"■""' 007<M-o| ftW.SL'.lOStfll
Health Systems Research, Inc.
2021 L Street NW Suite 400
Washington DC 20036
(202) 828.5100
Fax: (202) 728.9469
The Feasibility of Analyzing WIC Food Purchasing Patterns
Using Combined Issuance and Transaction Data:
Feasibility Volume
Final Report
Prepared for:
U.S. Department of Agriculture
Food and Consumer Service
Alexandria, VA
Prepared by:
Loren Bell, Lawrence Bartlett, Renee Schwalberg,
Katherine Clayton, Leanne Bel' igie, Hilary Bellamy and Renee Simon
Health Systems Research, Inc.
Washington, DC
April 1997
K-i*
FINAL REPORT
SUBMITTED BY:
Health Systems Research, Inc.
2021 L Street NW Suite 400
Washington DC 20036
(202) 828.5100
Fax: (202) 728.9469
Health Systems Research, Inc.
2021 L Street, N.W., Suite 400
Washington, DC 20036
CONTRACT TITLE: WIC Food Purchasing Study
Contract No. FCS: 53-3198-4-024
FCS PROJECT OFFICER: U.S. Department of Agriculture
Food and Consumer Service
Office of Analysis and Evaluation, Room 208
Attn: Marie Mitchell, Contracting Officer's
Representative
3101 Park Center Drive, 2nd Floor
Alexandria, VA 22302
CONTRACT OFFICER: U.S. Department of Agriculture
Food and Consumer Service
Contract Management Branch, Room 914
Attn: Karen Scott, Contract Specialist
3101 Park Center Drive
Alexandria, VA 22302
HSR PROJECT DIRECTOR: Loren Bell
Senior Associate
Health Systems Research, Inc.
2021L Street, N.W., Suite 400
Washington, DC 20036
a'I
Table of Contents
Volume I. Feasibility Study
Chapter I. Introduction and Overview 1
A. Background 1
B. Purpose of the Study 6
C. Potential Uses 7
D. Overview of the Methodology 11
E. Feasibility Questions 12
Chapter II. Feasibility of Using Scanner Technology to Examine WIC Participant Food
Purchasing Patterns 15
A. Introduction and Overview 15
B. Major Findings Related to the Feasibility of Matching WIC State
Agency Data With Store Transaction Data 16
Appendix A. Description of Merge Data Set
Health Systems Research, Inc. Table of Contents Page I
CHAPTER I
Introduction and Overview
This chapter provides a brief overview of the Special Supplemental Nutrition Program for
Women, Infants and Children (WIC), highlighting the need for more data about the food
purchasing patterns of WIC participants that led to the implementation of the feasibility study
described in this report. Potential uses for data on the food purchasing patterns of WIC
participants are described. A brief overview of the methodology of linking supermarket point-of-
sale data with information from the WIC agency database is presented. Feasibility questions
encountered in implementing this methodology are outlined.
A. Background
1. The WIC Program
The WIC Program was established in 1972 through an amendment to the Child Nutrition Act.
Its purpose is to provide low-income pregnant, breastfeeding, and postpartum women, infants,
and children up to age five with supplemental foods, nutrition education, and health care
referrals to counteract the adverse affects ofpoverty on their nutrition and health status.
There are four methods used by States to deliver WIC Program benefits to participants. The
majority ofWIC State and local agencies have relied primarily on the distribution of a "food
instrument (FI)," which can be in the form of a check or a voucher. Participants are given the
FI at their local WIC clinic for the purpose ofobtaining WIC-approved foods. In a State using
checks as food instruments, the check is accepted by a store in exchange for WIC-authorized
foods, and then deposited in the store's bank. The check clears through the Federal Reserve
system, and like other checks, is paid from the bank account managed by the State WIC
Health Systems Research, Inc. Chapter I Page t
agency. If a State uses a voucher, the store must send the voucher to the State or its fiscal
agent for payment. In either case, the FI is exchanged for food at a retail grocery store.
However, some States operate different types of systems. For example, six counties in
Wyoming are testing replacement of their check system with an electronic benefits transfer
(EBT) system that requires participants to use a "smart card," a card containing a computer
chip, to purchase WIC foods. Mississippi distributes food directly to participants through
State-operated warehouses. Vermont uses a home delivery service to provide WIC participants
with foods, while Ohio uses both a retail check system and a home delivery system (in 32 of 88
counties). Alaska uses a direct delivery system in areas of the State where retail grocers are not
available.
States that operate systems using a food instrument are known as "retail delivery States."
These States issue WIC FIs to participants who qualify for program benefits. WIC participants
use their FIs to purchase specific foods at authorized grocery stores. Relying on guidelines and
regulations established by the United States Department of Agriculture's (USDA) Food and
Consumer Service (FCS), each State develops a list of approved WIC foods from which
participants may choose. Each of the general food categories and the amount of food
authorized to be purchased is clearly displayed on the WIC FI. The exact type and quantity of
food a participant may purchase is based on one of five WIC participant eligibility categories:
pregnant women, postpartum women, breastfeeding women, infants, and children.
Table 1-1 displays the types of foods that participants in each of the five eligibility categories
are authorized to purchase. Because the foods are chosen to meet the specific nutritional needs
of persons in the different categories, the FIs are regarded by Federal regulation and program
staff as "food prescriptions."
Health Systems Research, Inc. Chapter I Page 2
Table ML
Food Type
Pregnant
Women
Postpartum
Women
Breastfeeding
Women
Infants Children
Milk X X X X
Cheese X X X X
Eggs X X X X
Cereal X X X X
Juice X X X X
Beans or Peanut Butter X X X X
Tuna X
Carrots X
Infant Formula X
Infant Juice X
Infant Cereal X
Special Formula* X X X X X
* Under special conditions only
Participants must redeem their WIC FIs at retail stores that are authorized by the program.
Some States operate a vendor-specific retail system, in which participants are required to select
a single authorized store and to purchase all of their WIC foods at that store. Other States
operate open systems, in which participants may use their FIs at any WIC-approved store.
Among geographic WIC State agencies with retail delivery systems, there are 11 that are
vendor-specific and 35 which operate open systems. Figure 1-1 displays the food delivery
systems of all SO geographic State agencies.
2. Procedures for Purchasing WIC Foods
Each month the WIC participant receives a series of FIs that may combine a variety of food
categories. For example, a participant may be issued certain FIs that authorize the purchase of
milk, eggs, cereal, and juice, and other FIs that allow the purchase of peanut butter and cheese.
Within each food category, participants are given a choice of food products. For example,
participants can select different types of cheese, such as Colby, Cheddar, or Swiss; or various
brands of WIC-approved cereals.
Health Systems Research, Inc. Chapter I Page 3
FIGURE 1-1.
STATE FOOD DELIVERY SYSTEMS
'%>o
] Open System States
| Vendor - Specific States
| Home Delivery State
■ Direct Delivery State
■ EBT State
| Partial Home Delivery State
] Partial Direct Delivery State
Health Systems Research, Inc. Chapter I Page 4
Each WIC FI has a number, which is recorded in the participant's issuance file at the WIC
agency where the FI is issued. A participant takes the FI to an authorized store, makes
selections, and pays for the foods by presenting the WIC FI to the cashier. The WIC agency is
required to use the FI number and the total dollar amount it was redeemed for to create a
redemption database.
3. The Need for Food Purchasing Data
In any State that uses a FI for delivery of WIC benefits, officials administering the program
have no means of conducting an examination of the purchasing patterns of WIC participants.
The only purchasing information available to program officials is the total expenditure amount
for each WIC transaction, which is reflected by the dollar amount entered on each FI by the
store cashier. Although WIC nutritionists encourage participants to purchase all their
prescribed foods, WIC staff have no way of tracking whether these foods are actually
purchased. In addition, even though officials know the range of foods available for purchase,
they do not know which brands and forms of food participants are purchasing. They have no
information about how many trips participants make to the store, what foods they do or do not
purchase, or the costs of individual food categories.
Information about WIC participants' food purchasing patterns could enhance the ability of FCS
and WIC State and local agencies to administer the program. Data indicating how often
participants shop, which foods they purchase or do not purchase, and the cost of each WIC
food category could better prepare WIC officials to make decisions about policies related to
food authorization, the content of nutrition education materials, and selection of foods to
include in their WIC food package. States could also project monthly food costs more
accurately and examine reasonable alternatives when considering which food products to
include in the program's authorized food list.
Health Systems Research, Inc. Chapter I Page 5
B. Purpose of the Study
With the introduction in recent years of bar-code scanning technology and Universal Product
Codes (UPCs) and Local Product Codes (LPCs), supermarket checkout systems can generate a
wealth of data about individual food purchases. UPCs and LPCs are unique codes assigned to
individual food products. Stores use these codes to create an individual product description in
their data systems that identifies the food product by such characteristics as product brand
name, type of food, container size, and price. When the product is scanned at the checkout
stand, the computer system recognizes the code and records the purchase of the food. These
data can be aggregated and analyzed in various ways to study food purchasing patterns of
individuals and groups, including WIC participants.
The purpose of this study, which was implemented in October 1994, was to test the feasibility
of combining data from a WIC agency's participant records, which include food prescription
and client demographic data, with information from local grocery stores' UPC and LPC data
files created by point-of-purchase scanning. By linking these databases, information about the
shopping patterns of WIC participants can be obtained.
In examining food purchasing patterns of WIC participants, the primary goal of FCS is to
ensure that participants obtain maximal nutritional benefit from the food package issued to
them. Data about food purchases will allow FCS to evaluate the food packages and whether
adjusting the types and amounts of food would improve the nutrition of participants. Thus,
FCS has two main research objectives: to determine the feasibility of using scanning
technology to describe food purchasing patterns of participants and to examine differences
among participants in such patterns. The differences in purchasing patterns examined in this
study were between rural and urban participants, between seasons, between WIC eligibility
categories, and among four ethnic groups: Blacks, Hispanics, Asians or Pacific Islanders, and
Whites. To explore feasibility under a variety of conditions and with different populations,
two States were selected to participate in the study—Wisconsin and Pennsylvania. To examine
Health Systems Research, Inc. Chapter I Page 6
seasonal purchasing patterns, data were collected over two time periods, one in the
spring/summer and one in the fall/winter of 1995.
It should be emphasized that the purpose of this project was to test the feasibility of linking the
two databases, as well as the feasibility of analyzing the resulting combined data set. This
study was not designed to lescrib? the purchasing patterns of a scientifically selected
representative sample of WIC participants or the stores in which they shop. The data reported
here are applicable only to the population included in the study. In addition, the chain stores
selected may not necessarily represent the types of stores where the majority of WIC
participants shop. Moreover, no attempt was made to determine reasons for shopping patterns
of WIC participants. Some of the data described in this report are highly suggestive of
particular interpretations; however, because no valid comparison groups were available, no
attempt was made to statistically test the significance of the findings or to examine the validity
of any conclusions that might be made from these data.
C. Potential Uses
If this process appears feasible and/or cost effective, the data could be used for several
purposes. This section will describe the potential uses of the data by Federal and State WIC
Program officials. As indicated earlier, data collected and analyzed using the technology
described in this feasibility study can be useful to Federal and State officials in evaluating the
food purchasing patterns of WIC participants. Four key areas in which these data may be
useful are described below.
1. Potential Uses: Food Selection and Benefit Utilization
a. Developing and Assessing the Impact ofFood Selection Restrictions
Under Federal WIC regulations, WIC agencies or State directors are authorized to
develop criteria for selecting specific brands and forms of food to be included in their
food package. In addition, State WIC directors may need to make decisions to restrict
participants' food choices to lower priced foods when necessary to remain within the
Health Systems Research, Inc. Chapter I Page 7
State's budget limitations. Within each food category, high-cost items can be
eliminated as long as other food choices within that category are available. For
example, during the months data were being collected for this study, the issuance of
peanut butter, a relatively high-cost item, was restricted in Wisconsin to odd-numbered
months. Participants were authorized to purchase dried beans, a nutritionally similar
food, during even-numbered months. States can also set stricter nutritional
requirements for authorized foods, for example, by further limiting the sugar content of
cereals offered.
In current WIC State retail delivery systems, the exact quantity of individual food items
purchased cannot be determined. Therefore, when WIC State directors make the
decision to restrict or remove a food from the program, they do not know what the
overall impact of that decision will be—the number of participants affected by the
change, or the total amount of savings that may be created or the impact on demand at
WIC vendors.
Data about the types of food and the forms of food selected by participants would allow
State officials to make more informed decisions about food restrictions and to better
assess the impact of their decisions on participants and stores. For example, data
concerning cereals and juices popular with certain ethnic groups would allow States to
evaluate the impact of eliminating these choices from the WIC food categories and
substituting other lower-cost items when necessary.
b. Developing Food Package Tailoring Policies
WIC local agency nutritionists have the option of tailoring participants' food packages
to better meet their nutritional needs. Under the current system, the nutritionist must
rely on information provided by individual participants to determine if they are
purchasing prescribed foods. By having information about the foods not purchased by
participants, the nutritionist could develop policies for tailoring food packages to
maximize the benefit to the participants. For example, if data reveal that certain ethnic
Health Systems Research, Inc. Chapter I Page 8
groups do not purchase all of the prescribed milk, but purchase all of the prescribed
cheese, the nutritionist will have additional information to develop guidelines and
policies for reducing the amount of milk issued and increasing the cheese issuance.
c. Developing Nutrition Education Activities and Materials
Most WIC nutrition education is focused on the importance of eating the appropriate
foods. If State and local nutritionists know that participants do not purchase foods
authorized, they can direct their efforts to encourage participants to purchase and use
those foods. For example, adolescents may not purchase dried beans because they do
not know how to cook them. Adults may not understand the importance of limiting fat
intake by drinking low-fat and skim milk rather than whole milk. With data about food
purchases. WIC nutritionists can better target nutrition education.
d. Improving Product Stocking Requirements for Vendors
Food purchasing data would also help WIC State and local agencies ensure that newly
authorized WIC vendors stock enough of the more popular food items to provide an
adequate supply for WIC participants. When a store applies to become an authorized
WIC vendor. States could better direct stores in stocking WIC foods. In addition,
States could use this information to develop indicators that a store is not stocking
sufficient foods. If data show that certain foods are not purchased in a particular store,
this may indicate that the store is carrying insufficient stock of the food product.
2. Potential Uses: Number of Shopping Trips/Checks Used
States can also use data regarding the number of shopping trips made by WIC participants. In
particular, States may use such data to assist them in developing policies related to the number
of FIs the food package is distributed over. States have an interest in reducing the number of
FIs issued to participants. Each FI issued by a program and processed by a bank or fiscal
intermediary incurs a cost to the State in a processing fee. The total amount of FI charges is
significant considering that millions of FIs are issued. Some participants may make frequent
Health Systems Research, Inc Chapter I Page 9
shopping trips due to transportation difficulties, lack of refrigeration, or the large number of
infant formula cans they must purchase. In addition. States may wish to explore the underlying
causes of a low number of trips to a specific store. While a low number of trips is not
necessarily a problem. States may wish to examine the reasons for participants making a small
number of shopping trips, including the convenience of the store location.
Data about the number of trips participants make would allow States to assess whether they are
issuing an appropriate number of FIs. If participants are making only a few trips for reasons
unrelated to grocer convenience. States may decide to reduce the number of FIs participants
receive by combining more foods on one FI. This would not only reduce the cost of FI
processing fees, but may also expedite the processing of transactions at the checkout counter.
3. Potential Uses: WIC Food Expenditures
An additional use of food purchasing data would be to address the cost of the WIC food
package and the amount spent or not spent. While there is limited use for expenditure data
from a single store or a group of stores, expansion of the methodology may have specific
benefits to WIC State agency fiscal staff.
a. Enhancing Food Package Expenditure Forecasting
States are required by FCS to project monthly WIC food expenditures and report actual
expenditures against projected expenditures. States usually project expenditures based
upon monthly redemption data that provide officials with an average cost per food
package. One problem with this method is that it relies on data captured several
months after the FIs have been redeemed. It is difficult to use these data to project
future expenditures if there are changes in the allowable foods during the fiscal year.
Having accurate and timely food product redemption rates would provide State officials
with an additional tool to estimate food costs based upon actual purchasing patterns for
individual food categories, thus allowing adjustments for changes in policy over the
course of the fiscal year.
Health Systems Research, Inc. Chapter I Page 10
b. Calculating Food Expenditures Associated With Caseload Expansion
Having expenditure data for each food item would assist State officials to project the
fiscal impact of caseload expansion. Knowing the differences in food package costs for
urban or rural stores and for participants with various demographics would allow States
to project the increased costs of adding participants in urban and rural areas or
increasing services in areas serving a particular group.
D. Overview of the Methodology
The study examined the feasibility of linking data from two databases: food purchasing data
from a supermarket point-of-sale transaction and data from the W1C State agency database. As
noted above, under the current system, WIC agencies can determine only that a FI was used or
not used, and the total dollar amount for which the FI was cashed. Detailed information about
the type and form of foods purchased are unavailable. With the introduction of supermarket
scanning and point-of-sale databases, it became possible for the WIC Program to tap into the
wealth of data about food purchases contained in supermarket databases.
The means used to link the supermarket data and the WIC participant data was the WIC FI
number. Each WIC FI has a unique number, and the WIC agency database contains a record of
thv numbered FIs issued to each participant. At the supermarket point of sale, after scanning
the WIC-authorized foods at the checkout counter, the cashier entered the WIC FI number into
the transaction record. The WIC FI number became the single common data element of both
databases and a key step of the methodology described in this report is matching the
supermarket transaction record to the participant data in the WIC database using this number.
References below and in Chapter II to the "match rate" indicate the extent to which each WIC
point-of-sale transaction was able to be matched by its check number to participant food
issuance and demographic data in the WIC agency database.
Processing the food purchasing data supplied by the supermarket and matching it to WIC
participant data required the building of a combined database. Periodically, the supermarket
Health Systems Research, Inc. Chapter I Page u
supplied the data processing agency with a file containing raw UPC/LPC data for all the WIC
purchases made at the study stores during the data collection period. The first task of the data
processor was to convert the raw data into usable information. For this task, a conversion table
was needed. A conversion table is a list of all the foods authorized for purchase by WIC
participants and their respective UPC/LPC codes, including different types of a food (e.g.,
Colby, Swiss, Cheddar, etc.) and the different forms of a food (sliced cheese, bulk, etc.), the
container size, and the price. Armed with these data and with the WIC food instrument (FI)
numbers for each of the supermarket transactions, the data processor was able to match an
individual shopping transaction with WIC participant data.
Appendix A presents a description of the merged data set, containing detailed information
about the foods purchased in the transaction and WIC participant issuance and demographic
data. These data can be analyzed in numerous ways, some of which have been described above
in the section on potential uses for the data.
The next section lists specific feasibility questions that were answered by implementing the
methodology described above.
E. Feasibility Questions
Because this was a feasibility study, the findings of most interest and value are related to the
lessons learned in selecting States and stores to participate in the study, in working with WIC
and chain store representatives in Pennsylvania and Wisconsin to link their databases, in
ensuring that the merged data set that resulted was flexible and user-friendly, and in learning to
examine the data in meaningful ways.
The feasibility findings are addressed in detail in Chapter II. These findings are based upon a
series of feasibility study questions detailed below.
Health Systems Research, Inc. Chapter I Page 12
1. Is technology available that will allow data from WIC State agency files to be
matched with point-of-sale transaction data at WIC-authorized stores?
1.1 What elements are necessary in a WIC State agency data system to make the
technology feasible?
1.2 What are the required characteristics of a store's point-of-sale data system to
allow transaction data to be matched with WIC State agency data?
1.3 What changes would stores need to make in their data systems to participate in a
project to match State WIC data with transaction data? Are theses changes
feasible?
1.4 What is the cost to the stores of making changes in their data systems to
accommodate the data collection requirements?
1.5 By what methods is it feasible to link WIC State agency data with store
transaction data?
2. Is using UPC/LPC data sufficient to identify the foods purchased by WIC
participants?
2.1 How can WIC-authorized foods be identified in the store's UPC/LPC database?
2.2 Once identified, are data in the UPC/LPC database complete enough to
determine the type and form of foods purchased?
2.3 What is the best method of converting UPC/LPC codes into useful transaction
data?
2.4 Can WIC transactions be separated from non-WIC purchases in the transaction
record?
3. How will the completeness of the data be affected by cashier point-of-sale data
entry?
3.1 How will cashier errors and omissions affect the completeness of transaction
data?
3.2 How will cashier errors and omissions affect the match rate between State data
and transaction data?
3.3 Does training of cashiers improve the data entry error rates?
3.4 Does cashier familiarity and experience over time improve the data entry error
rates?
Health Systems Research, Inc. Chapter I Page 13
4. Are algorithms available to compensate for data entry errors and improve the
match rate?
4.1 What sources of data are available to use in the development of the algorithms?
4.2 How effective are the algorithms in improving the match rate?
5. Are there methods available to improve point-of-sale data entry?
6. How do State food delivery systems affect the completeness of the data?
6.1 How many States use a vendor-specific food delivery system? An open system?
6.2 Are there differences between the completeness of data in vendor-specific States
and open system States?
6.3 What are the advantages and disadvantages of each system in collecting
complete matched data?
Health Systems Research, Inc. Chapter I Page 14
CHAPTER II
Feasibility of Using Scanner Technology to Examine WIC
Participant Food Purchasing Patterns
A. Introduction and Overview
This study examined the feasibility of merging data from WIC State agency files with grocery
store transaction data tapes built from scanned UPC/LPC codes, which contain a detailed
record of information about WIC participant purchases, to analyze the food purchasing patterns
of WIC participants. The merged database provides information about the types, forms, cost,
and amounts of food purchased by each WIC participant with each WIC FI, and, conversely,
what authorized foods participants did not purchase.
The purpose of a feasibility study is to separate what is realistically possible from what is
theoretically possible—and to account for the difference. The main focus of this chapter is on
the findings of the feasibility study. The findings are based on a study carried out over
approximately 18 months (October 1994-April 1996). At the beginning ofthe study, several
States and chain stores were considered as possible candidates to participate in the study. The
chain stores selected to participate, and selected local stores in the chain, met the requirements
for having the appropriate scanning technology, serving rural and urban WIC shoppers of
diverse ethnic backgrounds, and having a high and a low volume of WIC sales. WIC State
agencies and the selected store chains were briefed on the study's purpose and methods, and
staff were prepared for data collection.
Health Systems Research, Inc. Chapter II Page 15
B. Major Findings Related to the Feasibility of Matching WIC State Agency
Data With Store Transaction Data
This section presents the major findings related to the feasibility of collecting and matching
data from WIC State agency participant files with store transaction data. However, there are
two important considerations to take into account when examining the findings.
First, it is important to note that the success of the entire study is based on the ability and
willingness of WIC-authorized vendors to cooperate with the objectives of the study and
implement new procedures within their stores. Not all WIC-authorized stores will have the
necessary technology to participate in a study of this nature. As will be discussed in these
findings, the primary technology that must be available in a store is the capacity to scan
UPC/LPC codes to record purchase transactions into a database. In attempting to determine
how many stores serving WIC participants have this technology, researchers contacted various
trade associations, scanner companies, and the FCS Supplemental Food Program Division to
ascertain if data were available on the number of stores with scanning systems. However, no
data currently exist that provide information about the number of WIC authorized stores with
scan technology. Researchers were informed by FCS that such information will be collected
from WIC State agencies in 1998. Once these data are collected, WIC State agencies will be
able to determine the extent to which a study similar to this one can be conducted.
Second, the feasibility study was designed to assess the capacity of geographic State agencies
to collect and analyze food purchasing data. No attempt was made to determine the feasibility
of Indian Tribal Organizations (ITO) conducting a study of this nature. While some of the
findings are relevant to ITOs, differences in data system capability and types of stores used by
clients affect the general izability of the findings. In addition, the overall cost of conducting
this type of research may render it infeasible for these agencies.
Health Systems Research, Inc. Chapter II Page 16
Finding #1. Data from WIC State agency participant flies can be matched with store
transaction data to provide information about WIC participant food
purchasing patterns.
By developing a method by which food purchasing information collected at the point-of-sale
can be linked to WIC State agency files, a database can be developed and analyzed to provide a
description of the food purchasing patterns of WIC participants. Information contained in the
matched database provides information about WIC participant food purchasing patterns,
including the number of shopping trips, the type and form of the foods purchased, the foods not
purchased, and food cost information. Information can be analyzed by both participant and
store demographic characteristics. A complete description of the structure of the combined
database is included as Appendix A.
Once data are received from a WIC State agency and the selected stores, building a complete
database requires both computer programing and data processing time. To build the complete
database for Wisconsin, 455 programing and data processing hours were required for the first
data collection period and 154 hours were required for the second. To develop the
Pennsylvania database, 372 programing and data processing hours were required for the first
database and 184 hours for the second.
There are key elements that must be present in both the WIC State agency and the grocery store
to make this methodology successful. The WIC State agency must have an automated data
system that captures both client demographic data and FI issuance information for each client.
The agency must have redemption data that can link a paid WIC FI with the participant using it
and the store where it was cashed.
Stores selected for conducting this type of study must have several elements in place for the
methodology to be successful.
■ Stores must have a product scanning system that captures transaction data
at the point ofsale and links those sales to specific food products through
the use ofUniversal Product Codes (UPCs) or Local Product Codes (LPCs).
In order to create the automated database, a store must have a method by which
Health Systems Research, Inc. Chapter II Page 17
food purchasing information is captured electronically. Use of a product
scanning system that identifies food products by their UPC/LPC identifiers is
critical to capturing transaction data. Without the ability to identify food
products by individual type, form, and cost, the methodology will not be
successful.
■ The store's system must be able to distinguish WIC transactions from other
routine transactions. Stores with electronic scanning systems have methods
by which they record the type of payment being used to make a purchase. This
identification of a payment is known as a "tender." The most common tenders
are cash, check, or Food Stamp. In order to identify the transaction as being
paid for by WIC FIs, the store must have a method to include a "WIC tender" in
their system. This will usually require the store to reprogram their date, system
to include a WIC tender. Chain stores participating in this study reported that
there was little additional cost to conduct the reprogramming. In both States'
chain stores, the cost of reprogramming their computer systems to meet this
requirement was less than $500 for each chain.
If a purchase cannot be identified as a WIC transaction in the store's database, it
will be impossible to separate it from any other transaction in the files. Because
WIC foods are common foods purchased by most people, and because WIC
foods supplement a participant's monthly food purchases, the lack of a method
by which to identify the transaction as a WIC transaction makes it difficult to
determine if a WIC participant purchased her foods with a WIC FI or with
another form of payment.
■ The store's data system must have a method by which the WIC FI number
or client identification number can be entered into the transaction record.
Simply identifying a transaction as a WIC tender is not sufficient to link the
transaction to a particular participant. The store must be able to enter a WIC FI
number or a participant identification number into the database in order to link
the transaction to the participant's WIC food prescription. A WIC FI number is
the preferred method for linking the transaction, as this creates a direct link
between the purchase and the FI. When a participant identification number is
used, the method by which the transaction is linked to the issuance record
requires several additional steps.
Scanning systems have the capacity for entering numbers, such as a personal FI
number or a driver's license number, into their system. These number entry
systems can be used to enter a WIC identification number, whether it is the FI
number or the participant identification number. While this creates a small
additional burden to the cashier, stores participating in the study reported little
disruption to the checkout process. The total cost of reprogramming the stores'
data | /stems to capture the WIC FI numbers was less than $1000 per chain store
in each State.
Health Systems Research, Inc. Chapter II Page 18
:-wnm
However, having all of the above systems in place does not guarantee that complete food
purchasing records will be available for all WIC participants using the store. As is seen in the
remaining findings, there are a number of variables that affect the completeness of the merged
database.
Finding # 2 Developing a transaction file from raw UPC/LPC data requires refinement
on the part ofthe stores and the State data processing agency.
Building a transaction file from raw UPC/LPC data contained on the supermarket data tapes
requires significant refinement on the part of the stores and data processors. The data
processing system used by the stores collects data on all purchases made by any person
shopping at the store. WIC participants represent only a portion of those transactions. In order
to match food purchases to a WIC participant, a file must be created that separates the WIC
purchases contained in the database from those made by other customers. To do this, the store
must sort the files by the code developed to identify the purchase as a "WIC tender." All
records that are identified as a WIC purchase are then placed in a separate file for transmission
to the State data processing agency for matching with individual food issuance data. Two
specific factors influence how the transaction database was developed.
Finding 2.1 Conversion of UPC/LPC codes requires cooperation between the WIC
State agency, the data processing agency, and the selected stores.
The data provided to the WIC State agency contain the UPC/LPC coded information about the
type, form, date of purchase, and cost of each authorized WIC food purchased by the
participant during a single transaction. In order for the information to be useful, the code must
be converted to the actual product name, type of food, size of container, date ofpurchase, and
price. However, to provide a State data processing agency with the information to convert the
codes into useful information, the store must first have information about which food products
are authorized by the WIC State agency. Without this information from the State agency, the
stores cannot identify all ofthe WIC-eligible food products by their UPC/LPC codes.
Health Systems Research, Inc. Chapter II Page 19
To develop UPC/LPC conversion tables listing W1C food items and their corresponding codes
for this study, the data processing agency had to initially sort out all of the UPC/LPC codes
contained in the transaction data and provide the cnain store officials with a list of those codes.
The stores were then able to identify each product and send the list back to the data processing
agency with the necessary information for each food product listed in the conversion table.
States and stores will need to work out how information needed to create conversion tables will
be transmitted to the stores before data collection begins. States will need to pay close
attention to the data contained in the conversion tables to ensure that the food purchasing data
collected are as comprehensive and detailed as possible. Some stores do not identify food
products in as much detail as WIC State agencies may wish. There were three specific food
groups for which the lack of specific product information limited the ability to identify a food
product in detail.
■ Cheese Purchases. Data on cheese purchases collected from the chain store in
Pennsylvania could not identify the specific type of cheese if it was purchased
as a deli item. Cheese purchases were broken down by type of cheese (Colby,
Swiss, American, etc.) and form of cheese (sliced, block, shredded, etc.). Each
type and form was identified by a separate code. However, from 30 percent to
50 percent of the cheese sold in Pennsylvania was not able to be identified by
type or form because it was identified in the store's LPC system only as "deli
cheese." Deli cheese is typically, but not exclusively, block cheese, but can be
of any type.
■ Juice Purchases. Similarly, in both Pennsylvania and Wisconsin, only single-flavored
juices (orange, grape, apple, etc.) could be identified by their separate
codes. All mixed juices were identified by a single code "juice." In this case,
however, while the specific flavor of the juice could not be identified, the form
of the juice (canned, bottled, or frozen) was able to be identified. In both States,
these mixed juices were the type most frequently purchased by WIC shoppers,
and they could not be identified further by flavor.
■ Bean/Legume Purchases. In some cases, beans that were purchased in bulk
from bins in the stores were identified as "beans/legumes," rather than by the
specific type of bean.
Health Systems Research, Inc. Chapter II Page 20
Because the study used the initial UPC/LPC codes from actual WIC purchases during the first
round of data collection as the basis for building the conversion tables, a problem was
identified in the second round of data collection. During the second round of data collection,
new UPC/LPC codes appeared in the data sets. This was a result of WIC participants
purchasing some approved foods in the second ;ound that had not been purchased in the first
round. Therefore, new UPC/LPC conversion tables were built using the same process
described above to add the new products. Had a complete listing of all possible authorized
WIC food products been developed during the first data collection period, this problem would
not have occurred.
Finding 2.2 Including non-WIC transactions in the database may hinder accurate
reporting of foods purchased.
A second issue that required significant attention by the data processing agency was the fact
that more than 50 percent of the stores' raw transaction files contained information about non-
WIC purchases as well as WIC purchases. This situation occurred because WIC foods are
normally scanned separately from other foods the participant is purchasing, but only subtotaled
into a single transaction rather than counted as two separate transactions. Having these non-
WIC data on the record made accurate identification and analysis of the WIC purchases more
difficult. Foods that were not WIC-eligible were easily identified and were sorted from the
files. However, if clients purchased additional WIC-eligible foods with cash, the record
became more difficult to analyze. For example, if the issuance record authorized the purchase
of two gallons of milk, and the client purchased two gallons of milk with the WIC FI and one
additional gallon with cash, the transaction record did not distinguish between the transactions,
and it appeared that the client over-purchased milk. For purposes of this study, food products
in excess of the maximum allowed in the issuance record were removed from the transaction
record.
The only method to determine if the milk was purchased by using a WIC FI was to compare
the total prices of the issued food on the FI with the total dollar amount for which the FI was
deposited. In the example above, if the total dollar amount of the deposited FI was the same as
Health Systems Research, Inc. Chapter II Page 21
the total of the prices for foods issued on that FI, then the client did not use the W1C FI to
purchase the additional milk. If the total amount of the deposited FI included three gallons of
milk, then the client would have exceeded the authorized limit.
While the identification of UPC/LPCs created a problem for completing the transaction record,
the problem was solved through communication between the State, the store, and the data
processing agency. Approximately 20 hours of chain store and data processing staff time were
required to build a complete UPC/LPC conversion table. However, other problems affecting
the accuracy of the data occurred during the study, some of which could not be completely
resolved. These problems are discussed in Findi. g #3.
Finding #3. Data entry errors and omissions made at the point ofsale have a negative
impact on the completeness ofthe matched data set and the subsequent
analyses.
Complete, accurate, and matchable transaction data are critical to the success of the
methodology. Perhaps the single most important finding of the feasibility study was that the
accuracy ofthe cashier in entering the WIC FI number was the most critical element in creating
a successful merging of the WIC participant data and the store transaction data. Cashier error
played a large role wh< n store transaction data could not be matched to WIC issuance data.
The extent to which these errors affected the match rate are discussed in later findings and
displayed in Table II-1. The major errors identified include:
■ Cashiers Did Not Identify the Transaction as a WIC Transaction. In
Wisconsin, the cashier was required to enter the WIC tender before they could
complete the transactions. However, in Pennsylvania the system used by the
chain store allowed a cashier to bypass the process of identifying the WIC
transaction and still complete the sale. As a result, an unknown number of the
WIC transactions did not get identified as such. Thus, the WIC transaction
appeared as a normal cash or FI transaction, and was not included in the
matched database.
■ Cashiers Did Not Enter the WIC FI Number or Entered Incomplete FI
Numbers. In both States, even when a purchase was identified as a WIC
transaction, it was possible for the cashier to bypass entering the WIC FI
number into the system and still complete the sale. In addition, some records
Health Systems Research, Inc Chapter II Page 22
had fewer than the required number of digits entered in the FI number field,
indicating that a complete FI number was not entered.
■ Cashiers Entered Incorrect WIC FI Numbers. The majority of errors in
Wisconsin occurred in the entry of the FI number into the system. This was due
in part to the similarities between the participant identification number and the
WIC FI numbers. The two numbers are the same length, and both are printed
on the face of the FI. In 909 cases (14 percent) during the first data collection
period and 1045 cases (11 percent) in the second, the cashiers entered the
client's WIC identification number in place of the FI number.
Even with the problems associated with cashier error, there are methods by which some of the
transaction record errors can be resolved. These are discussed in the next two findings.
Finding #4. Training and experience seem to improve cashier error rates, thus
improving the ability to match transaction data with WIC participant data.
Senior representatives of the chain store in Pennsylvania preferred that cashier training be
conducted by customer service managers in the individual participating stores, rather than by
outside trainers, as this approach would be less disruptive to store operations. The purpose of
training is to inform store managers and cashiers about the study and review the keying
sequence for processing WIC transactions. Training materials were developed and delivered to
the chain headquarters in advance of the training sessions, which included:
■ A one-page summary ofthe study describing the data collection
methodology and the steps cashiers would need to take when handling
WIC transactions. The summary included information on the reason for the
study, an overview of the study design, and an explanation of the contractor's
role.
■ Two substantially enlarged facsimile WIC FIs, one handwritten and one
computer-generated. The purpose of these samples was to illustrate where FI
numbers are located on WIC FIs.
■ Five hundred cards that contained an explanation ofthe study. These cards
were sent to the store headquarters for distribution to store managers. They
were to be given to WIC customers who inquired about the study.
Health Systems Research, Inc. Chapter II Page 23
However, the chain store in Pennsylvania failed to carry out training before the first data
collection period began. As a result, cashiers failed to identify transactions using the WIC
tender, or failed to enter the FI number into the transaction record. The final match rate for the
first data collection period was 5,379 matching records from 17,531 eligible issuance records,
or 36 percent, much lower than expected.
Several steps were taken to develop a reliable training plan before the second data collection
period. In a memorandum to the customer service managers at the study stores, senior
management stated that they must inform cashiers to enter WIC FI numbers. The memo also
required each customer service manager to have all cashiers sign a form stating that they had
been informed of the study and understood what they were being asked to do.
In addition, the study protocol was reviewed individually with each store manager, and any
questions were answered. Each step in the keying sequence was reviewed, and it was
explained that if the cashiers did not complete the entire keying sequence, the transaction
would not be identified as a WIC sale. As a result, the final match rate during the second
round of data collection improved to 13,908 matching records from 21,829 eligible issuances
records, or 47 percent.
The contract for this study did not call for training of cashiers in Wisconsin. The chain store
selected in Wisconsin was already using a WIC tender system. Cashiers working in the
selected chain store simply were instructed to enter the WIC FI number into the system once
the tender was complete. The initial match rate for Wisconsin—based on matching only the
WIC FI numbers—was 60 percent of all FIs issued to the study stores.
However, as in Pennsylvania, the match rate between transaction and issuance records in
Wisconsin improved between the first round of data collection and the second round. During
the intervening months, even though data were not being gathered, cashiers continued to enter
WIC FI numbers into the system. As a result, by October, the cashiers had been entering the
WIC FI numbers for five months and the match rate improved to 80 percent. Store officials
Health Systems Research, Inc. Chapter II Page 24
attributed the higher match rate to the fact that checkers became more skilled at entering the
WIC FI number. Early training of cashiers prior to the data collection period increases the
likelihood of high-quality data entry at the point of sale, and accuracy of data entry skills
improves over time.
Finding #5. Ure ofan algorithm based upon information contained in the WIC State
agency issuance files and bank records can also improve the match rate
between issuance and transaction data.
Even with training and experience, it is likely that errors in the data e-^-y process will still
occur. To create the initial merged database, data from the WIC State agency files were
matched with data from the stores1 transaction records by l.ieans of the WIC FI number.
However, largely due to the data entry problems mentioned in findings #3 and #4, the match on
the FI number produced an unacceptably low rate of successful matches. Other methods of
matching the available data records were needed, and a matching algorithm was developed to
match issuance and transaction records using data elements other than the FI number.
The matching algorithm (i.e., the series of steps described below to improve the match rate)
was initially developed and tested on unmatched items from the first round of data from
Wisconsin and then used in all subsequent rounds of data collection. The steps in the
algorithm are as follows:
1. Matching on Participant's Identification Number. In some cases, a cashier
entered the participant's WIC identification number instead ofthe FI number.
The number in the "FI number" field in the transaction record was then matched
to the participant identification in the issuance record.
2. Matching on Bank Record. If the FI number from the transaction record did
not match either the FI number or the participant identification number in the
issuance record, the two records could sometimes be matched using information
provided when the store redeemed the FI at a bank and the canceled FI was
returned to the WIC agency. The bank record that is appended to the issuance
record contains the date the FI was cashed by the store, the identification
number of the store redeeming the FI, and the amount paid to the store. In the
feasibility study, to match an issuance record to a transaction record using this
information, for each unmatched issuance record, all transaction records that
met three criteria were identified: (1) the transaction took place at the store that
Health Systems Research, Inc. Chapter li Page 25
cashed the FI. (2) the total WIC amount matched the amount paid by the bank,
and (3) the date of the transaction preceded the date the FI was cashed by the
store. If only one transaction record matched a given issuance under these
criteria, the records were considered matching.
When these criteria are used, it is common for more than one transaction record
to emerge as a potential match for a given issuance. To narrow the field of
potential matches, one more step was added to the algorithm. This step limited
the match criteria to those records in which the date of the transaction was no
more than five days before the date the FI was cashed. Five days was selected
as a limit to take into account situations when a transaction took place on a
Friday evening of a holiday weekend and the FI could not be deposited until the
following week.
3. Manipulating Incomplete FI Numbers. In the feasibility study, many
transaction records had a number in the "FI number" field that had more or
fewer digits than an actual WIC FI number. If such records were not matched in
either of the steps described above, two additional efforts were made to match
them. First, to attempt to match transaction records in which the FI number
entered was shorter than a true FI number, FI numbers on all unmatched
issuances were searched for a substring of digits that matched those in the
transaction record's FI number. If such a record was found, it was considered to
be a match if the store number and transaction amount were the same and if the
transaction date preceded the date on which the FI was cashed by the bank.
Second, transaction records in which the FI number was too long were searched
for cases in which a double zero appeared. If it did, one of the zeroes was
eliminated and the record was compared with the remaining issuance records.
(This was intended to account for cases when a cashier hit the "00" key instead
of "0.") If a match was found, it was considered a true match only if the store
number and amount matched, and the transaction date preceded the date the r'l
was cashed.
4. Hand Matches. Finally, a printout of all remaining unmatched issuance and
transaction records was examined to find records in which the FI number was
inaccurately keyed by the cashier. By examining each unmatched issuance
against a list of candidate transaction records (ones that took place at the same
store, for the same amount, with the transaction date preceding the date the FI
was cashed), records in which the FI number was entered incorrectly were
identified and matched to the correct issuance record.
Once the matched database was created, the various steps in the matching algorithm were
tested by searching for records in which none of the items purchased (as described in the detail
Health Systems Research, Inc. Chapter II Page 26
records) were included in the WIC issuance. Such instances would be considered "bad
matches" and would be deleted from the final data set. Matches were considered invalid only
in cases in which the issuance and transaction records contained no items in common. No such
records were found in any of the data sets.
The final match rate can be calculated in either of two ways: (1) as a percentage of the total
number of eligible issuances to participants shopping at the study stores for which matching
WIC purchasing transactions were found, or (2) as a percentage of the total number of
identified WIC purchasing transactions for which matching issuances were found. The first
method reflects the feasibility ofthe study design from the perspective of the WIC State
agency's issuance records, the second presents a measure of the technological feasibility of the
project to identify those transactions identified as valid WIC transactions in a particular retail
chain. If all WIC transactions are correctly identified, the two percentages will be the same;
however, since some WIC transactions are not likely to be identified because of cashier errors
and omissions, and are therefore irretrievable, the number of identifiable WIC transaction
records in the store files represents the maximum number of records that may be matched.
In Wisconsin, the match rate was calculated for each round of data collection in both ways. In
Pennsylvania, only the percentage ofcashed issuances that were matched was calculated.
Because the study stores' data included all WIC transactions occurring during the month, the
Pennsylvania transaction data sets included transactions that used issuances from months other
than the study months. In Wisconsin, issuances from prior months were eliminated because
the data processing agency had all of Wisconsin's issuance information for the months around
the study period. These issuance data were not available to the data processing agency for
Pennsylvania, and could not be eliminated from the data set. Therefore, using the total WIC
transaction number as the denominator would produce an artificia'ly low match rate.
Table II-1 displays the results ofthe matching algorithm for both rounds ofdata collection in
Wisconsin and Pennsylvania. The number of matches found in each step was higher in the
second round of data collection. The number oftransaction records available for matching was
Health Systems Research, Inc. Chapter II Page 27
higher as well, indicating that cashiers were more thorough in identifying WIC transactions in
the second round.
As Table II-1 shows, the match rate increased in the second round of data collection in both
study States. In Wisconsin, the fact that the total number of WIC transaction records in the
first round of data collection is lower than the number of WIC issuances indicates that the
stores did not consistently code WIC transactions as such. In the second round, the two
numbers were similar, indicating improvement in this regard. This was also reflected in the
increase in the percentage of issuances that were matched.
Finding #6. A system that requires the cashier to enter the Fl number into the
transaction at the point ofsale would likely produce higher match rates.
As was indicated above, a major problem with regard to data entry was the fact that a cashier
could bypass either a WIC tender or enter an incomplete FI number into a transaction and still
complete the WIC sale. Because there was no built-in method requiring the WIC transaction to
be identified and a FI number to be entered into the data system, cashiers could circumvent the
store's policy.
The feasibility study initially planned to test the use of a computer prompt to remind the
cashier to enter the WIC FI number into the store's data system. The Pennsylvania chain store
management indicated a willingness to program their computer system to require the entry of
the WIC FI number into the data system. However, because the programming would have
affected all ofthe chain's outlets, doing so to accommodate data collection in the 10 study
stores was considered too disruptive. Had the cashiers been required to identify the WIC
transaction and enter the FI number before they could complete a WIC sale, the problem of
cashiers not entering the WIC FI number would have been eliminated.
Health Systems Research, Inc. Chapter II Page 28
Table04.
Results of Algorithm Used in the Feasibility Study to Increase tiie Match Rale
Between Issuance and Transaction Records
Wisconsin Pennsylvania
Round / Round II Round I Round II
Total Eligible Issuance Records 8,668* 9,435* 15.081* 29,651*
Total Eligible Transaction Records 6,742** 9,486** 17,531 * 21,829*
Step A: FI Number Match 3,762 5,803 3,009 12,163
Step B: Participant Identification Match 909 1,045 5 3
Step C: Bank Record Match (No Date Limit) 378 493 1,870 1,108
Step D: Bank Record Match (Date Within 5
Days)
102 107 471 601
Step E: Matches of Incomplete FI Numbers 0 0 24 33
Step F: Matches made by Hand 52 128 0 0
Total Matches 5,203 7,576 5,379 13,908
Final Match Rate 60% of
issuances
77% of
transactions
80% of
issuances
80% of
transactions
36% of
issuances
47% of
issuances
* FIs issued in the study month and cashed at one of the 12 study stores
** Identified WIC transactions taking place in the study months that did not use previous or later months'
issuances
'FIs issued and cashed in the study month(s) for any of the 10 study stores
'Identified WIC transactions taking place in the study months; includes those using previous months' issuances
Source: Health Systems Research, Inc.
Stores with scanning systems are already likely to be programmed to require the entry of
personal check numbers or check cashing card numbers into their system. Expanding this
technology to include the entry of WIC FI numbers would not necessarily constitute a major
system change. However, if a chain store operates in more than one State, it would need to
customize its systems to match different FI numbering systems in each State, increasing the
computer programming costs.
Health Systems Research, Inc. Chapter II Page 29
Finding #7. A State's ability to collect comprehensive data for a large number of
participants, as well as the completeness ofthe data regarding food
purchasing patterns for any individual, is affected by the nature ofthe
State's food delivery system.
The feasibility study demonstrated that data regarding food purchasing patterns of WIC
participants can be collected, matched, and analyzed in one selected chain store within a State.
However, the ability of a State to expand the data collection beyond a very limited number of
stores that may agree to "volunteer" to participate in this type of study is limited by the
approach States take to selling the use of the technology. The completeness of the food
purchasing data for an individual participant is affected by the number of authorized WIC
vendors that have the capacity to provide WIC transaction data to the WIC State agency. For
example, small neighborhood markets may not have the scanning technology required to
provide the necessary WIC transaction data to a WIC State Agency.
These issues are discussed below.
Finding 7.1 States using a vendor-specific system are more likely to collect
complete participant transaction data than States using an open
system.
One of the major findings of this feasibility study concerns the type of WIC FI system operated
in the State. To a great extent, the comprehensiveness of the data collected depends on
whether the State agency operates a vendor-specific FI system or an open system. As
discussed in Chapter I, vendor-specific systems require WIC participants to identify a single
store in which they will cash their FIs. In a vendor-specific State, WIC participants select one
store in which to do all oftheir WIC shopping. Thus, when collecting and analyzing data from
a single chain store, complete information on the participants' food purchase is available in a
vendor-specific State. When collecting data from a single chain store in an open system State,
complete participant food purchasing data is only available for participants choosing to shop at
one ofthe study stores. The only way to capture more complete food purchasing information
in an open system State would be to collect data from a number ofdifferent stores in a
geographic area.
Health Systems Research, Inc. Chapter II Page 30
There are currently 12 vendor-specific States operating retail delivery systems. However,
based upon data contained in the July 1996 State Agency Participation Report compiled by
FCS, these States represent 35 percent of the total WIC population being served in all retail
delivery States. This represents over 1.9 million participants shopping in States where
complete participant food purchasing information could be made available through the
selection of a single chain or group of stores.
Finding 7.2 Increased use of electronic benefits transfer (EBT) will likely
increase the feasibility of utilizing the study technology.
Future methods of providing WIC benefits will lend themselves to this methodology. The
increased use ofEBT systems will cause States to reevaluate their grocer authorization
practices. EBT will require stores to have in place the technology to read a participant's
benefit card. Because EBT systems automatically record a client identification number at the
point of sale, a clear link is established between the WIC transaction and the client, and cashier
error is bypassed. Testing of the methodology developed through this study in an EBT
environment would likely produce complete purchasing profiles of participants, regardless of
whether they live in a vendor-specific State.
Health Systems Research, Inc. Chapter II Page 31
Appendix A: Description ofMerged Data Set
3*
Description Of The Merged Data Set
The following is a description ofthe merged data set created from matching the WIC State
agency participant data with the store point-of-purchase transaction data.
1. The Header Record and the Detail Record
Table A-l presents the elements included in the merged data set, which is the end product ofthe
methodology described in this report. Deciding on the structure was a complex task, because
each WIC transaction can contain multiple food items and each WIC participant can have many
transactions. The final data set has to be flexible enough to facilitate analysis on the levels of
item, transaction, person, and family.
As shown in Table A-l, each WIC transaction record contains a header record and at least one
detail record. The header record includes demographic information about the participant to
whom the WIC FI was issued and information about the participant's WIC category, status, and
priority (column 1). The header record also shows the types and amounts of foods the participant
is authorized to purchase (column 2). These data are obtained from the demographic issuance
records maintained by the WIC State agency. Information about the transaction, such as the FI
number, store number, and dollar amount, is also contained in the header record (column 3).
This information comes from the records submitted by the retail store.
The detail records are files containing the specific food purchasing information obtained from the
stores. These detail records show the type and amount of a particular food actually purchased in
that transaction (column 4). It contains comprehensive information on the item purchased,
including brand, type, form, volume, quantity, and price, and the number ofthe FI used to
purchase the item. Each item purchased in a given WIC transaction will have its own detail
record describing the product and linking the purchase to a FI number. After the sets ofheader
and detail records have been created, the two data sets can be combined by linking a header to its
corresponding detail(s) using the FI number. Detail records can then be aggregated to produce a
Health Systems Research, Inc. Page i
summary record of all items purchased with a given FI. Similarly, an individual WIC
participant's records may be combined to produce a single record containing that client's total
issuance and purchasing information.
Dftfe EteBMHits Praia WIC Ageacy FH*s an! Sttperaaarkat Traatactian Recards
Header (Transaction) Record
Detail (Item)
Demographic Record
Information
Issuance Information Transaction
Information
Encrypted participant
identifier
Gallons of milk FI number UPC code
Date of birth Pounds of cheese Transaction date Description
Education level Dozens of eggs Store number Brand
Ethnic group Pounds of carrots Total purchase
amount
Food category
Migrant/refugee code Cans of tuna FI paid amount Type
Homeless code Pounds of beans/peas Draft type Form
Household smoking
code
Ounces of peanut
butter
Match step (for
algorithm)
Item price
Health care source Cans ofjuice and size
ofcans (in ounces)
Number of FIs used Item quantity
Sex Ounces of cereal Item volume
Family income Cans of formula FI number
Income period Cans of infant juice
Family size Ounces of infant
cereal
Breastfeeding status (for
infants)
Breastfeeding duration
Immunizations
WIC status
WIC category
Priority
Source: Health Systems FResearch, Inc.
Health Systems Research, Inc. n Page 2
The merged record was originally designed to represent one WIC FI, and to include all detail
records of items purchased with that FI. However, individuals occasionally use more than one FI
in a single transaction; in these cases, it was unclear which header record the associated detail
records should be linked to. Therefore, the header record was redesigned to represent a single
transaction, rather than a single check. When only one check is used in a WIC transaction, the
two will be the same. When more than one check is used in a transaction, the header record
represents the total amount of food in each category issued on all checks used in the transaction.
The check number used to link this record to the associated detail records is that of the last check
used in the transaction. A final data element in the header record represents the number of
checks used in the transaction. This allows identification of records in which more than one
check was used and entered into a single transaction record. The records could then be separated
by participant and transaction when a single transaction record reflected a purchase by multiple
participants, such as a pregnant woman who shopped for her if and an a child participant.
2. Presentation of the Data in the Merged Data Set
a. Identifiers
The data set was designed to exclude information with which to determine the actual
identity of WIC participants. Instead, the data included two encrypted identifiers: a five-digit
family identification code, created by numbering sequentially all families in the data
set from an undisclosed starting point, and an individual identifier, created by adding one
digit to the family identification. In this field, a "0" represents a pregnant or postpartum
woman; other numbers represent children, numbered in order of age, youngest to oldest.
b. Demographic Characteristics
A number of demographic data elements, including information relating to WIC
enrollment (WIC eligibility category, status, and priority) and the participant's ethnic
background and socioeconomic status (racial/ethnic code, income, and education level)
are included in the header records. Based on the participant's date of birth, the
participant's age was calculated as of the middle of the study period. Using the
Health Systems RReesseeaarrcchh,, Inc. // Page 3 1 3/
participant's reported family income, the family's poverty status was calculated as
defined by the 1995 Federal Poverty Income Guidelines.
c. Foods Issued
All WIC FIs issued in Wisconsin and Pennsylvania fall into more than 300 draft type
codes in each State, each of which describes a specific package of WIC foods. The
original issuance records contained a field for this "draft type" code rather than a list of
all foods issued. To create usable variables describing the amount of food issued in each
food category, each draft type was converted into 12 separate variables, each containing
the amount of one of the 12 WIC food categories issued on that type of check. Table A-2
presents the units ofmeasure used for each category.
d. Other Information in the hsuance Record
In addition to demographic data and information about food issued, WIC issuance records
contain information about the ultimate disposition ofthe WIC FI. If the FI is not cashed
within its valid time period, the record indicates that the FI had been voided. If the FI is
redeemed by the store, the record gives the date the FI was deposited at the bank and the
amount the store was paid. In most States, the record also contains the identification
number of the store depositing the check.
e. Foods Purchased
The detail records contain information about each food item purchased with one or more
WIC FIs. This information is based on the UPC and LPC codes received from the stores,
each of which contains the brand, type, and volume of the item purchased. For the
feasibility study, these codes were converted into three variables, each of which has
different possible values for each food category. The values for these variables are
presented in Table A-2.
Health Systems Research, Inc. Page 4
To compare the amount of food issued with the amount purchased, the amount of each
item purchased is presented in the data set ir. the units used for the food issued. In
addition to the volume of the item purchased, the detail record contains the number of
items purchased and their total price.
Health Systems Research, Inc. Page 5
^7
m^t-^ TMtf»*& ?
-':■:■:::::•''-.■.■.■.,■■■..•:■. Vtlw» for Dmripiivf Varj«bk« by Food Catego.7
Food Category Units Type Form Brand
Milk gallons l=skim
2=lowfat
3=whole
4=lactose reduced
1 liquid
2=powdered
3=evaporated
alpha
Eggs dozens n/a n/a n/a
Cheese pounds l=American
2=Colby
3=Swiss
4=Cheddar
5=mozzarella
6=Monterey Jack
7=Muenster
8=unknown
l=block
2=shredded
3=sliced
4=string cheese
5 ^cheese balls
(mozzarella
only)
not collected
Cereal ounces n/a n/a alpha
Juice/ cans l=orange 5=grapefruit 1 =firozen alpha
Infant Juice 2=grape 6=lemonade
3=apple 7=pineapple
4=tomato 8=berry
9=mixed/unknown
concentrate
2=canned single-strength
Peanut Butter ounces 1=crunchy
2=creamy
n/a alpha
Beans pounds I=navy 4=1 ima
2=pinto 5=lentil
3=black 6= unknown
l=dry
2=canned
alpha
Infant Cereal ounces I=rice 4=mixed
2=barley
3=oatmeal
n/a alpha
Infant Formula1 cans l=soy-based
2=milk-based
1 =powdered
2=concentrate
3=ready-to-feed
alpha
Tuna ' cans [white/light/dark
packed in water or oil]
[solid, chopped,
flaked, grated]
alpha
Carrots pounds n/a [canned, fresh] alpha
1 Foods issued in canswereiden titled by can size, in ounces.
Health Systems Research, Inc.
-/
Page 6
Home Economics and Household Collection
