The Impact of New Jersey's Urban Enterprise Zones on Local Employment: A Synthetic Control Approach

Abstract

The designation of enterprise zones is a place-based policy that seeks to revitalize economically blighted areas. The literature on place-based policies has found mixed results regarding their effects on local payroll employment. This paper examines the causal effects of five of New Jersey's Urban Enterprise Zones (UEZs) on local payroll employment: Bayonne, Gloucester City, New Brunswick, Roselle Borough, and The Wildwoods (Wildwood City, Wildwood Crest, North Wildwood, and West Wildwood). All were designated as UEZs by the state in the 2000s, and none have been previously evaluated in the academic literature. The program offers reduced local sales tax, tax credits for newly hired employees, subsidized unemployment insurance costs, worker training assistance, and tax-free purchases on capital equipment and facilities. A synthetic control approach is used with the industrial composition of local firms and poverty rate as the covariate group and no impact of UEZ status on local employment in the treatment periods of the five areas is found. These results suggest that enterprise zones may not be effective job creators for treated areas, particularly for those zones that were added long after the program's inception.

Keywords

place-based policies employment enterprise zones J48 R11 R12 R58

Place-based policies have been implemented throughout Europe and North America since the end of the Second World War to stimulate economic development in blighted areas. In charting the rise of economic development as a widespread practice and field of study, Deller and Goetz (2009) suggested that a role for government in spurring economic activity evolved out of several perceptions in the immediate postwar period. The memory of the Great Depression raised questions about the stability and resiliency of private markets while recent central planning successes of the U.S. Marshall Plan, as well as in Japan and Soviet Russia, suggested government intervention could aid economic growth. Following the first wave of economic development policy, which emphasized smokestack chasing or luring industry to a targeted locality, a second wave emerged in the 1970s that encouraged homegrown activity.

Rather than chase large companies to relocate, second-wave policies focused on the expansion and retention of small and medium firms that were already located in a blighted area. Enterprise Zones are an example of such a policy in that they rely on tax incentives to local firms for hiring, capital investment, and facility expansion. Hall (1977) noted that tax incentives could be used to encourage employment growth in blighted areas, which served as a blueprint for zones that would be established in the United Kingdom and the United States in the 1980s. Neumark and Simpson (2015) noted that these zones have been designated in one form or another in at least 40 states, as well as at the federal level, since the 1970s. A notable early adopter of the enterprise zone model, New Jersey established its Urban Enterprise Zone program (UEZ) in 1983 to encourage job creation in areas with high unemployment rates. New Jersey's program consisted of a 50% reduction in sales tax, tax-free purchases of capital equipment and facilities, and hiring subsidies for qualifying businesses (New Jersey Department of Community Affairs, 2019).

Most enterprise zones broadly target economic development as the intended outcome, with criteria encompassing, but not limited to, poverty reduction, business formation, unemployment reduction, and employment growth. Several studies of prominent enterprise zones have focused primarily on whether the policies were successful in encouraging employment growth. In a literature review of federal and state enterprise zone programs, Neumark and Simpson (2015) suggested that enterprise zone effectiveness in job creation ranges from nonexistent to limited. Freedman (2013) found a positive but largely insignificant effect (3% to 8% per year) on payroll employment growth for Texas’ enterprise zone employers. Neumark and Kolko (2010) and O’Keefe (2004) found conflicting evidence on the effectiveness of the zones in California at inducing payroll employment growth, with the former finding no effect and the latter finding a 3% payroll employment increase in the first 6 years that programs were operating. However, Busso et al. (2013) found a strong positive payroll employment effect (12% to 21%) in neighborhoods targeted by Federal Empowerment Zones. A recent review of the literature on enterprise zones by Neumark and Young (2019) concluded that state enterprise zones have been mostly ineffective at reducing poverty or improving labor market outcomes in the United States.

A primary challenge that evaluators of enterprise zones face, and even program evaluators in general, is identifying a methodologically sound set of controls to compare to treated areas that received enterprise zone designation (Boarnet, 2001). Few papers have used the synthetic control method (SCM) to evaluate the impact of enterprise zones in the United States despite the method offering a potential solution for handling the “endogenous selection” of zones discussed by Neumark and Young (2019). In other words, if zones are selected based on prior changes in labor market outcomes, such areas will tend to experience a negative trend in employment immediately before treatment, also known as Ashenfelter's Dip. This poses a major source of bias for traditional program evaluation methods such as difference-in-differences, which were primarily employed in earlier enterprise zone studies, due to violation of the parallel trends assumption in the pretreatment period. The ability of the SCM to match treated areas to a counterfactual with a similar trend prior to treatment may improve on earlier studies by addressing the endogenous selection issue for evaluating enterprise zone outcomes.

In this paper, I assess the impact of the UEZ model on payroll employment for zones that have not been previously evaluated in the academic literature using data from the U.S. Census Bureau. These zones became active in 2002 within Bayonne, Roselle Borough, and The Wildwoods (Wildwood City, Wildwood Crest, North Wildwood, and West Wildwood) and in 2004 within Gloucester City and New Brunswick. The methodology used is a SCM approach that develops a counterfactual for each of the five zones based on the industrial composition and poverty rates of the respective towns. SCM addresses the credible control dilemma brought up by Boarnet (2001) and the endogenous selection issue of traditional approaches (Neumark & Young, 2019). I investigate the effect of enterprise zones at the ZIP code level, which is a smaller geographic unit than has typically been used in the literature. The results largely conform with the literature in suggesting that the UEZs had no impact on payroll employment growth in the five zones studied.

New Jersey's Urban Enterprise Zone Program

New Jersey's UEZ program was established in 1983 with the first five municipalities (Bridgeton, Camden, Newark, Trenton, and Plainfield) joining the program in 1986 and the most recent municipalities (New Brunswick and Gloucester City) joining the program in 2004. All zones listed in Table 1 are set to expire at the end of 2023 at the earliest (New Jersey Department of Community Affairs, 2022). Figure 1 maps the state's Urban Enterprise Zones along with their encompassing ZIP codes.

Figure 1.

New Jersey's urban enterprise zones with encompassing ZIP code communities.

Table 1.

New Jersey Urban Enterprise Zones by Effective Year.

Municipality	County	Year
Bridgeton	Cumberland	1986
Camden	Camden	1986
Newark	Essex	1986
Trenton	Mercer	1986
Plainfield	Union	1986
Millville	Cumberland	1988
Vineland	Cumberland	1988
Elizabeth	Union	1992
Jersey City	Hudson	1992
Kearny	Hudson	1992
Orange	Essex	1992
Asbury Park	Monmouth	1994
Lakewood	Ocean	1994
Long Branch	Monmouth	1994
Passaic	Passaic	1994
Paterson	Passaic	1994
Perth Amboy	Middlesex	1994
Phillipsburg	Warren	1994
Carteret	Middlesex	1995
Mount Holly	Burlington	1995
Pleasantville	Atlantic	1995
Union City	Hudson	1995
East Orange	Essex	1996
Guttenberg	Hudson	1996
Hillside	Union	1996
Irvington	Essex	1996
Pemberton	Burlington	1996
North Bergen	Hudson	1996
West New York	Hudson	1996
Bayonne	Hudson	2002
Roselle Borough	Union	2002
The Wildwoods	Cape May	2002
Gloucester City	Camden	2004
New Brunswick	Middlesex	2004

The goal of the UEZ program is to “stimulate growth by encouraging businesses to develop and create private sector jobs through public and private investment” (New Jersey Department of Community Affairs, 2022). The three major benefits of the program are a reduced sales tax (half the state rate) as well as tax-free purchases on capital equipment, facility expansions, and upgrades. Certified businesses are also eligible for funding grants through the New Jersey Economic Development Authority. Further, participating firms may receive assistance from the New Jersey Department of Labor through its One Stop Centers for hiring, training, and retraining existing or new employees.

To receive benefits from the UEZ program, a firm must become a Certified UEZ Business. This process requires registering with the state, locating within 1 of the 32 designated zones, and being in tax compliance with the state. New Jersey Economic Development Authority (2011) estimated the number of certified businesses in 2011 at 6,639 out of 33,730 eligible ones across the state's 32 zones, a participation rate of 19.7%. Table 2 shows the participation rate for the five zones examined in this study, which at 22.1% is slightly higher than the state's rate. A survey of participating UEZ firms conducted by the New Jersey Economic Development Authority in 2010 indicated varying degrees of participation across the UEZ programs for which they were eligible. Across 1,003 surveyed firms, 59% indicated participating in the sales tax reduction benefit, 7% participated in the employee tax credit program, and 3% participated in the worker training benefit.

Table 2.

Urban Enterprise Zone Business Participation Rate in 2011.

Zone	Certified businesses	Eligible businesses	Participation rate (%)
The Wildwoods	255	549	46.4
Bayonne	229	986	23.2
Gloucester City	36	214	16.8
Roselle	41	274	15.0
New Brunswick	117	1,046	11.2
Total	678	3,069	22.1

Source: New Jersey Economic Development Authority (2011)

Another feature of the UEZ program was Zone Assistance Funds (ZAFs), which were flexible revenue sources that communities could use for economic development activities. ZAFs were funded by the sales tax generated by UEZ-certified businesses. They were used by participating municipalities to remediate properties, build infrastructure, and support economic development project-gap funding (New Jersey Department of Community Affairs, 2019). However, ZAFs were discontinued in 2011 after Governor Chris Christie suspended payments to the zones in favor of balancing the state's budget, which had come under pressure following the Great Recession (O’Dea, 2011).

The entirety of the academic literature on New Jersey's UEZs focused on the zones that became active in the 1980s. Boarnet and Bogart (1996) studied the impact of the UEZ designation on the first generation of targeted municipalities with data from 1980 to 1990 and found that the zones had no discernible impact on economic development, specifically payroll employment and municipal property values. Greenbaum and Engberg (2004) studied the impact of New Jersey's zones from the 1980s (in addition to those of California, Florida, New York, Pennsylvania, and Virginia) and found no effect on overall employment growth when matched to similarly distressed and economically similar areas. While Rubin (1990) found a 5% increase in employment over the first 2 years of a zone's active status, he did not use a control group in his analysis.

This paper focuses on the three zones that became active in 2002 within Bayonne, Roselle Borough, and The Wildwoods¹ and the two zones that became active in 2004 within Gloucester City and New Brunswick, all of which are labeled in Figure 1. Bayonne is a city of 72,000 east of Newark, over Newark Bay, and south of Jersey City. It is home to Port Jersey, an intermodal freight transport facility. Roselle is a borough of 21,000 west of Elizabeth that is famous for being the first town in the world to be lit with electric overhead wires by Thomas Edison in 1883. The Wildwoods are a group of “Jersey Shore” seaside resort communities north of Cape May with a collective year-round population of 13,000. Gloucester City is a city of 11,000 east of the Delaware River of Philadelphia and south of Camden. New Brunswick is a city of 55,000 located along the Raritan River and is home to Rutgers University.

Data

I use data from the U.S. Census Bureau to construct all variables at the ZIP code level. For the employment and industry share variables, I use data from ZIP Code Business Patterns (ZBP), which provide annual statistics for businesses with paid employees within the United States at the ZIP code level. ZBP are calculated using data from the Standard Statistical Establishment List, a business register of all known single and multi-establishment companies, as well as several other economic surveys (e.g., Annual Company Organization Survey, Annual Survey of Manufacturers, Current Business Surveys) and government administrative records (e.g., Internal Revenue Service, Social Security Administration, Bureau of Labor Statistics).

To construct the employment series, I use the ZBP series’ total number of payroll employees for the pay period including March 12. The raw employment series from ZBP are volatile on a year-to-year basis, which could be the result of non-sampling errors² from the various surveys utilized and the Census’ use of noise infusion methodology beginning in 2007. To remove noise and isolate medium-term trends from short-term noise, I apply a 3-year moving average from 1996 to 2012. I use raw data beginning in 1994 to construct the moving average so that the value of the dependent variable in 1996 is equivalent to

\frac{1}{3} \sum_{i = 1994}^{1996} Employment,

(1)

To construct the industry share³ variables I use the ZBP series’ total number of establishments and the number of establishments by North American Industry Classification System (NAICS) at the two-digit level in 2000. I identify the number of firms at the ZIP code level for eight broad industries using the industry groupings in Table 3 and calculate their shares out of the total number of establishments.

Table 3.

Firm Industry Classification.

Industry group	NAICS codes
Construction	23
Manufacturing	31, 32, 33
Trade, transportation, and utilities	22, 42, 44, 45, 48, 49
Information	51
Finance, insurance, and real estate	52, 53
Professional and business services	54, 55, 56
Education and health services	61, 62
Leisure and hospitality	71, 72

For the poverty share variable, I use data from the U.S. Census Bureau 2000 Decennial Summary File 3, Table P090. To construct the poverty share variable, I simply use the number of families living below the poverty line divided by the number of families in the ZIP code.

The final ZIP code level data set consists of employment series for 1996 to 2012, the share of firms in eight industry groupings in 2000, and the share of families in poverty in 2000 at the ZIP code level. Summary statistics for the variables can be found in Table 4.

Table 4.

Pretreatment Summary Statistics.

	Bayonne	Roselle	The Wildwoods	Gloucester City	New Brunswick	NJ
Employment	13,456	4,773	3,591	3,842	22,523	4,585
Poverty rate (%)	8.4	5.8	11.3	7.5	16.9	5
Construction (%)	10.6	11.1	6.6	10.3	3.6	12.5
Manufacturing (%)	3.4	14.8	1.5	9.6	7.2	4.9
Trade, transp., and util. (%)	29.6	26.5	17.6	31.8	21.4	23.6
Information (%)	0.1	0.1	0.1	1.5	1.5	1.7
Finan., insur., and real est. (%)	7.6	6.2	8.4	5	6.6	8
Prof. and bus. serv. (%)	10.4	8.3	6.1	9.2	19.6	17.4
Educ. and health serv. (%)	15.2	8.3	3.1	5.4	14.2	9.3
Leisure and hospitality (%)	9.6	6.5	46.9	14.6	12.9	9

Notes: Employment is mean from 1994 to 2001. All other variables indicate shares in 2000. NJ indicates mean statistics across all New Jersey ZIP codes.

Methods

I use the SCM to empirically evaluate the impact of UEZ designation on the local payroll employment levels of the five treated areas. The treated areas are the ZIP codes that encompass the five selected UEZs are labeled and referred to as Urban Enterprise Zone Communities in Figure 1. Ferman and Pinto (2019) suggested that standard econometric methods (e.g., difference-in-differences) are unlikely to detect program effects or produce reliable hypothesis tests in settings where there are few treated areas and recommended SCM as an alternative estimator for such cases. There are several examples of the SCM used to empirically evaluate the effect of a policy on employment (Castillo et al., 2017; Munasib & Rickman, 2015). Bundrick and Yuan (2019) used SCM to evaluate the impact of an Arkansas targeted business subsidy program on per capita income and poverty at the county level. Additionally, Chaurey (2017) and Gobillon and Magnac (2016) used SCM to evaluate place-based policies in India and France, respectively.

This study uses the SCM at a much smaller geographic scale than most of the literature. It is much more typical for synthetic control studies to apply the method to policy interventions implemented at an aggregate level affecting a small number of large units (countries, regions, or states). While recent studies (Acemoglu et al., 2016; Kreif et al., 2015) have extended the SCM to settings with a large number of units, Abadie (2021) warned that a large number of units in the donor pool may introduce bias to the estimator, so each of the units in the donor pool must be chosen judiciously. However, Ferman (2019) suggested that a large number of units in the donor pool may be beneficial in high dimensional settings such as ours, and that the SCM estimator becomes asymptotically unbiased as the number of pretreatment periods and donor pool units increases.

The SCM generates a synthetic version of the treatment area's variable of interest based on weights of untreated donor areas to be used as a counterfactual in evaluating a policy's effects. In our case, the synthetic version of a treated ZIP code's payroll employment will be constructed from a weighted average of donor ZIP codes from New Jersey. However, the donor pools will exclude ZIP codes that hold UEZ status (see Table 1). These donor ZIP codes will be selected to match the employment levels, poverty rates, and industrial composition of firms in 2000 (based on eight categories specified in Table 3) in each respective treated ZIP code before UEZ treatment occurred. Therefore, five separate synthetic control models will be estimated to generate synthetic controls for Bayonne, Gloucester City, New Brunswick, Roselle Borough, and The Wildwoods, respectively.

I follow the synthetic control methodology from Abadie et al. (2010), which I briefly outline below. $Y_{i t}^{N}$ represents the payroll employment level that would be observed for ZIP code i at time t in the absence of the enterprise zone treatment for ZIP codes i = 1, …, J + 1 ZIP codes and time periods t = 1, …, T, where J represents the number of untreated “donor” ZIP codes.

Let T₀ be the number of pre-treatment periods, with 1 <= T₀ < T . $Y_{i t}^{I}$ represents the employment that would be observed for ZIP code i at time t for the ZIP code exposed to the enterprise zone intervention in period T₀ + 1. I assume that the enterprise zone treatment has no effect on payroll employment before the implementation period, so for t ε (1, …, T₀) and all i ε (1, …, J + 1), I have that $Y_{i t}^{I}$ = $Y_{i t}^{N}$ .

Let α_it = $Y_{i t}^{I}$ − $Y_{i t}^{N}$ be the effect of the enterprise zone intervention for ZIP code i at time t, and let D_it be an indicator that takes value one if ZIP code i is exposed to the treatment at time t, and value zero otherwise. Therefore, the observed outcome for unit i at time t is

Y_{i t} = Y_{i t}^{N} + α_{i t} D_{i t}

(2)

Only the first ZIP code (ZIP code “one”) is exposed to the enterprise zone treatment after period T₀ so I aim to estimate (α₁_T_0 + 1, …, α_1T). For t > T₀,

α_{1 t} = Y_{1 t}^{I} - Y_{1 t}^{N} = Y_{1 t} - Y_{1 t}^{N}

(3)

Since

Y_{1 t}^{I}

is observed, in order to estimate α_1t I need only

Y_{1 t}^{N}

, which is the synthetic control, or counterfactual outcome.

The synthetic control estimator will estimate $Y_{1 t}^{N}$ using a linear combination of donor ZIP codes i ε (2, …, J + 1) using weights w = (w₂, …, w_J_+ 1) that solve a constrained optimization problem that matches the treated ZIP code on both pretreatment employment levels and pretreatment firm industrial composition and poverty rates (see Abadie et al., 2010 for a more detailed explanation of the estimation procedure). The weights are nonnegative and sum to one. Therefore, the synthetic control of the treated ZIP code is:

Y_{i t}^{N} = \sum_{i = 2}^{J + 1} w_{i}^{*} Y_{i, t}

(4)

where

w_{i}^{*}

are the optimally chosen weights. Ideally, the synthetic control should match a treated ZIP code on pretreatment employment levels and on pretreatment covariates. The synthetic control represents what the treated ZIP code would have experienced without the enterprise zone treatment. T₀ + 1 is 2002 for three models (Bayonne, Roselle, The Wildwoods) and 2004 for two models (Gloucester City, New Brunswick). The pretreatment payroll employment series begin in 1996 for all five models. The treatment horizons end in 2010 (Bayonne, Roselle, The Wildwoods) and 2012 (Gloucester City, New Brunswick) so each model has an 8-year treatment horizon.

To estimate the significance of the α_1t effects for the five models, I use a permutation method that compares the synthetic control estimates to a distribution of placebo estimates. This results in the estimation of the same synthetic control procedure in each model for the J donor ZIP codes. I provide standardized p-values for each of the years in the treatment horizon. Additionally, I provide an overall p-value that measures the proportion of placebos that have a ratio of posttreatment Root Mean Square Percentage Error (RMSPE) over pretreatment RMSPE at least as large as the ratio for the treated ZIP code. Please see Cunningham (2021) or Galiani and Quistorff (2017) for a more detailed discussion of the placebo methodology.

I use the local poverty rate as a covariate to ensure that the synthetic control matches the same level of economic blight as the treated areas. According to the original legislation that established the zones, a municipality must meet several criteria to qualify, such as high poverty, high unemployment, and high dependence on public assistance (New Jersey Department of Community Affairs, 2019). The five treated areas in this study placed in the 76th (Roselle), 82nd (Gloucester City), 85th (Bayonne), 88th (The Wildwoods), and 94th (New Brunswick) percentiles, respectively, for families in poverty across New Jersey ZIP codes in 2000.

The use of firm industry composition as the set of covariates for the selection of donor areas is based on how local payroll employment evolves over the business cycle (Rissman, 1999). The rationale for selecting donor areas based on industrial structure is that a good control area should experience roughly the same cyclical sensitivity to the national business cycle, or regional business cycle, as the treated area over the treatment horizon. Domazlicky (1980) identified industrial structure and trade relations as the primary drivers of differences in regional cyclical amplitudes. Industrial structure refers to the composition of output produced by an area. Trade relations refers to the buyers of the goods and services that a region produces, whether regional residents or from another area. Therefore, employment growth fluctuates largely with the nature of the business cycle in that national booms and busts have disparate effects on local economies depending on their industry mix. For example, during the COVID-19 pandemic areas that had a large share of workers in leisure and hospitality suffered the highest unemployment rates in the nation while manufacturing job loss was relatively intense between 2001 and the Great Recession, disproportionately affecting Rust Belt cities along the Great Lakes (Alder et al., 2014; Muro et al., 2020). Additionally, evidence from firm surveys in 2010 and 2019 suggest that firms from certain industries (retail, manufacturing) are more likely to participate in the enterprise zone program than others (professional and business services, construction; New Jersey Department of Community Affairs, 2019; New Jersey Economic Development Authority, 2011).

Results

Validity of Synthetic Controls

Table 5 compares the balance of the pretreatment employment levels and covariates between the treated area and the synthetic control for each of the five respective models. The synthetic controls are very similar to treated ZIP codes in terms of pretreatment employment levels. However, the synthetic control models provide higher shares of Professional and Business Service firms than their treated areas in four out of the five models (all except Gloucester City). Out of the five models, the SCM for New Brunswick appears to least resemble its treated area, while the model for Gloucester City most resembles its treated area in terms of pretreatment characteristics. Table 6 reports the donor ZIP codes that are assigned nonzero weight values in the estimation procedure.

Table 5.

Predictor Balance and Model Fit.

Predictor variables	Bayonne		Roselle Borough		The Wildwoods		Gloucester City		New Brunswick
Predictor variables	Treated	Synthetic	Treated	Synthetic	Treated	Synthetic	Treated	Synthetic	Treated	Synthetic
Employment (1996)	13,900	13,885	4,819	4,817	3,336	3,362	3,946	3,942	23,775	21,809
Employment (1997)	13,755	13,741	4,784	4,759	3,394	3,423	3,929	3,925	19,032	20,866
Employment (1998)	13,512	13,495	4,782	4,810	3,585	3,614	3,888	3,884	19,725	20,546
Employment (1999)	13,256	13,239	5,088	5,051	3,673	3,704	3,771	3,767	20,442	20,810
Employment (2000)	13,124	13,108	4,846	4,845	3,811	3,839	3,733	3,729	22,195	21,726
Employment (2001)	13,036	13,020	4,585	4,586	3,760	3,790	3,744	3,739	23,119	22,896
Employment (2002)	-	-	-	-	-	-	3,707	3,701	23,586	23,987
Employment (2003)	-	-	-	-	-	-	3,570	3,564	24,976	24,791
Poverty Rate (2000)	.084	.054	.058	.041	.113	.115	.075	.073	.169	.025
Construction (2000)	.106	.093	.111	.134	.066	.063	.103	.103	.036	.078
Manufacturing (2000)	.034	.053	.148	.099	.015	.029	.096	.096	.072	.037
Trade, transportation, and utilities (2000)	.296	.281	.265	.262	.176	.18	.318	.318	.214	.278
Information (2000)	.009	.013	.006	.011	.006	.009	.015	.015	.015	.022
Financial, insurance, and real estate (2000)	.076	.073	.062	.097	.084	.069	.05	.05	.066	.081
Professional and business services (2000)	.104	.166	.083	.155	.061	.097	.092	.092	.196	.265
Education and health services (2000)	.152	.137	.083	.054	.031	.042	.054	.053	.142	.079
Leisure and hospitality (2000)	.096	.077	.065	.035	.469	.412	.146	.145	.129	.070
Model fit
Pretreatment RMSPE	88.4		27.5		49.2		34.8		978
Posttreatment/Pretreatment RMSPE	4.86		8.49		11.2		6.28		4.0
p-value	.813		.612		.427		.718		.874

Table 6.

Synthetic Control Weights.

Bayonne		Roselle Borough		The Wildwoods		Gloucester City		New Brunswick
ZIP code	Weight	ZIP code	Weight	ZIP code	Weight	ZIP code	Weight	ZIP code	Weight
07631 - Englewood	.191	08067 - Pedricktown	.523	08751 - Seaside Heights	.715	08316 - Dorchester	.263	08817 - Edison	.338
07006 - Caldwell	.187	08033 - Haddonfield	.14	08016 - Burlington	.074	07970 - Mount Freedom	.107	08807 - Bridgewater	.214
07657 - Ridgefield	.133	08512 - Cranbury	.139	07041 - Millburn	.068	07029 - Harrison	.105	08002 - Cherry Hill	.183
08084 - Stratford	.133	07662 - Rochelle Park	.124	08343 – Monroeville	.037	07608 - Teterboro	.102	08534 - Pennington	.137
08401 - Atlantic City	.086	08084 - Stratford	.053	08002 - Cherry Hill	.03	07657 - Ridgefield	.091	08854 - Piscataway	.104
08089 - Waterford Works	.069	07058 - Pine Brook	.018	07058 - Pine Brook	.021	08751 - Seaside Heights	.078	07470 - Wayne	.024
08648 - Lawrence	.064	08520 - Hightstown	.002	07662 - Rochelle Park	.021	07857 - Netcong	.04
07041 - Millburn	.049			08837 - Edison	.019	07863 - Oxford	.037
08832 - Keasbey	.047			08402 - Margate City	.011	07660 - Ridgefield Park	.033
07109 - Belleville	.041			08817 - Edison	.006	08873 - Somerset	.032
						07734 - Keansburg	.025
						08078 - Runnemede	.023
						08809 - Clinton	.022
						08015 - Browns Mills	.018
						07662 - Rochelle Park	.009
						07960 - Morristown	.006
						08349 - Port Norris	.006
						08402 - Margate City	.001

UEZ Impact on Employment

Of the five models examined in this study, only Bayonne outperformed its synthetic control over the treatment horizon. However, the overall standardized p-values reported in Table 5 do not indicate a significant difference between the employment trajectories of the five treated areas and their respective synthetic controls.

Figure 2 plots the estimated synthetic control for Bayonne compared with its actual employment levels before and after enterprise zone treatment (vertical line indicates treatment year) in the left panel. The right panel plots year-specific p-values from the placebo test. The results suggest that Bayonne's payroll employment outperformed its synthetic control over most of the treatment horizon, with its most significant employment impact occurring 1 year after treatment. However, by the final year the treatment series matched the synthetic control. Figure 3 indicates that Roselle's payroll employment performed just below its synthetic control in the treatment horizon between 2002 and 2010. Figure 4 suggests that The Wildwoods’ payroll employment performed far below its synthetic control over the treatment horizon, with a major drop in employment occurring 2 years after enterprise zone designation. Figure 5 indicates that Gloucester City's payroll employment underperformed its synthetic control over most of the treatment horizon by about 200 to 300 jobs. Figure 6 shows that New Brunswick's payroll employment underperformed its synthetic control over most of the treatment horizon, but it suffered a much more substantial loss than its control from the Great Recession. However, by the end of the treatment horizon the city was only down 1,000 jobs from its synthetic control.

Figure 2.

Synthetic control results: Bayonne.

Figure 3.

Synthetic control results: Roselle Borough.

Figure 4.

Synthetic control results: The Wildwoods.

Figure 5.

Synthetic control results: Gloucester City.

Figure 6.

Synthetic control results: New Brunswick.

Firm participation in the program across the five zones might play a role in explaining their divergent paths of employment over the treatment period. Bayonne firms participated in the enterprise zone program at an above average rate in 2011 (23.2%) and was the only area that experienced better employment outcomes than its synthetic control. However, The Wildwoods had by the far the highest participation rate among zones in 2011 (46.4%), despite having the worst employment growth in the treatment period relative to its synthetic control (see Table 2.) It is possible that the proximity of other UEZs could be playing a role in enhancing area zone employment effects through spillover effects. Gloucester City is directly south of Camden (UEZ area), Bayonne is directly south of Jersey City (UEZ area), and Roselle is southwest of Elizabeth (UEZ area). These three areas performed comparatively better than the two study zones, which are not adjacent to another UEZ community (New Brunswick and The Wildwoods).

Robustness to Rescaling Dependent Variable

In this section, I estimate the five SCMs using a scaled version of the dependent variable. Abadie (2021) suggested that using synthetic controls with weights that sum to one may be valid only if the variables in the data are rescaled to correct for differences in the size between units. Since there are likely differences in employment levels between the treated and donor ZIP codes in the main results, it is worth investigating whether the results are sensitive to rescaling employment. Therefore, I index employment⁴ Y_i so that it equals 100 in year T₀, the year before treatment, for each ZIP code i:

Y_{t} = \frac{Y_{t}}{Y_{T_{0}}} * 100

(5)

Table 7 compares the balance of the pretreatment employment levels and covariates between the treated area and the synthetic control for each of the five respective models and Table 8 contains the nonzero area weights used in the models. Note that T₀ is equivalent to 2001 for Bayonne, Roselle, and The Wildwoods and 2003 for Gloucester City and New Brunswick. The match for pretreatment indexed employment across all models is stronger than the match for employment levels in the previous section. Additionally, the match across all covariates is much better for New Brunswick with the indexed employment variable. However, the model for The Wildwoods has a considerably worse match for the leisure and hospitality industry share.

Table 7.

Robustness Check: Predictor Balance and Model Fit.

Predictor variables	Bayonne		Roselle Borough		The Wildwoods		Gloucester City		New Brunswick
Predictor variables	Treated	Synthetic	Treated	Synthetic	Treated	Synthetic	Treated	Synthetic	Treated	Synthetic
Employment (1996)	106.6	106.9	105.1	105.0	88.7	88.8	110.5	110.4	95.2	95.2
Employment (1997)	105.5	105.7	104.3	104.3	90.3	90.3	110.1	109.9	76.2	76.3
Employment (1998)	103.6	103.8	104.3	104.3	95.4	95.4	108.9	108.8	79.0	78.9
Employment (1999)	101.7	101.8	111.0	110.9	97.7	97.7	105.6	105.5	81.8	82.1
Employment (2000)	100.7	100.8	105.7	105.7	101.3	101.3	104.6	104.5	88.9	88.9
Employment (2001)	-	-	-	-	-	-	104.9	104.8	92.6	92.6
Employment (2002)	-	-	-	-	-	-	103.8	103.7	94.4	94.6
Poverty Rate (2000)	.084	.084	.058	.050	.113	.095	.075	.075	.169	.083
Construction (2000)	.106	.106	.111	.112	.066	.076	.103	.103	.036	.125
Manufacturing (2000)	.034	.039	.148	.060	.015	.048	.096	.096	.072	.044
Trade, transportation, and utilities (2000)	.296	.278	.265	.302	.176	.255	.318	.318	.214	.233
Information (2000)	.009	.011	.006	.003	.006	.008	.015	.015	.015	.016
Financial, insurance, and real estate (2000)	.076	.070	.062	.062	.084	.069	.05	.05	.066	.064
Professional and business services (2000)	.104	.105	.083	.126	.061	.087	.092	.092	.196	.168
Education and health services (2000)	.152	.146	.083	.084	.031	.056	.054	.054	.142	.111
Leisure and hospitality (2000)	.096	.10	.065	.108	.469	.230	.146	.145	.129	.107
Model fit
Pretreatment RMSPE	1.49		2.41		2.23		0.246		0.155
Posttreatment/Pretreatment RMSPE	6.06		14.3		11.2		78.2		53.9
p-value	.70		.49		.57		.13		.18

Table 8.

Robustness Check: Synthetic Control Weights.

Bayonne		Roselle Borough		The Wildwoods		Gloucester City		New Brunswick
ZIP code	Weight	ZIP code	Weight	ZIP code	Weight	ZIP code	Weight	ZIP code	Weight
08046 - Willingboro	.302	08880 - South Bound Brook	.384	08751 - Seaside Heights	.286	07857 - Netcong	.19	07822 - Augusta	.203
08242 - Rio Grande	.219	08048 - Lumberton	.245	08349 - Port Norris	.227	08242 - Rio Grande	.134	08326 - Landisville	.176
08078 - Runnemede	.162	08514 - Cream Ridge	.117	08066 - Paulsboro	.155	08751 - Seaside Heights	.107	08066 - Paulsboro	.134
08010 - Beverly	.097	08401 - Atlantic City	.091	07440 - Pequannock	.123	07608 - Teterboro	.097	08402 - Margate City	.127
08015 - Browns Mills	.055	08067 - Pedricktown	.072	08752 - Seaside Park	.083	08066 - Paulsboro	.077	07970 - Mount Freedom	.124
08401 - Atlantic City	.049	08010 - Beverly	.037	07865 - Port Murray	.064	07863 - Oxford	.055	08048 - Lumberton	.095
07631 - Englewood	.033	07608 - Teterboro	.035	08048 - Lumberton	.041	07460 - Stockholm	.051	08535 - Perrineville	.066
08316 - Dorchester	.031	08402 - Margate City	.019	08343 - Monroeville	.017	07029 - Harrison	.051	07028 - Glen Ridge	.042
08069 - Penns Grove	.027			08402 - Margate City	.004	08014 - Bridgeport	.048	08755 - Toms River	.027
07462 - Vernon	.013					07420 - Haskell	.039	08316 - Dorchester	.007
08402 - Margate City	.013					08316 - Dorchester	.023
						08349 - Port Norris	.013
						07109 - Belleville	.005
						08343 - Monroeville	.001

Several of the models are highly sensitive to the indexed employment series. Using that series, I find that Bayonne performs substantially worse than its synthetic control over most of the treatment horizon when it had previously performed better (see Figure 7). While both Roselle and Gloucester City's employment performs slightly below their respective synthetic controls in the main results, they both perform much worse than their counterfactuals when employment is indexed (see Figures 8 and 9). However, the models for The Wildwoods and New Brunswick were not sensitive to rescaling employment (see Figures 10 and 11). Despite the trajectory of the synthetic controls being altered, Table 7 indicates that none of the differences between treated areas and their synthetic controls were statistically significant.

Figure 7.

Robustness check: Bayonne.

Figure 8.

Robustness check: Roselle Borough.

Figure 9.

Robustness Check: Gloucester City.

Figure 10.

Robustness Check: The Wildwoods.

Figure 11.

Robustness Check: New Brunswick.

Discussion

My examination of New Jersey's UEZs that became active in the 2000s suggests that the place-based policy did not produce a significant impact on local payroll employment in the encompassing ZIP codes of the five zones studied. Despite the paths of the respective synthetic controls for some areas experiencing sensitivity to the scaling of employment, there is little to no evidence of a positive employment effect from enterprise zone designation across the five treated areas. These results are consistent with other studies of the local employment impact of the zones that became active in the 1980s (Boarnet & Bogart, 1996; Greenbaum & Engberg, 2004). However, the paper addresses the credible control issue brought up by Boarnet (2001) by attempting to match a synthetic control that exhibits a similar level of economic distress as well as industrial composition as the treated area. Since the urban economics literature suggests that local payroll employment fluctuates throughout the business cycle based on industry mix, a credible control group should exhibit similar industry mix as treated areas for evaluations of place-based policies on employment. Otherwise, the selection of the treatment horizon might impact the results of the program evaluation due to varying regional business cycles between treatment and control groups. Additionally, firms in certain industries (retail, manufacturing) are more likely to participate and benefit from the enterprise zone program than others.

The evaluation procedure put forth in this paper offers a highly transparent framework to select controls for areas treated with a place-based policy based on observable factors. Ferman and Pinto (2019) suggested that the SCM is a suitable estimation strategy for policy settings with relatively few treated areas. Additionally, the SCM should be better able than traditional program evaluation methods to reduce bias from the endogenous selection of enterprise zones, noted by Neumark and Young (2019), by matching a treated area to a counterfactual with a similar trend prior to treatment. Furthermore, unlike approaches such as difference-in-differences, the control weights are reported in a standardized way that allows researchers to assess with a sniff test whether these control areas are valid. For example, the donor area with the largest weight in the synthetic control for The Wildwoods was Seaside Heights, a comparable Jersey Shore resort town. However, Abadie (2021) cautioned that small policy interventions can be difficult to detect with the SCM, which may be an issue given the size of the incentives and relatively low firm participation rate across the zones. Additionally, Abadie (2021) cautioned that many donor units in the pool can introduce bias into the SCM estimation procedure.

One potentially significant issue is an enterprise zone saturation effect in the state that could be biasing the treatment effect in these five models toward zero. In other words, by the time the five most recent UEZs were introduced in New Jersey, businesses around the state already may have received similar incentives in 27 other zones. Therefore, the relative desirability of these five zones is probably lower for potential firm entrants compared to zones that were established earlier when zones were scarcer. Additionally, there lies the corresponding concern that since these five areas were the last ones chosen, they were less likely to benefit from the UEZ program than zones that were designated earlier in its history. Therefore, the conclusion that one might reach from the lack of treatment effects in these zones isn’t that New Jersey's UEZs do not increase employment, but that there is no marginal employment benefit of adding zones to a saturated program 20 years after its inception. Furthermore, given that the New Jersey Department of Community Affairs (2019) suggested a substantial decline in firm participation rate in some of the state's older zones (e.g., Camden, Elizabeth, Vineland, Bridgeton) in recent years, the program's overall desirability has likely waned. On the other hand, the results from the section on UEZ Impact on Employment suggest that the three areas that performed best (relative to their respective synthetic controls) were adjacent to existing UEZ communities, which may suggest a positive cluster effect from the zones rather than a geographic saturation effect.

The results from the analysis largely conform to the literature on place-based policies that determined that their effects on payroll employment are minimal. There are several potential reasons why enterprise zones have not been particularly effective job creators. Gottlieb (1997) posited that cities and neighborhoods might not be the correct scale for enterprise zones due to commuting patterns. He suggested that if most employed residents of targeted neighborhoods commute outside of their immediate residential area, a county or regional policy might better stimulate job growth for area residents. Additionally, Rubin (1988) suggested that, due to environmental uncertainty, economic development practitioners often tilt the system in favor of the business community. Therefore, policy might not always be optimized in the best interests of the public to create quality jobs. Finally, it is worth noting that beyond job creation, the accompanying goals of the program were to stimulate investment and economic development activity within the zones (New Jersey Economic Development Authority, 2011). Therefore, it is possible that the zones could be affecting other indicators related to the local economy, such as local property values and business formation, which are not examined in this study. Future studies might utilize this empirical framework to study the impact of place-based policies on other local economic indicators.

Footnotes

Acknowledgments

The author would like to thank Paul Gottlieb, Michael Lahr, Alyssa Oshiro, and three anonymous referees for helpful comments on previous drafts of this paper.

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

ORCID iD

Adam Scavette

Notes

Author Biography

Adam Scavette is a regional economist at the Federal Reserve Bank of Richmond and a PhD candidate in public policy at Rutgers University–New Brunswick.

References

Abadie

(2021). Using synthetic controls: Feasibility, data requirements, and methodological aspects. Journal of Economic Literature, 59(2), 391-425. https://doi.org/10.1257/jel.20191450

Abadie

Diamond

Hainmueller

(2010). Synthetic control methods for comparative case studies: Estimating the effect of California’s tobacco control program. Journal of the American Statistical Association.vol, 105(490), 493-505. https://doi.org/10.1198/jasa.2009.ap08746

Acemoglu

Johnson

Kermani

Kwak

Mittone

(2016). The value of connections in turbulent times: Evidence from the United States. Journal of Financial Economics, 121(2), 368-391. https://doi.org/10.1016/j.jfineco.2015.10.001

Alder

Lagakos

Ohanian

(2014). The declines of the U.S. rust belt: A macroeconomic analysis. Federal Reserve Bank of Atlanta Working Paper, 14-05.

Boarnet

(2001). Enterprise zones and job creation: Linking evaluation and practice. Economic Development Quarterly, 15(3), 242-254. https://doi.org/10.1177/089124240101500304

Boarnet

Bogart

(1996). Enterprise zones and employment: Evidence from New Jersey. Journal of Urban Economics, 40(2), 198-215. https://doi.org/10.1006/juec.1996.0029

Bundrick

Yuan

(2019). Do targeted business subsidies improve income and reduce poverty? A synthetic control approach. Economic Development Quarterly, 33(4), 351-375. https://doi.org/10.1177/0891242419875502

Busso

Gregory

Kline

(2013). Assessing the incidence and efficiency of a prominent place-based policy. American Economic Review, 103(2), 897-947. https://doi.org/10.1257/aer.103.2.897

Castillo

Garone

L. F.

Maffioli

Salazar

(2017). The causal effects of regional industrial policies on employment: A synthetic control approach. Regional Science and Urban Economics, 67, 25-41. https://doi.org/10.1016/j.regsciurbeco.2017.08.003

10.

Chaurey

(2017). Location-based tax incentives: Evidence from India. Journal of Public Economics, 6(15), 101-120. https://doi.org/10.1016/j.jpubeco.2016.08.013

11.

Cunningham

(2021). Causal inference: The mixtape. Yale University Press.

12.

Deller

Goetz

(2009). Historical description of economic development policy. Targeting Regional Economic Development.

13.

Domazlicky

(1980). Regional business cycles: A survey. Journal of Regional Analysis and Policy, 10(1), 15–34. https://doi.org/10.22004/ag.econ.128371

14.

Ferman

(2019). On the properties of the synthetic control estimator with many periods and many controls. Journal of the American Statistical Association, 116(536), 1764–1772. https://doi.org/10.1080/01621459.2021.1965613

15.

Ferman

Pinto

(2019). Inference in differences-in-differences with few treated groups and heteroskedasticity. Review of Economics and Statistics, MIT Press, 101(3), 452-467. July. https://doi.org/10.1162/rest_a_00759

16.

Freedman

(2013). Targeted business incentives and local labor markets. Journal of Human Resources, 48(2), 311-344. https://doi.org/10.3368/jhr.48.2.311

17.

Galiani

Quistorff

(2017). The synth runner package: Utilities to automate synthetic control estimation using synth. The Stata Journal, StataCorp LP, 17(4), 834-849. December. https://doi.org/10.1177/1536867X1801700404

18.

Gobillon

Magnac

(2016). Regional policy evaluation: Interactive fixed effects and synthetic controls. Review of Economics and Statistics, 98(3), 535-551. https://doi.org/10.1162/REST_a_00537

19.

Gottlieb

P. D.

(1997). Neighborhood development in the metropolitan economy: A policy review. Journal of Urban Affairs, 19(2), 163-182. https://doi.org/10.1111/j.1467-9906.1997.tb00402.x

20.

Greenbaum

R. T.

Engberg

J. B.

(2004). The impact of state enterprise zones on urban manufacturing establishments. Journal of Policy Analysis and Management, 2(23), 315-339. https://doi.org/10.1002/pam.20006

21.

Hall

(1977). Green fields and grey areas. Proceedings of the Royal Town Planning Institute Annual Conference.

22.

Kreif

Grieve

Hangartner

Turner

A. J.

Nikolova

Sutton

(2015). Examination of the synthetic control method for evaluating health policies with multiple treated units. Health Economics, John Wiley & Sons, Ltd., 25(12), 1514-1528. December. https://doi.org/10.1002/hec.3258

23.

Munasib

Rickman

(2015). Regional economic impacts of the shale gas and tight oil boom: A synthetic control analysis. Regional Science and Urban Economics, 50, 1-17. https://doi.org/10.1016/j.regsciurbeco.2014.10.006

24.

Muro

Maxim

Whiton

(2020). The places a COVID-19 recession will likely hit hardest. Brookings Institution. https://www.brookings.edu/blog/the-avenue/2020/03/17/the-places-a-covid-19-recession-will-likely-hit-hardest/

25.

Neumark

Kolko

(2010). Do enterprise zones create jobs? Evidence from California’s enterprise zone program. Journal of Urban Economics, 68(1), 1-19. https://doi.org/10.1016/j.jue.2010.01.002

26.

Neumark

Simpson

(2015). Place-based policies. In Duranton

Henderson

Strange

(Eds.), Handbook of regional and urban economics, Vol 5. Elsevier.

27.

Neumark

Young

(2019). Enterprise zones, poverty, and labor market outcomes: Resolving conflicting evidence. Regional Science and Urban Economics, 78. https://doi.org/10.1016/j.regsciurbeco.2019.103462

28.

New Jersey Department of Community Affairs (2019). New Jersey Urban Enterprise Zone Program assessment 2019.

29.

New Jersey Department of Community Affairs (2022). New Jersey Urban Enterprise Zone Program. https://www.nj.gov/dca/affiliates/uez/

30.

New Jersey Economic Development Authority (2011). New Jersey Urban Enterprise Zone Program assessment. https://www.nj.gov/dca/affiliates/uez/publications/pdf/NJ%20Urban%20Enterprise%20Zone%20Program.pdf

31.

O’Dea

(2011). Restoring the State’s Urban Enterprise Zone Program. New Jersey Spotlight News. https://www.njspotlightnews.org/2011/12/11-1209-0200/

32.

O’Keefe

(2004). Job creation in California’s enterprise zones: A comparison using a propensity score matching model. Journal of Urban Economics, 55(1), 131-150. https://doi.org/10.1016/j.jue.2003.08.002

33.

Rissman

(1999). Regional employment growth and the business cycle. Federal Reserve Bank of Chicago: Economic Perspectives, Q IV(23), 21-39.

34.

Rubin

H. J.

(1988). Shoot anything that flies; claim anything that falls: Conversations with economic development practitioners. Economic Development Quarterly, 2(3), 236-251. https://doi.org/10.1177/089124248800200304

35.

Rubin

(1990). Urban enterprise zones: Do they work? Evidence from New Jersey. Public Budgeting Finance, 10(4), 3-17. https://doi.org/10.1111/1540-5850.00881