1 Introduction

The Great Recession was the worst economic crisis to hit the United States since the Great Depression. When compared to the recession of the early 1980s, the peak-to-trough decline in real GDP was 2% points lower with even more substantial declines when looking at payroll employment (Blinder 2015). The Great Recession was also prolonged. Unemployment rates remained above pre-recession levels through 2016 (Cunningham 2018) and real GDP did not return to pre-crisis levels until 2011 (Blinder 2015). This lackluster performance has led to a large literature examining the federal and monetary policies (or lack thereof) that could explain such an anemic recovery (Taylor 2014; Verick and Islam 2010; Wynne 2011).Footnote 1

While the entire country felt the negative fallout from the Great Recession, the impact and subsequent recovery was not homogenous across areas. For example, Bennett et al. (2018) find that rural areas experienced a milder recession with a slower recovery than urban areas. In a similar vein, when comparing the length of recent recessions across the 50 largest metropolitan statistical areas (MSAs), Arias et al. (2016) find that while the Great Recession had a negative impact across all cities, some MSAs experienced a relatively brief downturn. The peak-to-trough period was less than a year for cities such as Austin and San Antonio, while the other cities suffered declines for much longer (e.g., Richmond and Memphis). What explains these different recovery rates? Why did some areas experience such a sharp recession accompanied by an anemic, slow-moving recovery? These are the questions we explore in this paper.

The extant literature offers several explanations for the heterogenous impact and recovery of the Great Recession within the United States. Arias et al. (2016) find both education and housing supply elasticity to be important determinants of crisis severity across MSAs. Similarly, Piskorski and Seru (2021) highlight financial frictions associated with the housing market as a major factor in predicting a region’s recovery rate. Walden (2014), using state-level data, finds that certain industry characteristics (e.g., whether the state had a high concentration of financial services) tend to quicken recovery, while government intervention via income transfers and corporate taxes have the opposite effect. We expand upon this existing research by focusing on the role of local economic policies in determining both the initial impact and the rate of recovery of the Great Recession across 382 U.S. metropolitan statistical areas (MSAs). More specifically, we explore how economic freedom, defined broadly as an institutional or policy environment associated with voluntary exchange (Gwartney et al. 2019), influenced both the severity of the Great Recession and the speed of the recovery. In doing so we control for important factors like industrial structure, concentration, and housing costs. We additionally focus on within MSA effects such that important (largely) time-invariant factors like housing elasticity are differenced out.

A positive association between a country’s level of economic freedom and a number of (good) economic outcomes is well-established in the literature including growth (e.g., Heckelman 2000; De Haan 2003; Dawson 2003; Grier and Grier 2021), investment (Bengoa and Sanchez-Robles 2003; Kapuria-Foreman 2007), and entrepreneurship (Nyström 2008).Footnote 2 Similar evidence can be found at the U.S. state-level (Sobel 2008; Compton et al. 2011; Wiseman and Young 2013; Hall et al. 2019). There is also a growing body of literature connecting MSA-level freedom to economic outcomes such as personal income (Bologna et al. 2016), entrepreneurship (Bologna 2015; Bennett 2019, 2021), and patent activity (Wagner and Bologna Pavlik 2020). However, much of this literature is focused on uncovering general associations between economic freedom and economic outcomes outside of crisis years. It is unclear that economic freedom is beneficial throughout times of crisis. Economic freedom implies a lack of government involvement, where the latter may be necessary as a safety net and to facilitate recovery. Thus, the question remains: do the benefits of economic freedom outweigh these potential costs even in a crisis?

There is a blossoming literature highlighting the potential benefits of economic freedom on crisis recovery. In the context of global pandemics, Geloso and Bologna Pavlik (2021) and Candela and Geloso (2021) show that economic freedom can lessen the associated negative economic consequences. Similarly, Bjørnskov (2016) examines 212 major crises across 175 countries and finds a negative association between economic freedom and crisis severity. The intuition behind these studies is that economic freedom offers the flexibility necessary for entrepreneurs to make adjustments that support recovery. Piskorski and Seru (2021) study emphasizes the role of frictions in explaining crisis severity. A more economically free society tends to remove many of the barriers that inhibit growth and recovery.

We are the first to test whether economic freedom has an impact on crisis severity and recovery at the local level. We expand upon the analysis of Bjørnskov (2016) in that we are studying the within nation impact of a single (nationally) homogenous crisis. We also expand upon Walden (2014) by utilizing a more comprehensive measure of economic freedom as opposed to specific governmental policies (e.g., corporate taxes). Further, given the heterogeneity of the crisis even within states, our study focuses on the local (MSA) level as opposed to states.

Our analysis can be separated into two parts. First, we utilize an MSA-level economic freedom index developed by Stansel (2013; 2019) and relate this index to the MSA’s unemployment rate, employment per 100 people, and per capita income levels throughout the crisis period. Our focus here is whether economic freedom tends to lessen the negative impact of the recession. We also examine the relationship between economic freedom and total income per capita, as opposed to just productive income (i.e., income net of transfer payments) as is typically done in the literature (e.g., Higgins et al. 2006). A potential drawback of economic freedom in a crisis is the absence of a social safety net. It could be the case that while productive income is higher, total income is lower due to a lack of governmental transfers.

The second part of our analysis utilizes matching methods (Propensity Score and Mahalanobis Matching) where we relate changes in economic freedom (i.e., the treatment) to subsequent changes in our economic outcome measures from 2007 to 2012. We define a treatment as one where an MSA experiences a significant and sustained increase in economic freedom prior to the crisis.Footnote 3 We then match these treated MSAs to “similar” untreated areas and calculate an Average Treatment Effect on the Treated (ATET) as the average difference in the change of our outcome variable for our treated MSAs relative to their matched counterfactuals. Examining the effect of economic freedom on changes in our outcome variables has the intuitive benefit of focusing on MSA recovery from 2007 to 2012. It also has the practical benefit of differencing out time-invariant characteristics (such as housing elasticity), analogous to the fixed effect specifications of regression models (An and Winship 2017; Grier and Grier 2021).

This matching analysis is also an important robustness check of our regression estimates. Recent literature has shown that two-way fixed effect regression estimators, such as the one we use in our panel analysis, can result in a biased treatment estimate, particularly when the assumption of linear additive effects is violated (e.g., Imai and Kim 2021). Matching has been proposed as a useful alternative as it does not rely on functional form assumptions and focuses on the treatment’s impact using a simple difference in averages as opposed to using the less transparent and potentially non-convex regression weights (e.g., Imai and Kim 2019; Grier and Grier 2021).Footnote 4 This also has the benefit of matching upon both pre-treatment outcomes and initial economic freedom levels to ensure that we are only comparing “treated” MSAs (i.e., MSAs that experienced changes in economic freedom) with appropriately similar non-treated units. In other words, we are comparing MSAs that had similar economic environments prior to both the treatment and crisis helping to address the concern of selection bias. We then see how changes in this environment impact several economic outcomes that relate to economic recovery.

The MSA-level economic freedom index is available on a 5-year basis according to the available Census of Governments years. Thus, we first relate economic freedom to income using panel data in 5-year increments (2002, 2007, and 2012). In doing so, we include both period and MSA-level fixed effects. We then focus on cross-sectional results using the 2007 economic freedom level and average outcomes from the crisis period alone.Footnote 5 Our primary goal in this first part of the analysis is to test whether more economically free areas experience better economic outcomes, even during a crisis. A positive association implies a lower recovery burden – economic freedom could be viewed as a preventative measure to avoid major economic collapses.

We then address the question of whether economically free areas grow faster following an economic downturn using a matching analysis. We compare the recovery of “treated” MSAs with similar “untreated” MSAs in the post-crisis period (2007–2012). To define a treatment, we focus on large and sustained jumps in economic freedom from 2002 to 2007. We then construct for each treated MSA a plausible counterfactual against which to compare crisis recovery. We choose counterfactuals based on covariates that plausibly determine the probability of treatment and/or are otherwise correlated with crisis recovery. In other words, these counterfactuals are MSAs that were similarly likely to have received the treatment but did not.

Combining the results of the first part of our analysis and our matching estimates, we uncover an intuitive narrative. Economic freedom tends to be positively associated with key outcomes, even during times of crisis. A standard deviation increase in economic freedom (0.76) decreases unemployment by nearly 1% point, increases employment by 1.36 per 100 persons, and increases income by (at least) 3%. These are meaningful changes in economic outcomes associated with only modest changes in economic freedom – equivalent to moving from Buffalo, New York to Grand Rapids, Michigan, for example. Economic freedom also accelerated the recovery of treated MSAs in the aftermath of the recession. Treated MSAs (those that experienced increases in economic freedom) experienced faster income and employment growth and slower growth in the unemployment rate throughout the recovery period. Thus, not only can economic freedom dampen the negative effects of an economic shock, but it can help communities recover quicker.

The remainder of our paper is as follows: Sect. 2 discusses the literature on institutions and crisis recovery, Sect. 3 summarizes empirical methodology, Sect. 4 describes our data, Sect. 5 discusses our results, and we conclude with Sect. 6.

2 Institutions and Crisis Recovery

Understanding community resiliency and crisis recovery is an important area of research. Recent literature spurred by the COVID-19 pandemic has highlighted the role of institutions in creating environments that facilitate recovery and create growth. Geloso and Bologna Pavlik (2021), for example, study the 1918 flu pandemic and show that the induced economic crisis was less severe in countries with higher levels of economic freedom. Similarly, using data on 20 OECD countries, Candela and Geloso (2021) show that economic freedom lessens contractions and accelerates recoveries associated with the major influenza pandemics of the 20th century. Economic freedom has also been shown to be important to crises recovery more generally. Bjørnskov (2016), for example, uses data covering 212 crises across 175 countries and finds that economic freedom tends to reduce both the peak-to-trough ratio (i.e., make the crisis less severe) and the recovery time. Thus, economic freedom seems to be associated with quicker economic recoveries and smaller negative shocks in response to crises in general.Footnote 6

Our paper contributes to these studies concerning economic freedom and crisis recovery by focusing on small locales within a single nation. More specifically, we focus on the Great Recession. While the effect of the crisis varies across MSAs, the crisis itself is much more homogenous when making intra- versus inter-national comparisons. This is important because we can understand how economic freedom influences crisis recovery when the major characteristics of the crisis in question are effectively held constant.

Why does economic freedom improve resiliency in response to economic crises? One potential explanation relates to flexibility. Economic freedom gives entrepreneurs the ability to make the necessary adjustments to facilitate recovery. In the context of recovery associated with Hurricane Katrina, Boetkke et al. (2007) argue that overregulation inhibited entrepreneurs from reopening. Similarly, Smith and Sutter (2013) argue that the lifting of building regulations and zoning laws accelerated the recovery of Joplin, Missouri following the 2011 tornado.

A related explanation involves entrepreneurial alertness. Entrepreneurs in areas with high levels of economic freedom have an incentive to be more innovative (Kreft and Sobel 2005; Boudreaux et al. 2019), for example, show a positive association between entrepreneurial alertness and a country’s level of economic freedom. Regardless of the initial impact of the crisis, areas filled with ingenious and opportunistic individuals are likely to experience a swift recovery.

These arguments, however, run counter to the idea that governmental action is necessary to coordinate action that facilitates recovery. In the context of Katrina, for example, Burby (2006) argues that some form of government planning and/or intervention is crucial to recovery. More recently, in response to the COVID-19 pandemic, there is a renewed interest in the debate over the appropriate governmental response to crises. While some highlight the effectiveness of measures such as mask mandates or stay-at-home orders in reducing the spread of COVID (and potentially quickening recovery as a result) (e.g., Courtemanche et al. 2020), others call into question the necessity of such interventions and argue further that they could be harmful to social welfare relative to the alternatives (Boettke and Powell 2021). The latter argument emphasizes the importance of bottom-up solutions, local knowledge, and entrepreneurial nimbleness in navigating the COVID-19 pandemic and facilitating recovery.

Even if governmental action can improve upon crisis situations, there are several incentive incompatibilities that must be considered. Political motivations (e.g., reelection incentives) strongly influence political behavior and have been shown to impact federal spending allocation (Young and Sobel 2013), federal grants (Kriner and Reeves 2015; Stratmann and Wojnilower 2015), transfers (Tackett and Bologna Pavlik 2021), disaster declarations (Leeson and Sobel 2008), and corruption convictions (Bologna Pavlik 2017). In discussing constitutionally mandated power in response to an emergency, Bjørnskov and Voigt (2022) find that a (relative) boost in executive power during an emergency results in a greater number of disaster related deaths. Their explanation behind this finding is that these natural disasters are being used to expand their power rather than save lives.Footnote 7

More recently in the context of the COVID-19 pandemic, there is evidence that emergency orders were implemented based on underlying political and institutional factors as opposed to pure need. For example, Bjørnskov and Voigt (2021) find that the ability to gain discretionary power is a key determinant of whether a country declared a state of emergency. Similarly, within the U.S., McCannon and Hall (2021) find that states with Democratic governors and less economic freedom tended to implement stay-at-home orders quicker, even after controlling for important factors such as the date of the first COVID-19 related death in the state. Thus, though governmental involvement can improve economic conditions, there are significant political barriers that can inhibit welfare enhancing policy. For this reason, and because of the importance of local knowledge, Grube and Storr (2014) emphasize self-governance in determining community resiliency and explore how pre-existing self-governance systems aided recovery post-Katrina in two communities in New Orleans. Top-down (i.e., government) solutions are often devoid of local knowledge and are non-customizable. Bottom-up solutions may be better able to handle the complexity of the situation.

We directly consider the potential benefits of governmental intervention in response to the Great Recession. We focus not only on productive or net (income less transfers) income, but total (earned plus transfers) income. While the COVID-19 pandemic had significant health related externality concerns surrounding mask-wearing and individual behavior, the Great Recession was different. The biggest concern was economic stability, making transfers and government spending the most important focus of governmental intervention. Economically free areas, by definition, have a more limited government and tend to rely on individual decision-making efforts. Economic freedom yields more flexibility for entrepreneurial recovery efforts. Whether these benefits of economic freedom outweigh the cost of losing the governmental safety net throughout a crisis is an open question.

3 Empirical methodology

We are interested in understanding how economic freedom influences crisis severity and recovery. Identifying a causal effect of economic freedom (on crisis recovery) is challenging for (at least) two reasons. First, economically free areas are not selected at random. The determinants of economic freedom across metropolitan areas likely also affect economic outcomes (omitted variable/selection bias). Second, simultaneity could also be a concern if faster growing areas experience higher levels of freedom. Our empirical strategy aims to address these issues.

First, we estimate a panel model with both period- and MSA-fixed effects. This allows us to focus on within-MSA differences and eliminate many of the time-invariant factors that could bias our results. We also include a wide range of controls described in detail below. Second, we narrow in on the Great Recession years and conduct a cross-sectional analysis to get an estimate of the correlation between economic freedom and economic outcomes throughout periods of economic strife. Third, we employ a matching analysis that addresses both selection and simultaneity concerns, in addition to recent concerns with two-way fixed effect regression analyses (e.g., Imai and Kim 2019; 2021; Gibbons et al. 2019). This latter analysis also helps us focus more on recovery.

3.1 Panel regression

We start with a balanced panel regression of three time periods: 2002, 2007, and 2012. We estimate the following model:

$${Y_{i,t}} = {\beta _0} + {\beta _1}MEF{I_{i,t}} + {\beta _2}{X_{i,t}} + {V_{st}} + {V_i} + {V_t} + {\varepsilon _{i,t}}$$

where Y is our outcome variable (unemployment rate, employment per 100 people, or per capita income); MEFI is the MSA-level economic freedom measure; X is a set of relevant controls outlined below in Sect. 4; Vst is a state-specific linear time trend; Vi is an MSA-level fixed effect; and Vt is a period specific fixed effect. These latter three variables are only relevant for the panel results. Standard errors are clustered at the metropolitan area level. These specifications provide us with 1,146 observations (382 MSAs over three 5-year periods). Importantly, this panel includes both pre- and post-crisis years.

These panel regressions give us an estimate of the relationship between economic freedom and important economic outcomes using the standard workhorse regression model: two-way fixed effects (TWFE). These regressions include unit (MSA) fixed effects in an attempt control for time-invariant unobservable characteristics. However, serious concerns have been raised regarding TWFE models. First, they require strong functional form assumptions surrounding these unobservable characteristics (e.g., linearly additive) (Imai and Kim 2019). Second, regressions can extrapolate beyond the support of the data further demanding functional form assumptions. Relatedly, TWFE models can yield negative weights in the presence of treatment heterogeneity resulting in an estimated treatment that is potentially of the incorrect sign (de Chaisemartin and D’Haultfoeuille 2020). In addition to these TWFE concerns, it is also not clear that this association holds throughout crisis years alone. We therefore narrow in on the crisis years in the following section and lastly utilize matching methods to: (1) address selection bias, (2) remove functional form assumptions, and (3) focus on simple averages as opposed to weighted, and potentially non-convex, averages to estimate a treatment effect.

3.2 Cross Section Regression

To explore the relationship between economic freedom and crisis severity more directly, we run a cross-sectional OLS regression with 382 MSAs using outcome data from the peak of the Great Recession alone. We use the cross-sectional version of Eq. (1) with outcomes averaged using data from the most severe crisis years and economic freedom, along with all controls, held at their initial levels from 2007. Because the exact timing of the crisis differed across locales, we use two separate three-year averages for our outcomes: 2006–2008 and 2007–2009. Our goal here is to capture the trough across all MSAs on average.

If economic freedom and these economic outcomes are still positively associated, this suggests that economically free areas may fair better even during crisis situations. Of course, cross-sectional regressions suffer from significant concerns including omitted variable bias due a lack of fixed effects, simultaneity, and functional form restrictions (i.e., linear regression). Our matching analysis helps us overcome these concerns and is described in the following section.

3.3 Matching analysis

Our goal is to uncover a causal relationship between economic freedom and economic outcomes like unemployment and incomes. However, in doing so we need to address both selection bias and simultaneity concerns. Our matching method is analogous to two-way fixed effects regressions in the sense that we compare the within differences in our outcomes across “treated” metropolitan areas. However, matching is a non-parametric method that uses a simple average when estimating the treatment effects. We can therefore avoid the problem of making strong functional form assumptions and the potential of negative weights biasing our results.

Table 1 Cases of Jumps in MSA-Level Economic Freedom (2002–2007)

Our matching analysis first defines a treatment – a large and sustained increase in economic freedom prior to the Great Recession (between 2002 and 2007). We match these treated MSAs to MSAs that are “similar” but did not experience such reform. In this context, similar implies MSAs that were just as likely to have received the treatment but did not and thus alleviates selection bias concerns. We then compare the average change in the outcome over the subsequent five-year period (e.g., unemployment rate in 2012 – unemployment rate in 2007) for the treated group versus the constructed counterfactual. This latter step represents our Average Treatment Effect on the Treated (ATET) estimate and is important to highlight because it is focused only on a simple average and compares only changes in our outcome variables. Because the major concern with matching is the inability to match on unobservables, we follow An and Winship (2017) and difference out these important time-invariant characteristics by focusing on changes in our outcome variables as opposed to levels.Footnote 8 Thus, this is analogous to a TWFE model where our focus is on within unit changes in the dependent variable.

We define our treatment as an MSA that experienced a sustained jump in economic freedom as measured by the MEFI index of 0.5 points or greater between 2002 and 2007. The treatment must be “sustained” in the sense that their economic freedom scores did not substantially drop more than 50% of its original increase value in the next five years (2007–2012). The choice of 0.5 is arbitrary; we use 0.5 as an initial cut-off as it is roughly two-thirds of a standard deviation in MEFI (0.76) and leaves us with a reasonable number of treatments (28).Footnote 9 We explore other potential cut-offs and increase the threshold to a full standard deviation. However, in this case there are only seven treatments and therefore not enough to utilize matching methods. We also reduce the threshold to one-third of a standard deviation (0.25; 109 treatments) and re-estimate our results as a robustness check.Footnote 10 While these results generally support our main finding, they are largely insignificant suggesting that the boost in economic freedom needs to be substantial to have a meaningful impact.

Once we define our treatment, we compare treated units to those that did not experience a jump in economic freedom but are similar in important ways. We construct similar, but untreated counterfactuals by matching on important covariates including the industrial structure in 2002 for each MSA using employment shares, a measure of income inequality, industry concentration, initial (2002) levels of our outcome variablesFootnote 11, economic freedom levels in 2002, and the standard deviation of the MEFI component score in 2002Footnote 12. We also utilize a housing price index to match upon the rate of housing price growth from 2002 to 2007. Details surrounding the sources and construction of these covariates are discussed below in Sect. 4. These are all factors that could potentially influence both the likelihood of treatment (economic freedom reform) and changes in our outcome variables (e.g., income growth). Matching using the initial levels of economic freedom and each economic outcome is especially important given that it compares only MSAs that started from the same general economic environment. Note also that our treated MSAs are being matched to counterfactuals only in 2007, so we are comparing post-treatment outcomes between units from the same period.

To construct our counterfactual using the above-mentioned covariates, we employ two alternative matching methods: Propensity Score Matching (PSM) and Mahalanobis Distance Matching (MDM). The former matches each treated unit to a control based on the closest probability of treatment. In other words, the covariates are used to predict whether the MSA experiences an increase in economic freedom between 2002 and 2007. Those with the closest prediction probability are matched. The latter focuses on matching covariates such that they are as close as possible across the treated and control units. For both methods we use the nearest neighbor criterion and alter the number of neighbors from one to three. For PSM, we also use Kernel matching, which weights all untreated MSAs according to the closeness of their propensity scores. Following Grier and Grier (2021), we estimate our standard errors in the Propensity Score method with bootstrapping and utilize Abadie and Imbens’ (2011) method of bias-correction for Mahalanobis matching. It is important to consider the results of both matching procedures as the method of matching can result in a different degree of balance across covariates. PSM places a heavier weight on matching the covariates that are important predictors of the treatment, while MDM matches on the covariates directly. Thus, a difference in the results between the two methods could be due to a difference in covariate balance. We present covariate balance tables in Appendix A1-7 with net (less transfers) income per capita as a reference. Covariate balance tables for all specifications are available upon request.

4 Data

The data used in this paper is divided into three categories: main independent variable, outcome, and controls. We use data from the years 2002, 2007, and 2012. Thus, we include information from both the pre- and post-crisis periods.

4.1 Economic Freedom

We use the Stansel (2013; 2019) economic freedom index available at the metropolitan area level (henceforth referred to as MEFI) as the independent variable of interest. This data is available every 5 years from 1972 to 2012 for 385 US metropolitan areas. For the purposes of this paper, we use only the MEFI scores from 2002, 2007, and 2012 as we are focused on the periods immediately surrounding the Great Recession.

Based off the Fraser Institute’s global index (Economic Freedom of the World) and the state and province-level North American index (Economic Freedom of North America), MEFI attempts to quantify the level of economic freedom at the local level. Economic freedom, according to Gwartney et al. (1996), is defined as when “individuals have […] property they acquire without the use of force, […] they are free to use, exchange, or give their property as long as their actions do not violate the identical rights of others. Thus, an index of economic freedom should measure the extent to which rightly acquired property is protected and individuals are engaged in voluntary transactions” (pg. 12). This local-level economic freedom index uses three major areas: (1) size of government, (2) taxation, and (3) labor market regulations. Size of government and taxation quantify the ability to freely use property, while labor market freedom operationalizes the ability to engage in voluntary transactions within the workforce.

MEFI is constructed on a scale from 0 to 10, with higher scores indicating greater economic freedom. The index is a simple average of three areas. Area 1 (Size of Government) is based on government consumption, transfers and subsidies, and insurance and retirement payments. Area 2 (Taxation) collects data on income and payroll taxes, sales tax revenue, revenue from property tax, and tax revenue from each source except severance taxation (since this is levied at the state level only). Areas 1 and 2 are measured as a share of total metropolitan personal income. Lastly, Area 3 (Labor Market Freedom) scores MSAs based on minimum wage, government employment shares, and private union density. Minimum wage is the share of full-time income as a percentage of per capita personal income; government employment and private union density are shares of total MSA employment.

While MEFI attempts to quantify important variation in economic freedom across metropolitan areas, it is important to discuss its measurement relative to the economic freedom of the world (EFW) index. The most recent version of the EFW index measures economic freedom across 165 countries and is constructed using economic freedom scores across five areas: (1) size of government, (2) legal system and property rights, (3) sound money, (4) freedom to trade internationally, and (5) regulation. Arguably the most important component to long run economic growth and development is Area 2 (e.g., Carlsson and Lundström 2002; Rode and Coll 2012). Area 2 is often used to proxy for a country’s rule of law. While there is some variation in the rule of law within a country, this variation is minimal compared to cross-country differences. MEFI mostly relies on local level estimates of Area 1 – size of government – and Area 5 – regulation – to capture differences in economic freedom across areas. Differences in the rule of law, sound money, and trade policy are similar within the U.S.Footnote 13 Thus, any result we do find could be interpreted as a lower bound estimate for the benefits of economic freedom more generally.Footnote 14

Another important difference in the EFW versus MEFI measures is the potential variance across each component. The EFW measure is broad and can have substantial variance across component scores within a given country. For example, Bolen and Sobel (2020) highlight several examples where countries score poorly in the legal system and property rights category but receive high scores in other categories resulting in a reasonably high level of economic freedom overall. They compare these examples to other countries with similar levels of overall economic freedom, but with much more uniform scores across each respective component. Their hypothesis is that the variation in component scores should also matter for growth and development and find that the standard deviation in component scores negatively correlates with economic growth. Given this finding, we additionally include the standard deviation of MEFI components as a control in our regressions. However, we note that the standard deviation in MEFI components (0.757 on average) is much lower than that of the EFW components (1.439 using the most recent data; Gwartney et al. 2021). Moreover, as described in the preceding paragraph, metropolitan areas have implicitly similar scores across several of the EFW components (legal system, sound money, and trade). In line with much of the literature, Bolen and Sobel (2020) highlight the importance of the legal system and property rights score in their study of the variation in component scores. Given that this component is relatively constant within the United States, it is not clear that component variation will have the same impact on economic outcomes as the analogous measure at the country level. Nevertheless, we include the standard deviation as a control.

4.2 Outcome variables

We focus on three important indicators of crisis severity: unemployment rate, employment per 100 persons, and income per capita. Our unemployment rate data comes from the U.S. Bureau of Labor Statistics. The seasonally adjusted unemployment data is provided on a monthly basis; we average the unemployment rate across all twelve months to get annual estimates. Employment per 100 persons is simply total employment adjusted for population. While employment per 100 persons and unemployment rates are similar, there are key distinctions. Employment per 100 counts both full-time and part-time jobs, as well as self-employment. Unemployment rates, however, only account for those people actively in the labor force. It is important to consider both variables when examining crisis recovery.

We also use income per capita as an outcome variable using the Bureau of Economic Analysis’s (BEA) measure of personal income. Following Higgins et al. 2006 and Bologna et al. 2016, we exclude transfer payments and thus refer to this measure as a net income per capita estimate.Footnote 15 We convert income into 2015 US Dollars using the World Bank’s estimate of the U.S. GDP deflator. In some specifications, we take the average real net income per capita from 2006 to 2008 and 2007–2009 to focus only on crisis years. We also explore the effect of economic freedom on transfers and total (with transfers included) income per capita, using the same U.S. GDP deflator adjustments.

Table 2 The effect of economic freedom on the unemployment rate, employment per-100 people, and (logged) income per-capita; all controls included
Table 3 The effect of economic freedom on (logged) transfers per capita and total income per capita; all controls included.

The use of real net income per capita in studies relating economic freedom to economic outcomes within the United States is common. However, this is potentially problematic when examining the relationship throughout the Great Recession. A major component of the Great Recession was the housing market crash, which had drastic impacts on the cost of living across the U.S. These effects were likely not uniform. We therefore address this concern three ways. First, in our panel data estimates we include a state-specific time-trend in addition to year and metropolitan area fixed effects. This controls for general state level characteristics that change through time, including the potential for changes in the cost of living. Second, for the two cross-sectional specifications we include a control for the average change in the housing cost in the relevant period (2006–2008 and 2007–2009, respectively) using the Federal Housing Finance Agency’s Housing Price Index available at the state level. We also include the percentage change in this index from 2002 to 2007 as a covariate in our matching specifications.Footnote 16 Third, we utilize the Bureau of Economic Analysis’s (BEA) regional purchasing power index, available at the metropolitan area level starting in 2008, to adjust personal income (and transfers) per capita and re-estimate our cross-sectional and matching specifications as a robustness check. For the cross-sectional specifications, we utilize the 2008 value to adjust for regional price disparities. For the matching estimates, we utilize both the 2008 and 2012 values to calculate an RPP adjusted change in income and transfers per capita. This is not a perfect adjustment as the index does not exist prior to 2008 and there were likely important changes in the years leading up to the crisis, however this is an important robustness check. These latter results largely reflect our main findings and are available in Appendix C.Footnote 17

4.3 Other controls: Industry Shares and Inequality

Following Bologna et al. (2016), we include the share of industry employment as controls.Footnote 18 In particular, we employ shares from eighteen industries (collected from the BEA): construction, education, farming, federal government, finance and insurance, food, forestry, healthcare, information, manufacturing, mining and extraction, other services, professional, real estate, recreation, retail trade, transportation, and wholesale trade. We also include a measurement of inequality as Cyanmon and Fazzari (2016) argue that inequality explains the slow recovery following the Great Recession. However, here we are limited to a state-level measurement derived from Frank et al. (2015). Specifically, we take the income shares for the top 10% income earners in each state. Because some MSAs cross state borders, we include the level of inequality in the primary state, as labeled by Stansel (2013; 2019). A summary statistics table for all variables is reported in Table 4a, 4b.

Table 4a Summary Statistics
Table 4b Summary Statistics

5 Results

5.1 Regression analysis

We begin the discussion of the results with our regression analyses for the three main dependent variables: the unemployment rate (panel a), employment per 100 persons (panel b), and real (net) income per capita (panel c). We present these results both without controls (Table 5) and with controls (Table 6) as a comparison. To preserve space, we show estimates for our main variable of interest only (Economic Freedom). Full results are available upon request.Footnote 19 Column (1) of each table presents our panel estimates with a state-specific time trend and MSA and period fixed effects included. Columns (2) and (3) give the cross-sectional estimates for two alternative sets of crisis years (2006–2008 and 2007–2009, respectively).

Table 5 The effect of economic freedom on the unemployment rate, employment per-100 people, and (logged) income per-capita; no controls.

As can be seen in the tables, economic freedom is significantly related to a lower unemployment rate, higher levels of employment per 100 persons, and increased income per capita. This relation holds both with and without controls included; for both the panel and cross-section estimates. This relationship also appears to be economically meaningful. Using the most conservative estimates given in Table 2, a one standard deviation increase in economic freedom (0.76) corresponds to 0.61-point reduction in the unemployment rate, an increase of 1.36 jobs per 100 persons, and a 3.8% increase in income per capita. For unemployment rates, this explains approximately 25% of a standard deviation (Table 5). The increase in income seems to be the most efficacious impact in that the average 5-year growth rate of income hovered at just above the 4% level in the post-crisis period. Employment per 100 persons is positive but less meaningful explaining only 17% of a standard deviation change.

We also explore whether economically free areas experience better outcomes once transfer payments are included in our income estimates (Table 3)Footnote 20. Tables 5 and 2 show that economic freedom has benefits when considering income coming from productive sources only. However, transfer payments are a particularly important function of governmental intervention throughout times of crisis. As expected, the first panel of Table 3 (panel a) shows that transfer payments tend to be lower in economically free areas. Panel b, however, shows that despite the lack of this social safety net, total income per capita is higher in MSAs with more economic freedom. Thus, economic freedom can still benefit MSAs throughout a crisis period even when governmental intervention might be most valued.

5.2 Matching analysis

The second part of our analysis utilizes the matching methods described above. Recall that our matching analysis compares post-crisis (2007–2012) recovery in “treated” MSAs with similar “untreated” MSAs. We define an MSA to be treated if it experienced a meaningful jump in economic freedom from 2002 to 2007 (see Table 1 for a list of treated units). The five-year period following the crisis dictated how these metropolitan areas were able to recover; we therefore focus on changes in our outcome variables between 2007 and 2012. If improvements in economic freedom facilitated recovery, this is an important consideration.

Results for the unemployment rate, employment per 100 persons, and real (net) income per capita are given in Tables 6, 7 and 8, respectively. The top portion of the tables presents the Propensity Score Matching (PSM) results, and the corresponding Chi-squared test of overall covariate balance. The null in the Chi-squared test is that the covariates are balanced on average; thus, rejecting the null would imply imbalance. The bottom portion of the tables gives the Mahalanobis results. While an overall covariate balance test analogous to the Chi-square statistic given for PSM is not available for Mahalanobis, we show the pre- and post-match difference in means in the appendix for reference (Tables A5-7). For all cases, we first compare our treated MSAs with the full set of potential untreated controls. We then drop controls that were “almost treated”; i.e., had increases in economic freedom that were large (0.25), but did not meet the 0.5 cutoff.

Table 6 Effects of a jump in economic freedom on 5-year changes in the unemployment rate
Table 7 Effects of a jump in economic freedom on 5-year changes in employment per 100
Table 8 Effects of a jump in economic freedom on 5-year changes in (logged) real net (less transfers) income per capita

Starting with the unemployment rate, we see that the average change in the unemployment rate was consistently lower in areas that had experienced significant jumps in economic freedom (Table 6). Given the general increase in the unemployment rate across all areas (Table 6), this implies that MSAs that had reforms experienced less severe increases in unemployment. However, only one of these effects are statistically significant and three specifications yield an estimate of the opposite sign (though insignificant). Employment per 100 persons presents more of a mixed result (Table 7). The estimates are mostly negative, implying steeper reductions in employment per 100 persons for economically free areas. However, the signs switch directions across matching methods; these effects are positive for Mahalanobis Nearest Neighbor 2 and 3 in the baseline estimate and for Mahalanobis Nearest Neighbor 1 and PSM Nearest Neighbor 3 after dropping the “almost treated” controls. Three of the estimated effects are statistically significant and negative, suggesting that increases in economic freedom may not lead to consistent increases in employment.

An interesting pattern begins to emerge in Table 8. Economically free areas grew faster, in terms of real (net) income per capita. This finding holds for both PSM and Mahalanobis when looking at our baseline estimates; and for Mahalanobis even after dropping the “almost treated” controls. There is only one instance of a negative coefficient – Mahalanobis Nearest Neighbor 1 – and this effect is statistically insignificant. The magnitude of these estimates are on par with what is uncovered in the regression analysis – anywhere between three and five% increased income per capita. Similar to above, we explore how jumps in economic freedom impact transfers and total income per capita (Tables 9 and 10, respectively). We find that economic freedom has a consistent negative effect on transfers per capita and this effect is statistically significant in six specifications. We also find that total income per capita tends to be higher in areas that experienced economic freedom reform. While these effects are only significant in three of fourteen cases, they are positive across all specifications except for Mahalanobis NN1.

Table 9 Effects of a jump in economic freedom on 5-year changes in (logged) transfers per capita
Table 10 Effects of a jump in economic freedom on 5-year changes in (logged) total (income plus transfers) income per capita

Overall, our results suggest that institutional reform along the lines of economic freedom can bolster recovery within a local economy and improve resiliency. While we cannot say that areas with higher levels of economic freedom experienced more rapid recoveries, we can say that increases in economic freedom made several areas relatively better off. This is true even when considering only the most severe crisis years. The exception seems to be employment. Our matching estimates imply that there is some limited evidence that increases in economic freedom can lead to reductions in employment per 100 persons.

5.3 Components of Economic Freedom

Our focus is on the general economic environment of a metropolitan area and therefore we have used overall economic freedom as our main variable of interest. However, this index can be decomposed into its three main parts: the size of government, taxation, and labor market regulations. We can use these components to explore whether one (or more) of the three drive our key findings. These results are given in Appendix B1B7.

We find that the results using either area 1 (size of government) or area 3 (labor market regulations) as our independent variable of interest largely echo results using the overall index. That is, areas with a smaller government size and/or less labor market restrictions tend to experience better economic outcomes. However, these results do not hold when we explore the analogous matching estimates. Using the same rule as with overall freedom, we categorize an MSA as treated if they experience a 2/3rds of a standard deviation jump in the relevant component. This corresponds to 45 treatments for area 1 and 83 treatments in area 3 (labor market freedom). Area 2 (taxation) is incredibly stable as this same rule only yields 2 treatments; the lack of variation here likely explains the insignificant regression results.

These matching estimates yield little in terms of statistical significance over all (see Table B7). When digging into the data a bit more, there is surprisingly little overlap between jumps in overall freedom (Table 2) and these individual components (Table B8 & B9). This implies that these individual jumps may not be important enough to trigger major changes in the overall economic environment, rendering them insignificant on their own.

6 Conclusion

The Great Recession was a serious economic crisis that shocked the world – worse than the recession of the early 1980s in the United States. While all areas of the U.S. suffered to some extent, the impact of this crisis was highly variable across the U.S. (Walden 2014; Arias et al. 2016; Bennet et al. 2018).

In this paper, we explore the role of economic freedom in both determining the impact of the crisis and the subsequent recovery using metropolitan statistical area (MSA) data. Our results suggest that MSAs with high levels of economic freedom experienced lower unemployment rates, more employment per 100 persons, and higher levels of (net of transfer payments) income per capita. This is even true throughout the most intense of the recession years. We also find that areas that increased their levels of economic freedom before the crisis (2002 to 2007) experienced accelerated recoveries. Taken together, our results suggest that economic freedom can play a significant role in mitigating the impact of economic crises.

We believe this finding leads to many essential questions left to future research. First, what are the exact mechanisms that help economically free areas recover quicker? For example, did these areas experience more innovation and/or entrepreneurial pursuits? Second, what drives economic reform? Relatedly, are there important differences across party lines? Using a regression discontinuity design, Hankins and Hoover (2019) do not find any difference in economic freedom across Democratic versus Republican states. However, this finding does not address changes in economic freedom or differences in policy responses to crises. Understanding the process of reform amongst our 28 treated MSAs could yield useful insights into reform more generally. The answer to both questions is important to understanding local resiliency.