SOW SOURCES SOUGHT.docx

Statement of Work: Linking USDA nutrition data to IRI scanner data

General Information Scope The USDA seeks expertise to continue developing links between the IRI scanner data and USDA’s Agricultural Research Service (ARS) food group and nutrient data. In addition, USDA seeks expertise to estimate prices for foods reported consumed by participants in the What We Eat in America, the dietary component of the National Health and Nutrition Examination Survey (WWEIA/NHANES), using IRI scanner data, and develop retail ingredient quantities for the foods in WWEIA/NHANES.

Specifically, The USDA seeks to include additional years of both the IRI data and the USDA food composition and nutrient data. Expertise is sought to update these links and tools to estimate prices and ingredients for the following years:

· 2015-16: Link IRI 2015, 2016 to the FNDDS 2015-16 including conversion factors, and develop tool to estimate prices and ingredients for WWEIA/NHANES 2015-16

· 2017-18: Link IRI 2017, 2018 to the FNDDS 2017-18 including conversion factors, and develop tool to estimate prices and ingredients for WWEIA/NHANES 2017-18

Intended Use and Purpose The USDA uses the IRI household data, called the Consumer Network, and IRI retail sales data, called InfoScan, in a variety of purposes including estimating prices, preparing price forecasts, understanding retailer strategies, and modeling consumer behavior. To allow for expanded research on healthy food purchases, USDA has established a mechanism (Linkages) to import USDA food composition and nutrient data into the IRI InfoScan data (the Purchase to Plate Crosswalk -- PPC) and estimate prices for foods reported consumed by participants in the WWEIA/NHANES (the Purchase to Plate Price Tool -- PPPT). The PPPT is used to create national average prices to estimate the market basket for the USDA Food Plans, including the Thrifty Food Plan. An extension of the PPPT is the Purchase to Plate Ingredient Tool (PPIT) which provides retail quantities for the foods reported consumed by WWEIA/NHANES participants. This is used to update the cost of the USDA Food Plans each month, using the CPI’s for food.

Background USDA has expanded the research capacity of the IRI data to include expanded nutrition information from USDA, allowing researchers to measure how well food purchases follow the Dietary Guidelines for Americans by calculating HEI scores based on purchased food. This requires that researchers be able to easily assess the quantities of food purchased in the same units USDA uses in making dietary recommendations to consumers (e.g., the food pattern equivalents found in the USDA Food Pattern Equivalent Database or Food Pattern Ingredient Database (FPED/FPID)). Under USDA contract AG-32SB-12-0011 / AG-32SB-D-14-0281, and GS00F009DA - 1232SB18F0322, USDA established the PPC which includes links between the FNDDS and the IRI data, as well as conversion factors to convert the as purchased weight to the weight in the linked USDA code. The conversion factor includes the yield factor to account for the refuse weight included in the purchase weight and any cooking loss or gain when the USDA code represents a cooked product. While some conversion factors can be established over a broad group of items (e.g. crackers have a conversion factor of 1), others must be established at the IRI item level (e.g. whole carrots and baby carrots have a different conversion factor because whole carrots have skins that many people peel and inedible ends).

There are currently two versions of the PPC:

· PPC 2013:

· Links IRI InfoScan 2013 and FNDDS 2011-12

· Conversion factors for IRI items representing 95 percent of InfoScan sales available to USDA at the barcode level

· PPC 2015

· Links IRI InfoScan 2015 and FNDDS 2013-14

· Links IRI Consumer Network 2015 and FNDDS 2013-14

· Conversion factors for IRI items representing 95 percent of InfoScan dollar sales available to USDA at the barcode level or 95 percent of Consumer Network purchases in dollars In building these versions of the PPC, USDA learned that a combination of semantic, probabilistic, and manual matching works well, but this is not a requirement for establishing new matches. While matches established under the previous round provide a starting point, they are not necessarily correct in the new round. This is because the FNDDS revises the list of included foods each cycle, making some previously established matches incorrect in the new round. In addition, there are about 200,000 IRI barcodes with sales in the “95% of sales group” every year, and 250,000 in a two-year cycle (e.g. 2017 and 2018). The previous versions of the linking database included significantly more UPCs (2015 version includes 525,000 links). USDA also learned that the linking process is best done by breaking down both the IRI and FNDDS data into linking categories. The 2015 PPC development used the IRI “category” column as a basis of the matches. Another possibility is to use the ERS Food Purchase Groups.

This is a complicated linking project, both because there are no common identifiers between the two datasets, and because there are significant changes to the input data with each new set of data. Expertise is sought on improvements to methods since the input data changes every two years.

USDA also seeks expertise in providing a method for researchers to estimate food prices for WWEIA. These prices show the actual cost of consuming foods by taking into account food refuse and cooking factors and combining ingredients together to price mixed dishes. This is known as the Purchase to Plate Price Tool (PPPT). In addition, the Purchase to Plate Ingredient Tool (PPIT) estimates the purchase weight in grams of ingredients, based on a given quantity of a WWEIA/NHANES food. USDA uses the PPPT to generate prices to update the USDA Food Plans, as well as conduct research using the WWEIA/NHANES.

There are currently two versions of the PPPT and one version of the PPIT

· PPPT 2013: for WWEIA/NHANES 2011-12

· uses the PPC 2013 and 2013 InfoScan to estimate prices

· PPPT 2015: for WWEIA/NHANES 2013-14

· Uses the PPC 2015 and 2015 InfoScan to estimate food prices

· Uses the PPC 2015 and 2015 Consumer Network to estimate food prices (does not include Consumer Network random weight products)

· PPIT 2015

· Uses the 2015 PPPT to estimate the purchase weight of ingredients for a given quantity of a WWEIA food Available Resources from USDA

1. Technical Bulletin on PPC. Available at:

https://www.ers.usda.gov/publications/pub-details/?pubid=92570

2. Draft Technical Bulletin on PPPT – This is a partial draft of a future ERS publication. It includes methodology used to create the PPPT, but results and validation is withheld.

3. Documentation and statistical properties reports on the IRI data available to USDA. These can be found at:

https://www.ers.usda.gov/topics/food-markets-prices/food-prices-expenditures-costs/using-scanner-data/

4. The USDA Food and Nutrient Database for Dietary Studies (FNDDS) 15-16, is a database of foods, their nutrient values, and weights for typical food portions. The primary purpose is to estimate the nutrient profile of foods in the WWEIA/NHANES study. In the past, the FNDDS drew nutrient data from the National Nutrient Database for Standard Reference (SR) which has been replaced by Food Data Central. To work with this change, the FNDDS is making changes so that it can continue to provide a broad nutrient profile for WWEIA/NHANES. The 2015-16 documentation discusses what the 8-digit codes represent. Although not yet released, offerors should assume that changes between the FNDDS 15-16 and 17-18 will be like those between FNDDS 13-14 and 15-16. Please reference the FNDDS ingredient file for the categories that were revised in 15-16. The FNDDS is available at:

https://www.ars.usda.gov/northeast-area/beltsville-md-bhnrc/beltsville-human-nutrition-research-center/food-surveys-research-group/docs/fndds-download-databases/

5. The USDA Food Patterns Equivalents Database (FPED) and the USDA Food Patterns Equivalents Ingredient Database (FPID) converts foods and beverages in the Food and Nutrient Database for Dietary Studies to 37 USDA Food Pattern (FP) components. The FPID includes SR codes in the FNDDS-SR Links file. FPED was formerly known as the MyPyramid Equivalents Database (MPED).

https://www.ars.usda.gov/northeast-area/beltsville-md-bhnrc/beltsville-human-nutrition-research-center/food-surveys-research-group/docs/fped-overview/

6. Sample IRI product dictionaries will be available to offerors without current access to the IRI data through the NORC Data Enclave after signing the appropriate agreement letter in Appendix A or B. IRI provides detailed characteristics and nutrition information for all UPCs for which IRI records retail sales or household purchases. The product dictionaries contain detailed characteristics for 100% of UPCs and nutrition information for 20-65% of UPCs depending on the nutrient. Comment by Carlson, Andrea - REE-ERS, Washington, DC: I am still working on this file

7. Data files and computer programs associated with the 2013 and 2015 PPC and the 2013 and 2015 PPPT, and the 2015 PPIT. Items will be available to offerors via a secure file transfer after signing the appropriate agreement letter in Appendix A or B. These files also contain proprietary data, and include the decision log, user’s guides, delivery memos, the current set of matches, conversion factors, sample code, and price tool (SAS code and associated support files).

8. Documentation associated with the 2015 ERS Food Purchase Groups (FPG) linked to IRI data. The links between UPCs in the product dictionaries for 2016-2018 and the FPG’s will be available by December 10, 2020. Documentation describes what the groups are. These may be useful in either establishing matching categories, or establishing forms (e.g. fresh, frozen, canned, dry mix) of products for the PPIT.

9. Computer programs and matching tables used to generate the match between IRI 2013 and FNDDS 11-12 and IRI 2015 and FNDDS 13-14. Offerors will be given access through a secure file transfer after signing the agreement in the Appendix. These files also contain proprietary data. Note: The code does not contain the code used to build the matching tables, using semantic and manual matching. Creating the tables is most of the work and requires revision each time.

10. Food Intakes Converted to Retail Commodities Database (FICRCD), released by USDA/ARS converts foods in the FNDDS to 65 retail level commodities. This may provide support in establishing the crosswalk. However, the latest version is for WWEIA/NHANES 2007-8 so use may be limited.

http://www.ars.usda.gov/Services/docs.htm?docid=23887#

Task List

Technical Requirements Note: All documentation, data and computer programs used to conduct these tasks become property of USDA and at USDA’s discretion may be shared with researchers and the general public. USDA will not share files containing proprietary IRI data beyond those authorized to use it.

1. Pilot (Base)

a. Review past methods considering new data to determine which methods should be improved for higher quality matches and WWEIA/NHANES food prices.

b. Conduct pilot study to test linking methods.

c. Prepare a detailed plan to

i. complete optional tasks 2-3 by the following dates:

1. PPC, PPPT and PPIT using IRI data from 2015 and 2016 and FNDDS and WWEIA 2015-16 by June 1, 2021

2. PPC, PPPT and PPIT using IRI data from 2017 and 2018 and FNDDS and WWEIA 2017-18 by December 1, 2021

ii. Plan should also detail dates when USDA will need to review draft deliverables. The final deadline (December 1, 2021) is fixed and cannot be delayed.

iii. Standards for the PPC, PPPT and PPIT are listed below. Standards are the same for each version.

2. Purchase to Plate Crosswalk (PPC) –task 2 and 3

a. Crosswalk consists of at a minimum:

i. IRI product dictionary entry (UPC, UPC2, or GTIN) matched to a USDA food code contained in the FNDDSingred table (including 8-digit FNDDS code, and SR (NDB-No))

ii. Conversion factor to convert the purchase form (e.g., with refuse and possibly raw) to the form that the item is matched (without refuse and possibly cooked)

iii. Other optional information to assist researchers and policy makers understand the data

iv. Computer programs to assist researchers to integrate the FNDDS nutrient data and the FPED/FPID data into IRI data. Programs should be written in at least one of the following languages: STATA, R, python, SQL, and/or SAS. Previous versions used SAS, but IRI data users have access to all these languages.

v. User’s Guide

vi. Documentation of methods and assumptions.

vii. Note: UPC, UPC2, and GTIN are all character fields and the leading zeros must be maintained

b. Crosswalks cover the following IRI Product Dictionaries:

i. PD_POS_all – isolate UPCs to prescribed years based on column “year”

ii. PD_perish_all – isolate UPCs to prescribed years based on column “year”

iii. PD_PL_all – isolate UPCs to prescribed years based on column “year”

iv. PD_RWpanel (Note: UPCs in this product dictionary remain constant, so previous matches only need to be verified against new FNDDS.)

c. Crosswalk consists of all UPCs representing 95 percent of sales dollars for each IRI category from each year of InfoScan and 95% of purchases from each year of the Consumer Network

i. Create item codes by combining UPCs that differ only by package size or package type (e.g. can, carton, jar)

ii. 95% of sales calculated by item code

iii. 95% of sales calculated within the IRI category.

d. Links should be consistent with previous decision log.

e. Matching should have an error rate of less than 5 percent by matching category. An error consists of:

i. A better match exists

ii. An item is coded “no match” when a match exists

iii. An item is matched, but “no match” is more appropriate

f. The following deliverables delivered by March 31, 2022:

i. Group unmatched IRI items within the 95 percent cut-off into groups that likely have similar Food Pattern Equivalent Data for possible creation of new codes

ii. Decision log

iii. All computer programs and full documentation. Test code to ensure all required data files are available. There should be enough information that a user can recreate the PPC starting with IRI data available on NORC and public download of FNDDS. Instructions should include how tables are created and how manual matching is done. Programs should include final correction table to correct and update matches and to correct errors discovered during the creation of the PPPT and PPIT.

3. Purchase to Plate Price Tool (PPPT) – optional task 2 and 3

a. Price Tool consists of at a minimum:

i. Set of recipes that use the items in the PPC to estimate prices for foods reported consumed in WWEIA/NHANES survey.

ii. Computer programs written in at least one of the following languages: STATA, R, SQL, python, and/or SAS to allow users to use recipes to estimate prices from the IRI data.

iii. User’s Guide

iv. Documentation of methods and assumptions

v. Other optional information to support researchers

b. Two versions of the price tool for each optional task 2 and 3

i. Use all or a user-defined subset of IRI InfoScan calculate prices for WWEIA/NHANES

ii. Use all or a user-defined subset of IRI Consumer Network to calculate prices for WWEIA/NHANES.

c. Recipes:

i. Suggested starting point is the FNDDS ingredient file, but since designed to estimate the nutrient profile many “recipes” require modification to estimate prices

ii. FNDDS ingredient file may include a variety of preparation methods for a single FNDDS code to form a “statistical recipe”. Where possible this should be maintained in the price estimation

iii. Assume that all WWEIA foods are prepared at home

iv. Most preparation times should be about 30 minutes. Preparation times based on professional judgement since times are not available in the FNDDS.

d. The tool should estimate prices for foods reported ingested at least 10 times by WWEIA/NHANES participants in the respective year.

e. Price tool is not expected to use IRI products sold by “counts” or where no quantity purchased is recorded. However, for the InfoScan version, the price tool is expected to use random weight items where the amount sold is recorded in pounds.

f. National average prices should be estimated without anomalies, including but not limited to:

i. products that are similar should have a similar price. For example, school lunch pizza should have similar price to store-bought pizzas that are similar.

ii. The price for a WWEIA code in one year should be like prices in another year for the same code, unless there is a documented change in the WWEIA code. Price changes that are more than what would be expected under inflation should be documented.

4. Purchase to Plate Ingredient Tool - optional task 2 and 3

a. Purpose of the Ingredient Tool is to define the types and quantities of foods that are purchased to make foods in WWEIA/NHANES. It is used in estimating the USDA food plans to translate the model solution to the market basket groups in the published report. The market basket quantities are also used to adjust the cost of the USDA Cost of Food at Home each month following a set of Consumer Price Indices (CPI)’s for food.

b. Ingredient Tool consists of at a minimum:

i. Purchase weight of each ingredient for a given quantity of an FNDDS code covered by the PPPT

ii. User’s guide

iii. Documentation including methods and assumptions

iv. Other optional information that may be useful to the users

c. Ingredient is defined as:

i. a product available in retail food stores, but not at the barcode or item group level. For example, a can of tomatoes, bag of spinach leaves, frozen lasagna

d. Ingredients should be separated by forms such as fresh, frozen, canned, ready-to-serve or ready-to-heat

e. ERS Food Purchase Groups may be useful in sorting UPCs used to estimate prices for the recipes into different forms, but this is not a requirement

Key Deliverables

Table Notes: SOW – Statement of Work TR – Technical Requirements Government Furnished Equipment The government will not furnish equipment for this work. However, the budget should include data access to the proprietary IRI data. For more information please see “Data Access: at:

https://www.ers.usda.gov/topics/food-markets-prices/food-prices-expenditures-costs/using-scanner-data/ Security and Data Rights USDA purchases the proprietary scanner data to address policy and programmatic issues of interest to USDA. Offerors will be granted access to the data after signing a Third-Party Agreement with IRI (see “Data Access” at the above link). The agreement cannot be modified, and offerors must be willing to sign it within 1 week of the contract award. Offerors may not use the data for any other purpose than to conduct the tasks outlined in this SOW.

All documentation, data and computer programs used to conduct these tasks become property of USDA and at USDA’s discretion may be shared with researchers and the general public. USDA will not share files containing proprietary IRI data beyond those authorized to use it.

Place of Performance The offeror is expected to perform duties outside of USDA office space, other than in-person meetings.

Period of Performance The period of performance is from date of award through March 31, 2022.