Improving Data to Analyze Food and Nutrition Policies (2005)

As part of its data-gathering activities, the Panel on Enhancing the Data Infrastructure in Support of Food and Nutrition Programs, Research, and Decision Making hosted a workshop in Washington, D.C., on May 27-28, 2004. The workshop served as a forum for input from agencies with policy responsibilities, agencies that provide relevant data, private firms that produce relevant data, and independent researchers. (See Appendixes B and C for the workshop agenda and participants.)

Representatives from the U.S. Department of Agriculture (USDA) and from the Food and Drug Administration (FDA), the National Institutes of Health (NIH), and the Environmental Protection Agency (EPA) discussed current and emerging data needs related to food consumption for policy and decision making. Representatives from key federal statistical agencies that produce food consumption and expenditures datasets—the National Center for Health Statistics (NCHS) of the Centers for Disease Control and Prevention (CDC) and the Bureau of Labor Statistics (BLS)—discussed the strengths and limitations of currently available data. Representatives from private firms that produce data on food consumption and expenditures, the NPD Group and ACNielsen, also discussed current data and data gaps. Outside researchers responded to the presentations and suggested possible improvements to the data infrastructure. The workshop was geared toward national data only, so large state-level dietary databases were not discussed.

This appendix summarizes the workshop. The next four sections cover the sessions at which presenters from federal agencies, private firms, and

academia discussed existing data and data needs in four areas. The last two sections summarize the issues and questions raised by members of the panel and others on two forward-looking topics: use of proprietary data for policy purposes and possible data improvements.

This summary does not offer any conclusions or recommendations; those are in the main body of the panel’s report (Chapter 5). The panel’s mission in this workshop and in its deliberations was to consider modest data improvements that could be made to the current data infrastructure with little expense, such as adding new questions to existing surveys and linking existing datasets. The panel was not asked to consider a major overhaul of data systems.

Many topics were covered during the one-and-a-half day workshop. The workshop began with an overview of food consumption, expenditures, and sales datasets, focusing on the National Health and Nutrition Examination Survey, the Consumer Expenditure Survey, and proprietary datasets from food market research firms. The next three sessions of the workshop were devoted to the food consumption and expenditure data needs for different agencies in USDA and in other federal agencies: one session focused on food marketing and promotion and food market analysis at which there were presentations of the food consumption data needs of USDA’s Agricultural Marketing Service and a description of USDA’s World Agricultural Board’s economic forecasts; one focused on food consumption data and the evaluation of food assistance programs for monitoring and evaluation; and one covered food safety and food consumption data and featured presentations describing current uses of food consumption data by USDA’s Food Safety and Inspection Service, FDA’s Office of Food Additive Safety, and the Environmental Protection Agency’s Office of Pesticide Programs. The fifth session, on food consumption data and health, included presentations from USDA’s Center for Nutrition Policy and Promotion and from the National Cancer Institute. Each of these sessions included a discussion of data needs for that topic by individual researchers working outside of the federal government. The sixth session was devoted to the use of proprietary sources of data to address policy questions and included presentations from representatives of ACNielsen and the NPD Group, along with a presentation about the applications of these data. The final workshop session consisted of a panel discussion on possible data improvements or data linkages, with participation by four panel members. The topics of supplement intakes and food composition databases were discussed only briefly at the workshop.

During this session, presenters reviewed two public datasets, the National Health and Nutrition Examination Survey (NHANES) and the Consumer Expenditure Survey (CE), and data produced by market research firms. NHANES is a major source of food consumption data, but the NPD Group’s National Eating Trends (NET) also gathers food consumption data. Food expenditure data offer another view into food behavior. The CE and proprietary scanner data offer data on household food expenditures. This session of the workshop consisted of overviews of these datasets by Clifford Johnson of NCHS on NHANES, Steven Henderson of BLS on CE, and Abebayehu Tegene from ERS on proprietary sources of data.

The National Health and Nutrition Examination Survey (NHANES) is a cross-sectional survey conducted by NCHS. NHANES’ objective is to assess the health and nutritional status of adults and children in the United States, and the data are used in making public health policy. Clifford Johnson, director of the NHANES program at NCHS, gave an overview of NHANES specifically focusing on its nutrition component. NHANES data collection began in 1971 and became an annual survey in 1999. Information is collected from about 5,000 participants of all ages annually. Participant information is pooled for every 2 years of data collection. Topics covered by NHANES range from mental health to obesity to bone density to environmental exposures (see http://www.cdc.gov/nchs/nhanes.htm). In addition to obtaining information on respondents’ demographic background, basic medical, anthropological, and health status information is obtained. This information is collected when respondents visit Mobile Examination Centers (MECs), which travel around the country to administer the surveys.

NHANES nutrition assessments include dietary nutrient intake; anthropometric measurements; nutritional biochemistries and hematologic tests; physical examination; and interview. Anthropometric measurements include height, weight, and body mass index (BMI). Nutritional biochemistries and hematologic tests include measures of iron and folate status as well as other vitamins, minerals, electrolytes, cholesterol, and triglycerides. NHANES also includes body composition measurements. The physical

fitness assessment considers both daily living activities, such as walking to work or house and yard work, and leisure physical activities, such as muscle-strengthening activities and sedentary activities, including watching television.

The NHANES 1999-2001 dietary assessment included a 24-hour recall of food intake for all participants. Respondents were asked to record everything they ate and drank for 24 hours. A subset of persons was asked for a second day of recall. The day two sample was selected to have representative coverage of the full sample’s age/sex subgroups, but it may not have been fully representative of the total population because it only included about 8 percent of the original sample. During an interview, focused dietary questions were asked, along with detailed questions about supplements and medications and food security. The dietary behavior assessment asked questions about alcohol consumption, salt use at the table, and the frequency of consumption of vegetables, fish and shellfish, and skin on chicken and visible fat on meat. Other questions covered self-reported weight during a person’s life, self-perception of weight, and weight control practices.

NHANES was changed in 2002. Until 2001, the Agricultural Research Service of the USDA had conducted the Continuing Survey of Food Intakes by Individuals (CSFII). The last year of data collection for the CSFII was in 1996 for adults and 1998 for children. In order to reduce redundancy in national dietary data collection, CSFII was discontinued, and the USDA staff responsible for the CSFII began working on the dietary component of NHANES, beginning with the 2002 data collection. The dietary component of NHANES is now called “What We Eat in America,” and it includes 2 days of dietary intake data. The first day of the recall is done by an inperson interview in a mobile examination center, and the second day is done by telephone interview in the participant’s home. CSFII also contained the Diet and Health Knowledge Survey (DHKS), which studied respondents’ dietary knowledge and so provided information on why people choose certain foods and beverages. James Blaylock of ERS noted that the President’s 2005 budget includes a data initiative that will reinstate a diet and health knowledge survey. According to a workshop participant, these questions will add 5-7 minutes to NHANES. The original DHKS lasted 20-30 minutes by telephone, so the questions will be modified. Ronette Briefel also noted that the FDA is currently testing the Health and Diet Survey, which will be similar to the DHKS, but will not include any sort of dietary recall as does NHANES.

Beginning with NHANES 2003-2004, physical activity monitors have been used on participants aged 6 and older. Each participant is asked to wear a monitor for 7 days and then return it by mail to the researchers. The monitor measures the intensity and duration of locomotion activities and number of steps taken. A food frequency questionnaire, known as the Food Propensity Questionnaire (FPQ), will be used for participants aged 2 and over. The FPQ gathers information about food consumption probabilities on a given day, includes 134 questions on individual food items and food groupings, and helps in estimating usual intake for those foods. The FPQ was adapted from the National Cancer Institute’s (NCI) Diet Health Questionnaire. NCI collaborated with NHANES to develop and field a pilot test of the FPQ (see http://riskfactor.cancer.gov/studies/nhanes).

Steven Henderson from BLS gave a presentation on the Consumer Expenditure Survey (CE). The CE comprises two surveys—a diary survey, in which households keep an expenditure diary, and a quarterly household interview survey, which collects information on major purchases on a quarterly basis and background information on consumers. Information is collected continuously from 105 geographic areas of the United States. The diary survey is a record of daily expenses for a consumer unit that is kept by a respondent from each consumer unit for two consecutive 7-day periods. Each quarter, 7,500 consumer units participate in the interview, and 7,500 consumer units annually complete two diaries. U.S. Census Bureau interviewers explain to respondents how to fill out the diary and review the diaries for completeness when they collect them at the end of each week.

All daily expenses, except business expenses and expenses incurred while out of the home overnight, are included in the diary. It also collects demographic, work experience, and income data on household members aged 15 and over. The household interview survey also includes questions to double check food purchases reported in the diary portion of the survey, including questions about food purchased away from home. Data from the two surveys are integrated to provide information about both detailed day-to-day purchases and long-term, major purchases.

The CE annual data are usually available one year after they are collected in ten standard tables sorted by key demographic variables. BLS also produces unpublished tables, which are available on request, that contain more expenditure detail. For example, instead of a general “beef” category,

the unpublished table includes categories for ground beef, chuck roast, round roast, other roast, round steak, sirloin steak, other steak, and other beef. These tables are unpublished due to the higher variance associated with a more limited sample of persons making these expenditures. The CE provides the market basket of weights for the Consumer Price Index (CPI) and an annual snapshot of all spending by key demographic variables for researchers, analysts, and agencies.

BLS provides microdata in a CD format. The microdata include detailed food purchases reported in dollars (not quantity). Summary food statistics combine subcategories, such as fresh vegetables and fresh fruits. Demographics of household members include income, education, age, gender, race, home ownership, job status, plus Food Stamp Program participation, free meals, and poverty status. The CE does not collect data on who in the family made the purchase, who consumed the food, and the quantity purchased.

James Blaylock from ERS noted that public national datasets offer rich and important data, but they take a long time to produce and analyze.¹ Because of their relationships with retailers and their customer’s needs and willingness to pay for the absolutely latest trend information, datasets produced by marketing firms are produced on a more timely basis. These data typically include information about what food people buy, what they eat, and demographic information about households. Such data are a potential resource to the USDA to fill gaps in the current data infrastructure. Abebayehu Tegene from ERS gave an informative presentation on proprietary data. He discussed both scanner data, which provide purchase information, and survey data, which provide consumption information.

Scanner data come from two types of data collections: point-of-sale collections, which use the universal product codes (UPC) of products sold in retail checkout counters to identify products and quantities sold and their prices; and household scanner panels, which are usually random samples of households for which members are asked to scan in the items they have

purchased. Household panel scanner data are generally gathered from large samples. For example, ACNielsen has a household panel of 61,500. The samples are selected randomly on the basis of demographic and geographic targets. (Statistical weighting perhaps can expand the sample to the U.S. level.) The household panels provide information on who buys what and where; the data are gathered weekly; see Box 3-1 in Chapter 3 for the information Tegene noted was gathered by ACNielsen and IRI. There is a 12-day to 3-week lag between data collection and release, depending on the vendor.

The NPD Group and ACNielsen are two companies that provide survey data. The NPD Group has two ongoing surveys: the Consumer Report on Eating Share Trends (CREST) and the National Eating Trends (NET). CREST tracks consumer purchases of prepared meals and snacks at commercial restaurants with an online panel of 52,500. It is geared towards food service information and includes what is eaten, where and with whom, and how much is spent. It contains sales information by food type and outlet. Every day 3,500 questionnaires are sent—3,000 to adults and 500 to teens. The questionnaires ask for information for the day before. CREST represents each of the nine Census Bureau regions. Data are gathered monthly and are available 1 month after collection.

NET collects information on food and beverage consumption for all family members of 2,000 households during a 2-week period, using 14 daily paper diaries. Food and beverage consumption is recorded for each household member for 14 consecutive days. Questionnaires are mailed to 3,500 households that are selected from NPD’s panel of more than 50,000 households, with data collected continuously throughout the year. NET data are weighted by five key demographic variables: income, family size, age, employment status, and race. Data are available 3 months after collection. The diary does not collect information on food prices, but it does collect in-depth information about what and how much is consumed, who consumes it, and when and where food is consumed. The diary also collects information on diet status and type, height and weight, vitamin and mineral consumption, exercise habits, and nutritional attitudes. NET uses USDA’s information on nutrient composition and serving sizes to convert the collected data to food pyramid food groups.

Tegene noted that proprietary data have some shortcomings for some purposes. There is some concern about the sampling frames used to collect the data and how well they represent populations. Some survey data are gathered through the Internet, to which not all households, especially low-

income households, have access. Tegene also said that it is important to note that scanner data are not a direct measure of food consumption because they only include information about store purchases and only address purchase behavior, not consumption.

Helen Jensen of Iowa State University also offered some criticisms of the Nielsen HomeScan Panel during her workshop presentation. Members of the sample have higher incomes and smaller household sizes than the general population, and they are more likely to be married, more likely to be white, and less likely to be Hispanic. Jensen noted that when working with scanner data, it is important to think about whether the sample is representative and if low-income and minority populations are included in sufficient numbers. It is also important to think about whether store purchases are representative: purchases from convenience stores and other small stores may not be included. CREST does not include data on vending machine, bar and tavern, and social catering consumption behavior. It is also important to know if food assistance program purchases can be identified.

The household survey data also do not include information on the price a consumer paid for the goods. Proprietary data, especially very current data, are also more expensive to obtain than public data. Some researchers expressed concern that proprietary data will never be publicly available in tables or other raw form. While NHANES data are available for free download from the Internet in various forms, such as tables or microdata that have been protected for confidentiality, workshop participants questioned whether the firms will ever allow USDA to share those data publicly. There are also methodological issues with self-administered data: they are less accurate than interviewer-administered surveys because people may not completely understand what information to provide.

Despite their shortcomings, proprietary data could be very useful. Tegene and Blaylock noted that they are timely and detailed, providing current information on who purchases and consumes what and where. Blaylock used the case of mad cow disease in the United States as an example. If USDA had had access to private datasets, it would have been able to add questions to a private survey within days of the first news stories. The government would then have known almost immediately how consumers’ purchases and consumption of beef changed in response to the outbreak.

Proprietary data combine detailed product and household information, so researchers can study both retailer and consumer behavior. Proprietary

data lend themselves to different analytic approaches, which allow them to address emerging issues. For example, panel demographic and purchase behavior data can be used to address specific research and policy issues. Tegene noted that these data could be used to answer policy questions about obesity, such as: Do taxes on less healthy foods or subsidies on healthy foods change people’s purchases and consumption? What is the role of advertising in children’s obesity? Vendors could also be asked to provide focused surveys, like surveys on teenagers’ consumption behavior. They could also provide customized reports studying specific products or subgroups of items in greater detail.

During the workshop sessions, representatives of different USDA agencies and other federal agencies were asked to discuss key and emerging policy questions related to food consumption and data needed to answer those policy questions. The next four sections summarize these data needs.

Don Hinman of USDA’s Agricultural Marketing Service (AMS) began his presentation with a description of the agency’s focus on its food-related programs. AMS has both regulatory functions and marketing services: it establishes grade standards, provides grading and inspection services for a fee, provides market news and dairy marketing orders, and sets minimum milk prices; it also purchases commodities for school lunch and other feeding programs and provides oversight of generic promotion programs.

AMS uses consumption data to learn what consumers are looking for and to determine the effect of grade changes on purchases. AMS has an interest in having good consumption data available to researchers and for its own use in analysis of regulatory effects and marketing services. AMS also uses consumption data to analyze the effects of regulations and marketing orders. Panel data that measure purchase decisions in relation to quality attributes, such as fruit maturity, could facilitate better quantitative economic analysis. Consumption data can inform AMS when a commodity industry is in distress due to increased production or large inventories. They can also inform AMS if commodity demand is stagnant or declining. Consumption data that include preferences for food attributes are useful

for analyzing economic effects. For example, does country-of-origin labeling affect food purchases? To what extent are organic products a distinct market? What is the influence of USDA’s organic seal on consumer purchasing behavior?

AMS is most interested in per capita consumption data. This key variable is used in economic models designed to measure promotional effects (the way in which consumption changes after an advertising campaign, for example), so maintaining or improving the quality of per capita consumption data thus is critical to the agency. It is very helpful to distinguish where food is consumed—at home or away from home.

AMS used proprietary panel data from the NPD Group for an informative project on beef consumption. The project looked at servings of beef consumed per household member in a 2-week period. Information on the effects of prices, demographics, health and diet concerns, and the effect of promotions on the likelihood and amount of beef consumption were gathered. The unique panel data allowed AMS to analyze consumption before and after the promotion. These data are then used to estimate a rate of return to producers on promotional dollars expended. Beef promotion led to an increase in beef servings by 0.20 serving per household member. The estimated additional amount consumed due to the promotion was about 2 ounces per household member. The project found that more concern with cholesterol was associated with lower beef consumption and more fast food purchasing was associated with higher beef consumption.

One concern expressed with the use of such proprietary data was losing access to it. AMS has already ceased using the NPD Group’s eatings dataset because of high costs. NPD Group’s servings dataset may also become cost prohibitive. Researchers working on commodity promotion effects value future public investment in panel data and other sources of commodity-specific data on food consumption.

The World Agricultural Outlook Board (WAOB) is charged with coordinating the interagency process for preparing USDA’s economic forecasts of commodity supply and use. WAOB analysts must be knowledgeable about the latest trends in food consumption and the factors that affect the use of commodities. The WAOB’s George Bange began his presentation with a discussion of fundamentals. Commodity analysis uses a supply-and-

demand framework in which the total supply amount (carry-in stocks, production, and imports) and total disappearance amount (exports, domestic use, consumption, spoilage, and carry-out stocks) must equal. Accurate forecasts of domestic disappearance amounts, which are the dominant use category for most commodities, are essential.

Analysis of observed prices for agricultural commodities indicates that short-term changes in supply are typically inversely correlated with prices. In the short term, WAOB tracks changes along a demand curve. Shifts in demand occur over a longer time period, and identifying these shifts is difficult without sound knowledge of underlying factors, such as income, social and demographic characteristics, changes in animal feeding practices and feed technology, government policies, and dietary preferences.

USDA and WAOB project consumption or food use on both a short-term and a long-term basis. USDA publishes the World Agricultural Supply and Demand Estimates (WASDE) report on a monthly basis. This report has many users: farmers use it to project prices and market conditions; the Chicago Board of Trade uses the global supply and demand balances to determine what U.S. prices are likely to be; and the U.S. Secretary of Agriculture uses the report in establishing policy. The report includes projections of demand for a year ahead that usually rely on an extension of current trends of food consumption. USDA also projects long-run trends in consumption for the President’s budget. The current practice is to extend current trends. This is a reasonable action in the absence of better information, but it has become increasingly troublesome as commodity demand-shifting changes in dietary preferences, as well as other factors, have been observed, sometimes during relatively short time frames.

Because the WASDE report projects commodity disappearance about 18 months into the future, U.S. food demand is largely assumed to be static. Eating habits are assumed to change slowly over time and changing diets are difficult to document even after the fact. For example, how much has per capita beef consumption risen as a result of low-carbohydrate diets? Access to grocery store scanner data and private survey data could provide some insight into this and other such questions. The disappearance data do not allow such tracking of demand factors.

Various problems are encountered when gathering data about food consumption. For example, lack of demand information limits the ability of analysts to track emerging trends in the meat sector. The export market and away-from-home consumption are gaining importance in the meat sector. These segments likely have different demand elasticities from the

retail sector, and the price response from a change in supply may be different from those measured when the retail sector was the predominant source of meat consumption. Thus, it is important to have data on where food is consumed—something that is currently lacking for such analyses.

U.S. food and nutrition assistance programs are intended to provide a nutritional safety net. The aim of the programs is to eliminate hunger and guarantee freedom from want by eliminating the worry about lack of food. Food and nutrition programs also promote healthy life-styles through nutrition education and by offering guides to dietary choices. Expenditures on food and nutrition assistance programs in fiscal year 2003 totaled $41 billion. Several participants noted that such a large expenditure warrants evaluation and that enhancements to the current data infrastructure could help to determine whether existing programs accomplish their purpose, to identify new nutritional challenges, and to aid the design of policy.

Various administrative data are needed to evaluate food and nutrition assistance programs. Jay Hirschman from the Office of Analysis, Nutrition, and Evaluation of USDA’s Food and Nutrition Service gave as an example data needed to assess federal meal programs, such as school lunches and breakfasts. First, USDA must know what food was purchased by the institution, then how the food was offered (such as a meal, a la carte, through vending machines, or in the school store). Then it must know what food was consumed at the target meal and what food was consumed over 24 hours. Many factors influence what students are offered and what they are actually eating. Food and production costs, production methods, and menu planning systems affect what an institution offers. The availability and price of competitive meals and plate waste are also important issues for which more information would be useful.

David Smallwood from ERS also spoke about the data needs of food programs and of vendors and suppliers. Analysts need to know the cost to acquire and deliver benefits for food programs. They must understand client access, participation rates, and how participants use their benefits—that is, what food is being purchased and consumed. Vendors and suppliers need

to know how access and availability of foods and costs affect consumption. They must understand channels of distribution and the effect of food labeling. They need to know the effects of certain events on pricing.

Program monitoring and evaluation are essential to providing effective and beneficial food assistance programs. Hirschman gave as an example a description of the School Nutrition Dietary Assessment Study III. One of the charges to the study is to determine student dietary intakes and the effect of USDA meals on students’ specific meal and total dietary intake at school and over 24 hours. The study will compare the nutrient intake of participants to the nutrient intake of nonparticipants who consume meals brought from home; nonparticipants who purchase meals a la carte from the school cafeteria; and nonparticipants who obtain food from other sources. It will also study the effect of school meal participation on a specific day of nutrient intakes of school children. For policy development and program management, food consumption data are best collected in conjunction with program variables that provide insight on action likely to lead to improvement. For example, USDA does not make policy on what children must eat at school, but it does make policy on what students are offered and served; on what price students of different income levels pay; on education and promotion provided by schools; and on the school eating environment.

Hirschman also stated that general surveys, such as NHANES, provide valuable descriptive information, but are less useful for program evaluation because they do not involve random assignment of the intervention (in this case, program participation) and a control group. Special surveys and evaluations can be designed to more specifically target program use and effects of program use and are therefore more useful and meaningful information for program evaluation. An ongoing program of special surveys and evaluations could assess alternatives to current programs or program components. There are many gaps in information on food consumption data and evaluation of food programs. For example, what is the longitudinal relationship between Food Stamp Program participation and hunger and dietary status? What are the effects of free fruit and vegetables in schools on overall dietary intake? What are the very long-term effects of prenatal participation in the WIC (Women, Infants, and Children) program? What is the most effective balance of food assistance, nutrition education and promotion, physical

activity education promotion, and environmental change in order to simultaneously prevent both hunger and obesity?

Smallwood spoke about the major limitations of current food consumption research. Research dollars are limited. Currently available data are also relatively old. Public data take years to gather and release. Once data are available, a model and a framework are needed to appropriately analyze and understand them. Current public surveys, such as NHANES, used to gather only 1 day of food intake data, though NHANES began collecting 2 days of intake per person in 2002. While 1 day of food intake is useful for estimating the population’s mean intake, at least 2 days of food intake data are needed for estimating the distribution of usual nutrient intake and assessing nutrient inadequacy in relation to the recommended levels.² There are special data needs at the consumer level. Food choices, diet quality, food prices, food expenditures, and food security all affect what people eat. Food consumption can affect obesity, cognitive development, and health status and well being. Socioeconomic and demographic data can help researchers understand the interplay between such variables as race, poverty, and food consumption.

Program participation and eligibility status are key pieces of data for researchers. It is often difficult to determine whether people participate or are eligible to participate in the major food and nutrition assistance programs. For example, according to Smallwood, there is an underreporting of 20 percent or more of Food Stamp Program participation on the Current Population Survey, and, according to Steven Haider of the Department of Economics at Michigan State University, the participation rate for those eligible for the Food Stamp Program is probably under 50 percent. Understanding why people decide to participate in food programs requires an accurate estimate of eligibility. Most datasets are weak on eligibility information, especially those with the best outcome information. Some participants noted that NHANES offers the best information on nutrition outcomes, but it does not have detailed questions to estimate eligibility.

Haider discussed food expenditures and their relationship to nutrition outcomes. Food expenditure information is absolutely central to understanding nutrition outcomes. Are low-quality diets associated with low food

expenditures? If not, is this due to lack of dietary information or other constraints? The CSFII collected information on food expenditures, but when CSFII was merged with NHANES, food expenditure data were no longer collected. NHANES is now the major dataset for understanding nutrition outcomes, but the missing food expenditure data limit a full understanding of the relationships described above. Haider said that even a single question on food expenditures added to NHANES would be very useful.

Data linkage could improve nutritional program evaluation. Jay Bhattacharya from Stanford University stated that nutritional outcomes are the result of a complex process in which individual decisions interact with constraints imposed by the environment, such as family, job, school, and state and federal policy. The ability of researchers to link data together is critically important to study this complex process. By linking existing data and developing new data, more can be learned about particular populations, such as low-income people or children, and about institutions, such as schools, food vendors, agricultural markets, and government agencies.

Given limited resources to conduct population surveys, Bhattacharya enumerated two strategies. First, overlapping datasets could be developed, none of which cover everybody or everything, and linked together statistically. Different surveys could be aimed at different populations, such as the poor (Survey of Program Dynamics) or food stamp recipients (National Food Stamp Program Survey). Different surveys have different data strengths. For example, NHANES provides a medical and laboratory examination, CSFII included household food expenditures, and the CE includes food and other household expenditures. Second, a comprehensive data collection effort could be undertaken. A comprehensive strategy could include a health assessment module like NHANES; it could be collected as frequently as the Current Population Survey (CPS); it could include a longitudinal component like the National Longitudinal Survey of Youth; it could oversample needy populations, as does the Survey of Program Dynamics; it could include economic data like the Health and Retirement Study; and it could include a household food expenditure model, as did the CSFII.

If little extra money for surveys is available, Bhattacharya suggested exploiting state and federal administrative databases and making bold data linkages possible, while guaranteeing confidentiality. If lots of extra money were available, there should be a shift to a comprehensive strategy and larger sample sizes and economic modules in order to fill some key data gaps.

There are several agencies with food safety responsibilities in the federal government. Representatives of the Food Safety and Inspection Service (FSIS) of USDA, the Center for Food Safety and Applied Nutrition (CFSAN) of the Food and Drug Administration in DHHS, and the Office of Pesticide Programs (OPP) of EPA made presentations at the workshop. FSIS ensures that meat, poultry, and egg products shipped in commerce for human food are safe, wholesome, and accurately labeled. CFSAN is responsible for promoting and protecting public health by ensuring that the nation’s food supply is safe, sanitary, wholesome, and honestly labeled. OPP is responsible for regulating the nature and amount of pesticide residues in food.

Philip Derfler of FSIS said that his office needs data to help identify public health issues and trends, and to identify and react to public health risks. FSIS uses food consumption data to estimate benefits and costs in food safety and nutrition regulatory impact analysis; to perform risk assessments to assess exposure to meat and poultry products and how people obtain and use information about the safety and nutritional characteristics of products; and to develop and implement policies on labeling and related education programs. FSIS uses data from NHANES and the old CSFII and Diet and Health Knowledge Survey (DHKS), but there are gaps in the data. For example, perceptions of products of new technologies and some commodities are missing. The absence of recent and reliable data hampers the ability to demonstrate economic benefits of changes in nutrition, safe handling, and other labeling rules. DHKS data are currently 8 years old; the survey is no longer conducted. There are concerns about data quality when using these data because they are so out of date.

Available up-to-date data on food consumption of high-risk products by various populations would help FSIS better assess risks of food-borne illness from pathogens. Food production trends and marketing data could help FSIS assess products of new technology on the market and which population groups purchase such products. They would also help FSIS determine trends for allergen labeling requirements. FSIS could better evaluate breaches in safe food handling. Available up-to-date data on food consumption by various populations would help FSIS assess product recall situations. Regulatory activities would be more effective if FSIS had access

to production and marketing trends data, data on diet and health knowledge, and data on health status. FSIS uses survey data on nutrient intake and label use to correlate intake of nutrients with the use of existing nutrition facts. Data on how people obtain and use information on characteristics of meat and poultry products would help suppliers understand the economic effects of new requirements and assess enhanced health benefits to consumers. Finally, FSIS could use up-to-date data on health status and food consumption to determine the adequacy of petitions for nutrient fortification of meat and poultry products.

Michael DiNovi from CFSAN reviewed his center’s program needs. For its food and ingredient safety assessments, the center needs various exposure assessments before using new additives, amending uses of current ingredients, or issuing generally recognized as safe (GRAS) notifications. CFSAN also needs data for its contaminant risk assessments for both naturally occurring and human-made contaminants.

DiNovi said that CFSAN’s Total Diet Study (TDS), which determines levels of various contaminants and nutrients in foods, is derived from the CSFII or NHANES (depending on the most recent source of information) and includes a seasonal collection of market baskets, a monitoring of trends, and a monitoring of new substances of health concerns. The TDS involves purchasing samples of food throughout the United States, preparing the foods as they would be consumed (table ready), and analyzing the foods to measure the levels of contaminants or nutrients of interest. Food samples are purchased by FDA personnel from supermarkets or grocery stores in selected cities, and the samples are sent to FDA laboratories for analysis. Dietary intakes of these analytes are then estimated for the U.S. population by multiplying the levels found in the TDS samples by the amounts of foods consumed based on surveys, such as the CSFII, conducted by the USDA (see http://www.cfsan.fda.gov/~comm/tds-hist.html). As previously noted, the CSFII was discontinued, and the data are aging rapidly. CFSAN’s applied nutrition programs are using NHANES data for its Preventing Obesity Through Better Nutrition Project.

CFSAN’s future needs include continuing access to multiday food intake surveys. DiNovi and other participants noted that if any extra food consumption research money were to become available, it should be spent on increasing the number of days of dietary recall. But Susan Krebs-Smith of NIH and Barry Popkin of the University of North Carolina stated in a later workshop discussion that increasing 2-day dietary recall by 1 or 2 days would not increase understanding of usual dietary intake. They said that a

food propensity questionnaire provides more information on usual intake than 3 or 4 days of dietary intake recalls could provide. CFSAN would also like 24-hour dietary supplement recall.

David Miller of OPP gave a presentation on his office’s use of the CSFII. OPP estimates dietary exposure to pesticides based on two separate data sources: the CSFII 1994-1996/1998 and the amount of pesticide in and on food, which comes from field trial data, monitoring data, and market basket survey data. The CSFII is used because it is a nationally representative and statistically based survey that gathered data on the intakes of individuals all seasons of the year and all days of the week. It included a large number of individuals, and populations of interest were oversampled. Finally, CSFII was a high-quality survey, including in-person 24-hour dietary recalls, and it had a high response rate (roughly 75 percent for the 2-day dietary recall). It also included extensive ancillary and demographic data. Miller noted that OPP will rely more heavily on NHANES data in the future now that the CSFII has been discontinued. OPP looks forward to using the NHANES food propensity questionnaire. Miller is concerned that NHANES will not offer a large enough sample size to study small populations in depth.

Neal Hooker from the Department of Agricultural, Environmental, and Development Economics at the Ohio State University provided a data user’s perspective of food safety. The research community has been working for years to fill some data gaps. The Cost of Illness Calculator, available through USDA’s Economic Research Service (ERS), is a useful and customizable tool for estimating the cost of food-borne illnesses. CDC provides another very useful resource—FoodNet—which identifies emerging food-borne infections. FDA’s Operation and Administrative System for Import Support (OASIS) is the most important database for tracking potential problems with imports, but these data are not collected randomly, and the sample size is small relative to the volume of imports. Recall data (when products are recalled from the market due to food safety hazards) help highlight emerging problems. They could be even more useful if they were linked to other analyses, like food processing plant data or census data.

Food labeling is one of the most dramatic recent changes to diet and health knowledge in the United States. Direct surveys, focus groups, or experiments can be used to assess the effect of qualified health claims and other information on food labels.

Eric Hentges from the USDA’s Center for Nutrition Policy and Promotion (CNPP) began by saying that good data lead to good policy. Better data would lead to better policy. It is imperative to gather data on diet knowledge and attitudes. By gathering this type of data, informed decisions can be made, for example, about what kinds of policies will lead people to eat more fruits and vegetables. Both the food guide pyramid and food labeling methods are built on consumption data. Quality data are essential to policy and intervention programs.

Hentges noted that baseline and longitudinal data on such markers as weight and cholesterol are needed to better understand problems like the obesity epidemic. Hundreds of millions of dollars are spent on research trials, but little or no money is spent on gathering data from these trials, according to Hentges.

Currently, many food assistance programs run on performance-based budgets. If an agency cannot prove that its program succeeds at its mission, the program can lose some or all of its funding. If agencies do not have foundational, baseline data, they cannot possibly begin to show that their programs have had an effect on the target population. This is yet another reason why good consumption data are needed, Hentges noted.

Susan Krebs-Smith from the Risk Factor Monitoring and Methods Branch of the NIH’s National Cancer Institute gave a presentation on her organization’s interest in food consumption data. The mission of the Risk Factor Monitoring and Methods Branch is to contribute to reducing cancer in the U.S. population by serving as a critical link between research on the causes and origins of cancer risk factors and targeted interventions for prevention. Many of the risk factors that the agency studies are risk factors not only for cancer, but for other chronic diseases as well. The agency develops and improves methods to assess such factors and provides data to assist in formulating public policies.

The agency’s areas of research include tobacco use; diet; weight, height, and related measures; physical activity; genetics, family history, and individualized risk assessment; sun exposure; and pharmaceutical use. The agency also studies many diet-related topics. It conducts surveillance, such as the prevalence of consuming a certain number of fruits and vegetables daily. It also monitors health objectives, health policy, program evaluation, and health disparities. Krebs-Smith noted that the national surveys do not have a sufficient sample size to study many health disparities. It is difficult to determine differences among groups in dietary patterns with a sample size of only about 5,000 persons per year.

Food consumption survey data are the most direct measure of dietary food intake. Food consumption data can measure nutrients, foods as eaten, food guide pyramid servings, and agricultural commodities. Food consumption data can provide details about timing, patterns, and combinations of foods.

National food consumption survey data have limitations. The annual sample size in current surveys is inadequate for many policy-relevant analyses, especially since the CSFII was discontinued. Discontinuing the CSFII was a loss in terms of national dietary data. Measurement error is a problem with 24-hour dietary recalls. There needs to be a way to assess usual intake because there are great within-person variations. Usual intake of rare foods is impossible to examine with only one 24-hour recall. Dietary supplement intakes are not as well quantified as foods, so they are difficult to incorporate into nutrient intake measurements. The final limitation Krebs-Smith noted was food cost information. It would be helpful to have cost information tied with diets.

Food supply data fill some gaps in understanding the American diet that cannot be filled with food consumption survey data. Food supply data address aggregate consumption and provide researchers with upper bounds on food intake. Also, a consistent methodology has been applied to the food supply data over time so it is possible to study trends over time. This is not the case for food consumption data. Food supply data can also reveal the agricultural implications of eating according to the food guide pyramid recommendations.

Food supply data have some limitations. They cannot be related to health disparities, the relation of diets to other health factors cannot be studied, and food supply data do not offer details about how foods are consumed.

Krebs-Smith noted that NIH is also interested in filling gaps in assess-

ment of diet, weight, and physical activity. To that end, the agency has developed two new modules for use in NHANES. Although 24-hour recalls provide important food consumption information, in order to attain a good estimate of the distribution of intake of specific food and nutrients it is imperative to assess usual intake. With this in mind, the agency developed a food propensity questionnaire that has recently been added to NHANES. The agency also saw the need for an objective measure of physical activity, so the agency has supported the addition of a physical activity monitor to NHANES. The agency is also working to add some questions on personal weight-loss efforts, health practitioner advice on weight loss, and weight loss history.

Subpopulations are difficult to monitor, even with a survey as large and inclusive as NHANES. Krebs-Smith mentioned the idea of a community HANES, which would include scaled-down mobile examination centers that would be able to capture variables such as diet, physical activity, anthropometrics, and biomarkers in one visit. Populations defined by race and ethnicity or by various geographic areas could be studied in depth. Because these populations would be geographically concentrated, collection of community-level variables would be possible. Census-tract level information on neighborhoods—such as the availability of food sources and access to walking areas—could easily be collected.

Barry Popkin of the Department of Nutrition at the University of North Carolina at Chapel Hill gave a presentation focusing on the importance of data linkage and a critique of NHANES for understanding diet and health links. The CSFII and other past USDA surveys had no state or local identifiers or ability to link to price and other contextual data. This lack makes it very difficult to study such issues as state or county-level WIC programs or to evaluate school lunch programs. One of the challenges for NHANES is to be disaggregated to a level at which the richness of other datasets can be fully used. Geocoding could be used to achieve this. In order to study the determinants of dietary behavior, it is necessary to be able to link individual and household data to food price data at the smallest geographic units possible. A vast array of other contextual issues need to be studied to understand how the broader environment affects food choice. Many researchers would like linkage to the actual address of interviewees, and there are ways to do so that would protect the privacy and confidentiality of human subjects.

Popkin then spoke about issues relating to NHANES. Changes in NHANES coding affect trend, program, and policy analysis. He also noted

that NHANES never did bridging studies to understand the effects of changes in methods used to collect data between the key phases of diet collection design in the 1980s and the 1990s.

Popkin noted that there are no national datasets in the U.S. that collect dietary data longitudinally. USDA has funded some hunger questions on the Early Childhood Longitudinal Survey (ECLS), but there are no questions relating to diet. The National Institute of Child Health and Human Development, which cooperates with the Department of Education to conduct the ECLS, has proposed a new birth cohort study, the National Children’s Study. It would examine the effects of environmental influences on the health and development of more than 100,000 children across the United States, following them from before birth until age 21. The goal of the study is to improve the health and well-being of children (see http://nationalchildrensstudy.gov/about/overview.cfm). Unfortunately, the study does not currently include plans for collecting dietary data. This is a missed opportunity, Popkin said.

Abebayehu Tegene began this session by noting that public sources of data may not be sufficient for research or for policy analysis, for which timely data are essential. The private sector could provide some data files of use to researchers and policy makers. The private sector’s infrastructure allows it to conduct focused surveys very quickly and provide customized reports. Private-sector data may not be best for answering questions of diet and health, but they can help look at how market forces influence diet and health. A third provider of proprietary data, IRI (Information Resources, Inc.), was invited to speak at the workshop, but was unable to send anyone. Food consumption data include both dietary intake data generally gathered through dietary recalls and food propensity (or frequency) questionnaires.

John Green, vice president of industry strategy at ACNielsen, presented an overview of his company’s work relating to food consumption and purchases. ACNielsen gathers information about both retailers and consumers. Virtually all retailers except Wal-Mart send ACNielsen price and item information, on a weekly basis. ACNielson edits and processes the data. The company works mostly for manufacturers, retailers, food brokers, and

wholesalers. The company does occasional work for the federal government on an ad hoc basis.

ACNielsen also does survey work through its household panels. Through this survey work, attitudes and behaviors can be linked. Household panel participants are given a scanner that they keep at home. Each time a household member shops, he or she uses the scanner to record each item purchased. For items without a universal product code (UPC), such as some fresh meats and produce, respondents are given a vocabulary code book to identify the item, and they are asked for the product’s weight. Once a week participants send their information to ACNielsen by telephone lines. Information gathered by ACNielsen includes what store was shopped in and the age and sex of the shoppers. Currently, ACNielsen processes 61,000 households every week, and the company is planning a major expansion of the household panels. One of the limitations of these data, however, is that the company has difficulty recruiting certain kinds of households, such as those of minorities, low-income families, and mobile singles.

Cindy Beres, operations manager-Foodworld of the NPD Group, presented an overview of her company’s work relating to food. The company’s Consumer Report on Eating Share Trends (CREST) tracks consumer purchases of prepared meals and snacks from commercial restaurants. CREST is a daily online survey of about 3,000 adults and 500 teens. Behavioral and attitudinal survey questions are included. The CREST survey captures what participants ate yesterday, where they purchased it, where they ate it, who they were with, and how much money they spent.

The NPD Group’s National Eating Trends (NET) tracks food and beverage preparation and consumption habits, including end dishes, ingredients, additives, and cooking aids. NET has been continuous since 1980; it includes 14 consecutive daily food and beverage diaries, which are returned daily. About 60 households begin 14-day diaries every Monday. Data are accessible 3 months after the close of the quarterly data collection period. NPD is currently conducting a supplement of 500 Hispanic households. The NET database variables include description of the food (kind, flavor, type), how it was served (topping, main dish), the brand name, where it was obtained, how it was prepared, the ingredients, and who consumed the food.

After collecting the 14 days of food intake, NPD collects the following information:

• diet status

• diet type

• medical conditions

• height and weight (self-reported)

• vitamin and mineral supplement usage

• vegetarian or not

• exercise (number of days per week, type, and history)

• nutritional attitudes (concerns about sugar, carbohydrates, taste, trans fat, etc.).

NET is similar to national government surveys in that it samples about 5,000 individuals annually, collects similar food details and brand names, and has similar reporting capabilities. However, NET is different in other ways: the primary food preparer reports for all household members; it features a longitudinal design (the 14 days of data collection); and it offers continuous data collection and quarterly data releases. NET’s food journal eliminates interviewer bias, but there is no opportunity to prompt for forgotten foods.

Data from the NPD Group’s nutrient intake database on associated average serving size and nutrient composition are combined with data from NET on eating frequencies to estimate an individual’s intake of macro and selected micro nutrients. The NPD Group mapped NET eating frequencies to the CSFII’s average serving sizes by gender and age and the CSFII Survey Nutrient Database to create the Nutrient Intake Database. This database is currently available for 1998-2003.

Beres ended her discussion by mentioning two other NPD Group services—food safety monitor and dieting monitor. The food safety monitor regularly measures consumers’ level of concern about various food safety issues. The dieting monitor regularly measures awareness and participation of popular diets.

Helen Jensen of Iowa State University began by describing the current consumer market. Incomes have been rising and labor markets are changing. Demographics are changing; the country is becoming more ethnically

diverse. Scientific discoveries of food consumption and health links are making people look more closely at specific foods. New food technologies create challenges for traditional food groupings. For example, orange juice can now be fortified with calcium, so people may now be getting calcium from an unexpected source. The food industry’s influence on dietary choices is also an emerging area of study. Because of all of these new issues, food product detail is often required to support research and decision making.

Household scanner data provide a good amount of product detail and space and time purchase information. They also allow purchases to be matched with demographics. However, there are some problems with data quality. Products with standard UPC codes are easily picked up and recorded with scanner data, but such products as fruits, vegetables, and some cheeses and meats lack standardized UPC codes. The level of product detail can also be a disadvantage because researchers must narrowly define their categories and then seek out the appropriate products from long lists.

Scanner data can be used in evaluation and policy analysis. Scanner data can shed light on market participation: for example, what percentage of consumers is purchasing this product? Researchers and policy makers can also use scanner data to estimate demand parameters for a single product or a group of products. Scanner data can be used to determine both market and nonmarket evaluation: that is, the data can reveal both the market price and the value that consumers place on attributes in products. Scanner data include information on expenditures and expenditure sales, which can be particularly useful if a household receives food stamps or WIC benefits. Policy makers can then study how these households spend their money on food. Scanner data can also provide insight into infrequently purchased products or products only purchased by a few households. Scanner data can inform research and policy issues related to the introduction, adoption, purchase patterns, and demographic factors of new products.

Jensen also offered some criticisms of the ACNielsen HomeScan Panel. The panel members have a higher income and smaller household size than the general population, and they are more likely to be married, more likely to be white, and less likely to be Hispanic. When working with scanner data, it is important to think about whether the sample is representative and if low-income and minority populations are included in sufficient numbers. It is also important to think about whether store purchases are representative: purchases from convenience stores and other small stores may not be included. It is also important to know if food assistance program purchases can be identified.

During this session, four members of the Panel on Enhancing the Data Infrastructure in Support of Food and Nutrition Programs, Research, and Decision Making discussed possible data improvements or data linkages. Ronette Briefel enumerated six areas in which USDA could apply additional funding. First, USDA could augment existing data collections: for example, household assets and expenditure data could be added to NHANES, though it would be important not to overload that survey. Second, USDA could work to more thoroughly analyze existing data collections: there is a lot of untapped information in NHANES 1999-2000 and NHANES III (1988-1994). Third, work could be done to enhance methodological research areas, like dietary knowledge and attitudes, which lags 5-10 years in comparison with other areas, such as dietary intake methodology and physical activity assessment. Fourth, the advantages and disadvantages of cross-sectional and longitudinal data could be studied. What population groups and issues should be studied cross-sectionally or longitudinally? Fifth, information could be gathered about the school nutrition environment, such as what children are offered at meals, what foods are available in vending machines or other sources, what foods are purchased, and where children consume meals. The Youth Risk Behavior Survey or the Behavioral Risk Factor Surveillance Survey could be useful monitoring tools. Finally, datasets could be linked, and they could be developed using common definitions and survey questions on dietary behavior, attitudes, and sociodemographic factors. Briefel added that the discontinuation of the CSFII resulted in a loss of a sample of 5,000 persons. This loss of sample size limits researchers’ abilities to add new modules to food consumption surveys and to study subpopulations.

Laurian Unnevehr raised the issue of linking macrodata with microdata: for example, food disappearance data could be linked to the NHANES to see whether or not one predicts the other. If a prediction is found, this could help researchers understand how short-term microlevel data collection could predict outcomes for national agriculture trends. Household scanner data, including surveys regarding concerns and attitudes, could be linked to national sales trends to see if attitudes and beliefs really affect people’s purchases. Assessing the strength of such linkages would demonstrate whether one data source could be substituted for another. She emphasized that the loss of consumption data linked to both economic variables and diet/health/knowledge information—because of the demise

of CSFII—is an important gap in the ability to answer policy questions. Some way to link economic data with food consumption data could also be considered, as well as a way to link health knowledge with actual consumption behavior. Alternative public investments in a survey that links food purchases with household economics, knowledge, and behavior need to be evaluated, she said. It is possible that a partitioned NHANES could meet these needs, but if not, then some new survey including all three kinds of information, and possibly with more limited consumption detail, could be considered.

Alan Kristal raised the issue of linking NHANES data to Social Security information in order to learn more about respondents’ Medicare participation and income information. He discussed “conceptual linkage,” meaning to develop sets of items that allow certain kinds of parallel analyses across different kinds of surveys. This approach would take a bit of scientific effort: diet knowledge, attitude, and behavior are difficult to measure, and there is currently no agreement on their definitions. Kristal also suggested consideration of overlapping sampling units across some of the large surveys: for example, have NHANES and another large government survey overlap in the same geographic area.

William Eddy of Carnegie Mellon University urged consideration of increasing the sample size and number of questions related to diet and demographics in NHANES. He also suggested consideration of interagency cooperation regarding food consumption data.

The Workshop on Enhancing the Data Infrastructure in Support of Food and Nutrition Programs, Research, and Decision Making covered various topics related to food consumption over the course of its one-and-a-half day meeting. The workshop began with descriptions of key datasets, such as the NHANES, CE, and proprietary datasets. Representatives from various government agencies spoke about the specific food data needs of their offices. Researchers working outside of the federal government gave presentations that voiced their concerns about food consumption data and possible ways to improve the data infrastructure. The workshop provided an opportunity for people from government, private industry, and academia to come together and share concerns and ideas about data on food consumption and expenditures.

One presentation, by Jay Bhattacharya, gave two alternatives for how the data infrastructure could be improved. Overlapping datasets could be developed, none of which cover everybody or everything. Different surveys could be aimed at different populations, such as the poor or food stamp recipients. Or a comprehensive data collection effort could be undertaken that would include a health assessment module that is collected frequently, oversamples needy populations, includes economic data, and includes a household food expenditure model. Many participants expressed concern about the discontinuation of the CSFII—the loss of the 5,000-person sample and the loss of diet and health knowledge questions and questions on food expenditures.

The Panel on Enhancing the Data Infrastructure in Support of Food and Nutrition Programs, Research, and Decision Making considered each of these topics and others for its final report. The panel considered priority areas for new questions to surveys such as the NHANES that could fill gaps in knowledge about how people make food consumption and expenditure decisions. It also considered how alternative data sources, such as those from proprietary firms, could be used to fill gaps.