Documentation

The Food-at-Home Monthly Area Prices (F-MAP) data product provides monthly U.S. food price data for 90 food-at-home (FAH) categories across 15 geographic areas of the United States. The F-MAP enables comparisons of food prices over time, across food groupings, and across geographic areas. The F-MAP contains two types of price measures—an average unit value price and a price index—for each food group, geographic area, and month combination for 2012–18.

The methods used to develop the F-MAP data are described in the following publication:

Development of the Food-at-Home Monthly Area Prices Data

This page provides the following information:

Data Sources

The F-MAP is constructed using Circana OmniMarket Core Outlets retail scanner data. Circana retail scanner data is a commercial dataset that contains nominal (not adjusted for inflation) dollar sales (revenue) and quantities of food items sold at FAH retail establishments. The data are based on a sample of about 50,000–60,000 retail stores per year, including grocery stores, supermarkets, supercenters, club stores, drug stores, and convenience stores. Stores that are in the retail scanner data are not a representative sample of stores, so store-level weights that were developed by USDA, ERS (through a contract with RTI International) for the retail scanner data are applied to adjust the store-level sales data to be representative of the population of U.S. stores. More information about the characteristics and statistical properties of the Circana (formerly Information Resources, Inc. (IRI)) retail scanner data are available on the Using Proprietary Data page and in the following publication:

Understanding IRI Household-Based and Store-Based Scanner Data

Disclaimer: The findings in this data product or stated on this page should not be attributed to Circana.

The 90 food categories in the F-MAP are based on the ERS Food Purchase Groups (EFPGs), a categorization system developed by USDA, ERS to classify foods based on characteristics such as ingredients, nutritional content, and convenience level. The EFPGs were created to correspond with the food groups used in both of USDA’s 2015–20 and 2020–25 Dietary Guidelines for Americans document and to capture price premiums for convenience and processing. The categories are hierarchical and flexible so that researchers can aggregate, disaggregate, and move categories around to meet individual research needs. See appendix A of the Development of the Food-at-Home Monthly Area Prices Data report, which provides more information on the EFPGs.

Coverage of Data

The F-MAP provides data across the following dimensions:

  • Monthly, 2012–18
  • 15 geographic areas
    • Nationally
    • 4 Census regions: Midwest, Northeast, South, West
    • 10 metropolitan areas: Atlanta, Boston, Chicago, Dallas, Detroit, Houston, Los Angeles, Miami, New York, and Philadelphia
  • 90 ERS Food Purchase Groups (EFPGs) (Note: A full list of the 90 EFPGs and their descriptions can be found in appendix A of Development of the Food-at-Home Monthly Area Prices Data)
    • 8 groups for grains
    • 23 groups for vegetables
    • 8 groups for fruit
    • 8 groups for dairy and plant-based milk products
    • 14 groups for meat and protein foods
    • 4 groups for prepared meals, sides, and salads
    • 25 groups for other foods

For each of these month, area, and food group combinations, the F-MAP includes the following value variables:

  • Purchase_dollars_wtd: Total weighted sales in U.S. dollars (nominal, i.e., not adjusted for inflation)
  • Purchase_dollars_unwtd: Total unweighted sales in U.S. dollars (nominal, i.e., not adjusted for inflation)
  • Purchase_grams_wtd: Total weighted quantities in grams
  • Purchase_grams_unwtd: Total unweighted quantities in grams
  • Number_stores: Number of stores in geographic area
  • Unit_value_mean_wtd: Weighted mean unit value per 100 grams
  • Unit_value_se_wtd: Standard error of weighted mean unit value
  • Unit_value_mean_unwtd: Unweighted mean unit value per 100 grams
  • Price_index_GEKS: Weighted price index value, constructed using Gini-Eltetö-Köves-Szulc (GEKS) formula

The supplemental price index files for 2016–18 also include six weighted price index variables constructed using different index formulas: Price_index_Laspeyres, Price_index_Paasche, Price_index_Törnqvist, Price_index_Fisher_Ideal, Price_index_GEKS, and Price_index Caves-Christensen-Diewert (CCD).

Methods

This section describes the methods for constructing the price measures in the F-MAP.

Data Preparation

The process to prepare the datasets for creating the F-MAP data product are as follows:

  1. The retail scanner data report sales on a weekly basis. Weekly sales are grouped into the respective months that the sales occurred. In cases where the week straddles 2 months, sales units and values are allocated proportionately based on the number of days in each month.
  2. Unit value outliers are eliminated using the interquartile range (IQR) method. The IQR is the difference between the 25th and 75th percentiles of the price distribution, in this case across all unit values by store and week for each item. A unit value is considered an outlier if the value is below the 25th percentile minus 1.5 multiplied by the IQR or above the 75th percentile plus 1.5 multiplied by the IQR.
  3. The weights of each package are converted into grams to calculate unit values on a per 100-gram basis:
    1. Convert from ounces: gram weight = 28.35 X ounces per package
    2. Convert from pounds: gram weight = 28.35 X 16 X pounds per package
    3. Convert from fluid ounces: gram weight = 29.57 X fluid ounces per package
  4. Individual food items sold in the scanner data (about 600,000 per year) are identified and categorized into 90 food groups, based on the EFPG classification system.
  5. Retailer Marketing Area (RMA) sales data are disaggregated to individual stores. In the retail scanner data, most retailers release data by individual store location. However, some retailers only release data by RMA, a grouping of stores in a retailer-defined geographical area. To disaggregate the RMA sales data to individual stores, the RMA sales data are proportioned to individual stores based on the store sales values in the store-level weight files developed for the retail scanner data. For more information about the store weights, see the Using Proprietary Data page.
  6. Store-level survey weights are applied to each store. Stores in the retail scanner data are not a representative sample of stores, and store-level weights adjust the sales data to be representative of the population of stores nationally and for each geographic area in the F-MAP. In the F-MAP datasets, unit values are provided as both weighted and unweighted estimates, and the price indexes were calculated using weighted data.

Unit Values

Sales (in U.S. dollars) and quantity (in grams) are summed over each month, EFPG, and geographic area. Mean unit values in the F-MAP are calculated by dividing the food group sales by the food group quantity and are standardized to the price per 100 grams. This process is completed as follows: (1) calculate the total purchase values in dollars and in grams for each EFPG in a given month and geographic area, weighted by the store weight for that year of data (note, weight is 1 for unweighted estimates); and (2) divide the total (weighted or unweighted) purchase dollars by the total (weighted or unweighted) grams to get the unit price.
The F-MAP also includes standard errors for the weighted unit values. Standard errors are a measure of the precision of survey estimates and can be used to construct confidence intervals for an estimate. Confidence intervals represent a range of values that are likely to include the actual population mean. Standard errors of the weighted unit values are calculated by re-estimating the weighted unit values 200 times, using replicate weights, and then using the general formula for standard errors. These calculations are described in more detail in the Development of the Food-at-Home Monthly Area Prices Data report.

Price Indexes

Price indexes are a unitless measure of the cost of a basket of goods and are used to measure price changes over time. A price index converts many item-level price comparisons into a single value that quantifies the overall price of the basket at a time and location relative to a base period. The base period for the F-MAP is the national average for each EFPG from 2016 through 2018. Index values lower than 1 indicate prices lower than the national average from 2016 through 2018, while index values higher than 1 indicate prices higher than the national average from 2016 through 2018.

The primary F-MAP price index is constructed using a weighted GEKS index formula (named for contributors Gini, Eltetö, Köves, and Szulc). GEKS is a multilateral price index specifically designed to compare prices over time and space. A GEKS index can also be extended for future years without revising the index numbers that have already been published. A GEKS price index is available for all years of the F-MAP (2012–18). A set of supplemental indexes is also available for 2016–18 as a research series, which includes the bilateral Laspeyres, Paasche, Törnqvist, Fisher Ideal indexes and the multilateral Caves-Christensen-Diewert (CCD) index.

Multilateral price indexes are transitive, which means that any month-area pairing (or entity) can be compared directly with another pairing or through a third pairing, and the ratio between any two pairings is independent of the choice of base period. Transitive indexes are advantageous if the mix of goods being measured is dynamic; that is, if the basket of goods changes due to product turnover. Indexes based on scanner data are dynamic because the indexes include all goods sold in stores, which may change in each time period, rather than a sample of goods selected through a survey. Indexes that are transitive also allow spatial comparisons, regardless of the choice of area used for the base.

As additional years of price data become available beyond the base period, the GEKS index can be updated using a rolling window, or the time period over which the index is calculated. In standard multilateral indexes, as new data become available beyond the initial base period, the index numbers for existing entities must be recalculated because the multilateral index compares product prices in an entity with prices in all other entities. A rolling-window GEKS index compares product prices in a new entity with prices of entities within a rolling window. The F-MAP GEKS uses a 1-year rolling window, which allows maintaining published indexes without revising historical numbers.

Bilateral indexes with a fixed base period can become less representative of the cost of food as the indexes move further away from the base, due to the effects of product turnover, as products are discontinued and new products are introduced. Although bilateral price indexes can be updated using chained indexes, which capture product substitution, chained indexes are subject to drifting. Chain drift is a phenomenon in which the price index drifts lower even as item-level prices return to their base levels. Multilateral indexes are fully transitive and free of chain drift. While chain drift is possible in rolling-window multilateral indexes, if a wide window length is chosen, the rolling-window index will be largely free of drift despite not being fully transitive. The 1-year rolling window used in the F-MAP GEKS has been found to be sufficient to remove chain drift caused by high-frequency data and seasonal variation in variety.

The GEKS index builds upon bilateral indexes as elements. The multilateral GEKS index is the geometric mean of all possible Fisher Ideal index month-area pairings. The Fisher Ideal index is, in turn, based on the geometric mean of the Laspeyres and Paasche indexes. Therefore, constructing a GEKS index requires first calculating Laspeyres, Paasche, and Fisher Ideal indexes. For more information about the construction of the F-MAP GEKS, the rolling-window GEKS, and the five supplemental indexes (i.e., Laspeyres, Paasche, Törnqvist, Fisher Ideal, and Caves-Christensen-Diewert (CCD)), see ERS’s report on the F-MAP methods:

Development of the Food-at-Home Monthly Area Prices Data

Combining EFPGs Into Broader Food Groups

The EFPG-level price indexes can be aggregated into an overall food-at-home Stone price index by summing across the weighted log of each EFPG’s price, using dollar sales shares as weights. This calculation is described in more detail in the Development of the Food-at-Home Monthly Area Prices Data report.

Additionally, this approach can be used to create an index or unit values for a targeted bundle of foods by aggregating across multiple EFPGs proportionally to total sales volume. The dollar sales share can be used as weights to proportionally group the EFPGs into higher tier EFPG groupings or into customizable food groupings, such as combining whole and reduced-fat milk into a single milk grouping.

Related Data

The monthly Consumer Price Index (CPI) produced by the U.S. Department of Labor, Bureau of Labor Statistics (BLS) is the most widely used measure of price changes of consumer items. BLS publishes a national monthly CPI for total food at home (FAH) and for many FAH categories, as well as monthly or bi-monthly price indexes for six aggregate FAH categories for certain metro areas and geographic regions. In addition to the price indexes, the BLS CPI program also publishes monthly average unit price data for about 70 food products at the national level and for the 4 Census regions (i.e., Midwest, Northeast, South, and West regions).

The USDA, ERS Fruit and Vegetable Prices data provide the average retail price per pound and per cup for more than 150 commonly consumed fruits and vegetables. This data product includes standardized prices for individual fruits and vegetables and is more detailed compared with the 31 aggregated fruit and vegetable categories in F-MAP.

The USDA, ERS Purchase to Plate National Average Prices (PP-NAP) data product is a unique price dataset that provides prices for foods in the form that the foods are consumed (e.g., cheeseburger), whereas most food price datasets measure the prices of foods in their purchased form (e.g., ground beef, bun, cheese). The PP-NAP provides prices for thousands of items compared with the 90 aggregate food categories in F-MAP, and the data are advantageous for diet quality and nutrition research. Each PP-NAP price series covers a 2-year period (e.g., 2017–18).

Strengths and Limitations

Existing sources of retail food price data each have specific uses and strengths, and the F-MAP’s unique characteristics contribute to the landscape of public food price datasets. The strengths of F-MAP include a higher frequency of data than many food price datasets, price indexes appropriate for comparing price change over time and across geographic areas, and food groupings designed to facilitate food and nutrition research. However, the F-MAP also has limitations, including timeliness and the level of product detail compared with some available data sources. Users should carefully consider the best resource for their needs.

The monthly CPI data offer limited food categories at a subnational level, and prices in the CPI are not comparable across areas. CPI data are published on a 1-month lag and have an extensive historical series, so CPI data are valuable for monitoring both recent and long-term trends. The F-MAP publishes annual data releases on a greater lag and does not have the length of history compared with the CPI. However, the F-MAP includes greater food category detail for each geographic area, and the F-MAP index is specifically designed for price comparisons across areas, so the F-MAP has advantages for retrospective and spatial food price research.

The F-MAP unit value prices are calculated using all individual products sold at retailers, standardized by weight and categorized into EFPGs (e.g., whole milk, all sizes), to provide a comprehensive coverage of FAH products. In contrast, individual food item prices in the BLS average price data are numerous and varied, but the data do not cover all FAH products purchased by consumers. Prices are tracked for products of a specified type and unit of size, such as 1 gallon of whole milk, so the average price data do not capture prices of all varieties and sizes of products purchased by consumers.

Food price datasets often require a tradeoff between covering more products or providing prices at a higher frequency. The Fruit and Vegetable Prices and PP-NAP data offer richer product detail than F-MAP and CPI, but prices are averaged to an annual or biennial level, respectively, compared with F-MAP’s and CPI’s monthly data. The Fruit and Vegetable Prices data product includes standardized prices for more than 150 individual fruits and vegetables and is more detailed compared with the 31 aggregated fruit and vegetable categories in F-MAP. However, the Fruit and Vegetable Prices data product covers only select years and is not appropriate for monitoring price changes over time, while the monthly nature of F-MAP makes the data well suited to track seasonal price trends for these products. Similarly, the PP-NAP provides prices for thousands of items, compared with the 90 food categories in F-MAP, but the PP-NAP is available biennially and should not be used to track changes over time.

Prices in the F-MAP, Fruit and Vegetable Prices, and PP-NAP data are all derived from Circana retail scanner data, whereas the BLS price products are derived from the BLS CPI commodities and services survey and the BLS Consumer Expenditure survey. Retail scanner data contain more products and price observations compared with the CPI but are based on a nonprobability sample (i.e., not a random sample) of food retailers. CPI contains fewer price observations but is based on a multistage probability sample to construct a representative basket of goods and services. Because the data used to construct the F-MAP are based on a nonprobability sample, some F-MAP data series may exhibit variation across months or years due to the makeup of retailers and categorization of products in the underlying data. In particular, this may affect the sales values (i.e., total dollars and grams) and unit price variables, as well as certain geographic area-food category combinations such as those that are based on fewer observations. 

The primary F-MAP price index is a multilateral GEKS index. Multilateral methods are superior for constructing price indexes using high frequency transaction data such as scanner data. Price indexes are often preferred to using unit values for tracking inflation and in economic analyses because the index values are comparable over time and across areas. Also, price indexes explicitly define assumptions about product substitution and product entry and exit. Price measures based on unit values may be influenced by changes in the product mix or quality differences among items, in addition to changes in prices. Unit values are valuable for understanding prices faced by consumers at a point in time, but price indexes are recommended for tracking price changes over time.

Resources

The methods used in the initial development of the F-MAP data for 2016–18 are described in the following publication:

Information about subsequent data releases and changes to the data product are described on the Update and Revision History webpage. The F-MAP replaced the Quarterly Food-at-Home Price Database (QFAPHD). QFAHPD data for 1999–2010 remain available as archived files. The Update and Revision History page provides a discussion of the differences between the QFAHPD and F-MAP, whereas the following publication describes the QFAHPD:

Resources for understanding the retail scanner data used in the F-MAP are available through the Using Proprietary Data page.

Recommended Citation

U.S. Department of Agriculture, Economic Research Service. (2024). Food-at-Home Monthly Area Prices [data product].