Effects of immigration on native entrepreneurship in the US: an analysis of self-employment over 1980–2018

Abstract

This paper examines the causal impact of immigration on entrepreneurship among US-born non-Hispanic whites in non-agricultural private sectors from 1980 to 2018. Using self-employment as a proxy for entrepreneurship and distinguishing between incorporated individuals and unincorporated self-employed workers, I find a sizable negative impact of immigration on native self-employment. Importantly, this effect is consistent across incorporated and unincorporated self-employment and remains robust across demographic groups based on gender, age, and education. Results are robust to the choice of controls and estimation methods.

Plain English Summary

Immigration has long been a contentious issue among economists and policymakers, and this paper contributes to the debate by examining its effects on the self-employment of US-born non-Hispanic whites in non-agricultural private sectors over the past four decades. The findings reveal a negative impact of immigration on native self-employment, with each self-employed immigrant replacing approximately one native self-employed worker. These findings contribute to the previous studies that argue that immigration poses severe challenges to local-labor markets.

1 Introduction

Immigration to wealthy countries has grown dramatically over the last 50 years and thus has significantly changed these countries’ racial and demographic composition. For example, according to the US Census data, there were fewer than 10 million foreign-born individuals in 1970 (less than 5% of the US population), but that size steadily rose to about 45 million in 2016 (around 15% of the population). A similar upward trend has also occurred in other advanced countries (OECD, 2017). The socioeconomic effects of immigration have been intensely debated among economists and policymakers. Researchers have investigated the impact of immigration on several labor-market outcomes, but its effects on native entrepreneurship have attracted limited attention.

Economists have long recognized entrepreneurial activity as a critical engine behind innovation, job creation, and economic prosperity. Recent empirical studies have shown that expansions of start-ups constitute a significant source of job creation (Decker et al., 2014). Immigration becomes relevant to entrepreneurship among natives in several important ways. First, immigrants’ cheaper labor makes businesses more profitable, encouraging entry into entrepreneurship. Second, immigrants demand goods and services that may lead to more entrepreneurial activity. Thus, immigrants can create new business opportunities for natives. Finally, several studies have found that foreign-born individuals have higher business ownership (Fairlie & Lofstorm, 2015). In this case, increased competition from immigrant entrepreneurs may crowd out native entrepreneurs. Therefore, the net impact of immigration on entrepreneurship is ambiguous, and addressing it is an empirical question.

This paper investigates the causal impact of immigration on entrepreneurship among US-born non-Hispanic whites over the 1980–2018 period. I utilize the Census Bureau’s decennial surveys and the American Community Surveys in my analysis and use self-employment as a proxy for entrepreneurship, following previous studies (Borjas & Bronars, 1989; Hamilton, 2000; Fairlie, 2014). However, Levine and Rubinstein (2017) show that incorporated and unincorporated individuals differ in their cognitive and non-cognitive traits; as a result, my analysis also distinguishes between these two groups of self-employment.

Before presenting the findings, three remarks are in order. First, some studies have used establishment counts as a proxy for entrepreneurship (e.g., Decker et al., 2014). However, employer-based counts represent a small share of entrepreneurial activity. For instance, they constitute only about 10% of all new business entries (Fairlie, 2014). Further, self-employed individuals play a significant role in employment dynamics (Levine & Rubinstein, 2018) and (especially incorporated ones) hold a substantial portion of the US wealth (De Nardi et al., 2007; Chatterji et al., 2014). Finally, using self-employment is more consistent with theory, which identifies entrepreneurship at the individual level (Hurst & Pugsley, 2017).

Second, I consider only non-Hispanic white Americans for the following reasons. First, they have the largest population in the sample, allowing me to conduct a detailed analysis of them in several dimensions.¹ Second, non-white Americans include distinct groups (such as Black/African American, American Indian, Hispanic, Chinese, South Asians), each may have a different attitude towards entrepreneurship and thus deserve a separate analysis. Lastly, several other factors can affect non-whites decision to become self-employed business owners. For example, Asiedu et al. (2012) show that minorities face significant discrimination in getting loans to finance their businesses.

Finally, the geographic unit used in the analysis is the commuting zone (CZ), a cluster of counties with strong commuting ties between businesses and workers. My study covers 722 CZs across the entire mainland US. I use an instrumental variable approach to address the possible problem of simultaneity and omitted variable bias, building on the shift-share instrument of Altonji and Card (1991) and Card (2009). The instrument is an imputed immigrant share, constructed by taking the total new immigrants from each source country and reallocating them across CZs based on the distribution of earlier immigrants settled in those areas. The rationale behind this instrument is that their historical settlement across CZs drove the distribution of immigrants, as previous studies have shown (Patel & Vella, 2013; Peri et al., 2015).

The results of this analysis can be summarized as follows. First, immigration has a sizable negative impact on self-employment among white Americans. Specifically, my estimates imply that each self-employed immigrant results in a reduction of about one self-employed American. In addition, the effects are very similar on incorporated and unincorporated business owners. Exploring heterogeneity across different demographic groups (based on gender, age, and education) suggests that the impact of immigration does not vary substantially across these groups. Finally, an extensive set of sensitivity checks show that the results are robust to the choice of controls, measurement of variables, and estimation methods. By considering various channels through which immigration can affect native entrepreneurship, these findings are broadly consistent with the explanation that increased competition from highly entrepreneurial immigrants dominated possible positive effects of immigrants and thus crowded out native entrepreneurs.

The paper is organized as follows. The following section reviews the previous studies and puts the paper’s contribution in perspective. Section 3 introduces the econometric methodology employed in the paper and discusses the instrument. Section 4 discusses the data and provides summary statistics on key variables. Section 5 presents and evaluates the results, and Section 6 implements an extensive set of sensitivity checks. Section 7 concludes the article.

2 Literature review

This paper first contributes to a growing body of research studying entrepreneurship and its determinants. One strand of this literature investigates how preferences and personality traits can affect the decision to become an entrepreneur (Levine & Rubinstein, 2017; Hurst & Pugsley, 2017; Chanda & Unel, 2021). A larger volume of research has investigated how outside factors can affect entrepreneurship. This literature has examined factors, including access to financial capital (Cagetti & De Nardi, 2006), business cycles (Levine & Rubinstein, 2018; Unel & Upton, 2023), family background and inheritance (Hurst & Lusardi, 2004), immigration (Fairlie & Meyer, 2003; Kerr & Kerr, 2020), tax policies and politics (Cullen & Gordon, 2007; Beland & Unel, 2019), and international trade and investment (Unel, 2018; Eren et al., 2019), among others.

In this literature, my paper contributes to studies investigating the nexus between entrepreneurship and immigration. Some studies have explored the entrepreneurial aspect of immigrants and how they contribute to economic prosperity and job creation. Fairlie and Loftstorm (2015), Kerr and Kerr (2020), and Azoulay et al. (2022), among others, have documented that business ownership is higher among immigrants than non-immigrants in the USA and many other developed countries. For example, Kerr and Kerr (2020), using the Survey of Business Owners in 2007 and 2012, show that immigrants create about 25% of new firms. Similarly, using administrative data for the US economy, Azoulay et al. (2022) investigate how often immigrants start companies and how many jobs these firms create. They find that immigrants act more as job creators than job takers, and the wages immigrant entrepreneurs pay are similar to those paid by native-founded firms. In exploring the underlying mechanism, Azoulay et al. (2022) argue that their findings are more consistent with immigrants being positively selected on entrepreneurial acumen. Other studies have identified factors (such as access to financial capital, home country business experience, ethnic enclaves) that explain immigrants’ decision to become entrepreneurs.²

A complementary set of studies have investigated how immigration has impacted entrepreneurship, especially among natives. Using the 1980 and 1990 decennial surveys across 132 metropolitan areas, Fairlie and Meyer (2003) examine the impact of immigration during the 1980s on the likelihood of becoming self-employed among US-born non-blacks.³ They find that immigration has large adverse and highly significant effects on self-employment, and the impact on males is more substantial. This paper differs from their study in several aspects, including the measurement of variables, sample size and period, and the econometric methodology used. For example, considering only a sample of large metropolitan areas during the 1980s may not fully capture the long-run impact of immigration on native entrepreneurship. In addition, they use the 1980 immigrant share as an instrument for the change in immigration share over 1980–1990. However, the initial immigrant shares are likely to be correlated with the subsequent economic shocks.

Some recent studies have investigated the impact by exploring different aspects of immigrants. For example, Duleep et al. (2021), using the CPS annual data over the 1996–2013 period, show that states with a higher share of skilled immigrants had experienced more subsequent entrepreneurship among natives. They estimate the effects using a simple OLS approach and thus do not address the endogeneity problem stemming from simultaneity and omitted variable bias. Ghimire (2021), using the availability of the State Children’s Health Insurance Program (SCHIP) for newly arrived immigrants as an instrument for the supply of immigrant entrepreneurs, finds that self-employed immigrants did not significantly impact native self-employment over 1997–2008. Since the SCHIP provides coverage only for children from low-income families, his analysis essentially exploits the impact of self-employed immigrants with low incomes.

This paper also relates to the extensive literature on immigration economics. Previous studies have investigated the implications of immigration for various issues (such as local-labor-market conditions, crime, schooling, innovation, productivity, politics), and the complete rendering of this literature is beyond the scope of this paper. A quick review of this literature reveals that there has been no consensus about the impact of immigration on the US economy. In his influential book, Borjas (2014) studies immigration and its effects on labor markets, and his overall message is that immigration poses severe challenges to local market conditions. Several other studies, however, provide a more positive assessment of immigration. For example, Card and Peri (2016) argue that Borjas’s book offers a one-sided view of immigration. Their survey concludes that immigration is mostly beneficial for host countries, and its adverse effects are limited and transient (see also Fairlie and Loftstorm 2015, Peri 2016).

3 Econometric specification

Let $E_{zt}$ denote the number of self-employed white Americans in the non-agricultural private sector in a commuting zone (CZ) z in year t, and $M_{zt}$ the number of immigrants. I estimate the effect of immigration on entrepreneurship using the following model:

$$\begin{aligned} \triangle e_{zt}=\beta \triangle m_{zt}+\gamma X_{z,t-1}+\eta _{rt}+\varepsilon _{zt}, \end{aligned}$$

(1)

where $\triangle e_{zt}=(E_{zt}-E_{zt-1})/N_{zt-1}$ is the decadal change in the number of entrepreneurs normalized by the start-of-period CZ working-age native population $N_{zt-1}.$ Similarly, $\triangle m_{zt}=(M_{zt}-M_{zt-1})/N_{zt-1}$ denotes the decadal change in immigration normalized again by $N_{zt-1}.$ Changes in entrepreneurs and immigration over the 2010–2018 period are multiplied with 10/8 to put all periods on a comparable decadal scale. The normalization controls for trends and variation in population across local-labor markets over time. Note that both entrepreneurship and immigration variables are normalized by the same variable, which makes the coefficient of interest ($\beta$) easy to interpret: a 1000 increase in immigrants increases the number of entrepreneurs by 1000 $\beta .$

Several remarks about the measurement of immigration are in order. First, some previous studies (Fairlie & Meyer, 2003; Borjas, 2014) have measured the impact of immigration by changes in the immigrant share, i.e., $\triangle m_{zt}=M_{zt}/L_{zt}-M_{zt-1}/L_{zt-1},$ where $L_{zt}$ is the total population. However, this approach is problematic because it confounds native supply changes with immigrant-supply shocks, as immigration may also affect population growth (Card & Peri, 2016). Second, assuming no out-migration, internal migration, or death of previous immigrants implies that $\triangle m_{zt}$ equals the number of new immigrants between $t-1$ and t (normalized by $N_{zt-1}$). Card (2001, 2009) and Jaeger et al. (2018) consider only the number of new immigrants in estimating the impact of immigration on local wages because previous immigrants’ location decision is more likely to be determined by local economic conditions. Following their lead, I also measure $\triangle m_{zt}$ as the number of new immigrants to CZ z between $t-1$ and t normalized by $N_{zt-1}.$ I will show that analysis based on this approach yields more consistent and robust estimates. However, I will also present results based on changes in the stock of immigrants. Finally, one may argue that not all immigrants would have the same impact on entrepreneurship. For example, it is plausible that immigrants from more entrepreneurial countries will have a more substantial impact. I will address this issue in Section 7 and show that results remain mostly the same.

The set $X_{z,t-1}$ represents the start-of-period CZ-level controls, including the percentages of ages 16 to 24 and 25 to 39; the percentages of whites, blacks, Hispanics, and the college-educated among the working-age natives; and the percentage of women employed, the percentage of workers performing routine tasks, and the average offshorability of tasks that workers perform. I include the latter two variables to control for the impact of automation and globalization on local-labor markets. The set also consists of the lagged shares of total employment in construction and mining, manufacturing, and the private service sectors to control shifts in sectoral demand.

I include $\eta _{rt}$ denotes region-by-period fixed effects to control trends in self-employment and common economic shocks.⁴ Regions here refer to nine geographic divisions defined by the US Census. Note that $\eta _{rt}$ is more general than using region and period fixed effects ($\eta _r$ and $\eta _t$) separately because $\eta _{rt}$ allows for common shocks to have a differential impact across regions. Finally, $\varepsilon _{zt}$ denotes the error term. In estimating Eq. (1), I cluster robust standard errors at the state level, which allows for the correlation of errors across CZs within states and yields more conservative statistical significance. I also weight regressions with the start-of-period CZ working-age US-born population to generate population estimates. Since variables are less precisely measured in small areas, weighting regressions with the CZ population also improves the precision of estimates.

Despite including several controls and considering only new immigrants, estimating Eq. (1) is subject to reverse causality and omitted variable bias. For example, new immigrants may still choose to live in CZs, where entrepreneurial activities are more common. To address this endogeneity problem, I use an instrumental variable (IV) approach. I construct the instrument utilizing the distribution of earlier immigrants at some base year (Altonji and Card 1991; Card 2009). Following Card (2001, 2009), I partition foreign-born individuals into 18 country groups where they are originally from.⁵ Let $\triangle M_t^c$ denote the total number of new immigrants from country group c to the United States between $t-1$ and t, and $s_{zt-2}^c$ the share of immigrants from country group c living in CZ z in $t-2$ (i.e., two decades ago). The instrument is defined as follows:

$$\begin{aligned} \triangle \widehat{m}_{zt}=\frac{1}{N_{zt-1}}\sum _{c} s_{zt-2}^c\triangle M_t^c. \end{aligned}$$

(2)

Note that $\triangle M_t^c$ does not depend on z.⁶ I use a base year sufficiently lagged to reduce concerns about the instrument’s correlation with the error term. As a robustness check, I will also consider an alternative formulation in which the base year remains the same at 1970.⁷

This shift-share instrument only uses the distribution of earlier immigrants by CZ, which was driven by their historical settlement across commuting zones. It then exploits the changes in the national population of immigrants by origin to generate variation likely exogenous to local conditions. Previous studies have shown that earlier immigrants from the same country strongly influence the locations of new immigrants and that economic factors have a relatively small effect on these decisions (Patel & Vella, 2013; Peri et al., 2015).⁸

4 Data

Data on employment are from the Census Bureau’s surveys, available on the Integrated Public Use Micro Samples (IPUMS) website (Ruggles et al., 2020). For my analysis, I use the 1970, 1980, 1990, and 2000 census decennial surveys and the 2009–2011 and 2017–2019 American Community Survey (ACS) files. To increase sample size and the precision of estimates, I pool the ACS 2009–2011 and 2017–2019 files and relabel them as 2010 and 2018, respectively.⁹

The geographic unit used in the analysis is the commuting zone (CZ), which has been increasingly used in the recent labor literature. Using CZs in the present study has two main advantages. First, each CZ represents a cluster of counties with strong commuting ties between employers and employees (Tolber and Sizer 1996). Second, they are the smallest geographic unit that can be consistently constructed over a long period and cover the entire USA. County groups and Public Use Microdata Areas (PUMAs) reported in the surveys are mapped to 722 CZs in the mainland USA.¹⁰

Each census survey includes several million individuals and records their class for their primary job, gender, race, birthplace, age, education, industry worked, occupation held, etc. The analysis includes all working-age individuals (ages 16–64), not residents of institutional group quarters. Further, in calculating self-employment statistics, I exclude those with imputed/missing employment status and worker class. About 0.2% of birthplaces are recorded as not specified, and I exclude these observations in constructing immigration variables. The study focuses on self-employment among US-born non-Hispanic whites working in non-agricultural private sectors. In my analysis, I refer to them simply as white Americans.

Census surveys classify individuals in the labor force as wage and salary workers or self-employed, with the latter classified as incorporated and unincorporated. Previous studies identified all self-employed individuals as entrepreneurs (Hamilton, 2000; Fairlie, 2014). However, Levine and Rubinstein (2017) show that incorporated and unincorporated self-employed individuals differ in several aspects. For example, incorporated self-employed business owners are more educated and earn more than comparable unincorporated self-employed and wage workers. They also tend to score higher on learning aptitude tests, exhibit greater self-esteem, and engage in more aggressive and risky activities when young. Therefore, my analysis distinguishes between these two groups.

Table 1 Summary statistics on worker class among white Americans, 1980–2018

Table 1 reports summary statistics on the class of workers among white Americans employed in the non-agricultural private sector over the 2000–2018 period. In this table, “All” covers both incorporated and unincorporated self-employment, and “College” represents individuals with at least some college education. The last row represents the share of each worker class in the working-age population, and note that the share of self-employment is about 6%. Observe that incorporated self-employed business owners are primarily males, older, and more educated. Further, consistent with Levine and Rubinstein (2017), they work longer and earn more than unincorporated self-employed and wage workers.¹¹

About 22% of self-employed individuals perform routine tasks, and not surprisingly, this ratio is lower among incorporated business owners.¹² As a comparison, the share of wage workers who perform routine tasks is about 30%. The last two rows show the fraction of each worker class working in the construction and private service sectors where self-employment is more prevalent. Around 18% of self-employed individuals work in the construction sector, whereas only 6.6% of wage workers employ in that sector. At least three-quarters of self-employed individuals work in the private service sector.

Fig. 1

Immigrants share in the working-age population (%). Notes: The data draw on the census decennial surveys and ACS Files from IPUMS (Ruggles et al., 2020), and cover all working-age (16–64) population

The surveys also record individuals’ birthplaces, citizenship status, and years of immigration. I use this information to construct the immigration data. Neither decennial surveys nor the ACS has a citizenship, green card, or visa requirement for inclusion into the sample universe; as a result, these surveys include undocumented immigrants as well. However, it is impossible to determine undocumented immigrants from the surveys’ records. The information about birthplaces and the year of immigration is crucial in constructing the instrument described in the previous section. Figure 1 shows the share of immigrants in the working-age population over the 1970–2018 period. The share was about 5.3% in 1970 and rose to 17.1% in 2018. Note that immigrant growth has slowed considerably after 2000.¹³

Self-employment is relatively more common among immigrants. About 6.9% of working-age immigrants are self-employed, and the share of incorporated self-employment is about 2.4%. However, self-employment rates show substantial variation across country groups. For example, self-employment is relatively less common among Mexicans, Caribbeans, and the Philippines (usually less than 5% of their working-age population) but more common among Europeans, Japanese, and Koreans (generally more than 10% of their working-age population). In this case, the composition of immigrants may play a crucial role in the impact of immigration on self-employment among white Americans. I will address this issue in Section 6.

5 Results

This section presents the effects of immigration on self-employment among US-born non-Hispanic whites using model Eq. (1) and discusses their implications. It also explores whether the impact has been different across demographic groups and time and compares the results with previous studies. But before presenting and discussing results, I will first discuss the instrument’s suitability by analyzing the first-stage regressions.

5.1 Quality of the instrument

To assess the predictive power of the instrument, I estimate the following model:

$$\begin{aligned} \triangle m_{zt}=\theta \triangle \widehat{m}_{zt}+\delta X_{z,t-1}+\alpha _{rt}+u_{zt}, \end{aligned}$$

(3)

As in the benchmark model Eq. (1), I weight regressions with the start-of-period CZ working-age US-born population and cluster standard errors at the state level. The coefficient of interest is $\theta$, which measures the instrument’s strength. For brevity, I do not show estimates on CZ-level controls in the table (see Tables 8 and 9 in the Appendix).

Table 2 Impact of immigration on self-employment in the US

Table 3 Impact of immigration on self-employment in the U.S

Table 2 shows the results from this exercise across different periods. In column 1, I estimate the relationship over the full sample, whereas, in other columns, I estimate it for each period separately. Panel A reports results when $\triangle m_{zt}$ is measured as the number of new immigrants between $t-1$ and t (normalized by the initial working-age native population), and $\triangle \widehat{m}_{zt}$ is the corresponding instrument defined by Eq. 2. Panel B presents results when $\triangle m_{zt}$ is measured as the change in the stock of immigrants (again normalized by the initial working-age native population), and the instrument uses the decadal change in the stock of immigrants (rather than the flow of immigrants). The last row in each panel shows the (Kleibergen-Paap) F-statistics, commonly used in the applied literature, to test the weakness of instruments. The conventional cutoff used in the literature is 10, i.e., the instrument is not weak if F-statistics is at least 10.

Panel A shows a strong relationship between $\triangle m_{zt}$ and its instrument $\triangle \widehat{m}_{zt}.$ Point estimates are all positive and statistically highly significant, suggesting that the predictive power of the instrument is considerably high.¹⁴ Results in panel B, however, suggest a less powerful instrument. Point estimates are markedly lower than those reported in panel A. Note that the instrument has a statistically less significant effect during the 1990–2000 period and no impact over 2000–2010. Indeed, the estimate over 2000–2010 has a wrong sign. Thus, the predictive power of the instrument is weak in these periods, as also indicated by the low F-statistics in the last row. For these reasons, I measure the immigration variable by the new immigrants in my econometric specifications Eqs. (1) and (2).

5.2 Main results

Table 3 reports the regression results using Eqs. (1) and (2). All regressions include region-by-period fixed effects and start-of-period CZ-level controls, as discussed in Section 3. For brevity, I do not report estimates on CZ-level controls here but present them in the Appendix in Table 8. As in Table 1, “All” represents any self-employed individuals (i.e., both incorporated and unincorporated). Columns 1–3 report the OLS estimates, and columns 4–6 are the IV estimates.

Estimates in columns 1–3 imply that exposure to immigration negatively impacts both incorporated and unincorporated self-employment among white Americans, but the effect is small and insignificant on incorporated business owners. For example, the point estimate in column 3 implies that a 1000 immigrant flow to a CZ would have decreased unincorporated self-employment there by 22. Since the unincorporated self-employment rate among immigrants is about 4.5%, this point estimate further implies that each unincorporated self-employed immigrant results in a decrease of about 0.5 ($\approx 0.022/0.045$) unincorporated self-employed white Americans. However, the OLS estimates are likely to be biased, as discussed in Section 3.

Columns 4–6 report results when I instrument $\triangle m_{zt}$ by the predicted inflow of new immigrants, $\triangle \widehat{m}_{zt}$, defined by Eq. 2. As shown in Table 2, the (Kleibergen-Paap) F-statistics from the first-stage estimates related to regressions is 24.8, well above the conventional cutoff of 10 used in the literature. The effects are larger and more precisely estimated than the OLS results. The point estimate in column 4 implies that a 1000 immigrant flow to a CZ would have decreased self-employment by 67. It further implies that each self-employed immigrant decreases by 0.97 ($\approx 67/69$) self-employed workers among white Americans—a sizable reduction in self-employment. Similarly, estimates in column 5 (6) imply that each self-employed immigrant results in a decrease of about 0.35 (0.62) incorporated (unincorporated) self-employed white Americans.

As mentioned in the introduction, immigration can affect native entrepreneurship in several ways. First, previous studies have documented that immigrants earn substantially lower than native workers even after controlling individual characteristics (Beland & Unel, 2018). Cheaper labor provided by immigrants makes businesses more profitable, encouraging more entrepreneurship. Second, immigrants demand goods and services that may lead to new business opportunities for natives. Third, several studies have shown that immigrants are highly entrepreneurs (Kerr & Kerr, 2020; Fairlie & Lofstorm, 2015). Using the monthly CPS data, Fairlie and Loftstorm (2015) show that the share of business startup activity generated by immigrants is substantially higher than the immigrant share of the workforce (24.9% vs 15.6%). They also find that immigrant-owned businesses have lower average sales and incomes than non-immigrant firms. If immigrant and native entrepreneurs concentrate in complementary sectors, the former will most likely have a positive impact on the latter ones. However, if they are in the same sectors, the increased competition created by immigrant entrepreneurs will have a crowding-out effect on native entrepreneurs. The findings in Table 3 suggest that the adverse impact of immigration has dominated its positive effects.

5.3 Effects across demographic groups

This section examines the effects of immigrants on self-employment among different demographic groups based on gender, education, and age. The dependent variable in Eq. 1 now represents the change in the number of entrepreneurs in a subgroup normalized by the start-of-period CZ working-age native population. In this case, the sum of estimated coefficients across subgroups equals the corresponding point estimate on $\triangle m_{zt}$ in Table 3.

Panel A reports the effects of immigration on self-employment among females and males. Estimates are all negative and statistically highly significant. For example, the point estimate in column 1 implies that a 1000 immigrant flow to a CZ would have decreased self-employment among females and males by 25 and 42, respectively. Thus, each self-employed immigrant results in a decrease of 0.36 (0.65) self-employed female (male) workers among white Americans. Note that the decline in female self-employment constitutes about 37% ($\approx 25/67$) of the reduction in total self-employment. Since the share of females in self-employment is 36.5% (as shown in Table 1), the findings in panel A further imply that the impact of immigration on female and male self-employment is mostly the same.

Table 4 Impact of immigration on self-employment across groups and periods

These findings differ from Fairlie and Meyer (2003), who investigate the relationship between changes in immigration and native self-employment across 132 of the largest metropolitan areas (MAs) over the 1980–1990 decade. They find that immigration decreases self-employment among non-black natives, and the impact is more significant among males. Specifically, their IV estimates imply an additional self-employed immigrant displaces about 0.60 self-employed natives, and 0.17 of them are women (see column 6 of Table 4 in their paper). Again, these are lower than my findings.

Geographic areas, population, years, and methodology used in this paper are different from Fairlie and Meyer (2003).¹⁵ For example, they use the 1980 immigrant share as an instrument for the change in its value from 1980 to 1990. However, the initial share is likely to be correlated with the error term. In addition, they use changes in unemployment, population, and average income as controls. However, these are outcome variables (e.g., immigration also affects unemployment and income) and are considered bad controls (Angrist & Pischke, 2009).

Panel B in Table 4 reports the impact of immigration on self-employment among younger (ages 16 to 39) and older (ages 40 to 64) adults. As in the previous panel, all estimates are negative and statistically significant. Note that the point estimate in column 1 is about 32% of that in column 4 in Table 3. In the data, younger adults constitute about 30% of the total self-employed individuals. Consequently, estimates imply that immigration has mostly the same impact on self-employment across the two age groups. Estimates in columns 2 and 3 suggest that immigration has a slightly larger (smaller) impact on incorporated (unincorporated) self-employed young adults (since the fraction of incorporated self-employed individuals who are younger adults is about 25.6%).

Panel C reports the impact of immigration on self-employment across different education groups. College-educated refers to individuals who have at least some college education. Effects are negative and statistically significant at the 1% level. The point estimate in column 1 is about 38% of that in column 4 in Table 3. In the sample, about 32.5% of self-employed individuals do not have any college education (see Table 1), which implies that immigration has a slightly more negative impact on self-employment among less educated workers. Estimates in other columns, combined with the averages in Table 1, imply that differences in the effects of immigration on each self-employment across these education groups are small.

5.4 Effects across CZs and periods

I now turn to examine whether the effects differ across CZs and periods. Immigrants have tended to live in metropolitan areas (MAs), and one may wonder whether the effects differ in MAs. Using the 1990 census classification of MAs, I divide the sample of CZs into two groups: CZs that include some MAs and CZs that do not include any MAs.¹⁶ I call these groups metropolitan CZs and rural CZs, respectively. I choose 1990 because the CZs used in my analysis are based on the 1990 Census CZ definition and fix the year to make the panel balanced. The metropolitan (rural) group includes 273 (449) CZs, and Table 5 reports results. The first-stage F-statistics associated with columns 1–3 and 4–6 are 21.7 and 18.4, respectively.

Table 5 Impact of immigration on self-employment across CZs

Table 6 Impact of immigration on self-employment across periods

Panel A in Table 5 indicates that immigration harmed both self-employment groups in metropolitan CZs. Point estimates are highly significant and close to benchmark results. Panel B reports the effects on self-employment in rural CZs. Note that estimates have the same signs as in panel A. A comparison with benchmark results indicates that estimates are larger and less precisely estimated. For instance, the impact on incorporated self-employment is statistically insignificant. Large standard errors most likely stem from measurement errors because data from smaller areas are usually less precisely measured.

How have effects changed over the years? Panels A and B in Table 6 present the results when I split the sample into two periods, 1980–2000 and 2000–2010, and estimate the relationship separately for each period. The first-stage F-statistics associated with regressions in columns 1–3 (4–6) are 13.5 (21.5). Estimates imply that immigration on self-employed white Americans had a more substantial negative impact during the first period: a 1000 immigrant flow to a CZ during the first (second) period would have decreased unincorporated self-employment there by 92 and 41. The effect difference is statistically significant and mainly stems from the impact on unincorporated self-employment. Although the estimate in column 2 is larger than in column 5, they are not statistically different.

Table 7 Impact of immigration on self-employment, robustness

One may wonder why the effects became smaller during 2000–2018. To address this question, I estimate the model for each period separately, as shown in panels C–F in the table. Note that the magnitudes of estimates are not monotonically decreasing over the periods. In addition, immigration had a substantially less negative impact on native entrepreneurship during the 2000–2010 period. This latter finding may be related to the Great Recession, which created a severe credit crunch for businesses, especially small ones. However, financial constraints are likely to be more severe for minorities (including immigrants) and their small businesses because of discrimination in financial markets. For example, using data from the 2003 Survey of Small Business Finances, Asiedu et al. (2012) find that the denial rate in a sample of loan applications is higher for minority-owned firms than for firms owned by white males: 37% vs 8.8%. In addition, minority-owned firms paid a higher interest rate on approved loans. The discrimination in credit markets would create an advantage for white Americans relative to other groups. Consequently, there would be less competitive pressure from immigrants on native entrepreneurs.

6 Sensitivity analysis

This section presents a series of exercises to check the sensitivity of results to alternative measures of immigration and the choice of controls. Table 7 shows the results, and as in the previous tables, I do not show estimates on controls.

As discussed in Section 3, in measuring $\triangle m_t$, immigrants from different countries are treated equally. However, entrepreneurship is more common among some country groups (e.g., Europeans, Koreans) than others (e.g., Mexicans, Africans). Therefore, areas more exposed to immigrants from more entrepreneurial countries may experience different effects. To address this concern, following Fairlie and Meyer (2003), I weigh the number of new immigrants from a country with the self-employment share in that country. More precisely, let $\mu ^c$ denote the average share of self-employment among immigrants from country c; I define the weighted number of new immigrants as follows:

$$\begin{aligned} M_{zt}^w=\psi _t\sum _{c\ne US}\mu ^c M_{zt}^c, \end{aligned}$$

(4)

where $M_{zt}^c$ is the number of new immigrants from country c to CZ z between $t-1$ and t.¹⁷ Here, $\psi _t$ is a scale factor that the total number of weighted new immigrants and the total number of actual new immigrants (i.e., $\sum _{z} M_{zt}^w=\sum _{z}M_{zt}=M_t$). Using the above definition, it then follows that $\triangle m_{zt}^w=M_{zt}^w/N_{zt-1}.$ I follow the same approach in constructing the instrument.

The above approach reallocates immigrants so that immigrants from high self-employment countries weigh more than those from low self-employment countries. Panel A reports the results of this exercise. The first-stage F-statistic is 26.7, indicating that the instrument is still powerful. Note that point estimates are very close to the benchmark results reported in panel B of Table 3. However, the impact is a bit lower than those implied by the benchmark results because the share of self-employment is higher under this weighted sample (7.5% vs 6.9% in the original sample). Specifically, each self-employed immigrant individual now results in a decrease of 0.84 ($\approx 63/75$) self-employed workers among white Americans.

In the sample, Mexicans constitute a substantial fraction of immigrants. In addition, during the 1980s and 1990s, there was a large influx of immigrants from Mexico. The uneven distribution of countries may affect the estimates. To address this issue, I estimate the model Eq. (1) by excluding all new immigrants from Mexico. The results are presented in panel B. Note that point estimates declined slightly, indicating that Mexicans were not the driving factor behind the results.

In constructing the instrument, I allocated the national change in the number of immigrants from each country of origin across CZs using the distribution of earlier immigrants at time $t-2$ as weights. Panel C presents results when the distribution of earlier immigrants does not change over time. In particular, I only use the 1970 distribution of immigrants across CZs when calculating Eq. (2). Panel B reports the results and notes that the effects are considerably higher than the benchmark case (cf. panel B in Table 4).¹⁸ However, the overall conclusion does not change.

Panel D reports results when I measure $\triangle m_{zt}$ in Eq. 1 by the change in the stock of immigrants normalized by the start-of-period native population. The impact of immigration on self-employment gets substantially higher. Each self-employed immigrant now results in a reduction of 1.53 ($\approx 0.106/0.69$) self-employed workers among white Americans. However, this approach does not yield consistent results if one looks at the effects in different periods. Table 11 in the Appendix reports results for each subperiod in the sample. Note that, consistent with results in Table 2, impact fluctuates widely across the periods. In addition, estimates are noisy (i.e., having large standard errors).

In estimating Eq. (1), I include region-by-period fixed effects, allowing common shocks to have differential impacts on regions. I now estimate the equation by replacing these fixed effects with state-by-period fixed effects, and the results are shown in panel E. I did not include state-by-period effects in my benchmark specification because too many fixed effects might create an over-controlling problem.¹⁹ However, estimates in panel E show that results are not too sensitive to the fixed effects included in the model.

Finally, panel F reports regression results when observations are not weighted by the CZ population. Following previous studies (Card, 2009; Autor et al., 2013), I weighted regressions by population to generate population-average causal effects and increase efficiency (i.e., making estimates more precise). However, Solon et al. (2015) argue that weighting by population may not help in these dimensions, and researchers should be careful in interpreting their results if weighted and unweighted regression results differ significantly. Estimates in this panel show that immigration still hurts native entrepreneurship, and the results are broadly consistent with those in Table 3. However, note that the effects are less precisely estimated, suggesting that weighting by population improves efficiency.

Table 8 First-stage regressions based on new immigrants

7 Concluding remarks

Immigration has been a hotly debated issue among economists and policymakers in the USA and other wealthy countries because it has generated a variety of benefits as well as challenges to the host economies. This paper contributes to the extensive literature on immigration economics by studying the impact of immigration on native entrepreneurship in the United States. Specifically, using survey data from the US Census across commuting zones from 1980 to 2018, I investigate the causal effects of immigration on self-employment among US-born non-Hispanic whites. In addition, I examine the impact on incorporated and unincorporated self-employment separately and utilize an instrumental variable approach to address the possible endogeneity problem.

I find that immigration has had a sizable negative impact on self-employment among US-born non-Hispanic whites over the last forty years. I also explore heterogeneity across different demographic groups based on age, gender, and education and find that its effects on self-employment are similar across these groups. These findings contribute to the literature on immigration by showing that immigration may pose challenges to native entrepreneurship in local economies.

Table 9 First-stage regressions based on changes in stock of immigrants

The present work can be extended in several ways. First, the current analysis only considered non-Hispanic white Americans. The excluded population includes other distinct races such as Black/African American, American Indian, Chinese, and South Asians. For example, entrepreneurship is lower among African Americans and Latinos but relatively higher among East and South Asians (Fairlie & Meyer, 2003). Investigating how immigration has affected each group will be an interesting extension to understanding their business and labor-market dynamics. Second, using the current econometric framework, one can also analyze the impact on native business owners’ income. Unfortunately, the decennial surveys and ACSs are unsuitable for addressing this problem because data on business income are pretty noisy. For example, several unincorporated self-employed individuals report zero business income. For this purpose, using the Survey of Business Owners data would be a better approach. Finally, immigrants come from different countries and have diverse skills. Exploring heterogeneity across immigrants based on their skills and country of origin (e.g., from developed or less-developed countries) may reveal much richer results. It can also shed light on better understanding the underlying mechanism that derives the negative impact of immigration on native entrepreneurship.

Appendix

 

Table 10 Impact of immigration on self-employment, 1980–2018

Table 11 Impact of immigration on self-employment, changes in stock of immigrants