference period in FoodAPS, which was 1 week, was short relative to the monthly cycle of SNAP purchases that prior researchers have noted (Tiehen et al., 2017; Shapiro 2005; Wilde and Ranney, 2000; Gregory and Smith, 2019; Dorfman et al., 2018). Luckily, Beatty and colleagues (2019) document that this interview period frequently spanned likely disbursal dates for SNAP and other programs. Some data users raised the possibility of eventually having some type of panel data structure in FoodAPS; while cost

___________________

15 See http://www.norc.org/Research/Capabilities/Pages/data-enclave.aspx.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

considerations may make this challenging at the household level, repeated sampling of some geographic areas in repeated rounds of the survey should be considered.

In the independent assessments of FoodAPS, several challenges were noted:

  • Response burden and response rates. Unit nonresponse could occur at several stages (Petraglia, Kerckhove, and Krenzke, 2016), including the initial screening, initial agreement to participate, initial interview, and final interview. The weighted response rate was 41.5 percent, sufficiently low to require nonresponse analysis (Petraglia, Kerckhove, and Krenzke, 2016; Page et al., 2019).
  • Response fatigue and underreporting. Because FoodAPS involved several distinct survey instruments over the course of more than a week, respondent fatigue and underreporting were serious concerns (Page et al., 2019). Respondent incentives helped ameliorate the problem, but independent assessments found some indication of systematic underreporting in households that might be expected to have higher respondent burden, such as larger households with more events to report (Maitland and Li, 2016).
  • Confirming the status of nutrition assistance program participation. State administrative data files were used in FoodAPS at several stages, including for the initial sampling frame for the SNAP participant sample and, later, for checking SNAP participation status. For 20 of the 27 states that participated in FoodAPS, state SNAP Quality Control agencies provided administrative files that could be used to corroborate participation status (Page et al., 2019). An independent assessment modeled SNAP participation in the remaining seven states that did not provide data files (Maitland and Li, 2016). Once weighted participation counts from FoodAPS were compared to USDA’s SNAP Quality Control files, FoodAPS appeared to underrepresent SNAP participants, particularly at the lowest income levels (Wilde and Ismail, 2018). For WIC, overall weighted participant counts in FoodAPS were lower than expected from national administrative data (Wilde and Ismail, 2018), but this is not surprising given that WIC participation was self-reported and undocumented individuals who receive WIC benefits may be reluctant to report. The facts that many sampled SNAP households were drawn from administrative SNAP records, the recall period was short, and respondents were primed to think about SNAP and food might explain why self-reports of SNAP participation were much more highly correlated with the administrative measures than in most other large surveys with self-reports of SNAP participation.
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
  • Measuring income. FoodAPS included a complex battery of income questions designed to be sufficient to determine SNAP eligibility in most cases. In addition, an initial screener had a simpler set of income questions that was designed to allow triage for purposes of selecting sampled households to meet targets for SNAP participants and low-income nonparticipants. This initial screener appears to have generated underestimates of household income, leading to the misclassification of households at the time of recruitment, which complicated efforts to achieve sample-size goals even when the income variables could later be corrected based on the longer full battery of income questions (Page et al., 2019). Of course, it is hard to adequately measure gross and net income, and most surveys suffer from the challenges of using self-reports for this purpose.
  • Food identification. One of the most complex tasks for FoodAPS was identifying individual food items acquired, both from grocery retailers and from other sources such as restaurants and food pantries (Page et al., 2019). FoodAPS respondents were given a barcode scanner, which allowed 59 percent of food-at-home items to be matched to Universal Product Code (UPC) codes, and another 16 percent of items to be matched to a project-specific, random-weight barcode sheet provided by FoodAPS to the households. Another 20 percent of items were identified based on event receipts, and 4 percent were identified from respondent descriptions. Food-away-from-home items proved even more difficult to identify than food-at-home items. Overall, the use of the hand-held scanners was valuable for many items, although a large part of the data processing burden remained in identifying the many items that could not easily be scanned and matched.

FoodAPS has been useful for providing descriptive information about how people acquire food, and it is unique in tracking food consumed both at home and away from home, including food at work, at school, and elsewhere. In general, it has been good for studying food supply and demand at a point in time. Its greatest strength is its support of systematic descriptive information about where households buy food, what they pay, and where they get food without monetary payment. Much of the value of FoodAPS stems from the way it enables researchers to compare the choices of SNAP recipients with the choices of eligible nonrecipients as well as those who are not eligible. However, it has some weaknesses in identifying eligible SNAP nonparticipants related to broad-based categorical eligibility.16

___________________

16 See https://www.fns.usda.gov/snap/broad-based-categorical-eligibility.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

FoodAPS also supports the study of the average distances from respondents’ homes to the stores where they buy food (although not necessarily the distances of their actual shopping trips) and the study of mean expenditures by food retailer category. FoodAPS data were used to estimate a choice model and to simulate removal of shopping options; this modeling was done to estimate average and heterogeneous willingness to pay at different food retailers (e.g., regular retailers, farmers markets) and to estimate what distances people are willing to travel to acquire food in a structural model (Taylor and Villas-Boas, 2016a).

While there is a substantial and primarily descriptive literature using FoodAPS to analyze aspects of the SNAP program (Wilde and Ismail, 2018; Page et al., 2019), the first round of the survey was less successful in supporting analysis of the WIC program. This is due to the fact that WIC participation was measured using self-reports rather than administrative data, and the groups likely to be users of WIC were not oversampled. ERS plans to correct this problem in future versions of FoodAPS by using administrative data as a basis for sampling WIC participants. Assuming that frames of SNAP and WIC participants will be drawn independently from administrative data, individuals who participate in both programs may require special treatment to generate nationally representative numbers. (Note that ERS presented the proposed enhancements to FoodAPS-2 to the panel in 2018.17 These enhancements are summarized in Box 2.3.)

In addition to information about food acquisition, a great deal has been learned from FoodAPS about the use of SNAP (cross-checked with administrative data), receipt of cash transfers, wages, salary and self-employment income, and receipt of other benefits. The largest contributions of FoodAPS to research have been about stylized facts, such as what is the average distance from home to the places people shop for food and information leveraging the random assignment of the start of the food acquisition week, which has led to studies about the SNAP cycle (Kuhn 2018; Beatty et al., 2019). Many such topics were discussed at the Symposium, “Food Access, Program Participation and Health: Research Using FoodAPS,” held in 2017.18

The lack of panel data in FoodAPS makes it unsuitable for modeling approaches that could estimate the causal nutrition and health impacts of policy changes over time. Tracking even a subset of households over time could help meet this need, especially if the time span coincided with key policy changes. The repeated cross-sections available from FoodAPS

___________________

17 The presentations by Thomas Krenske and Laurie May to the panel are summarized in Appendix B.

18 Papers presented at the symposium are available in the Southern Economic Association Journal, vol. 86, no. 1 (July 2019).

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

and FoodAPS-2 could enable researchers to examine how policy changes implemented during the time between the surveys will affect outcomes, provided the two surveys are conducted in some of the same states. For future iterations, including a subset of repeated cross-sections, would allow researchers to study the effects of state- or county-level variables, for example.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

Data quality is important to all surveys. The underreporting of self-reported income, program participation, and food acquisition and food consumption plagues most surveys that collect such data, and FoodAPS is no exception. One advantage of FoodAPS is that information on SNAP participation was drawn from state administrative data for participating states, which made it possible to check self-reported versus actual participation.

At the same time, FoodAPS crowded out investments by ERS staff in other data products; for example, the Quarterly Food at Home Database has not been updated since Version-2, which covers 2004–2010. Panel members have heard that this is because of the work required to update the code to use IRI databases and that the extra resources were consumed by FoodAPS. This leads to a question as to whether the first FoodAPS was too ambitious. An in-depth discussion of recommended solutions to these shortcomings, including limiting the scope of future rounds of FoodAPS, is presented in Chapter 4.

Use of Supplemental or Specialized Modules

ERS has actively expanded its data system portfolio by creating and sponsoring or cosponsoring specialized modules that could be added to surveys conducted by other agencies.19 This important strategy has been used to enhance old products and develop new ones. Using add-on modules also imposes less survey burden on participants than carrying out a new standalone survey on any given topic. Such add-ons are most useful if they permit novel descriptive measures to be presented or if they are done consistently across time and place, as this enables causal research about how outcomes are affected by policy changes or by changes in the food environment.

ERS sponsors three modules on surveys conducted by agencies other than USDA. First, the Food Security Module, fielded each year since 1995 on the Current Population Survey, tracks household food security over time. This module has also been added to many other federal surveys (listed in Box 2.4). Two more recent modules have allowed researchers to study where people buy things and how they spend time to do so: the Flexible Consumer Behavior Survey, added to NHANES, sponsored by NCHS, and the Eating and Health Module of the American Time Use Survey, sponsored by the BLS.

___________________

19 FNS pays for some of the food security data collections (NHIS, Medical Expenditure Panel Survey, National Survey of Children’s Health, PSID). Nonetheless, these projects require ERS investment in staff time to work with the agencies to obtain the food security items on the surveys, carry out data checks, recode the composite food security variables, etc.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

The Food Security Module

The Food Security Module was developed in response to the National Nutrition Monitoring and Related Research Act, passed by Congress in 1990 (PL101-445), which led to the development of a 10-year plan for assessing the dietary and nutritional status of the U.S. population (NRC and Institute of Medicine, 2013, p. 7). The Food Security Measurement Project developed and tested the food security module survey questions, which was first fielded with the December Current Population Survey (CPS) in 1995. Hamilton and colleagues (1997) reported the first of the national prevalence estimates for food insecurity and hunger. A key innovation in the 1990s, the Food Security Module improved on previous unscientific generalizations about hunger, replacing them with a monitoring data source akin to the poverty rate and unemployment rate.

Adding the food security module to many federal sources has helped to make the food security measurement and research program a success. It is also worth noting that ERS’s food security measurement and research program has become a model for other agencies. For example, the U.S. Department of Housing and Urban Development (HUD) borrowed from the ERS experience by creating and implementing a housing security module. ERS also continues to collaborate across federal agencies to institutionalize food security as a key measure of well-being—for example, in the indicators for America’s Children and the goals for the Healthy People Program.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

The answers to the survey questions in the Food Security Module lead to categorizing a household as either food-secure or food-insecure. Being food-insecure means being unable, at some time during the year, to provide adequate food for one or more household members due to a lack of resources. Another household categorization is very low food security, meaning the normal eating patterns of some household members were disrupted at times during the year and their food intake was reduced below levels they considered appropriate. Evidence suggests that while there may be a subjective component to these measures, they correlate with other measures of hardship and poor nutrition (Gunderson and Ribar, 2011; Bhattacharya, Currie, and Haider, 2004).

The advantage of having a wide range of surveys with the food security module is that together they provide the ability to correlate food security with a variety of other characteristics, depending on the focus of the specific survey. In addition, by tracking food insecurity in various settings, researchers can show how policy changes affect food insecurity as well as a host of other income sources, other measures of program participation, and other health, human capital, and economic outcomes. The long (since 1995) annual history of the food security module with the CPS makes it most useful for looking at the effects of policies. The NHANES data are most useful for cross-checking how food security is related to food intake, program participation, and objective measures of health. The food security module has also been included on numerous nonfederal surveys.

The federal government’s experience with food security measures on surveys has served to illuminate household-level experience with episodes or symptoms of food-related hardship, such as cutting or skipping meals or going a whole day without food because there was not enough money for food. The module has supported monitoring food security trends and contributed to causal studies of program evaluations by including appropriate covariates for use with econometric modeling techniques. Descriptive and monitoring research topics have included, for example, the relationships between disability and food security, between medical hardship (e.g., medication underuse) and food security, between chronic disease and food security, and between kinds of disability and food insecurity. Gundersen and Ziliak (2008) provide a comprehensive review of food security research.

A number of references examine the causal relationship between food program participation and food insecurity. Focusing on recent papers from the past few years that examine the effects of SNAP, we summarize the following: Yen and colleagues (2008) used an instrumental- variables approach with the National Food Stamp Program Survey to suggest that SNAP participation reduces the severity of food insecurity; Gundersen and colleagues (2017) used partial identification models with SIPP data to find that SNAP reduces the prevalence of food insecurity in households with children; and Swann and colleagues (2017) used data from SIPP in a bivariate probit model to explore

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

the relationship between food insecurity, the household’s history during the previous year, and SNAP participation. The results indicate that negative income shocks, moves, and both increases and decreases in household size increase the probability of being food insecure, while SNAP participation is estimated to reduce the probability of being food insecure. Arteaga and Heflin (2014) used variation in state kindergarten eligibility dates to explore the protective effects of national school lunch program participation on household food security among households with a kindergarten-age child in the Early Childhood Longitudinal Study—Birth cohort (ECLS-B), showing support for the contention that the National School Lunch Program reduces food insecurity; and Schmidt et al (2016) found that among nonimmigrant, low-income single-parent families, $1,000 in potential cash or food benefits from a safety net program reduces the incidence of food insecurity.

Of course, the module has limitations. The quality of its resulting data hinges on the ability of households to accurately report assessments that are somewhat subjective. For some research questions, the limitations stemming from small sample sizes are exacerbated by the rare nature of some of the items of interest, such as very low food security. Questions have been raised about how the item response theory models, including the Rasch model, were used to establish the scaling of the module’s questions (NRC, 2006; Wilde, 2004). For studies of policy impact, a problem arises because these surveys have a reference time period, typically annual, that does not match well with administrative data reference periods, which are frequently monthly.

The Flexible Consumer Behavior Survey (FCBS)

FCBS has been fielded as a module on NHANES since 2007. The module supplements the NHANES dietary and health measures with economic information (income, assets, food expenditures) and self-reported information on participation in food assistance programs (SNAP and WIC). It also contains a flexible set of questions that provide information on dietary habits and behaviors, which is useful for linkage to food intake and nutrient data. The module is designed to change according to proposed or current policy climates—to continue providing timely national data to inform food and nutrition policy-making decisions.

This has allowed NHANES data to be used, for example, in high-profile studies that compare outcomes for SNAP participants, low-income nonparticipants, and higher-income nonparticipants, although it is likely that SNAP participation is underreported. FCBS has also collected information on the use of packaged food product labeling, self-assessed diet quality, diet attitudes and behaviors, awareness of MyPlate,20 knowledge about calorie

___________________

20 See https://www.choosemyplate.gov.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

intake needed to maintain current weight, and use of restaurant nutrition labeling when dining out. Data from the module may be supplemented with American Community Survey data on demographic characteristics and local food policy through restricted-use geographic identifiers. A key innovation has been to link food economics, food consumption, and health outcome variables in NHANES, making it possible to conduct research to determine how food security is associated with objective health measures.

One of the primary objectives of adding FCBS to NHANES is to provide national data on both health variables and consumer use of food policy initiatives (such as packaged food product or restaurant menu labeling) to evaluate the impact of federal regulations.21 However, to estimate policy impacts it is necessary to examine repeated measures of outcomes before and after policy changes within the same locations, and this is difficult to achieve with NHANES because the survey does not usually revisit the same geography and is very unlikely to revisit the same persons.

The Eating and Health Module (EHM)

EHM is a supplement to the BLS’s American Time Use Survey (ATUS) that has been fielded twice, from 2006 to 2008 and again from 2014 to 2016. It is cosponsored by ERS, FNS, and the National Cancer Institute. The objectives of this module were to collect data to analyze relationships among time use, eating behavior, and obesity as well as time-use patterns of important subpopulations such as SNAP and WIC participants, grocery shoppers, and meal preparers. The module collects information on eating patterns, grocery shopping preferences, fast food purchases, meal preparation, food safety practices, general health, height and weight, physical activity, and income (Restrepo and Zeballos, 2019; Zeballos, Todd, and Restrepo, 2019). Although these data allow for a useful assessment of effects of policy change or environment change on changes in outcomes, as with other modules their value is limited by the fact that they use self-reports of program participation.

Use of ATUS is motivated by the need for information on how individuals decide to make use of their 24 hours each day, specifically in their decisions that carry short- and long-run implications for their income and earnings, their health, and other aspects of well-being. Understanding time-use patterns can provide insight into economic behaviors associated with eating patterns as well as the diet and health status of individuals. EHM

___________________

21Restrepo, Minor, and Peckham (2018), for example, show that restaurant menu label users consume fewer daily calories than do consumers who notice but do not use the menu labels to decide what to order in restaurants, https://www.ers.usda.gov/publications/pub-details/?pubid=88530.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

data facilitate understanding whether participants in food and nutrition assistance programs face different time constraints than nonparticipants face, thereby informing the design of food assistance and nutrition policies and programs.

An innovative investment that ERS made in EHM was to include time spent eating (while watching TV, for example) among the secondary activities it surveyed. By doing this, EHM paints a fuller picture of how much time Americans spend eating. In a recent report, Zeballos, Todd, and Restrepo (2019) compare the number and timing of eating occasions reported in the 2014–2016 ATUS-EHM to the information reported in the dietary intake component of the 2013–2016 NHANES, which is considered to contain the best available data for estimating average daily dietary intake among the U.S. population. Their findings show that the core ATUS captures only a small share of all daily eating occasions. EHM helps to reduce—but does not eliminate—the gap in the estimated number of total daily eating occasions between ATUS and NHANES.

As mentioned above, EHM collects information on SNAP and WIC participation with the intent to shed light on time use in food-related activities among these groups. This is important, since there is an ongoing debate about whether the SNAP benefit is adequate, a debate that appears to center on claims about participants’ needs for more food spending for processed or prepared foods due to time constraints on home cooking.22 A recent report shows that SNAP participants waited 6.6 minutes longer between primary eating and drinking events than non-SNAP participants did. When looking at food-related activities, the report shows that on an average day in 2014–2016, non-SNAP participants, including low-income non-SNAP participants, spent less time in food preparation and food-related clean-up than SNAP participants. The report also finds that non-SNAP participants spent 36.4 percent more time purchasing nongrocery food than SNAP participants (Anekwe and Zeballos, 2019).

Finally, time use in connection with topics such as geographic access to retailers is also of interest. As with other data sources, EHM has offered an innovation, but more research is needed to better integrate it into research programs in a way that taps into its potential for policy use.

2.2. ADMINISTRATIVE DATA SOURCES

Administrative data are collected by government agencies (state, federal, or local) for purposes of administering a program. Administrative datasets may consist of individual or household applications to participate in a program or denials of eligibility. They may cover information on the

___________________

22 See Ziliak (2016) for a discussion of the adequacy of SNAP benefits.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

distribution of benefits, information on the use of benefits, or information used to assess program quality. Many of the descriptive, monitoring, and program evaluation goals discussed in this report have been well served by expanded use of administrative data residing within USDA (Larimore et al., 2018).23 Administrative data writ large have been used for a wide range of quasi-experimental and observational studies and are particularly useful for answering questions about a program’s impacts. As noted by Prell (2016):

Administrative data contain complete and reliable information on who participates in a program, how long he or she participated, and the amount of benefits received. In addition, because administrative data have already been collected to operate the program, a re-use of the data for statistical purposes does not incur the cost of launching a new survey to collect comparable data. Linking administrative data with data from large, nationally representative Federal surveys leverages the strengths of the two data sources, gaining results that could not be obtained using either source separately.24

The virtues and shortcomings of administrative data for studying program trends and impacts—whether examined on their own or used to supplement survey data—are well known.25 For example, while these data can identify individuals receiving program benefits, they cannot on their own identify those who are eligible for program benefits but did not apply. So, while they allow researchers to study program trends, they cannot be used on a stand-alone basis to study take-up or program effects. They do offer the potential to study program effects when they are linked to population data in order to study the eligible population for programs. But not all participants can be identified from administrative data with the PIKs, which allow them to be linked to Census Bureau survey and administrative data.

Participation dynamics and intensity can be depicted accurately and in more detail with administrative data than would be possible with survey data alone, because the measurement error found in self-reports of program participation on surveys can be avoided and benefit receipts can be observed directly. However, data on the set of people who could participate in a

___________________

23 The U.S. Office of Management and Budget defines administrative data as data collected by government entities for program administration, regulatory, or law enforcement purpose. They include such records as employment and earnings information on state unemployment insurance records, information reported on federal tax forms, Social Security earnings and benefits, medical conditions and payments made for services from Medicare and Medicaid records, and food assistance program benefits (U.S. Office of Management and Budget, 2014).

24 See https://www.ers.usda.gov/amber-waves/2016/november/illuminating-snap-performance-using-the-power-of-administrative.

25 These strengths and weaknesses are well summarized in Prell (2016), which assesses SNAP performance using the power of administrative data.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

program but do not—those eligible for SNAP, for example—are crucial for modeling take-up of programs, a necessary prerequisite for studying how programs affect health and nutrition outcomes in a causal sense.

For ERS, the most significant administrative data are those developed to support the USDA’s SNAP and WIC programs, in part because obtaining data on the school meals program (also large programs) is even more complicated than obtaining data on SNAP and WIC, given that eligibility for the school meals program is determined by individual school districts. Informing food and nutrition program policy is a particularly important part of the CFDS mandate. For expensive programs such as SNAP, WIC, and school meals programs, it is critical to have reliable evidence quantifying the gains in improving food security as well as minimizing health issues such as obesity. The completeness and accuracy of information on the program participant population are known strengths of administrative data, although recent evidence from FoodAPS suggests that self-reports and administrative data on food program participation are similarly correlated with expenditure and disbursal data.26

The administrative data available for SNAP vary by state but can include state-level participant information, including components of income necessary for determining eligibility for SNAP, data on benefits by month, program participants’ mailing and physical addresses, and likely locations where their benefits were used. They also include SNAP quality control data, which is a sample of applicants’ data on inputs to the eligibility determination process and STARS, which has information on stores in the SNAP program (and those disqualified or investigated for potential fraud) as well as on program benefit redemption. WIC data can include state-level participant information as well as data on stores participating in WIC and on redemptions. The Integrity Profile (TIP) presents a summary of authorized food vendors disallowed from the program because of program violations. WIC also compiles data from the universe of WIC recipients in April of even-numbered years (WIC PC data).

SNAP and WIC are state-administered programs, and there are differences among states. This requires researchers to understand the details of the programs as administered in each state. ERS has compiled the SNAP policy database27 for the purpose of assisting researchers with this, providing details on SNAP policies in each state over time. This database is a key resource for causal research on SNAP. Using integrated or linked survey and administrative data approaches, researchers have used the SNAP

___________________

26 Several papers using FoodAPS, which combined administrative data from several sources with acquisition data and self-reports of SNAP participation, showed that survey data and administrative data from multiple sources had similar levels of discord (e.g., Courtemanche, Denteh, and Tchernis, 2019; Kang and Moffitt, 2019; Meyer and Mittag, 2019).

27 See https://www.ers.usda.gov/data-products/snap-policy-data-sets.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

policy database as a source of instruments for estimating program impacts using the Instrumental Variable estimation method.28 While it is recent, ERS has also created a dataset of the timing of SNAP disbursals, which also offers the promise to provide causal evidence.

Challenges in using administrative data in conjunction with other data sources, such as surveys, for analysis include the fact that definitions of variables may be different and that the populations represented may not overlap. Income may be measured relative to different time periods (e.g., monthly versus annual), households may be defined differently (e.g., those who are living in the home versus those who are sharing expenses, including meals). Surveys often provide data on the U.S. population that lives in households, but they are unlikely to include the homeless, those who live on military bases, or those who live in group quarters.

In research with longitudinal data where the unit of analysis is defined by geography (such as a state, county, or city) at a point in time (such as a year), it is common for some of the variables to be drawn from administrative sources and others to be drawn from survey sources. An archetypal example is the large literature on whether SNAP policy changes and the unemployment rate affect SNAP participation (e.g., Kabbani and Wilde 2003; Ganong and Liebman, 2018). Administrative data can be linked at the household level with survey data to correct or improve certain variables, such as SNAP participation. This was done in part in FoodAPS. One challenge is that data are available only for states that chose to participate by providing their administrative data. There are great opportunities for enhancing use of administrative data to better understand topics such as program participation—these are explored in Chapter 4.

ERS initiatives in the use of administrative records include the Next Generation Data Platform as well as FoodAPS. For FoodAPS (as described in section 2.1) state-level SNAP participant lists (where available) were used as part of the sampling frame; current SNAP population lists were linked to respondent data to verify self-reports of SNAP participation for those selected into the sample from a non-SNAP stratum; and linkage with STARS was used to estimate the distance from each respondent’s home to an authorized SNAP retailer.

The Next Generation Data Platform

Much of the progress in the use of administrative data in the federal statistical system is accomplished through the Census Bureau’s authority to collect and use administrative records. Title 13 of the U.S. Code established the Census Bureau’s legal authorities for collecting, accessing, and protect-

___________________

28 See, for example, Rigdon and colleagues (2017); Miller and Morrissey (2017).

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

ing information about the nation’s population and economy. It specifies that the Census Bureau should acquire and utilize records to the greatest extent possible (§ 6); engage in reimbursable studies and joint statistical projects (§ 8); and protect confidential individual and establishment data, limiting data access to statistical uses (§ 9). Title 13 also authorizes the swearing in of researchers to assist the Census Bureau to achieve its mission (§ 23).

Applications of administrative records for use in Census and survey operations have been developed for purposes of imputation, evaluating coverage, and sampling frame improvement. Under Title 13 authority, multiple data sources have also been linked using the Census Bureau’s Data Linkage Infrastructure to create new statistical products that enable innovative social science research. For information about nonprogram participants, researchers often turn to probability sample survey sources, such as the Census Bureau’s American Community Survey.

In 2012, ERS and FNS formed a strategic partnership with the Census Bureau called the Next Generation Data Platform. This joint project is a long-term effort to acquire state-level administrative data for USDA nutrition assistance programs, especially SNAP and WIC, and to make those data available for linkage to administrative files from other agencies and to surveys conducted by the Census Bureau, already available in the Census Bureau’s Data Linkage Infrastructure. FNS has contacted state SNAP and WIC offices to encourage them to share their USDA administrative data for this project, and the Census Bureau has then contacted those offices to solicit their interest and participation. As of mid-2017, 19 SNAP agencies and 39 counties in California and 11 WIC agencies had agreements to participate in the Next Generation Data Platform. Reaching agreements with states to share confidential administrative data with the Census Bureau is a long-term effort, requiring that a separate agreement be signed by each state. Some states are concerned about unauthorized access if they share data, despite strong security and confidentiality protections at the Census Bureau. The costs associated with preparing the data and its documentation for Census use are also deterrents, although Census has offered to offset such financial burdens. Finally, some states fear reputational harm if their practices and results look less favorable than those of neighboring states in comparison studies.

When such data are available in the Next Generation Data Platform, they will be linkable to survey data collected by the Census Bureau, as well as administrative data from non-USDA sources, which include 17 state TANF agencies, the Veterans Administration (VA), HUD, and HHS (Medicare and Medicaid).29 The resulting combined data are available to sworn

___________________

29 Drawn from Larimore and colleagues (2018, p. 10). For additional details about the Next Generation Data Platform and its application to food assistance research, see https://www.usda.gov/media/blog/2018/01/05/collaboration-across-agencies-supports-food-assistance-research. Also see Prell (2018) for a summary of state-level participation.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

Census Bureau agents for use in analysis at approved FSRDCs, provided their work has been approved as serving Census Bureau purposes.

One of the benefits of this program to USDA was anticipated to be the ability to evaluate SNAP and WIC participation and nonparticipation by county within a state, as well as by various demographic and other data from the American Community Survey, a large probability sample survey of households conducted by the Census Bureau. ERS researchers Newman and Scherpf (2013) accomplished this, linking SNAP participation data to the American Community Survey and developing a measure of SNAP participation rates and SNAP access rates for state-level geographic regions. They presented their data findings for Texas. Using the same methodology, the Census Bureau produced a visualization for New York State that was made available in 2017.30

This joint project addresses questions including: What types of people are likely to be eligible? Of those likely to be eligible, what types are likely to participate? How do caseloads, entries into, and exits out of the program change over time? And, how do the answers to these questions differ across counties? Notably, none of these questions satisfies the burden of providing causal answers, but increased access to data holds the promise to do so.

As noted above, often the greatest value from administrative data is created when they are blended with survey data. The Next Generation Data Platform enables linking of administrative and survey data to improve USDA models of SNAP eligibility and participation rates. Through this collaboration, program administrative records, while imperfect, have been found to accurately reflect information about participants. These sorts of linked data are also being used by a variety of researchers—primarily those internally based at the Census Bureau—to study underreporting of programs and errors in poverty measurement caused by underreporting. The American Community Survey also adds value by including annual income data to model SNAP eligibility, as well as demographic information, so that it is possible to compare the group estimated to be eligible with that estimated to be ineligible. Broader access to these data and information about what is included in them would allow an expansion of this kind of research.

One of the promises of the Next Generation Data Platform is the potential for analysis of linkages of survey data to a number of different administrative datasets. For example, with such a linked dataset, researchers could also learn about interaction effects associated with multiple program participation, such as by comparing those participating in SNAP alone, those in SNAP plus Medicaid, those in SNAP plus TANF, and those in SNAP plus unemployment insurance. One of the challenges in using the Census Data

___________________

30 The visualization is available at https://www.census.gov/library/visualizations/interactive/snap-profiles.html.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

Linkage Infrastructure is that researchers must apply to use a FSRDC and demonstrate how their work would serve a Census Bureau purpose.

2.3. PROPRIETARY COMMERCIAL DATA SOURCES

Proprietary data are collected, owned, and made available by commercial firms. To date, ERS has acquired and used commercial/proprietary data that fall into one of three categories: retail scanner data, household panel and scanner data, and food store and restaurant data.31 Specific companies providing these data to ERS are listed in Box 2.5. One of the challenges with the use of any outside data source is understanding its quality and coverage, key to understanding how the data can best be used. ERS has actively evaluated the quality of proprietary/commercial databases and their fitness for use. Following are the results of that evaluation, along with descriptions of the products.

Retail Scanner Data

Store scanner data capture transactions for purchased products with a Universal Product Code (UPC) on their labels, as well as random-weight products (e.g., fruits and vegetables that are weighed). In so doing, scanner devices can detect and record exactly which products are purchased, the number of items, total dollars spent after discounts (if any), and total gross amount (before discount). As a consequence, researchers can infer the average price paid as the ratio between dollars spent and units purchased, since many retailers do not share individual-level purchase prices with the data aggregators (Nielsen and IRI) but prefer to share average prices within a store or across geographic areas. Of course, this means the price data are not individual prices but are averages, and this may reduce their usefulness for research. InfoScan retail scanner data from the years 2008 through 2017 have been purchased and used by ERS. According to Muth and colleagues (2016), InfoScan captures weekly food sales data from 48,000-plus stores that generate more than 6.6 billion observations per year on expenditures and quantities of UPC and random-weight food products, covering 20 percent of all store locations and 50 percent of total food sales.

Data that ERS has purchased from InfoScan include sales data from individual stores or retailer marketing areas, which represent an unprojected (unweighted) subset of total store data.32

___________________

31 Mary Muth presentation to the panel. See Appendix C for a summary.

32 From the perspective of the firms IRI and Nielsen, store data are seen as a census. Whether or not this is accurate, their methods do not treat these sales data as a sample. IRI data available to ERS include only those stores that have agreed to share their data. This is an obvious research limitation (see Appendix B summary of the presentation by Okrent). Infoscan, for example, does not include all large retailers (including Costco).

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

Private retailers and manufacturers have a long history of collecting consumer data, often for market research purposes, and the value of these data is being extended to consumer food and health research. Granularity is among the strengths of scanner data that motivated ERS to purchase them.33 InfoScan data have become more and more detailed over time; these data are currently available for more than 1 million items identified at the individual UPC level, to which descriptions and attributes are attached, providing information about hundreds of product characteristics (e.g., brand, size, weight, type of packaging). Food price data can be pinpointed geographically to individual stores or market areas (except that some price data are averages reported by retailers), and the data are often available on a weekly basis, with the caveats mentioned above. Such information would be difficult or expensive to obtain in any other way.34 At the same time, these data are collected for marketing or other purposes, are not nationally

___________________

33 See presentation to the panel by Levin and Schweitzer summarized in Appendix A.

34Larimore and colleagues (2018, pp. 6–7).

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

representative, and are not well documented, and store coverage is not equal across all geographic areas.35

Infoscan also includes data that originate on product labels.36 Information on calories, nutrient quantities, daily values, serving size, product claims, and (sometimes) ingredient lists can be culled from label data and attached to data on purchases. Such information allows researchers to examine health- and nutrition-related claims about products acquired—such as, that they are gluten-free, or made from whole grain, or organic, or preservative-free, or hormone- free. With the Purchase to Plate Crosswalk,37 product label data also allows researchers to study the healthfulness of purchases, looking at nutrients or indexes such as the healthy eating index, which measures how healthy a group of foods is per 1,000 calories. (Of course, this approach may miss food waste, and it does not measure the calories consumed.)

Household Panel and Scanner Data

The second category of commercial information is household panel and scanner data. The National Consumer Panel, a joint venture by Nielsen and IRI, is used by both these firms in their household panel data products. It comprises more than 120,000 households, which provide information on their demographic characteristics in addition to food purchase information.38 Around half of these households provide sufficient purchase data to be included in the IRI statistical panel.39 The same households can participate in the panel every year.

Unlike retail scanner data collected at check-out, household scanner data are collected using hand-held scanning devices provided to participating households or using a mobile cellphone app. In this way, purchases can be captured for the panel of households. Again, this source includes products with barcodes and, for a portion of the panel, random-weight products. Data obtained by ERS represent the entire panel, both static households (with weights) and non-static households (without weights). The weights are created by IRI/Nielsen to make the demographics of the

___________________

35 Mary Muth presentation to the panel (see Appendix C for a summary).

36 Initially, these data were available as part of USDA’s Gladson UPC Information Database. See https://data.nal.usda.gov/dataset/gladson-gladson-upc-information-database.

37 The “crosswalk” uses “a combination of semantic, probabilistic, and manual matching techniques to establish a purchase-to-plate crosswalk between the 2013 IRI scanner data and the 2011–2012 USDA nutrient databases” (Carlson et al., 2019).

38 Households self-select to participate in commercial panels, and low-income households are underrepresented.

39 Weekly food purchase data from these households generate 72+ million food product observations from 65 metropolitan statistical areas and 8 nonmarket areas.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

panel match those of the geographic area where the households live. The household data contain geographic information, including the ZIP Code and Census block indicating where each household is located (for IRI) or the three-digit ZIP Code prefix (for Nielsen). This allows researchers to append food environment information from other datasets to household panel data to look at questions about food environment or macroeconomic conditions on household purchasing patterns. Unfortunately, the exact prices paid by the household are not available for all transactions, because often IRI and Nielsen substitute averages across time, space, or chain for retailer store data.

Geospatial Information on Food Stores and Restaurants

The third type of commercial data used by ERS provides geospatial information on food stores and restaurants. ERS has made use of Nielsen’s TDLinx, (2004–2017), NPD Group’s ReCount (1998–2017), and IRI’s InfoScan, summarized above under retail scanner data. The InfoScan retail data also include store information, including store name and corporate parent, address, and retail outlet type (i.e., grocery, convenience, dollar, drug, liquor, mass merchandiser, and club stores).40

TDLinx provides names and geospatial locations of food stores in the United States with sales greater than $1 million. The database is designed to provide universal coverage of grocery, club, convenience, and small-format food-selling stores, although in practice not every unit in the universe may be included. TDLinx comprises two broad retail channels, namely the grocery and convenience channels, and 10 narrower subchannels. In addition to store name, the database includes store address, geocodes, channel and subchannel, chain status, parent company name, sales volume, square footage, number of checkouts, number of employees, and indicators of sales of specific non-food items (e.g., gas, pharmacy, liquor).

ReCount is designed to cover nearly the whole universe of brick- and-mortar food-away-from-home establishments operating in the United States and includes their names and characteristics. In 2018, this included 650,000 restaurants, 130,000 convenience stores, and 450,000 noncommercial locations. To collect information on food service locations, NPD Group reviews chain directories from company headquarters, restaurant guides, industry magazines, and various business lists and conducts Internet and phone verifications. Data collection for a given establishment occurs on a rolling basis so that any one restaurant will be examined biannually. Firm-level characteristics include establishment name, exact geographic

___________________

40Muth and colleagues (2016) found that about 20 percent of store locations are included in InfoScan.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

location, segment (i.e., quick-service versus full-service), restaurant type (e.g., hamburger, Mexican), chain membership, open date, and close date (if applicable).

Abigail Okrent, in her presentation to the panel (see Appendix B), observed that when ERS research (Levin et al., 2018) compared store counts across TDLinx, the National Establishment Time-Series (NETS) database, and InfoScan, the authors found that for the period of 2008–2012 the numbers of stores and food sales found by InfoScan were considerably lower than the numbers found by TDLinx and NETS. A comparison of these totals to totals from the 2012 Economic Census indicates that the version of InfoScan purchased by ERS covered about half of all sales at the store level.

The next two sections address, first, the strengths of commercial/proprietary data and, second, their drawbacks and disadvantages.

The Strengths of Commercial/Proprietary Data

Data originating from commercial sources provide assets for consumer food and health research and evaluation not available elsewhere. For example, retail scanner data have the advantages of providing granular food prices subject to the caveats above, geographic distribution across individual stores or markets where there is coverage, and product-level details such as brand, size/weight, type of package, health and nutrition claims (e.g., gluten-free, type of sugar added, and “good for reducing risk” of heart disease or diabetes). In the near future, such data will also likely provide information about vitamins and minerals, hormone use, and other detailed health improvement claims.41 They also often have the advantage of providing longitudinal data and cross-time measures.42 At the same time, as discussed below, there are weaknesses to these data.

The Food Economics Division (FED) of ERS has played a substantial role in the history of using proprietary data to estimate detailed food prices and quantities of purchases, retail sales, and consumption and purchases of food for both at-home and away-from-home eating. Data on consumer purchase transactions, retail point-of-sales, and information in food labels have been used to help answer questions about the cost of eating a healthy diet and about how the nutrient content of food products changes over time.

In collaboration with other parts of USDA, ERS has been instrumental in integrating scanner data into cost estimates and evaluations of a number of programs. Commercial data have also been applied to policy-oriented and somewhat descriptive research questions about WIC, specifically the composition of WIC-household versus non-WIC household food purchases

___________________

41 See summary of presentation to panel by Brian Burke in Appendix B.

42 See summary of presentation to panel by Abigail Okrent in Appendix B.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

(e.g., types of products, such as breakfast cereals); use of WIC benefits by identified food items; effects of WIC program participation on food purchases; and effects on food purchases of program changes over time. Of course, identifying WIC recipients through self-reporting is less reliable than identifying them from administrative data. Projects are in place to estimate SNAP food package weights and the retail value of the average food distribution program on Indian reservations (FDPIR) using these data.43

Public products produced by ERS that rely on proprietary geographic data (summarized in Box 2.6) include the Food Access Research Atlas (FARA), Food Environment Atlas (FEA), and the Quarterly Food at Home Price Database (QFAHPD). These new geospatial databases provide new measures of food access and the food environment, such as supermarket availability, food choices, health and well-being, community characteristics, and food prices.

FARA provides a spatial overview of access to a supermarket, super-center, or large grocery store, and thereby supports estimation of proximity to stores, both for the overall population and for subgroups of interest, such as low-income people, households without vehicles, or SNAP participants. FARA includes similar estimates refined down to the census-tract level and includes four measures of low-access census tracts, which can be overlaid with low-income tracts. This database supports the mapping of food deserts, as defined by the USDA, HHS, and the U.S. Department of Treasury as low-income census tracts with a substantial number of residents who have little access to retail outlets selling healthy and affordable foods.44

FEA includes more than 200 indicators, aggregated mostly at the county level, that measure both a community’s ability to access healthy food and its success in doing so. Indicators include store and restaurant availability, food assistance use, food prices and taxes, local food initiatives, and residents’ health and physical activity. Many of this atlas’s indicators are culled from already published external data sources, but some are based on ERS data analysis.

Wilde, Llobrera, and Ver Ploeg (2014) used FARA to examine the local food retail environment in the United States. Rhone and colleagues (2017) described the changes in low-income low-food-access census tracts from 2010 to the 2015 updates to FARA. QFAHPD was used along with the ECLS-K class by Wendt and Todd (2011) to show that higher prices of sodas, 100 percent juices, starchy vegetables, and sweet snacks are associated with lower BMI, and that lower prices for dark green vegetables and

___________________

43 See Hastings and Shapiro (2018) and Beatty and Tuttle (2015) on how SNAP benefits are spent.

44 See https://www.ers.usda.gov/amber-waves/2011/december/data-feature-mapping-food-deserts-in-the-us.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

low-fat milk are also associated with reduced BMI; by Todd and colleagues (2011) to describe geographic differences in the prices of healthy foods; by Gregory and Coleman-Jensen (2013) to show that food insecurity was higher in areas with higher food prices; and with NHIS by Bronchetti and colleagues (2019) to show that lower SNAP purchasing power (because of higher prices) leads to a lower utilization of preventative care among children and more days of school missed due to illness.

The databases summarized in Box 2.6 also support enhancing sample designs and adding new variables to survey data. ERS added important additional value by extensively evaluating coverage—both geographically and across other dimensions—through comparisons of these data with Census data on sales in the retail trade and other sources. Comparisons that have been made with total sales from the Census of Retail Trade and other sources suggest that total spending is under-reported in scanner data.45 While some of these databases span nearly the whole country, others are limited in their geographic coverage. For example, Nielsen’s Homescan predominately covers large markets.46

The application of commercial data to the study of food, nutrition, and health topics is now commonplace. Mary Muth, in a presentation to the panel,47 identified more than 150 peer-reviewed publications using some form of scanner and/or label data for food policy research topics. Because no other comparable data source provides the same level of granularity, detail, and frequency, which is needed for many types of food policy analyses, scanner data will continue to be important in a range of research policy areas. Specifically, scanner data can be used

  • to analyze the effects of federal regulations on the healthiness of food acquired (e.g., new Nutrition Facts Labels, revised serving sizes, and the banning of partially hydrogenated oils [trans fatty acids] as an ingredient);
  • as inputs in analyses of new labeling regulations assessing the benefits of changing consumption and the costs (estimated elsewhere) of implementing these changes;
  • to analyze the effects of local regulations, such as taxes on sugar-sweetened beverages, on consumption and to evaluate the incidence of such policies;

___________________

45 Mary Muth presentation to the panel; see Appendix C.

46 For excerpts from an ERS report, see http://qed.econ.queensu.ca/jae/2007-v22.7/hausman-leibtag/Homescan-data.

47 See Appendix C for a summary of this presentation.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
  • to analyze, in a descriptive manner, the effects of voluntary industry initiatives, such as the Healthy Convenience Store Initiative;48
  • to analyze impacts of food contamination outbreaks on sales; and
  • to calculate price indices for a broad range of research studies, to the extent that they incorporate individual prices as well as commodities.

Drawbacks and Challenges in Using Commercial/Proprietary Data

While commercial data will certainly play a growing role in food research, measurement, and assessment, there are considerable hurdles to their use that will need to be overcome before such data can be used as the “gold standard” and in longitudinal assessments. These hurdles include access issues, bias in coverage and representation, perpetually dynamic algorithms, lack of documentation and transparency, fake data and bots, limited scope of organic data sources, and privacy concerns. Each is discussed next, in turn.

Access Issues

One of the most difficult aspects of using commercial or nonfederal administrative data is the challenge of accessing data sources (see National Academies of Sciences, Engineering and Medicine, 2017a). The potential roadblocks are many: propriety information concerning how a dataset was created and unwillingness to share or sell the details, limiting conditions of privacy, restrictions that limit use by some public universities (discussed in Chapter 4), lack of a central data repository, and disparate versions of seemingly similar data (such as refrigerator sensor data that vary across makes and models).

Among the easiest types of data to access are nonfederal administrative data from open data sources; commoditized data; and certain types of social media data (in particular, Twitter). Next easiest to access are data that can be obtained from a single proprietor, although these often require considerable negotiation, contract use restrictions, and time. This latter category includes administrative records from states or organizations; commercial transactions and e-commerce; and health or medical records. Among the most difficult data to access consistently are data derived from social media (such as Facebook or Instagram), web logs, and the so-called Internet of things.

___________________

48 See https://midsouthgreenprint.org/greenprint-20152040/subplanning-projects/healthy-convenience-store-initiative.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

Coverage and Representation Bias

For any research to produce valid and reliable conclusions, it is critical that the data and assessments be representative of the populations or subpopulations of interest and that the degree of representativeness be known or noted. For this reason, ERS has conducted or commissioned research to better understand the characteristics of commercial data sources. In some commercial databases there are gaps in coverage. For instance, Leibtag and Kauffman (2003) and Lusk and Brooks (2011) have documented underrepresentation of lower-income consumers in the Nielsen panel. At the retail level, some data exclude smaller independent stores or private-label products, which make up nearly 18 percent of all food purchases (Cuffey and Beatty, 2019). In terms of scanner data, there can be difficulties in identifying critical groups of interest, such as low-income households, working parents, WIC recipients, or even those who are WIC eligible but not participating in the program (Jensen, 2018).

The method or mode by which data are collected can also be a concern, leading to underrepresentation. For instance, for the populations that ERS wants to follow, it remains an open question whether online data collection methods are a viable option, for example because of unfamiliarity with the use of computers and online methods. Another issue may be that individuals are afraid of having their information in a database and/or fear that may lead to some form of reprisal (such as deportation). Nonresponse or lack of participation by program participants can lead to underrepresentation in the statistics produced.

Using a comparison of sales from the Economic Census to the IRI/Nielson consumer panel data, Muth49 showed that the consumer panel underrepresents sales, particularly for random-weight products such as fresh fruits and vegetables. There are also challenges with using a dataset like the consumer network panel, whose proprietors themselves use “projection factors” or weights derived from proprietary sources or constructed in a proprietary fashion to make the data comparable to national totals for demographics. If these factors or weights are suppressed when the data are made available for use, this is limiting. Comparisons of retail proprietary data and Nielsen Homescan data have also been reported to show discrepancies, as found by Einav, Leibtag, and Nevo (2010), who propose corrections for researchers using Nielsen Homescan data.

Perpetual Dynamic Algorithm

While nonsurvey/nonadministrative data can often provide useful analyses, it is important to remember that many of these data are derived

___________________

49 See summary of Muth’s presentation to the panel in Appendix C.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

from information required to run a system or carry out a process. Regular changes in the platform mechanics and algorithms used to drive these systems reflect the reality that these systems are in place for a business or platform purpose, not for the end goal of generating high-quality research data.

The problem this presents is that changes in data resulting from engineering or programming modifications are often (i) unknown to the researchers and (ii) impossible to disentangle from actual changes in human behaviors, attitudes, or transactions (Lazer et al., 2014). There is a similar issue when proprietary data products change or offer different versions over time.50 There can be a loss of—or significant change in—a data source or a production system, which then leads to different conclusions being drawn.51 One practical example involves the difficulty in making cross-time comparisons when manufacturers assign a new barcode to an existing product (which may be done when a product undergoes a substantial change of some sort). This can make it difficult to separate new products on the market from older ones that have a new label.52

Lack of Documentation and Transparency of Method

Organic data often lack the traditional types of documentation researchers are used to having or may have no documentation at all. This applies not only to the potential fields of data but also, perhaps more importantly, to the ability to trace the origins of the data or changes made to the data at various points before reaching the researcher. This can fundamentally undermine the ability to fully understand what the data actually represent, both conceptually and population-wise, and also limits assessments of data quality.

Fake Data/Bots

The problems caused by bots and fake accounts are ubiquitous within the social media and Internet space. They cause contamination by generating false information either automatically by machine or through use of a cadre of people and are generally designed to push a particular perspective or piece of information. This is particularly problematic in the realm of social media, where platforms are generally open and have fairly low thresholds for entry, leaving themselves vulnerable (Japec et al., 2015). Researchers interested in leveraging social media or scraping websites to gain greater

___________________

50 See Appendix C for a summary of the presentation by Alessandro Bonanno.

51 See Appendix C for a summary of the presentation by John Eltinge.

52 See Appendix C for a summary of the presentation by Mary Muth.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

understanding of food-relevant issues need to take such approaches with care. Although there are commercial packages that claim to help identify fake accounts and bot-generated data, the results of such efforts often conflict across various software packages, thereby rendering such services unreliable for assessing data quality.

Limited Scope of Organic Data Sources

While organic data can often provide very granular and timely data, the information they offer is often of limited scope, being rich in just a small set of variables. Researchers typically have much broader needs, wanting to understand a range of concepts, interactions, and often motivations. The need to understand the “why” behind attitudes and behaviors is still quite germane, but often it is not knowable from organic data alone. For example, scanner data tend not to include whole classes of goods, such as non-UPC products like fresh fruit and vegetables (Jensen, 2018), so it is difficult to assess attitudinal and behavioral changes related to the selection of these mostly healthy alternatives. To remedy this, such data are often best utilized in combination with richer data from surveys or more complete administrative records.

Privacy Concerns

As with nearly all forms of data related to individuals, there is concern about the privacy of the individuals whose data are used. This is a particularly complex issue when commercial and other forms of organic data are used. In those instances, the individuals are rarely (if ever) notified about the potential uses of their data—and even when they are informed, such as through a use agreement, they rarely understand the ultimate implications of potentially sharing their data with others. This is also an area where many laws and regulations have not kept pace with the technology and forms of data generated from these systems and devices. Researchers are therefore urged to approach such usage with caution and take what steps they can to protect the privacy of those whose data are being used.

2.4. NUTRIENT/FOOD COMPOSITION DATABASES

ARS, in collaboration with ERS, FNS, USDA’s Center for Nutrition Policy and Promotion, NCHS, the National Cancer Institute, and others, develops and maintains nutrient/food composition databases or “crosswalk” databases (tables or databases that show the relationship between variables in other tables or databases). Selected food composition databases are listed in Box 2.7. For example, the Food and Nutrient Database for

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

Dietary Studies (FNDDS) identifies the nutrient profiles for 8,000 foods and beverages reported on NHANES. The Food Intakes Converted to Retail Commodities Database (FICRCD) crosswalks the foods and beverages on FNDDS into 65 food commodities (foods directly related to agriculture). These databases are or will be available in USDA’s FoodData Central.53 They are used to add new variables to already collected survey data.

Andrea Carlson, in her presentation to the panel (see summary in Appendix A), described the collaborative project with USDA’s Center for Nutrition Policy and Promotion and Agricultural Statistics Service (ARS) to integrate ARS food composition databases with IRI scanner data to support creation of prices for foods consumed (as collected in NHANES) (Carlson

___________________

53 Link to FoodData Central: see https://fdc.nal.usda.gov/index.html. Link to Food Surveys Research Group, ARS: see https://www.ars.usda.gov/northeast-area/beltsville-md-bhnrc/beltsville-human-nutrition-research-center/food-surveys-research-group/docs/fndds-download-databases.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

et al., 2019). One purpose was to evaluate the prices and nutritional composition of foods associated with the Dietary Guidelines for Americans, especially MyPlate recommendations. The project created the purchase-to-plate crosswalk via the Food Purchase and Acquisition Groups, now called ERS Food Purchase Groups (EFPG), which assign IRI UPC codes to USDA-related food groups based on ingredients, nutritional content, convenience to consumer, and store aisle. An early version of this database and the Nielsen scanner data were used to prepare the QFAHPD to compute quarterly estimates of the prices of 52 food categories. These categories include three categories of fruit—fresh or frozen fruit, canned fruit, and fruit juices—and nine categories of vegetables for 35 regional market groups at several points in time. ERS plans to expand scanner data applications. For example, EFPGs could be included in future iterations of FoodAPS for purposes of food environment studies.

The project also created a price tool for the FNDDS, which estimates prices for the individual foods in FNDDS using scanner data. This tool supports analysis of the relationship between food prices and nutritional content. FNDDS and similar tools have been used to augment publicly available data from existing surveys.

The USDA Branded Food Products Database (BFPDB) was described by Alison Krester and Kyle McKillop at the panel’s second workshop (see Appendix B). The goal of the project is to enhance public health and the sharing of open data by complementing the ARS National Nutrient Database with information on the nutrient composition of branded foods and private label data provided by the food industry.54 The BFPDB covers 229,064 branded products from 238 food categories. Linking the BFPDB to specific years of NHANES surveys, if possible, could more accurately assess dietary intake within the United States. Having a historical record of branded and private-label foods enables comparisons of current and past consumption.

In her presentation to the panel (see Appendix B), Susan Krebs-Smith of the National Cancer Institute illustrated the value added by these crosswalk databases by explaining their application in estimating HEI scores. The HEI is designed to measure conformance of the diets of the U.S. population with the Dietary Guidelines for Americans, which is published every 5 years and which USDA is a partner in creating. The index can be computed for any given basket of food items, whether that is foods consumed, foods purchased, or food commodities.

The HEI is made up of 13 food group components: total fruits, whole fruits, total vegetables, greens and beans, whole grains, dairy, total protein foods, seafood and plant proteins, fatty acids, refined grains, sodium, added sugars, and saturated fats. Weights for constructing the index were derived

___________________

54 See https://data.nal.usda.gov/dataset/usda-branded-food-products-database.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

from the Dietary Guidelines for Americans. To generate a total HEI score for a person or group, information on the quantities of all food groups consumed is needed. The index’s aim is to determine the balance among food groups, including the nine food groups to encourage, such as fruits and vegetables, and the four to reduce, such as discretionary fats and added sugars. A person (group) can improve their HEI score by consuming more of the foods to encourage (these enter the HEI with a positive weight) and by decreasing consumption of the foods to discourage (these add to the HEI if consumed in moderation).

One advantage of the HEI is that scores can be constructed at different levels in the food supply chain, from the agricultural commodities produced by farmers to the food based on those commodities consumed by ultimate consumers and anything in between. In order to examine the HEI at different levels of the food chain, the foods or commodities at that level need to be identified and classified into the 13 categories of the HEI. In his presentation to the panel, Biing-Hwan Lin described translating the data from the ERS Food Availability (per capita) Data System (FADS) to assess the nutritional value of agricultural products produced by farmers and delivered for consumption (see Appendix B). FADS includes data on commodity flows from producer to end user to produce national estimates of the amounts of commodities that are available for consumption by end-users through all channels. It is a proxy for consumption.55 Lin noted that agricultural producers are interested in knowing who consumes their commodities, where they are consumed, and how they are served. Whereas food consumption surveys cover store-bought foods, including fresh produce and meats but also boxed and prepared foods (e.g., cake mix and apple pie), they do not cover the constituent commodities (e.g., apples, wheat, butter, sugar) of those prepared and boxed foods. FADS measures 200 food commodity supplies through the supply chain from the farmer to domestic consumption. The project described by Lin combined food consumption data from NHANES for 2007–2010 with the Food Intakes Converted to Retail Commodities Database and Food Patterns Equivalent Database to estimate food consumption by food groups as specified in the 2010 Dietary Guidelines for Americans. ERS has published statistics covering the years 1994 through 2008 in Commodity Consumption by Population Characteristics,56 using FADS and NHANES data to generate information about the roles of

___________________

55 The ERS loss-adjusted food availability (LAFA) data are derived from the FADS data by subtracting out estimates of food spoilage, plate waste, and other losses to more closely approximate consumption. LAFA is called a preliminary series by ERS because the loss estimates could be improved.

56 See https://www.ers.usda.gov/data-products/commodity-consumption-by-population-characteristics/documentation.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

agricultural commodities versus food and food policy effects on producers and consumers for various demographic characteristics.

For some levels of the food chain, the linkage between foods (whether grown, sold, or consumed) and nutrients or the HEI can be made using one or more of the nutrition databases. For example, constructing the HEI for FADS requires using data from NNDSR, FICRCD, and the U.S. Salt Institute. Constructing the HEI for food consumption data from NHANES requires using data from the FPED and the FNDDS. Crosswalk databases are still needed for food processing and for the community food environment. For example, packaged brownie mix and macaroni and cheese in a box need to be translated into food-pattern equivalents along with nutrient data.

With the appropriate crosswalks, the HEI can be used to evaluate the “diet quality” associated with grocery store purchases, with grocery store circulars, with the places where food is obtained (e.g., different kinds of restaurants or fast food outlets), with schools, with food pantries, and so on. In addition to being used for surveillance and monitoring, the HEI could be used to analyze the relationships between diet patterns and health outcomes. One example of the latter analysis is that of Mancino and colleagues (2018), who use the HEI to assess the quality of food acquired. Fang and colleagues (2019) also use FoodAPS to describe the healthfulness of food acquired by WIC recipients (although they call it the Health Purchasing Index), and Frisvold and Price (2019) use the HEI to characterize the healthfulness of school meals offered by the bulk of schools.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×

This page intentionally left blank.

Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 37
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 38
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 39
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 40
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 41
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 42
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 43
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 44
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 45
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 46
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 47
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 48
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 49
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 50
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 51
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 52
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 53
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 54
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 55
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 56
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 57
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 58
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 59
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 60
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 61
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 62
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 63
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 64
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 65
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 66
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 67
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 68
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 69
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 70
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 71
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 72
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 73
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 74
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 75
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 76
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 77
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 78
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 79
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 80
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 81
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 82
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 83
Suggested Citation:"2 ERS's Current Consumer Food and Nutrition Data Infrastructure." National Academies of Sciences, Engineering, and Medicine. 2020. A Consumer Food Data System for 2030 and Beyond. Washington, DC: The National Academies Press. doi: 10.17226/25657.
×
Page 84
Next: 3 Data and Knowledge Gaps »
A Consumer Food Data System for 2030 and Beyond Get This Book
×
Buy Paperback | $70.00 Buy Ebook | $54.99
MyNAP members save 10% online.
Login or Register to save!
Download Free PDF

Patterns of food consumption and nutritional intake strongly affect the population's health and well-being. The Food Economics Division of USDA's Economic Research Service (ERS) engages in research and data collection to inform policy making related to the leading federal nutrition assistance programs managed by USDA's Food and Nutrition Service. The ERS uses the Consumer Food Data System to understand why people choose foods, how food assistance programs affect these choices, and the health impacts of those choices.

At the request of ERS, A Consumer Food Data System for 2030 and Beyond provides a blueprint for ERS's Food Economics Division for its data strategy over the next decade. This report explores the quality of data collected, the data collection process, and the kinds of data that may be most valuable to researchers, policy makers, and program administrators going forward. The recommendations of A Consumer Food Data System for 2030 and Beyond will guide ERS to provide and sustain a multisource, interconnected, reliable data system.

  1. ×

    Welcome to OpenBook!

    You're looking at OpenBook, NAP.edu's online reading room since 1999. Based on feedback from you, our users, we've made some improvements that make it easier than ever to read thousands of publications on our website.

    Do you want to take a quick tour of the OpenBook's features?

    No Thanks Take a Tour »
  2. ×

    Show this book's table of contents, where you can jump to any chapter by name.

    « Back Next »
  3. ×

    ...or use these buttons to go back to the previous chapter or skip to the next one.

    « Back Next »
  4. ×

    Jump up to the previous page or down to the next one. Also, you can type in a page number and press Enter to go directly to that page in the book.

    « Back Next »
  5. ×

    Switch between the Original Pages, where you can read the report as it appeared in print, and Text Pages for the web version, where you can highlight and search the text.

    « Back Next »
  6. ×

    To search the entire text of this book, type in your search term here and press Enter.

    « Back Next »
  7. ×

    Share a link to this book page on your preferred social network or via email.

    « Back Next »
  8. ×

    View our suggested citation for this chapter.

    « Back Next »
  9. ×

    Ready to take your reading offline? Click here to buy this book in print or download it as a free PDF, if available.

    « Back Next »
Stay Connected!