Abstract
This paper offers an up-to-date review of Enterprise Zone programs across states in the U.S. An Enterprise Zone program often focuses on target locations, populations, and/or industries. Enterprise Zone program designs and implementations vary between states. The expected outcomes are likely to differ from the desired outcome, which is to revitalize Enterprise Zone areas. This paper also provides a review of the existing Enterprise Zone empirical literature on program effectiveness assessments. There are conflicting recommendations, even for the same Enterprise Zone program. To assess whether research design contributes to this situation, the paper explores several research design factors, including outcome measure, time span, spatial unit, data choice, and control of endogeneity. Information on these research factors in the existing literature is collected and coded. Descriptive and correlation analyses suggest that outcome measure, spatial unit, post-program time span, and the choice of an endogeneity control method have significant impacts on program effectiveness assessments.
Keywords
Introduction
Many government programs target specific locations, populations, and/or activities. For instance, public housing programs such as the Community Development Block Grants (CDBG) and Housing Choice Vouchers have a sharp focus on low- and moderate-income households. Redevelopment programs have strong geographical targets, depending on the purposes of these programs (e.g. redeveloping downtown, distressed areas, or brownfields). A state Enterprise Zone (EZ) program has multiple targets on location, population, and/or industries. Program effectiveness evaluation is important because it provides inputs for future program improvements and for the justification to renew or terminate a program.
It is challenging to evaluate a public program, due to the multi-factorial nature of urban activities, the diversity of program setup, and the lack of quality data to support such an evaluation. As for EZ programs, the concept first appeared in the United Kingdom in the 1970s (Hall, 1982), when most western cities were suffering urban blight in central areas. A majority of U.S state governments quickly embraced the concept in the 1980s. The federal Empowerment Zone (EPZ) program was introduced in 1993, sharing a similar structure with state EZ programs. The strategy of both EZ and EPZ is to use incentives to encourage firms to move into or stay in EZ areas, and hire local residents. Policymakers and administrators want to know whether such a program can stimulate growth and improve the economic environment in poor neighborhoods. In the past two decades, researchers have echoed this policy question with a great deal of empirical studies about EZ effectiveness in stimulating growth. Nevertheless, these studies point to mixed conclusions about EZ program effectiveness. For instance, Elvery (2009) and Bostic and Prohofsky (2006) both studied the California EZ program. Elvery concludes that the program is not effective, while Bostic and Prohofsky conclude that it is. Such contradictory conclusions can be found for many state EZ programs, including Florida, Kentucky, New Jersey, New York, Ohio, and the Federal EPZ program. Partially because of this lack of consistency, many states have terminated their EZ programs (Appendix 1).
This research explores the diversity in state EZ programs and in EZ research design. Its purposes are to (1) uncover the diversity of state EZ practice, (2) assess the relationship between research design and research conclusion, and (3) make suggestions about research design for future program evaluations. In contrast to many existing EZ review papers (Fisher and Peters, 1997, Greenbaum and Landers, 2009; Kolko and Neumark, 2010), this paper has a sharper focus on research design factors.
The remainder of the paper is organized as follows. The next section provides an up-to-date summary of state EZ program designs and analyzes expected EZ outcomes given the design of such a program. The “The EZ literature” section provides a brief review of the existing EZ empirical studies. The “Research design factors” section proposes several research design factors and discusses their relations to possible EZ effectiveness conclusions. The “Research design and EZ effectiveness assessments” section provides further quantitative assessments of the relations between these factors and EZ outcome conclusions in the existing EZ literature. The final section concludes the paper.
The EZ program and the expected outcomes
The literature argues that EZ programs adopt a place- and people-specific strategy to attract jobs and investments into target locations and for target populations (Greenbaum, 2004; Peters, 2002; Turner and Cassell, 2007). Most states hope that such programs can revitalize their economies and create more jobs. For instance, the goal of the Alabama EZ program is to “stimulate business and industrial growth in depressed areas,” California expected its program to “encourage business investment and promote the creation of new jobs,” and Iowa’s EZ goal is to “promote new economic development in economically distressed areas.” An EZ program may not achieve these desired goals simply because of their program design. In the following, I will explore the EZ program structure and discuss the expected outcomes given the program design.
I surveyed available active or terminated state EZ programs, and summarized program information in Appendices 1 and 2. Most states enacted their EZ acts and created EZ programs in the 1980s, and many of these EZ programs have been terminated or have expired without renewal. Some states may have different names for their EZ programs. For instance, North Dakota has the Renaissance Zone Program and New Hampshire the Revitalization Zone Program. Regardless of names, these programs share a similar structure, which includes benefit bundles, designation of EZ areas, and restrictions for receiving benefits, but with great diversity in detailed criteria and implementation processes.
The incentive bundle
The incentive bundles of most states offer job-related and/or investment-related benefits to retain and attract businesses. Job-related benefits are tied to the number of jobs created. The Colorado program offers $500 income tax credits for every new job in the first year. Most other EZ programs offer $1000–$3000 credits per job. In addition, training assistances and subsidies are provided by many EZ programs to reduce potential training costs to hire target low-income and low-skilled populations. The Colorado and New Jersey programs offer additional assistance on employees’ insurance, and this is expected to encourage the creation of quality jobs.
Investment-related benefits include tax credits or exemptions that are related to the amount of new investments, property-related investments, and sales/purchasing activities within EZs. Most state EZ programs match 1–10% of any new private investments. Many states allow for tax credits for activities related to property (e.g. new construction, rehabilitation, and rental) and for exemptions of sales tax on purchases. A few states (Colorado, Iowa, and New Jersey) provide tax credits on research and development activities.
Other benefits are often direct financial assistance offered in the form of income tax credit and/or general excise tax exemptions, if employers can meet basic requirements on business operation, job creation, and/or expansion. For instance, the Hawaii program allows for a general excise tax exemption and a credit on a portion of a business income tax, if this business is a new or expanding one of the target activities. Businesses within EZ areas may also directly receive local benefits in the form of local utility tax and regulation fee abatements, reduced local regulatory requirements, and/or better access to government grants and services.
Similar incentives can have varying outcomes, depending on how these incentives are awarded over time. Most EZ programs distribute benefits to business owners over a period of four to five years, and much of this distribution is front-loaded in the first one or two years. For example, the Alabama program allows business owners to use their credits for 80% of their tax liability in the first year, 60% in the second year, 40% in the third year, and 20% in the fourth year. The Arizona program is one of the very few that budget the distribution of benefits over time, allowing for up to $500 of job credit in the first year, up to $1000 in the second year, and up to $1500 in the third year. Such an EZ program is expected to keep businesses within EZ areas over a longer term.
Many state EZ programs offer both job- and investment-related tax credits. These two types of incentives have different impacts on job creation, depending on the rate of substitution between labor and capital in a specific industry (Hanson and Rohlin, 2011a). Job-related incentives have positive impacts on job creation. However, the impacts of investment-related incentives are subtle, because investment can substitute labor. Bondonio and Greenbaum (2007) study 1987–1992 establishment-level employment dynamics. Their results suggest that job-related incentives stimulate job creation, while investment-related incentives have negative impacts on job creation, supporting the existence of a substitution effect between investment and labor. Further, there is a scale effect between labor and investment (i.e. more workers leading to more investments and vice versa), and this effect varies between industries. According to the 2017 U.S. Input–Output Statistics, the intermediate input (that relies on capital investments) and labor input ratio is 3.51 for manufacturing, suggesting that for every one unit of labor input there are 3.51 units of capital inputs. This ratio is 1.16 for retail and 0.79 for professional and related services. The lower scale effect for service-oriented industries makes it easier for these industries to take advantage of job-related tax credits.
The EZ benefits could be unattractive for existing mature firms. The law of diminishing marginal return on investments suggests an optimal level of investment (Arrow and Kruz, 2013). When increasing investments cannot bring in the desired profits, firms do not have strong motivations to participate. New and small-sized firms are more likely in their growth phase, and possibly would take advantage of EZ benefits. The empirical evidences support that EZ programs have stronger impacts on new and small-sized establishments/firms than on existing establishments/firms (Billings, 2009; Bondonio and Greenbaum, 2007).
EZ designation
An EZ designation must involve a process to evaluate local distress level. Almost all states consider people- or job-based distress indicators, such as unemployment rate, poverty rate, per capita/household/worker income, and job/population loss. Nevertheless, the actual EZ distress criteria vary across states. In most states, the unemployment rates for EZ areas must be 50% higher than the state average. Some states (e.g. Connecticut, Georgia, Nebraska, Oregon) are stricter, requiring a 100% higher rate, and some (e.g. Colorado, Illinois, Michigan, New York, Vermont) are less strict, only requiring a 20–33% higher rate. West Virginia, New Hampshire, and North Dakota are the only three states that do not consider distress level for EZ designations.
Eighteen states control the size of EZ areas. This is important, as maximum and minimum area criteria can promote economies of scale and, at the same time, restrict EZ coverage to the most distressed target areas. Other EZ designation criteria can be grouped into two categories: growth potential and local involvement. Michigan’s and Utah’s designations directly require development plans. New Hampshire and North Dakota consider the availability of commercial, industrial, and/or residential land-use activities but not necessarily declining neighborhoods. All these criteria focus on growth potential. Local involvement is another common criterion. For instance, an EZ may be required to be located within one single jurisdiction boundary for the convenience of local involvements and local–state collaborations in regulating EZ-related issues. Most state governments expect local commitments in improving designated areas through local tax abatements, infrastructure plans, and supportive land-use regulations. A few states require the creation of a local legal entity, such as the Urban Enterprise Associations in Indiana.
These EZ designation criteria constitute a geographic filter to sift out qualified distressed areas with higher poverty, unemployment, vacancy rates, and lower per capita income and housing values (Elvery, 2009; Greenbaum, 2004; Greenbaum and Engberg, 2004; Hanson, 2009; Zhang, 2015). These poor local conditions discourage activities. Poorer areas have fewer resources (e.g. capital, infrastructure) and lower political will to attract more investments (Hill and Nowak, 2002; Peters and Fisher, 2004). Declining neighborhoods often have other social issues, such as limited mobility and social network, crime, segregation, and/or poor human well-being (Ludwig et al., 2012). More importantly, distressed neighborhoods may not have a labor force and a financial structure to support re-growth. Hill and Nowak (2002) find that about half of the residents of Camden, New Jersey (which the authors argue is a failed city) are not of working age. With the evidenced association between poverty and education (Ladd, 2012), it can be reasonably argued that distressed neighborhoods lack human talents to support growth. Further, the literature indicates the existence of insurance and mortgage redlining (Hernandez, 2009; Squires, 2003). After filtering through this geographical requirement, desired program outcomes will be reduced.
This geographical filter is stronger in some states than in others. For instance, EZs designated on the basis of county-level information, such as in Iowa and Kansas (see spatial unit information in Appendix 1), have weaker geographical focus than if designated on the basis of information for smaller spatial units. In the case of Iowa, the county level per capita income varies between $386 and $3708, with a standard deviation of $539, according to 2006–2010 American Community Survey data. The census tract level per capita income varies between $0 and $20,333, with a standard deviation of $1728. If the Iowa EZ program allowed for census-tract-level designation, the program would have had a sharper focus on distressed areas than when using county-level information. All else equal, the tighter the geographic filter, the more distressed the EZ neighborhoods are.
The people filter
To receive benefits, businesses need to hire target workers. Criteria for determining target workers vary across states. Most states have objective measures that consider the employment/income status and residential locations of workers. For instance, Alabama defines target workers as people who have been formerly unemployed for more than 90 days. Maryland’s definition of an economically disadvantaged individual is based on unemployment longer than 30 days. Kentucky required not only unemployment and low-income status, but also EZ locations. The state maintained a detailed list of qualified addresses. A few states also include some subjective criteria for describing qualified workers, for instance, “people unemployable by traditional standards” in the Louisiana program.
The people filter, requiring businesses to hire and continue to hire low-skilled workers, underestimates the impacts of technology on economic production and the increasingly important role of human capital in a firm’s performance (Crook et al., 2011; Shepherd, 2015). It may be easier for some firms to absorb this people requirement than for others. For instance, the Kentucky EZ program required employers to have at least 25% of their new hires from target populations. If there is no industry target, it is easier for small-sized businesses, which are often service-oriented and have lower requirements for labor quality, to participate in this program. For many other firms, it may be difficult to deal with the operational difficulties caused by employing a large number of target workers with low- or no-education, even with short-term employee training assistance. Further, there are arguments and evidences for the negative association between firm size and firm growth (Axtell, 2001; Beck et al., 2005). The cost of creating the same job is higher for a large firm than for a small one. In this sense, smaller firms are more motivated to create more jobs to take advantage of EZ benefits.
The industry filter
EZ benefits may only be eligible for specific industries (requirement II in Appendix 1). Many programs have a strong focus on manufacturing activities. The Alabama program requires that a qualified business has to be manufacturing, transportation, warehousing, or headquarters of such activities. The programs of Arkansas, Colorado, Georgia, Illinois, Kansas, and Oregon have a similar focus.
The industry focus limits potential benefit recipients to a narrow group of businesses, which often are manufacturing or manufacturing-related activities. Since the 1950s, there has been an observed trend of manufacturing activities moving from central to suburban areas, for more and cheaper land, and better highway access (Scott, 1982). This suburbanization of manufacturing activities has continued into the 1990s. 1 It could be operationally difficult for manufacturing activities to take advantage of EZ benefits because the geographic filter requires the program to target declining areas, which are more likely to be in central areas. At the same time, the manufacturing focus makes the existing activities in central areas, which are non-manufacturing, unable to participate in the program.
The expected EZ outcomes
EZ benefits subsidize a firm’s capital and/or labor costs, but the place, people, and industry requirements add operational costs or disqualify many local businesses to participate in the program. After filtering through these filters, it is unclear whether there will be detectable desired outcome. Policymakers hope for immediate outcomes of noticeable increases in investment, jobs, population, and consequently, a revitalized neighborhood (Figure 1). Direct financial assistant may help retain existing target businesses, but plays limited role in job/investment growth. Property-related benefits may benefit local property markets. However, the size of such benefits may be limited due to low property values of EZ neighborhoods. The impacts of investment- and job-related assistances have been extensively discussed in this section. The place, people, and/or industry filters discourage local businesses to take advantage of an EZ program. While an investment- or a job-related incentive is expected to increase local investment or job, it may have no or negative impacts on each other, and the impacts vary across industries.
EZ incentives, requirements, and expected outcomes.
The EZ literature
EZ program effectiveness has drawn much academic attention, with many empirical results available. Figure 2 shows the number of empirical publications by time period. Appendix 3 explains the process to collect these EZ papers. State EZ programs received an increasing level of attention over 1991–2005, a period immediately after the creation of most states’ EZ programs in the 1980s. There have been a declining number of empirical papers on state EZ programs since 2006, and, during this same period the federal EPZ program began to gain academic attention.
Number of publications by publication year. EPZ: Empowerment Zone; EZ: Enterprise Zone.
Several EZ review papers are available, all focusing on assessing the EZ structure and program effectiveness. Wilder and Rubin (1996) point to the positive effect of an EZ program on job creation, primarily based on program effectiveness assessments conducted by governments. Fisher and Peters (1997) find that state EZ incentives vary in size and nature, and there is no consensus about program effectiveness in the studies published in the 1990s. Buss (2001) focuses on the literature of state tax incentives, which do not have to be geographically targeted. His review points to conflicting conclusions about whether these programs have improved local economies. Greenbaum and Landers (2009) include more recent EZ studies and also conclude that the literature offers conflicting recommendations about program effectiveness.
Table 1 provides an up-to-date summary of available EZ empirical research, most of which uses regression analyses and makes inferences with statistical significance. A gray cell indicates a research that did not find significant EZ program impacts when the corresponding outcome indictor was used. A gray cell with P indicates a research that found positive and significant impacts (at the 5% significance level). A gray cell with N indicates a research that found negative and significant impacts (at the 5% significance level). A gray cell with M indicates mixed outcomes, with both negative and positive impacts (both significant at the 5% level). Further explanations about outcome indicators are provided in the next section.
Summary of EZ effectiveness studies.
EZ: Enterprise Zone.
EA1: total employment; EA2: workers income; EA3: total investment or inventory; EA4: number of establishments or number of firms; EA5: total shipments.
ED1: indicators by industry; ED2: indicators by size of firms; ED3: indicators by firms’ business cycle.
D1: total residence employment; D2: total population; D3: household or personal or family income; D4: unemployment reduction; disparity; D5: poverty reduction; D6: income disparity.
Table 1 echoes the conclusions of existing EZ literature reviews about conflicting conclusions on program effectiveness. Kolko and Neumark (2010) suggest that the heterogeneity in local economic conditions and administration could be the source for varying program effectiveness conclusions. Boarnet (2001) and Bartick (2002) raise the question of how much does methodology matter in EZ evaluation.
Many studies focus on the same EZ program, but draw different or conflicting conclusions. Examples include studies of the California program (Bostic and Prohofsky, 2006; Moore, 2003; Neumark and Kolko, 2010), of the Kentucky program (Lambert and Coomes, 2001; Zhang, 2015), and of the Ohio program (Ham et al., 2011; Hultquist, 2015; Landers, 2006). These results support the argument that methodology or research design may have possible roles in EZ research conclusions.
Research design factors
Cost–benefit, shift–share, and regression analyses are the three methodologies used in EZ empirical studies. In cost–benefit analysis, researchers compare total program costs and benefits to assess the net benefits. For instance, Sridhar (1996) uses this approach to assess the Illinois EZ program, which offers property tax abatements and public service supports and has stimulated $5653 million in investment and 79,119 new jobs in the state over 1984–1991. Sridhar can calculate the total program costs and benefits on the basis of wages of these new jobs and the public budget spent for tax abatements and service support. This is theoretically sound, but practically difficult because of a lack of complete cost and benefit information (Bartick, 2002; Rubin, 1990). Sridhar (1996) had to make assumptions about workers’ wages and administrative costs, and developed different scenarios to study the program net benefit.
A shift–share analysis compares industry-specific growth within EZ areas with growth in the region (city or county) and the nation, and attempts to separate the growth within EZ areas due to policy stimulation from the growth due to other national and regional factors (Dowall, 1996; Lambert and Coomes, 2001; Rubin and Wilder, 1989). Such studies often uncover positive impacts on some industries, but negative impacts on others. For instance, Dowall (1996) suggests positive zone effects on manufacturing, construction, and wholesale, but negative effects on other industries, and an overall negative effect in terms of total job growth in California. The results from a shift–share analysis could be biased, because this approach attributes the entire local component to the influence of the EZ policy, while ignoring that other localities in the EZ areas also affect growth.
Most studies use regression analyses to assess and quantify the impacts of EZ programs on selected outcomes. For such studies, several research design factors matter, including the choices of outcome indicators, spatial units, research design, data quality, the control of EZ endogeneity.
Outcome measure used in the literature can be grouped into three categories: economic, demographic, and housing market, as presented in Table 1. EZ programs are expected to stimulate economic growth within EZs, which can be directly translated into total employment/establishment/firm/wage growth (EA1–EA5). These indicators are at the aggregate level, and ignore heterogeneity in both program design and economic activities. For instance, the Arizona program focused more on encouraging job creation, while the Oklahoma program encourages investments (Appendix 1). An outcome measure related to job creation fits the Arizona program better than the Oklahoma program.
People hope for EZ programs to reverse the declining economic trends in the targeted areas. However, most EZ programs have strong industry targets, and different businesses react differently toward the same EZ program. Target industries may begin to grow, while non-targeted industries may continue to decline. Given the possible counterbalancing impacts of the same EZ program on different businesses, an aggregate economic indicator is likely to uncover null effects. Further, because of the unattractive economic environment in EZ areas, non-targeted industries may decline faster than targeted industries may grow. In this case, an assessment based on an aggregated economic indicator will find negative impacts from the program.
Many researchers have decomposed economic impacts by industry type, establishment size, and business status (ED1–ED3). Billings (2009) studies the impact of the Colorado program on new establishments in 10 industries. Zhang (2015) evaluates industry-specific impacts of the Kentucky program on agriculture, manufacturing, trade, public administration, and service activities. Hultquist (2015) focuses on the impacts of the Ohio EZ program on manufacturing and transportation, and uncovers positive impacts on manufacturing growth. Moore (2003) suggests that the California EZ program had positive impacts on small firms with less than 9 employees, or large firms with 500–999 employees. Bondonio and Greenbaum (2000) differentiate EZ impacts based on the business cycle (new, existing, and vanishing establishments), and conclude for a stronger impact on new establishments than on existing and vanishing ones.
If EZ programs promote economic growth in the long run, they are expected to increase population, household income, residence employment, and housing values, and to reduce poverty level and housing vacancy. Demographic (D1–D6) and housing market indicators are another set of commonly used indicators.
Popular spatial unit in urban studies are political boundaries (e.g. county, municipality), census designated areas (e.g. block groups, census tracts, traffic analysis zones), and zip codes. Most states designate EZs by evaluating distress level based on census tract information (Appendix 1). Census tracts, block groups, and traffic analysis zones are often consistent with EZ boundaries within a city. In contrast, political boundaries and zip codes are not very good choices for EZ research. Boarnet and Bogart (1996) point out that a New Jersey EZ typically occupies about 30% of the municipality area. Zhang (2015) presents a map comparing traffic analysis zone, zip code boundaries, and the EZ coverage in Louisville, and recommends traffic analysis zones over zip codes.
EZ empirical studies are available at various geographical levels. At the individual level, Bostic and Prohofsky (2006) study the impacts of the California EZ program on individual workers’ incomes; Billings (2009) and Lynch and Zax (2011) explore employment growth of individual establishments; and Landers (2006) explores changing values of commercial and industrial properties. All these studies have drawn positive conclusions about EZ programs. Census tract and zip code are commonly used spatial units, partly because of the availability of demographic data at the tract level and economic data at the zip code level. It is unclear how the selection of a spatial unit influences evaluation outcomes. For instance, Elvery (2009), O’Keefe (2004), and Ham et al. (2011) all study the California EZ program at the census tracts level. Ham et al. (2011) find that the program increased household income and decreased unemployment, but the other two studies do not provide any positive assessments. Similarly, Greenbaum and Engberg (2000) and Moore (2003) study the California program at the zip code level and draw different conclusions.
Time span and experiment design. It is unknown how long it will take for an EZ program to demonstrate itself. Time is needed to set up the program, and once it is set up, it may take much more time for businesses to become aware of its benefits, and to act accordingly. Individual actions will stimulate further growth, but it is unclear when observable improvements can emerge in EZ areas. Further, time span matters for experiment design for evaluating a policy, which often involves pre- and post-program comparison. Generally, the availability of pre- and post-program data can justify a pre- and post-program comparison. Figure 3 presents the information on research time span, in relation to program creation time. The average time spans for state EZ and federal EPZ research are 8.8 and 10.2 years, respectively. The time span of most studies does cover program creation times. However, it is difficult to differentiate pre- and post-program dynamics, because of the lack of data between decennial census years. Only a few studies consider experiment design. Dabney’s (1991) study differentiates 1980–1982 and 1982–1984 as pre- and post-program periods. Zhang (2015) assumes that 1980–1990 and 1990–2000 are pre- and post-program periods for the Louisville (Kentucky) EZ program.
Study time, program time, and research conclusions. EZ: Enterprise Zone.
As for the relationship between study time span and EZ effectiveness assessments, many existing studies evaluate 1980–1990 or 1990–2000 dynamics, simply because of the availability of the Decennial Census in 1980, 1990, and 2000. Almost all the studies focusing on 1980–1990 (41 out of 44) indicate negative or null effects. About half of the studies over 1990–2000 (15 out of 33) draw positive conclusions about EZ program effectiveness. These results support the previous discussion about that an EZ program needs time to demonstrate itself. Another possible reason could be that policymakers are learning from the failures of earlier EZ experiences.
Data. Table 2 presents the major data sources used in the existing literature. The Bureau of the Census is the primary provider. The Standard Statistical Establishment List database provides establishment-level statistics of economic activities annually. The Longitudinal Research Database provides longitudinally linked data for all establishments in the Standard Statistical Establishment List. County Business Patterns (CBPs) provide annual information on employment, numbers of establishments, and payroll at the county or zip-code level. The Bureau of Labor Statistics provides quarterly Census of Employment and Wage data. These data are available for the public at the county level. Some researchers were able to obtain establishment-level information from the Bureau of Labor Statistics (Billings, 2009; Lynch and Zax, 2011). Additional survey data collected by governments (e.g. HUD), individuals, or private companies (e.g. Dun and Bradstreet, Inc.; Walls and Associates) have often a sharp focus on local economic activities at small geographic scales. Economic data enable the creation of economic indicators.
Datasets used in EZ studies.
EZ: Enterprise Zone.
It can be challenging to obtain satisfactory economic data for an EZ study. CBPs and Census of Employment and Wage are available to the public at the county and/or zip-code levels. These boundaries are not consistent with and are generally larger than EZ areas. Data users have to apportion these data to the EZ area, and this process reduces data reliability. Establishment-level data can be aggregated within any EZ boundaries, but these data are of restricted use. Researchers need to first apply for access to these data, and then have to conduct their analyses in one of the Federal Statistical Research Data Centers.
Decennial Population Censuses provide residence-oriented socio-demographic information and have been commonly used in EZ research (Elvery, 2009; Engberg and Greenbaum, 1999; Ham et al., 2011). The Panel Study of Income Dynamics dataset is produced by the University of Michigan and focuses on residential activities, rather than on economic activities. The Census Transportation Planning Package and the Longitudinal Employer-Household Dynamics data have both employment and residence focuses. In addition, state and local administrative data may have mixed focuses. These demographic data are often available at small geographic scales, e.g. blocks, block groups, and census tracks, and can be easily regrouped to fit an EZ boundary.
Control of endogeneity. It is implausible to design a natural experiment to identify whether an EZ policy is the local cause for growth and to quantify its effects (Bartick, 2002). With EZ programs targeted at distress areas, any random or uncontrolled economic comparison between EZ and non-EZ areas could be naturally biased toward finding negative effects. On the other hand, when neighboring non-EZ areas are used for comparison, the EZ impacts may be overestimated. Because of the spill-over effects of jobs moving from neighboring non-EZ to EZ areas, as confirmed by Hanson and Rohlin (2013), the decline of neighboring non-EZ areas is endogenous to the growth in EZ areas. In order to reduce these biases, it is necessary to control for unobservable factors and the endogeneity issue related to EZ designation. Popular methods include the difference-in-difference (DID) approach, the instrumental variable approach, and constructing quality comparison groups.
There are three common groups of outcome measures: activity level, activity growth or growth rate, and the change in activity growth or growth rate. The DID approach assesses the change of growth after and before an EZ program. This approach indeed involves a research design that differentiates pre- and post-treatment periods and enables a researcher to evaluate whether there is a causal relation between the treatment (i.e. the EZ program) and the outcomes. This approach is also expected to eliminate some unobservable time-invariant local effects that could cause estimation bias. Nevertheless, an appropriate use of the DID approach highly depends on the availability of both pre- and post-program data. Unfortunately, with most states adopting their EZ programs in the 1980s, it is difficult to obtain pre-program economic data before the 1980s.
The instrumental variable approach can be used to deal with a possible endogeneity. An appropriate EZ instrumental variable should be associated with EZ designation, but does not directly cause the outcome measure. Bondonio and Engberg (2000) point out that it is difficult to find such a variable, because EZ zone designations involve a comprehensive survey of local socio-economic and physical conditions that affect both zone status and future growth. Instrumental variables that have been used in EZ studies include political variables (Hanson, 2009; Hanson and Rohlin, 2011b), and the probability for a zone to be designated as an EZ (Boarnet and Bogart, 1996; Zhang, 2015).
Two popular methods are used to select control groups in existing empirical studies. One is to select similar areas based on a direct comparison. Ham et al. (2011) select non-EZ tracts with socio-demographic characteristics similar to EZ tracts. Bostic and Prohofsky (2006) select business owners who are similar to EZ participants in terms of tax return filing status and number of dependents. Moore (2003) uses second-round EPZ-designations to study the impacts of EPZ policy on the first-round EPZs. The second method is using the propensity score matching approach (Billings, 2009; Elvery, 2009; Greenbaum and Engberg, 2000; O’Keefe, 2004). A probability model for a zone to be designated as an EZ is first estimated based on EZ designation criteria (e.g. poverty level, unemployment rate). Based on the estimated probability model, comparison zones are selected, which are non-EZ zones with similar EZ-designation probability as existing EZs.
Research design and EZ effectiveness assessments
To further assess whether and how research design influences EZ program effectiveness assessments, the information in Table 1 was converted into data records. Every gray cell corresponds to one record. For instance, O’Keefe’s (2004) study of the California program generates two records. The first record uses total employment growth (EA1) as the outcome indicator. There are mixed recommendations about program effectiveness, with significant and positive impacts in the first six years after zone creation, but significant and negative impacts over years 7–13. The second record uses total establishment growth (EA4) as the outcome indicator, and concludes that the program did not have significant impacts on establishment growth. Additional information related to research design (e.g. data, spatial unit) is introduced. There are a total of 127 records.
Table 3 provides information for assessing whether there is a relationship between outcome indicator choices and research conclusions. Eighty-one studies use non-economic (demographic and housing) indicators, and 19 (23.4%) find positive or mixed impacts. Twenty-five studies use aggregated economic indicators, and nine (36.0%) find positive or mixed impacts. When disaggregated economic indicators are used, 9 out of 21 (42.9%) studies draw positive conclusions about EZ programs, and 10 (47.6%) draw mixed recommendations.
Outcome indicators versus EZ effectiveness assessments.
EZ: Enterprise Zone.
There are two groups of factors related to research design. One group is about experiment design, including SPAN, POST, PRE, and UNIT, and another about specific statistical methods in regression analysis to deal with the endogeneity issue, including PSCORE, INSTRU, and DD. These factors are defined as follows: REC—Conclusion about the effectiveness assessment of an EZ program. =1, EZ program has negative impacts; =2, EZ program has no impacts; =3, EZ program has mixed impacts; and =4, EZ program has positive impacts; SPAN—The time span of the research; POST—The post-program time span that the research covered; PRE—The pre-program time span that the research covered; UNIT—Spatial unit. =1, for a metropolitan area, municipality, or county; =2, for a tax district; =3, for a zip code; =4, for an EZ boundary; =5, for a census tract; =6, for a census block group or traffic analysis zone; and =7, for an individual observation; PSCORE—Propensity score method dummy. =1, if use the propensity score method; =0, otherwise; INSTRU—Instrument method dummy. =1, if use the instrument method; =0, otherwise; DD—Difference method dummy. =1, if use the difference method; =0, otherwise.
Table 4 presents Spearman’s Rank Correlation Coefficients to explore how research design factors are related to research conclusions (REC). It is very interesting that the time span (SPAN) is not significantly correlated with research conclusions (REC), with p-value = 0.6573. However, POST is significantly and positively correlated with REC, while PRE is significantly and negatively correlated with REC. The positive and significant correlation between POST and REC supports the argument that it takes time for an EZ program to demonstrate itself. A longer post-program time increases the likelihood for research to detect EZ program impacts. The negative correlation between PRE and REC suggests the need to be cautious when assessments involve a pre-program time period. For instance, Lambert and Coomes (2001) assess the effectiveness of the Louisville (KY) EZ program by exploring 1980–1990 industry-specific growth. The majority of the EZ areas were designated after 1986. By having a comparatively long pre-program time, it is very likely that the research captured the impacts from declining neighborhood rather than EZ program impacts.
Spearman’s rank correlation coefficients.
*Correlation significant at 1% significance level; ** at 5% significance level; *** at 10% significance level.
The positive and significant correlation between spatial unit (UNIT) and outcome conclusion (REC) suggests that the smaller the spatial unit, the more likely it is to draw a more positive conclusion about a program. As discussed in the “Research design factors” section, data at geographical level with UNIT ≥ 5 (EZ boundaries, census tracts, block groups, traffic analysis zones, and individual observations) are consistent with EZ boundaries or can be aggregated into the existing EZ boundaries. Geographical precision affects research outcomes. When UNIT ≤ 4, i.e. the spatial unit is larger than an EZ boundary, neighboring non-EZ areas will be counted as EZ areas. There are 66 studies with UNIT ≤ 4, 16 (24.4%) of which draw conclusions of mixed or positive impacts. In comparison, out of 61 studies with UNIT ≥ 5, 31 (50.8%) draw conclusions of mixed or positive impacts.
As for statistical methods for controlling endogeneity, using an instrument does not have a significant effect on outcome conclusions (REC). This is not surprising. It is challenging to find a variable that is related to EZ designation, but not to local growth, because EZs are designated based on past growth data. There is a positive and significant correlation between the use of difference approach (DID) and the outcome conclusion (REC). After the difference approach taking away time invariant factors, it is more likely for a research to detect a meaningful outcome of an EZ program.
There is a negative and significant correlation between PSCORE and REC. There are different ways to explain this. One possibility is that this approach enables the creation of a suitable comparison group, with which a researcher can detect a more significant, but negative, impact from an EZ program. Another possibility could be the opposite. The comparison units selected based on these results may not be truly comparable. The maps of O’Keefe (2004) show that comparison units selected by using the propensity score approach are scattered tracts with demographic and economic characteristics close to EZ tracts. It is unclear whether sites of scatter poverty are good comparison for sites of concentrated poverty to study growth. A third possibility could be related to data limitation. The use of a propensity score approach can limit the choice of a study unit to be larger than census tract, because it is difficult to obtain social-economic data at a small geography before the 1980s. The negative and significant correlation between PSCORE and UNIT suggests that a research using the propensity score approach tends to have large study unit. Further, using a portion of data for propensity score analysis reduces the time span of a program effective analysis, as supported by the negative and significant correlation between PSCORE and SPAN and POST. As discussed earlier, the length of a time span affects a research conclusion.
Conclusions and implications
State EZ programs share a similar structure, using incentives to encourage businesses to move to depressed areas and to hire local people. The comparison between existing and past state EZ programs in the U.S. (Appendix 1) suggests that EZ programs differ greatly from one another. Seventeen state EZ programs, as well as the federal EPZ program, have been evaluated by researchers. There are conflicting program effectiveness conclusions about the same program (Table 1). It is clear that the program itself and state socio-economic conditions do not have determining influences on the conclusion of a given research. This naturally leads to ask why different studies draw conflicting conclusions for the same EZ program, and how research design contributes to this situation. This research has assessed whether and how outcome measure, data choice, spatial unit, time span, and control of endogeneity affect research outcomes.
State EZ programs have a common policy goal to revitalize distressed areas. Many researchers translate this goal into outcome measure of economic growth, such as total job growth, total investment growth, or total income growth (Boarnet and Bogart, 1996; Bostic and Prohofsky, 2006; Lambert and Coomes, 2001). Some attempt to explore program impacts on demographics and housing market (Freedman, 2013; Greenbaum and Engberg, 2000). When the outcome measures are related to aggregate economic conditions or demographic changes, the majority of the studies indicate a null effect. When the outcome measures are disaggregated economic indicators, almost all studies suggest significant impacts, which is either positive or mixed. This is consistent with a state EZ program setup. Many of them have strong industry targets, but not panacea for stimulating every industry. When selecting an outcome measure, one should start with the incentive structure of the program, analyze the expected outcomes, and adopt proper outcome indicators, rather than directly using the policy goal of stimulating growth.
The creation of a set of suitable indicators depends on the availability of right data at a right geographical scale. Economic data are better choices than demographic data, but they are not publically accessible for small geographies (e.g. block groups, census tracts). In the U.S., much economic data for small geographies are restricted-use data available from the Bureau of Census. Any researcher has to go through the Census Bureau proposal process to obtain access. However, it is worth making the effort, because this research clearly suggests that the choice of spatial unit is significantly correlated with research conclusions (Table 4).
The choice of time span is important and a longer time span is preferable. Further, studies should explore the time span in relation to the EZ program initiation time. Many studies in the 1990s and early 2000s explore 1980–1990 dynamics, simply because the 2000 census had yet been released. As the majority of EZ programs were created in the 1980s, the period 1980–1990 is not a good choice for time span. This analysis further suggests the need to be cautious when the study time period involves pre-program time. Without thoughtful experiment design to differentiate pre- and post-program growth, a longer pre-program time (PRE) brings in stronger unobservable impacts that are endogenous to the EZ designation and counteract EZ benefits.
It is important, but challenging, to control for endogenous factors related to the EZ designation. To deal with this endogeneity issue, researchers use such methods as DID, instrumental variable, and propensity score approach (Ham et al., 2011; Rich and Stoker, 2010; Smith, 2015). The correlation analyses (Table 4) suggest that the instrumental approach has no influence on research results. This can be explained by the difficulty of finding an appropriate instrumental variable for EZ designation. The DID and the propensity score approaches both are significantly correlated to research results, but in opposite directions. The DID approach can eliminate impacts from some unobservable factors that are endogenous of EZ designation and affect local growth in the negative direction. Studies using this approach tend to draw a positive conclusion of an EZ program. On the contrary, the propensity score approach has a negative association. There are different possibilities to explain this. It could be that with a better control group, a research tends to find the unbiased effect of an EZ program. It also could be the selected control group may not be suitable and created a bias in the estimation toward the negative direction. A third explanation for this negative correlation could be related to other research factors, such as the size of study unit and the length of research time span. To further explore this, a meta-regression analysis can be a continued study of this research.
The above findings can be useful for researchers and policymakers to evaluate the recently implemented Opportunity Zone program in the U.S., which is similar to state EZ programs. They also can be useful for evaluating other people- and/or place-target programs, such as the CDBG program, brownfield redevelopment, and local revitalization programs. A good research design should select an outcome measure based on the analysis of the program design, identify a proper geographic scale, and allow time for the program to show its impacts.
Footnotes
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.