Rethinking Social Welfare Fraud from a Complex Adaptive Systems Perspective

Kim, Y., & Maroulis, S. (forthcoming), Administration & Society

Rethinking Social Welfare Fraud from a Complex Adaptive Systems

Perspective

Abstract

Despite efforts to control fraud in public assistance programs, the perception and realities of the

problem persist. Serious barriers related to data collection and research methods impede the

understanding of how and why fraud occurs, thereby limiting options for improving program

integrity. This paper argues that, based on a complex adaptive systems perspective, social

welfare fraud can be understood as a collective outcome emerging from repeated interactions

among stakeholders during the routinized business processes of public assistance programs.

When dealing with fraud, great attention must be paid to how it occurs and persists, not just how

serious the problem is or who commits these crimes.

Key Words: Social Welfare Fraud; Complex Adaptive Systems; Agent-Based Modeling

  1  

Introduction

Public assistance programs such as The Special Supplemental Nutrition Program for

Women, Infants, and Children (WIC) and The Supplemental Nutrition Assistance Program

(SNAP, also known as Food Stamps) were established to assist individuals and families with

their nutritional needs. Unfortunately, these programs have experienced fraud (Government

Accountability Office (GAO), 1997, 1998, 1999, 2010). The existing literature and government

reports address fraud along with abuse and waste.1 This paper focuses on fraud alone. Fraud

involves “dishonesty, illegality and the intentional wrongful obtaining of either money or

benefits from governmental programs” (McKinney & Johnston, 1986, p. 5). In the literature, and

in the general public, this phenomenon has been known as “welfare fraud,” a term that has

primarily been associated with the public image of welfare recipients who are illegally collecting

welfare benefits (Chunn & Gavigan, 2004).

Contemporary public assistance programs are more complicated than what this public

image implies in terms of who is involved in the programs and how they engage in fraudulent

activities. Over the past few decades, it has become common practice for government agencies to

contract out social services to private organizations or third parties (Hodge & Greve, 2007;

Romzek & Johnston, 2005; Warner & Hefetz, 2008). Private retailers have become integral parts

of public assistance programs, such as WIC and SNAP, by delivering benefits to the needy on

behalf of the program. This complicates the issue of welfare fraud because the intermediate party

between government and benefit recipients may also engage in wrongdoing (United States

                                                                                                               1  McKinney and Johnston (1986) define abuse as “administrative violations of departmental,

agency or program regulations” and waste as “the unnecessary costs which result from

inefficient or ineffective practices, systems or controls” (p. 5).  

  2  

Department of Agriculture (USDA), 2001, 2003, 2007). Although it is rare, fraud committed by

government employees has also been documented (GAO, 1999).

Fraud in public assistance programs is difficult to understand, detect, and cope with.

Above all, there is an inherent data issue. Fraud information and statistics largely come from

government reporting systems or federal agencies’ enforcement efforts to estimate the

prevalence of the issue or any monetary loss that may be incurred (e.g., GAO, 1999; USDA,

2007). As is the case with many police- and investigation-generated crime data and statistics

(Maguire, 2002; Skeen, 2003), fraud statistics collected from government agencies are likely to

be biased downward because they include only those cases that were detected or charged.

Aggregated information on fraud can show the seriousness of fraud in a program, but it does not

reveal the underlying processes that lead us to observe fraud.

As mentioned above, many public assistance programs are implemented as a system

involving several loosely-coupled entities such as state agencies, local agencies, private

organizations, and benefit recipients. When fraud is committed in such a system, it takes a long

time for the fraud to affect other entities and the entire system. Especially in the current “pay and

chase” system, in which states reimburse benefit providers first and later determine the

likelihood of fraud, it becomes difficult to trace, identify, and tackle the source of fraud in a

timely manner using payment information. Further, fraudulent behavior is dynamic in nature.

Once a fraud detection method is in place, it immediately begins to lose its effectiveness (Bolton

& Hand, 2002). Managing fraud in public assistance programs is not just a matter of identifying

a set of simple deviance issues. With the introduction of new systems, such as Electronic Benefit

Transfer (EBT) cards, the misbehavior may evolve, transform, and become harder to detect in

ways that public managers have never before seen (USDA, 2012).

  3  

While aiming to understand, from a managerial point of view, how fraud keeps occurring

within public assistance programs, this paper acknowledges that the underlying causes of welfare

fraud, especially by benefit recipients, are complicated and require a deeper understanding of the

socioeconomic and historical contexts that welfare recipients deal with (Swan et al., 2008;

Regev-Messalem, 2014). As long as poverty and its associated constraints remain severe realities

that people on public assistance face on a daily basis, social welfare fraud may not disappear.

This paper does not intend to ignore the importance of understanding why people in public

assistance programs engage in fraud, exaggerate the issue of welfare fraud, or undermine the

value of public assistance programs. Instead, it intends to help to rethink government efforts to

manage fraud by presenting a fresh look at the problem. The issue of fraud is of major concern to

policymakers and program managers because it undermines the integrity and efficiency of

government programs and can result in public distrust of government. In addition, it hurts the

majority of honest people who need public assistance and support. It is important that these

programs run with integrity so that they continue to serve people in need. Below we provide a

more focused look on welfare fraud in the context of a public assistant program.

Fraud in a Public Assistance Program

The Women, Infants, and Children (WIC) program

WIC aims to safeguard the health of low-income women, infants, and children up to age

five who are at nutritional risk.2 The program provides nutritious supplemental foods, nutritional

education, and referrals to healthcare and other social services. This program is available in all

50 states, the District of Columbia, 34 Indian tribal organizations, and the United States (U.S.)

territories. These 90 WIC state agencies administer the program through approximately 1,900

                                                                                                               2  http://www.fns.usda.gov/wic  

  4  

local agencies and 10,000 clinic sites, working with 47,000 authorized retailers. WIC is a system

of heterogeneous players who have different functions and purposes.

WIC is not an entitlement program but rather a federal grant program for which the U.S.

Congress authorizes a specific amount of funds each year. Congress appropriated $6.6 billion for

WIC in fiscal year 2010 for the combination of food costs and nutrition services and

administration costs. By comparison, the WIC program spent $10.4 million in 1974, $700

million in 1980, $2.1 billion in 1990, and $3.9 billion in 2000.3 During the past 40 years, the

program has grown approximately 100-fold in its population size, from 88,000 in 1974 to 9.1

million in 2010.

The United States Department of Agriculture (USDA) was given the responsibility of

administrating the program, and WIC now operates through a federal-state-local partnership.

State agencies are responsible for the program’s operations. They contract with local WIC

sponsoring agencies, allocate funds to them, and provide assistance to the local agencies. Local

WIC agencies provide services to WIC participants either directly or through local service sites

(e.g., clinics). The clinics certify applicants, provide nutritional education, make referrals to other

social services, and distribute food vouchers to be used at WIC-participating retail stores. States

can use any combination of the three delivery systems: authorized retail outlets; home delivery;

and WIC storage facilities (GAO, 1999).

WIC’s vulnerability to fraud

The program’s vulnerability to fraud, abuse, and waste has long been known to WIC

managers, but only piecemeal empirical information exists. In the late 1990s, GAO (1999)

conducted a study and concluded that the USDA did not have overall estimates of fraud within

                                                                                                               3  The information was retrieved from the USDA website on May 29, 2013.  

  5  

the WIC program. Using various data sources, the study reported that about nine percent of all

vendors in the program committed fraud as of September 1998, and 7,074 participants engaged

in serious fraud, such as trafficking (0.14 percent of the average monthly participation in fiscal

year 1998). GAO’s own survey showed that four percent of all local agencies identified

documented cases of employee fraud in 1998.

The USDA has independently conducted studies focusing on vendor management. Early

information on the program’s vulnerability to vendor fraud appeared in two major national

studies conducted in 1991 and 1998 (USDA, 2003). The recent studies (USDA, 2007, 2013b)

report that the frequency and estimated dollar losses due to vendor violations, including

overcharges, were lower than indicated in previous studies.

These studies on WIC integrity by the GAO and the USDA have identified some crucial

findings. Here, findings are summarized with four major points, not necessarily in chronological

order. First, the program is vulnerable to fraud that can be committed by any player of the

program. According to the GAO (1999):

Vendors, participants, and employees can engage in a variety of fraudulent or abusive

activities. For example, vendors could charge the WIC program more for a food item than

the item’s shelf price. Participants could have misrepresented facts affecting their

eligibility, such as income, in order to receive program benefits. Finally, employees could

obtain benefits for friends or family who are not eligible for the program (p. 4).

It can no longer be assumed that welfare fraud is committed simply, or even primarily, by

welfare recipients (Luna, 1997).

Second, state and local WIC agencies have reported detecting “higher levels of vendor

fraud than of participant fraud or employee fraud” (GAO, 1999, p. 5). The types of vendor fraud

  6  

are numerous. The intentional deliberate actions that vendors take to violate program regulations,

policies, or procedures include, but are not limited to: trafficking, the exchange of public service

benefits for cash; overcharging, charging more than the shelf price or exceeding the maximum

price allowed by WIC; and substitution, providing credit toward the purchase of unauthorized

items that can be initiated by vendors or recipients. As an illustration, Table 1 summarizes

statistics on WIC vendors’ overcharging practices and the consequences, as identified in the four

previously mentioned national vendor management studies. It is also worthwhile to note here that

it is not reasonable to assume that all types of fraud in the program may be known to government

agencies.

<Insert Table 1 Here>

Third, studies have found that certain characteristics of vendors can signal their

likelihoods of engaging in vendor fraud (USDA, 2003; 2007, 2013b). According to the recent

study (USDA, 2007), small vendors with low business volume were eight times more likely to

overcharge than were large vendors. Vendors who did not provide a required receipt were eleven

times more likely to overcharge than were vendors who provided a receipt. Vendors who did not

scan food items were more than twice as likely to overcharge than those who did. The similar

patterns were identified from the latest study (USDA, 2013b).

Fourth, as a result of the findings from two previous USDA studies conducted in 1991

and 1998, major regulatory changes to WIC vendor management were made in 1999 and 2000.

The first change to the vendor disqualification rule was published on March 18, 1999, to

mandate uniform sanctions across state agencies for the most serious WIC program violations,

such as trafficking. The second change, which was published on December 29, 2000, aimed to

strengthen vendor management in retail food delivery systems (USDA, 2007). The USDA’s

  7  

2005 evaluation (2007) concluded that considerable improvement has been made in vendor

overcharging and undercharging due to these regulatory changes in 1999 and 2000.

Altogether, in spite of the accumulation of knowledge regarding WIC fraud and efforts to

control it, the problem continues to occur. Fraud, committed by service providers rather than

benefit recipients, has been recognized as a serious problem within the program. In addition, it is

found that welfare fraud can be committed by neutral third parties, as can be seen from the City

of Newark employee case of WIC fraud (October 2, 2011, release from the State of New Jersey

Office of the Attorney General).4

This paper draws from a systemic perspective in deriving insights into social welfare

fraud. Johnston (1986) notes that “[a] systemic approach … simply suggests that a full

understanding of the problem and possible remedies for it, must be based upon an analysis which

reaches well beyond immediate persons and cases” (p. 17). This task has been challenging for

both academics and practitioners. The question asked here is how a complex adaptive systems

(CAS) approach (Holland, 1995, 1998; Miller & Page, 2007) helps us to undertake the task and

changes our understanding of welfare fraud as well as the effectiveness of the policies that

address it.

Welfare fraud has not been the subject of much academic study, as pointed out by

McKinney and Johnston (1986) and Swan, Shaw, Cullity, and colleagues (2008). In the public

management literature, this issue has been tangentially touched upon in the discussion of other

topics, such as privatization (Amey, 2012; van Slyke, 2003). The complexity of fraud in

contemporary social welfare service programs requires an innovative perspective to make sense

of the problem and to increase effectiveness in responding to the challenge. In the following

                                                                                                               4  http://www.nj.gov/oag/newsreleases11/pr20111005c.html. Retrieved on July 15, 2013.  

  8  

section, some highlights of the CAS approach will first be discussed to more completely reflect

the current understanding of fraud in a public assistance program and to indicate how this

approach can be used to cope with the problem.

Insights from Complex Adaptive Systems Studies

The study of CAS —Complexity Science—is an intellectual child of early systems theory

and its subsequent developments, such as chaos theory. Holland (1995) defined CAS as “systems

composed of interacting agents described in terms of rules. The agents adapt by changing their

rules as experience accumulates” (p. 10). The system of interacting agents is necessarily dynamic,

and the agents can exhibit recognizable patterns of organization across spatial and temporal

scales (Holland & Miller, 1991; Parker, Manson, Janssen, Hoffmann & Deadman, 2003).

Manson (2000) briefly contrasts early systems theory with CAS theories. Early systems

theory studied entities linked by flows and stocks, which can lead to the discovery of a nonlinear

behavior of a system. Modern CAS theories take similar but more disaggregated approaches by

directly specifying the rules of interaction among constantly changing entities. The CAS view

and its analytical tools (e.g., agent-based modeling) have been applied to various topics in other

social science contexts, such as crime (Liu & Eck, 2008), civil violence (Epstein, 2002),

organizational behavior and dynamics (Kitts & Trowbridge, 2007, Maroulis & Wilensky, 2014),

and land-use and land-cover change (Parker et al., 2003).

The CAS perspective offers several insights that would be helpful for policymakers and

public managers in dealing with the problems they face. The insights most applicable to welfare

fraud are highlighted below, utilizing research examples where the CAS approach has been

meaningfully applied in the field of public policy and management.

Macroscopic regularities emerge from local interactions between heterogeneous agents

  9  

CAS scholars are interested in studying emergent phenomena that cannot easily be

predicted or even envisioned from knowledge of the properties of each agent in the system (Casti,

1997), and they are interested in understanding the mechanisms underlying these phenomena

(Holland, 1998; Miller & Page, 2007). In CAS studies, emergence refers to “macroscopic

societal regularities arising from the purely local interaction of the agents” (Epstein, 2002,

p.7245). Here, “local” implies that interactions take place between neighboring agents. Agents

are not limited to people or individuals, but they can include cells, words, trees, animals, and

organizations that can exercise limited or full decision-making power. The interaction can be

direct, indirect, or mediated social or physical exchanges among agents at an individual level

(Railsback & Grimm, 2012). It can be in the form of strategies, resource exchanges, decisions, or

communications between agents. In short, emergence, or what Resnick (1994) called “emergent

objects” (e.g., traffic jams), occurs from the interactions of lower-level agents (i.e., cars).

Consider environmental injustice (EJ), a macroscopic regularity characterized by a

disproportionate collocation of racial or ethnic minorities near environmentally harmful sites or

facilities (i.e., disamenities). This has been a concern of policymakers and researchers for the

past several decades (Liu, 2000; Ringquist, 2005). In search of explanations of race and

ethnicity-based environmental injustice, early studies have focused on the discriminatory

intention of environmental disamenities in siting processes as a main cause of the problem

(Bullard, 1996). In reality, environmental injustice occurs in communities where multiple

heterogeneous and autonomous actors, such as firms and residents, dynamically make location

decisions (Banzhaf & Walsh, 2008). The causes of environmental injustice are likely to be more

delicate than the simple behavior of one actor. However, due to the limitations of commonly

used research methods for EJ problems (e.g., the difficulties of considering multiple actors and

  10  

the dynamism of urban systems), existing research has primarily focused on assessing the

relationship between disamenity’s siting decision and environmental injustice. This focus has

limited our understanding of how and why race-based environmental injustice exists.

In their study of causes leading up to societal environmental injustice, Eckerd, Campbell,

and Kim (2012) examined the siting decisions of actors such as firms and residents and the

uncoordinated dynamic interactions between them within urban settings. The siting decisions

were informed by actors’ preferences and neighborhood conditions at the time of siting. The

researchers found that environmental injustice occurred when: disamenities intended to locate

near minorities or residents preferred to live near other residents who were like them. The study

suggests that environmental injustice is the result of “a race-conscious society rather than just a

collection of race-conscious firms” (p. 945). In addition, understanding how environmental

injustice occurs requires consideration of the residential choice constraints, such as similarity

preferences, as well as community characteristics, such as a growth rate and racial parity (Kim,

Campbell & Eckerd, 2013). These studies suggest that race- and ethnicity-based environmental

injustice is an emergent phenomenon that cannot be attributed to one actor or one factor. This

implies that the underlying mechanisms that enlarge or lessen societal environmental injustice

must be understood in order to take effective policy actions to address the problem (Campbell,

Kim & Eckerd, 2015).

Agents learn and adapt from each other, retaining rules that increase fitness

A distinguishing characteristic of CAS is that they are able to adapt or evolve (Mitleton-

Kelly, 2003). According to Holland (1995), “adaptation is the sine qua non of cas” (p.8), and

adaptation in CAS includes learning and related processes. CAS consists of large numbers of

agents that interact by sending and receiving signals (Holland, 2006). The system appears to be

  11  

adaptively intelligent because agents are able to learn and improve performance by acting upon

signals (Holland, 1995). Learning occurs when the behaviors of agents change as a result of

interactions with other adaptive agents. Learning agents are able to modify their own decision-

making or behavioral rules according to their experiences in reacting to others and the

environment, or as their experience accumulates (Holland, 1995). The more successful rules are

selected, whereas the less successful rules are cast aside (Axelrod & Cohen, 2000). Adaptive

agents can create a new way of working, new structures, and different relationships (Comfort,

1994; Mitleton-Kelly, 2003).

Working together (i.e., collaboration) is recognized as a necessary mode of governance,

especially for problems that are not easily tackled or managed by traditional governance modes

(McGuire, 2006; Koliba, Meek, & Zia, 2010; Kalu, 2012). Collaborative governance is defined

as a myriad of dynamic interactions among diverse players from multiple sectors who are

interested in addressing multi-faceted problems together (Ansell & Gash, 2008). Stakeholders in

collaboration are actors in a networked environment (Kim, Johnston & Kang, 2011). Each actor

has attributes, such as resource availability and expectations of others’ actions, and functions that

make up his or her decisions during the collaboration. Collaborative governance also involves

mechanisms, defined as a collection of processes that allow agents to perform their specific

functions, and structure, defined as a collection of relations between agents (Choi & Robertson,

2012).

Significant collaborative governance research based on a CAS perspective has been

conducted (e.g., Choi & Robertson, 2012; Johnston, Hicks, Nan & Auer, 2011). In this research,

agent-level decision rules and processes are considered a focal point of analysis, whether they

are in the form of deliberation and negotiation processes (Choi & Robertson, 2012) or of

  12  

inclusion processes (Johnston et al., 2012). Choi and Robertson (2012) argue that processes such

as deliberation seem to be more important than decision rules (e.g., a majority rule), in terms of

generating desired outcomes in collaborative governance. Deliberation processes create

opportunities to enhance mutual understanding and appreciation for others’ preferences (Choi &

Robertson, 2012). As a result, stakeholder attitudes toward others’ preferences and interests

evolve. In other words, the interactions during multiple processes facilitate stakeholders’

learning. Understanding the stakeholders’ adaptive preference can shed light on the roles of

various processes in collaborative governance that can help to collectively, and effectively, deal

with multi-faceted problems.

Mutual dependence of agents can lead to unexpected results

Not only are the aggregate patterns of CAS the result of interactions of multiple agents

but also those interactions frequently occur in ways that cannot be accurately captured by rules at

the aggregate level. When agents in a system have behaviors that are mutually dependent,

heterogeneity in agents makes it impossible to capture the outcome of those interactions without

looking at each interaction individually. In social and economic systems, agents’ mutual

dependence often leads to policy resistance, a situation “where our policies are delayed, diluted,

or defeated by the unforeseen reactions of other people, or of nature” (Sterman, 2000, p. 3).

According to Meadows (2008), policy resistance “comes from the bounded rationalities of the

actors in a system, each with his or her (or “its” in the case of an institution) own goals…. Such

resistance to change arises when goals of subsystems are different from and inconsistent with

each other” (p. 113).

Choice-based reforms in education are an illustration of a policy context where outcomes

are contingent upon the mutually dependent decisions of the agents within the system (Maroulis

  13  

et al., 2010). Choice-based reforms aim to change the catchment area approach (popular in the

United States) that requires students to attend a school in their neighborhood. That is, giving

parents the freedom to choose their children’s school can serve as a key driver to improving

educational outcomes by introducing competition among schools (Hoxby, 2003). One might

expect that in such a reform, the more that parents and students emphasize achievement in the

choice of schools, the more likely it would be that the reform brings about the desired academic

improvements.

Maroulis and colleagues (2014) illustrate how individual preferences do not necessarily

aggregate in an expected way in education systems. In their CAS model, individual-level

emphasis on school achievement relative to geographical proximity in school choice did not

always lead to higher mean achievement for the district. Moreover, they found that in addition to

the quantity and quality of new schools that enter the district, the timing of the entrance of new

schools matters. New schools entering the system learn from existing schools. Under certain

conditions, too much emphasis on achievement can lead to a “front-loading” of new school entry,

which short-circuits this learning process. A rapid introduction of new schools forgoes the

benefit that comes from late entrants who have an opportunity to learn from a population of

higher-quality schools. This paradoxical result is more likely to occur when the fraction of

households who participate in the choice program is low, low-performance schools do not close

very easily, and new schools are introduced to the education system at a high rate (Maroulis et al.,

2014).

These examples illustrate that urban systems, governance systems, and education systems

are complex adaptive systems. The CAS approach can shed light on understanding the behaviors

of these systems. What is happening in these systems, such as injustices, collaboration, and

  14  

reforms, is of primary concern to policymakers and public managers. Asking why and how these

phenomena emerge are logical follow-up questions in CAS studies. They are explored in greater

detail in the following discussion of rethinking fraud in a public assistance program.

Rethinking WIC Fraud from a CAS Perspective

In rethinking WIC fraud, the benefits of the CAS approach are twofold: Fraud can be

understood as a systemic phenomenon emerging from interactions among various players, and

the system can be scrutinized using the CAS approach—i.e., the use of agent-based modeling—

to draw useful insights for policymakers and public managers. Table 2 summarizes CAS’s

insights and implications for WIC and social welfare fraud. The remainder of the section

elaborates on each item.

<Insert Table 2 Here>

Recurrent activities in WIC business provide fraud opportunities

How does fraud occur? Or, more specifically, why does fraud keep occurring in the

program? According to Johnston (1986), two theoretical views on fraud exist: 1) Fraud is the

wrongdoing of people with criminal intention and 2) the program is flawed and has loopholes

that create fraud. The CAS perspective can be viewed as a complementary perspective that helps

to foreground the connection between the individual and institutional views. Moreover, this

connection, and its potential underlying mechanisms, can be explored using CAS simulation

models.

When inherent limitations in data collection and availability exist and the underlying

processes are important but hidden, simulations have been suggested as invaluable alternative

approaches (Liu & Eck, 2008). Simulations are built for various purposes, including the testing

of theories. Liu and Eck (2008) contend that if simulation models can mimic observed crime

  15  

patterns, the theories encapsulated in the model may be sufficient for explaining the crime

patterns. For instance, the routine activity theory of crime (Felson, 2002) has been tested to see

whether or not it can explain what is observed in crimes. CAS’s simulation models that test this

theory seem to be robust in replicating crime patterns, such as street robbery (Groff, 2007).

The focal point of the analysis in Kim and Xiao (2008) was not how one group of players

in the WIC program committed fraud but rather how observed fraud levels emerged from the

system as a whole, including program recipients, stores, and government agencies. Their CAS

model built in the core arguments of the routine activity theory (Felson, 2002) and the situational

action theory (Wikström, 2010). Both of these theories are variants of the systemic view of crime

(Bunge, 2006). The level and distribution of observed vendor fraud were replicated by

simulating the risk-taking behaviors of some players in WIC’s routine business processes, e.g.,

benefit recipients visiting a vendor for food redemption or asking for voucher redemptions by

vendors to the state (Kim, 2012; Kim & Xiao, 2008).

The underlying mechanisms of WIC fraud in the CAS model are as follows: recurrent

and prevalent activities in the program’s business process (e.g., when a program recipient is

visiting a store to exchange vouchers for supplemental foods) and fraud opportunities perceived

by some actors with a high-risk propensity for rule-breaking behaviors. For example, interactions

between the program recipient and the vendor lead to acts of rule-breaking in some instances,

such as when both actors with a high fraud propensity meet and agree to engage in illegal benefit

exchanges. The failure of negotiation in a fraud opportunity leads the program recipient to search

for another vendor. As a result of an actor’s choice in the fraud opportunity, the actor’s fraud

propensity changes, and this change influences his/her next choice in illegal benefit exchanges.

These iterative processes at a micro-level led us to observe a certain level of fraud in the system.

  16  

Because the motivation for an actor to commit fraud occurs in a setting that consists of

opportunity, friction, and monitoring, not every individual engages in rule-breaking behavior,

nor does every actor become corrupted. The system’s fraud level is stabilized at a certain point

but persists until policy changes are introduced.

The CAS view suggests that fraud has its roots in the program’s basic business processes

and in their relationships to their social environments. Efforts to control fraud require an

understanding of how fraud has occurred in light of the way that programs conduct business

(Johnston, 1986). It is necessary for public managers to be cognizant of pathways that produce

fraud within the program. For instance, the WIC service delivery method that uses private

vendors is expanding the scope of welfare fraud from the issue of welfare recipient’s individual

action to the problem of deliberated and organized vendor misbehavior involving deceptive

devices and planning. In a recently indicted WIC fraud case, a store operator who was not

eligible to own a WIC store changed her name, and her new name was used to open a store and

to apply for the WIC program as a vendor. Using the changed name, she attended a WIC training

program and presented herself as a store manager.5 In another case, many of the defendants that

conspired to open purported grocery stores in Georgia “canvassed low-income neighborhoods

and solicited WIC and Food Stamp participants to illegally exchange their benefits not for food

but for cash.”6 These identified potential pathways that are enabled due to the delivery method

must be of interest to WIC managers so that they can design appropriate managerial and

educational strategies.

Interactions are a vulnerability point

                                                                                                               5 http://www.justice.gov/usao/gan/press/2012/12-19-12c.html 6 http://www.fbi.gov/atlanta/press-releases/2014/fifty-four-defendants-charged-in-18-million-wic-and-food-stamp-fraud-conspiracy

  17  

In private retail stores, theft, or “shrinkage,” is largely due to customers or employees

acting as individuals (see the National Rail Security Surveys7). The case of WIC fraud shows

that fraud in public assistance programs is more complicated than that of fraud within the private

sector. Program recipients can commit fraud with or without malicious intentions, but feasibility

can be enhanced by complicity on the part of vendors or employees. Both vendors and program

participants can initiate and commit benefit substitutions (USDA, 2007). The multiplicity of

interactions among program players and others can complicate the issue, thus increasing the

likelihood of unintended non-routine problems in the system.

Sutherland (1940), who coined the term “white-collar crime,” suggested that “white-

collar criminality, just as other systematic criminality, is learned; that it is learned in direct or

indirect association with those who already practice the behavior” (p. 10). As mentioned,

adaptive actors can create new ways of working and forming different relationships; however,

WIC tends to approach the problem by separating out fraud by different groups, such as vendor

fraud, participant fraud, and employee fraud, and then it directs focus on these groups based on

the frequency or severity of the fraud (USDA, 2003, 2007). This limits our understanding of

learning processes of actors within and between groups.

The limitations of this approach to WIC fraud become apparent in how WIC monitors

fraud. WIC has focused on identifying potential and actual rule violations, especially those

committed by vendors (USDA, 2007). State agencies must “identify high-risk vendors at least

once a year using criteria developed by FNS and/or other statistically-based criteria developed by

the state agency” (USDA, 2013, p. 415). High-risk vendors are those who have a high

                                                                                                               7 http://lpportal.com/academic-viewpoint/item/2805-2012-national-retail-security-survey-first-glance-at-the-results.html; also see http://users.clas.ufl.edu/rhollin/srp/srp.html for the recent survey.

  18  

probability of committing vendor violations (USDA, 2013). Vendor violations are not only

limited to administrative violations but also to illegal monetary transactions. The criteria that

have been used to identify high-risk vendors include: store characteristics; WIC sales volume;

participant complaints; and field investigations (USDA, 2013). These procedures are fairly

simplistic. They are static, are less effective in uncovering the interaction between corrupt agents,

and are easy to evade because of familiarity with detection procedures (Kim, 2012).

If WIC fraud emerges from interactions among actors with a high-risk propensity during

business processes, the necessary approaches to detect wrongdoing must be updated based on an

understanding of the interactions among them. For instance, Kim (2007) developed a fraud

detection method for a state WIC program based on how recipients and vendors geographically

interacted. Linking several program data sources, the study first empirically examined interaction

patterns between WIC stores and program recipients in an Ohio county. As expected, most

recipients redeemed benefits at the store that theoretically had the highest probability of being

visited; however, some WIC stores and recipients interacted quite unexpectedly. Some vendors

were not only attracting recipients who had a very low chance of visiting the vendor but also

they were also dispelling recipients who had a very high probability of visiting the vendor. Such

vendors were more likely to be confirmed as fraudulent in field investigations.

This finding suggests that the program needs to focus on the evolution of interactions of

actors within and between groups, and technologies are available to aid in this task. For example,

Oregon recently received a USDA’s funding to combine analytics and Geographic Information

Systems (GIS) to better target fraud, and the state of Washington is using innovative strategies to

monitor and investigate fraud occurring via social media and e-commerce websites.8 These

                                                                                                               8 http://www.fns.usda.gov/pressrelease/2014/fns-000914

  19  

approaches are innovative in that they focus on interactions as a way of detecting abnormalities,

but they still suffer from other limitations mentioned at the beginning of the paper. Technical

solutions for dynamic social systems are often limited and easily become out of date. It is better

to educate policymakers about how important healthy interactions among actors for all parties

concerned are developed so that the collective responsibility for program accountability is

understood (O’Connell, 2005).

Policies can be delayed, diluted, or defeated by the unforeseen reactions of others

What would happen if a state agency considered changes in how it manages fraud? WIC

regulations mandate that state agencies must conduct compliance investigations on all high-risk

vendors, identified using the criteria mentioned, up to a five percent minimum each fiscal year

(USDA, 2013). If fewer than five percent of vendors are identified as high-risk, then the agency

must randomly select additional vendors, up to five percent, to conduct the investigation. If

actual violations are identified from the procedure and evidence of a pattern of fraud is

accumulated, a mandatory sanction—e.g., warning letters, monetary fines and penalties, and

disqualification from the program—is imposed on the vendor, depending upon the seriousness of

the vendor’s violations (USDA, 2013).

WIC’s sanction policies base their rationale on deterrence theory, which assumes that

punishment influences a criminal’s behavior (Becker, 1968). A plethora of empirical literature

exists on whether or not it is the probability of being punished or the severity of the punishment

that more strongly influences criminal decisions (Pratt, Cullen, Blevins, Daigle & Madensen,

2008). However, the role of a punishment’s imminence has been largely missing from the

literature (Selke, 1983). When would swiftness matter? How would swiftness interact with other

sanction dimensions, such as certainty and severity? In the WIC context, what if the percent of

  20  

vendors in compliance investigations (i.e., the probability of being caught) needs to be changed

due to some other reason, such as a shrinking resource? And what if sanctioning of fraudulent

vendors takes a short, versus a long, time after the investigation due to various administrative

reasons in a state agency? Under the assumption that states do not have the freedom to change

the severity of the sanctions in the regulations, thought experiments for these questions can still

be performed in order to envision the consequences of the changes in WIC.

Overall, the intervention involving a small fraction of fraudulent vendors (e.g., one

percent) worked very differently from the intervention involving a slightly higher fraction of

fraudulent vendors (e.g., five percent or ten percent) in the WIC system (Kim, Zhong, & Chun,

2013). The simulation first demonstrated what we would normally expect. When a state agency

has limited resources to enforce investigation and sanction policies among only a small fraction

of fraudulent vendors, their punitive action or intervention must be accompanied by a

promptness to achieve a significant reduction in fraud. Swiftness matters. However, even if a

state does not have many resources to take prompt action, a similar level of reduction may be

achieved once a relatively higher fraction of fraudulent vendors is sanctioned. Prompt action

matters less when five or ten percent of vendors are sanctioned. The effect of swiftness is

nonlinear after the threshold of five percent sanctioned in the experiment.

Using CAS models, nonlinear systems can be explored for understanding, but the systems

still remain unpredictable even after some understanding of them is achieved (Gilbert &

Troitzsch, 2005). CAS studies have provided an explanation of the behaviors of various

nonlinear systems, but they do not necessarily imply prediction (Epstein, 2008). Alternatively, it

is necessary for public managers and policymakers to pay attention to conditions that lead to

surprising or unexpected outcomes and to continue to assess whether or not the program is

  21  

appropriately attending to such conditions. The federal regulation that requires compliance

investigations up to five percent of high-risk vendors each fiscal year appears to be a sound

guideline, given that an increase or decrease in the percentage of vendors for compliance

investigations higher than that does not bring much deterrence effect regardless of the

promptness of punitive action. However, a plan to decrease the percentage of vendors for

investigations must be considered along with the promptness of action. This is an insight to help

policymakers and public managers to consider when the program plans a change in fraud

management rather than a specific guideline to follow.

Conclusion

This paper brings attention to the CAS perspective in expanding an understanding of

social welfare fraud and the possible responses that might be designed to address it.

Policymakers and public managers must often deal with undesirable events, such as fraud, in

public assistance programs. These programs consist of heterogeneous actors who interrelate, and

dynamically interact, for specific functions and purposes. These interrelationships and

interactions, shaped by the programs’ business processes, can also provide an opportunity for

fraud despite efforts to prevent it. The CAS perspective suggests that fraud is an emergent

property of public assistance programs. The system’s holistic characteristics are not necessarily

reflected in average individual behaviors, nor are they the simple aggregation of individual

behaviors. Indeed, the aggregation of individual behavior can lead to surprising outcomes and

unintended consequences. To increase the likelihood that policy prescriptions do not fall prey to

such “policy resistance” (Sterman, 2000; Meadows, 2008), an understanding of the system that

incubates the problem is needed first.

  22  

Comprehensive restructuring of government programs is not necessarily what the CAS

perspective would suggest. Instead, it highlights that the problem in contemporary public

assistance programs needs to be put into a broader perspective that attends to interactions and

pathways to fraud rather than what the programs currently do. Fraud cannot be “eradicated by

institutional tinkering, if it can be eradicated at all” (Johnston, 1986, p. 28) or by blaming certain

groups. The daunting task of a systemic approach that requires an analysis that goes beyond

immediate persons and cases can now be better undertaken with the CAS perspective and its

analytical tools. This approach can enable policymakers to continue to devise creative measures

to manage and alert others to fraud without unfortunate consequences.

In sum, this paper provides examples and a focused-case look into the field that shows

how a CAS perspective can mediate the link between what happens in the real world, what we

think, and how and why it happens. The CAS perspective tasks policymakers and program

managers to be creative, flexible, and holistic in dealing with social welfare fraud.

  23  

Table 1: A Summary of Statistics on WIC Vendors’ Overcharges

1991 1998 2005 2013*

% of vendors overcharged

(total weighted vendors in

study; #)

9.9%

(34,033)

7.0%

(36,754)

3.5%

(39,374)

5.6%

(40,634)

Gross overcharge in 2004

dollars (million)

$42.62 $42.87 $6.06 $13.8

Annual redemption in 2004

dollars (billion)

$2.9 $4.48 $4.47 N/A

Percentage of overcharge

relative to total redemptions

1.5% 0.9% 0.1% N/A

Data sources: USDA (2007), Table II-1 (p.10), Table V-2 (p.48), and Table V-9 (p.54); * USDA

(2013b), Figure VI-1 (p.46) and V-7 (p.54).

  24  

Table 2: Summary of CAS Insights and Implications for Welfare Fraud

CAS insight WIC fraud from a CAS perspective

Practical implications

Macroscopic regularities emerge from local interactions between heterogeneous agents.

Fraud emerges from the interaction of vendors, program participants, and local health clinics during the routinized business process of WIC.

Understanding pathways that produce fraud from the interaction between different stakeholders is more important than parsing blame across categories.

Agents learn and adapt from each other, retaining rules that increase fitness.

Routine interactions between participants and vendors evolve into financially beneficial but illicit activities as vendors and participants build trust and learn about loopholes from each other.

Repeated interactions are vulnerability points. While limited, deviations from regular interaction patterns between vendors and program participants can be monitored via data analytic techniques.

Mutual dependence of agents can lead to unexpected results.

WIC fraud levels drive the severity, swiftness, and certainty of fraud management efforts, which, in turn, influence an agent’s likelihood of WIC fraud. Additionally, the certainty, severity, and swiftness of punitive action interact to determine the effectiveness of fraud management efforts.

Widespread, swift, and mild responses to suspected fraud may be more beneficial than targeted, delayed, and extremely harsh punishments of fraud indicted.

  25  

References

Amey, S.H. (2012). Commentary: Contract design failure lead to bad deals. Public

Administration Review, 72(5), 697-698.

Ansell, C., & Gash, A. (2008). Collaborative governance in theory and practice. Journal of

Public Administration Research and Theory, 18, 543-571.

Axelrod, R., & Cohen, M.D. (2000). Harnessing Complexity: Organizational Implications of a

Scientific Frontier. New York: Basic Books.

Banzhaf, H., & Walsh, R. (2008). Do people vote with their feet? An empirical test of Tiebout’s

mechanism. American Economic Review, 98, 843-863.

Becker, G. S. (1968). Crime and punishment: an economic approach. In R. Febrero and P. S.

Schwartz. (Eds.). The Essence of Becker (pp. 463-517). Stanford, CA: Hoover Institution

Press.

Bolton, R. J., & Hand, D. J. (2002). Statistical fraud detection: A review. Statistical Science,

17(3), 235-255.

Bullard, R. (1996). Environmental justice: It’s more than waste facility siting. Social Science

Quarterly, 77, 493-499.

Bunge, M. (2006). A systemic perspective on crime. In P.H. Wikström and R.J. Sampson, The

Explanation of Crime: Context, Mechanisms and Development (pp.8-30). New York:

Cambridge University Press.

Campbell, H., Kim, Y., & Eckerd, A. (2015). Rethinking Environmental Justice: Insights from

Agent-Based Modeling. New York: Routledge.

Casti, J. L. (1997). Would-Be Worlds: How Simulation is Changing the Frontiers of Science.

New York, NY: John Wiley & Sons, Inc.

  26  

Choi, T., & Robertson, P.J. (2012). Deliberation and decision in collaborative governance: A

simulation of approaches to mitigate power imbalance. Journal of Public Administration

Research and Theory. doi: 10.1093/jopart/mut003.

Chunn, D.E., & Gavigan, S.A.M. (2004). Welfare law, welfare fraud, and the moral regulation of

the ‘never deserving’ poor. Social & Legal Studies, 13, 219-243.

Comfort, L.K. (1994). Self-organization in complex systems. Journal of Public Administration

Research and Theory, 4(3), 393-410.

Eckerd, A., Campbell, H., & Kim, Y. (2012). Helping those like us or harming those unlike us:

Illuminating social processes leading to environmental injustice. Environment and

Planning B, 39, 945-970.

Epstein, J. (2002). Modeling civil violence: An agent-based computational approach.

Proceedings of the National Academy of Sciences, 99(3), 7243-7250.

Epstein, J. (2008). Why Model? Journal of Artificial Societies and Social Simulation, 11(4)12.

http://jasss.soc.surrey.ac.uk/11/4/12.html.

Felson, M. (2002). Crime and Everyday Life (3rd ed.). Thousand Oaks, CA: Sage Publications.

GAO. (1997). Food Assistance: Reducing Food Stamp Benefit Overpayments and Trafficking,

GAO/T-RCED-98-37.

GAO. (1998). Food Assistance: Computerized Information Matching Could Reduce Fraud and

Abuse in the Food Stamp Program. GAO/T-RCED-98-254.

GAO. (1999). Food Assistance: Efforts To Control Fraud And Abuse In The WIC Program Can

Be Strengthened. GAO/RCED-99-224.

GAO. (2010). Supplemental Nutrition Assistance Program: Payment errors and trafficking have

declined, but challenges remain. GAO-10-956T.

  27  

Gilbert, N., & Troitzsch, K. G. (2005). Simulation for the Social Scientist. London: Open

University Press.

Groff, E.R. (2007). Simulation for theory testing and experimentation: An example using routine

activity theory and street robbery. Journal of Quantitative Criminology, 23(2), 75-103.

Hodge, G.A., & Greve, C. (2007). Public-private partnership: An international performance

review. Public Administration Review, 67(3), 545-558.

Holland, J.H. (1995). Hidden Order: How Adaptation Builds Complexity. Cambridge, MA:

Addison-Wesley.

Holland, J.H. (1998). Emergence: From Chaos to Order. Cambridge, MA: Addison-Wesley.

Holland, J.H. (2006). Studying complex adaptive systems. Journal of Systems Science and

Complexity, 19(1), 1-8.

Holland, J.H., & Miller, J. H. (1991). Artificial adaptive agents in economic theory. Learning

and Adaptive Economic Behavior, 81(2), 365-370.

Hoxby, C.M. (2003). School choice and school productivity. Could school choice be a tide that

lifts all boats? The Economics of School Choice (pp. 287-342). University of Chicago

Press.

Johnston, E.W., Hicks, D., Nan, N., & Auer, J. (2011). Managing the inclusion process in

collaborative governance. Journal of Public Administration Research and Practice, 21(4),

699-721.

Johnston, M. (1986). Systemic origin of fraud, wastes and abuse. In J.B. McKinney and M.

Johnston (Eds.). Fraud, Waste and Abuse in Government: Causes, Consequences and

Cures (pp.15-29). Philadelphia: Institute for the Study of Human Issues Publications.

  28  

Kalu, K.N. (2012). All that glitters: Competing narratives and transaction costs in complex

collaborative environments. Administration & Society, 45(4): 420-442.

Kim, Y. (2007). Using spatial analysis for monitoring fraud in a public delivery program. Social

Science Computer Review, 25(3), 287-301.

Kim, Y. (2012). Simulating a fraud mechanism in public service delivery, In A. Desai (Ed.),

Simulations for Policy Inquiry (pp. 121-138), New York: Springer.

Kim, Y., & Xiao, N. (2008). FraudSim: Simulating fraud in a public delivery program. In L. Liu

& J. Eck (Eds.), Artificial Crime Analysis Systems: Using Computer Simulations and

Geographic Information Systems (pp. 319-338). Hershey, PA: IGI Global.

Kim, Y., Campbell, H., & Eckerd, A. (2013). Residential choice constraints and environmental

justice. Social Science Quarterly, 95(1): 40-56.

Kim, Y., Johnston, E., & Kang, H.S. (2011). A computational approach to managing

performance dynamics of networked governance systems. Public Performance &

Management Review, 34(4), 580-597.

Kim, Y., Zhong, W., & Chun, Y. (2013). Modeling sanction choices on fraudulent benefit

exchanges in public service delivery. Journal of Artificial Societies and Social Simulation,

16(2)8, http://jasss.soc.surrey.ac.uk/16/2/8.html.

Kitts, J., & Trowbridge, P.T. (2007). Shape up or ship out: Social network, turnover, and

organizational culture. Computational and Mathematical Organization Theory, 14(3),

333-353.

Koliba, C., Meek, J.W., & Zia, A. (2010). Governance Networks in Public Administration and

Public Policy. CRC Press.

  29  

Liu, F. (2001). Environmental Justice Analysis: Theories, Methods, and Practice, Boca Raton:

CRC Press.

Liu, L., & Eck, J. (2008). Artificial Crime Analysis Systems: Using Computer Simulations And

Geographic Information Systems. New York: Information Science Reference.

Luna, E.G. (1997). Welfare fraud and the forth amendment. Pepperdine Law Review, 24(4),

1235-1290.

Maguire, M. (2002). Crime statistics: The ‘data explosion’ and its implications. In M. Maguire,

R. Morgan, & R. Reiner. (Eds.). The Oxford Handbook of Criminology (pp., 322-375).

Oxford, NY: Oxford University Press.

Manson, S.M. (2000). Simplifying complexity: A review of complexity theory. Geoforum, 32,

405-414.

Maroulis, S., Guimera, R., Petry, H., Gomez, L., Amaral, L.A.N., & Wilensky, U. (2010).

Complex systems view of educational policy research. Science, 330(6000): 38-39.

Maroulis, S., Bakshy, E., Gomez, L., & Wilensky, U. (2014). Modeling the transition to public

school choice. Journal of Artificial Societies and Social Simulation, 17(2)3.

Maroulis, S., & Wilensky, U. (2014). Social and task interdependencies in the street-level

implementation of innovation. Journal of Public Administration Research and Theory,

doi: 10.1093/jopart/mut084.

McGuire, M. (2006). Collaborative public management: Assessing what we know and how we

know it. Public Administration Review, 66(s1), 33-43.

McKinney, J.B., & Johnston, M. (1986). Fraud, Waste and Abuse in Government: Causes,

Consequences and Cures. Philadelphia: Ishi Publications.

  30  

Meadow, D.H. (2008). Thinking in Systems. White River Junction, Vermont: Chelsea Green

Publishing.

Miller, J.H., & Page, S.E. (2007). Complex Adaptive Systems: An Introduction to Computational

Models of Social Life. Princeton: Princeton University Press.

Mitleton-Kelly, E. (2003). Complex Systems and Evolutionary Perspective on Organizations:

The Application of Complexity Theory to Organizations. Oxford, UK: Elsevier Science

Ltd.

O’Connell, L. (2005). Program accountability as an emergent property: The role of stakeholders

in a program’s field. Public Administration Review, 65(1), 85-93.

Parker, D. C., Manson, S. M., Janssen, M. A., Hoffmann, M. J., & Deadman, P. (2003). Multi-

agent systems for the simulation of land-use and land-cover change: a review. Annals of

the Association of American Geographers, 93(2), 314-337.

Pratt, T.C., Cullen, F.T., Blevins, K.R., Daigle, L.E., & Madensen, T.D. (2008). The empirical

status of deterrence theory: A meta-analysis, In F.T. Cullen, J.P. Wright, & K.R. Blevins

(2008). Taking Stock: The Status of Criminological Theory (pp. 367-395). New

Brunswick, NJ: Transaction Publishers.

Railsback, S.F., & Grimm, V. (2012). Agent-Based and Individual-Based Modeling. Princeton:

Princeton University Press.

Regev-Messalem, S. (2014). Trapped in resistance: Collective struggle through welfare fraud in

Israel. Law and Society Review, 48(4): 741-772.

Resnick, M. (1994). Turtles, Termites and Traffic Jams: Explorations in Massively Parallel

Microworlds. Cambridge, MA: The MIT Press.

  31  

Ringquist, E. J. (2005). Assessing evidence of environmental inequities: A meta-analysis.

Journal of Policy Analysis and Management, 24(2), 223-247.

Romzek, B.S., & Johnston, J.M. (2005). State social services contracting: Exploring the

determinants of effective contract accountability. Public Administration Review, 65(4),

435-449.

Schelling, T.C. (1978). Micromotives and Macrobehavior. New York: W.W. Norton &

Company.

Selke, W.L. (1983). Celerity: The ignored variable in deterrence research. Journal of Policy

Science and Administration, 11(1), 31-37.

Skeen, J.W. (2003). Health care fraud and industry structure in the United States. Social Policy

& Administration, 37(5), 516-529.

Sterman, J.D. (2000). Business Dynamics: Systems Thinking and Modeling for Complex World.

Boston: Irwin McGraw-Hill.

Sutherland, E.H. (1940). White-collar criminality. American Sociological Review, 5(1), 1-12.

Swan, R.S., Shaw, L.L., Cullity, S., Roche, M. Halpern, J., Limbert, W.M., & Humphrey, J.

(2008). The untold story of welfare fraud. Journal of Sociology & Social Welfare.

XXXV(3), 133-250.

USDA. (2001). WIC Vendor Management Study 1998. Report No. WIC-01-WICVM.

USDA. (2003). A Comparison of WIC Vendor Management Practices in 1991 and 1998. Report

No. WIC-03-WICVM.

USDA. (2007). 2005 WIC Vendor Management Study. Report No. WIC-06-WICVM-02.

USDA. (2012). Supplemental nutrition assistance program: Trafficking controls and fraud

investigation. Federal Register, 77(104). 7 CFR Parts 271 and 274.

  32  

USDA. (2013a). The Code of Federal Regulations, Federal Register. 7 CFR. Part 246.

http://www.gpo.gov/fdsys/pkg/CFR-2013-title7-vol4/xml/CFR-2013-title7-vol4-

part246.xml.

USDA. (2013b). 2013 WIC Vendor Management Study. Retrieved from

http://www.fns.usda.gov/sites/default/files/2013WICVendor.pdf.

Van Slyke, D.M. (2003). The mythology of privatization in contracting for social services.

Public Administration Review, 63(3), 296-315.

Warner, M.E., & Hefetz, A. (2008). Managing markets for public service: The role of mixed

public-private delivery of city services. Public Administration Review, 68(1), 155-166.

Wikström, P.H. (2010). Activity fields and the dynamics of crime: Advancing knowledge about

the role of the environment in crime causation. Journal of Quantitative Criminology, 26,

55-87.