Postbellum electoral politics in California and the genesis of the Chinese exclusion act of 1882

Abstract

Why did the United States move from having nearly open borders from the 1840s to the 1870s to passing the Chinese Exclusion Act in 1882, the first law in American history to ban people from entering the United States solely based on race? We argue that the standard story of nativist backlash based on wage pressure explains the demand for immigration restrictions, but not their timing or their racial focus. The demand for immigration restrictions was largely inchoate until the political restructuring that followed the Civil War. Finding themselves uncompetitive in much of the country, the Democrats seized on immigration restrictions, most notably in growing California, as a wedge issue. Chinese residents were unable to vote, thus making restrictions on Chinese entry an especially effective strategy in political economy.

1 Introduction

The topic of immigration in postbellum America has been one of longstanding interest among both historians and economic historians (Abramitzky et al., 2012, 2013, 2014; Abramitzky & Boustan, 2017; Chen, 2015; Dunlevy & Hutchinson, 1999; Tabellini, 2020). This is for two main reasons. First, the volume of migrants was historically unprecedented-more than 22 million foreigners arrived between 1865 and 1910 (Bohanon and Van Cott, 2005, p. 529). Second, there was a strong nativist backlash against immigrants that culminated in a series of legislative actions, first against Asian immigrants (in the 1880 s) and later against Eastern Europeans (from the 1890 s onward) (Timmer & Williams, 1998). To explain these backlashes, economists and historians have tended to emphasize how native workers perceived the effects of immigration on their wages (Goldin, 1994) or cultural difficulties in managing the integration of large numbers of immigrants (Tabellini, 2020). These non-mutually exclusive explanations are seen as the main incentives behind the political mobilization to reduce immigration (Wong et al., 2018).

However, there is a problem with this narrative. Large numbers of immigrants were entering the United States as early as the 1840s. The perception of their negative effects on wages was present as early as the 1850s (Margo 2000, p. 144). The nativist backlash was immediate (Alsan et al., 2020; Cohn, 2000; Fogel, 1992; Ferrie, 1999), but it never translated into any form of federal anti-immigration policy until the 1880 s. All of the policies adopted were at the state-level. These were often overturned by courts (Kanazawa, 2005, p. 787) and even the most extreme policies were relatively modest in both scope and effect. Most importantly, there are close to four decades separating the beginning of a nativist backlash and the adoption of anti-immigration rules in 1882 (Fong and Markham, 1991, 2002; Lee, 2002; Kanazawa, 2005; Seo, 2011; Peng, 2021). The Chinese Exclusion Act of 1882 was “the first law in American history to ban any group of people from entering America solely based upon race” (Seo, 2011, p. 826). The delay becomes even more puzzling when one considers the 1868 Burlingame Treaty between the United States and Qing China, which sought to ease limits on Chinese immigration and protect the rights of already-established Chinese. This suggests that there was still strong opposition to immigration control in the late 1860 s. The adoption of the Chinese Exclusion Act a scant fourteen years later entails a remarkably fast flip.

Why such a long lag between backlash and policy action? Why such a rapid reversal in policy? Our answer is that the Civil War altered political incentives enough to make immigration a potential wedge issue in swing states.

Following the end of the Civil War, the Democrats were stuck with a major electoral disadvantage. States that had contributed large contingents of Union soldiers were unlikely to vote for the Democrats-a party associated with the initiation of the war. As a result, Democrats perceived a large block of demographically important states as being unwinnable (either because they were Republican strongholds before the war or because the war’s burden alienated voters against Democrats). Michigan, Pennsylvania, Ohio, Wisconsin, Illinois, Maine, New Hampshire, Vermont, Connecticut, Rhode Island, and Massachusetts were Republican strongholds (and remained so until the 1890s) that amounted to roughly a third of the Electoral College. In each election, Republicans would wave “the bloody shirt” (Arrington, 2020; Luthin, 1960) evoking the blood of Union soldiers spilled because of the Democrats who had favored secession. This had great electoral costs for the Democrats. The only way forward was to peel off a few states such as California, Indiana, New Jersey, and New York (Arrington, 2020).¹ Anti-immigration policies offered them a means to peel off these states from the Republican camp and negate the Civil War-induced disadvantage.

California was particularly appealing to Democrats for three reasons. First, the state was growing rapidly (its number of electoral college votes doubled from four to eight between the 1860 and 1884 elections). Its importance was large enough to overturn the 1876 election and tighten significantly the 1868 election. Second, the main foreign-born population of the state-Asian immigrants mostly from China (roughly 10% of California’s population)-was the object of intense hostility from native whites. Third, because they were not white, Asian residents could not vote. This limited the risk of alienating a portion of the state’s electorate. In other states where immigrants disproportionately came from Europe, adopting a strong nativist position meant alienating voters. Asian voters were a safer target. Soon after the Civil War ended (as early as 1868), the Democrats had already put forward anti-Chinese (and what they called Mongolians) proposals in an attempt to compete in California to great electoral success. This transformed California into a swing state (Seo, 2011, p. 825). Realizing that the Democrats were making headway, Republicans quickly followed suit in the 1870s. By the 1880 election, both parties had developed strong anti-Chinese planks that were virtually indistinguishable from one another.

Using county-level electoral results in California for 1868, 1872, 1876, and 1880 presidential elections, we find that larger Chinese/White ratios made a county more electorally competitive in all elections.² These effects did not exist for pre-Civil War elections. We also find that the effect of ratios on margins of victory became smaller over time. This is consistent with the fact that Republicans started emulating the Democrats in their anti-Chinese policies so as to negate their electoral competitiveness. Altering our results to control for the Chinese/White ratios of neighboring counties, our results are similar, and it is likely that by 1880, the Republicans no longer suffered any electoral penalty.We also replicate our empirical strategy on pre-Civil War elections and post-1882 (i.e., after the adoption of the Chinese Exclusion Act) elections. Our theory suggests that, prior to the Civil War, the Chinese/White ratio should be at least not different from zero. We find that this is the case which lends further credibility to our claim that the War rearranged political incentives in ways that ultimately led to the adoption of the Chinese Exclusion Act. Similarly, once the Act was adopted, we should fail to find any effects on electoral competitiveness. Using the 1884 and 1888 elections, we find that the Chinese/White ratio does not predict competitiveness after the Act’s adoption. We also replicated our results using eight different panel combinations. The reason for the use of multiple combinations is motivated by the fact that our control variables are taken from three censuses (1860, 1870, and 1880), which are associated with the nearest election. As such, no panel can have more than three time periods. However, because there is more than one election associated with each census, there are multiple combinations of three time periods that we can employ. In the present case, there are eight combinations. This is a serious limitation to the credibility of a panel specification. However, we can use these combinations to check if they yield conceptually similar results to the consecutive cross-sections from our baseline specification. We use an interaction effect between a postwar year dummy and the Chinese/White ratios. A negative coefficient on this term implies that a higher Chinese/White ratio after the Civil War generates greater electoral competitiveness. Moreover, we can employ an interaction between fixed-year effects and the Chinese/White ratios (without including the postbellum year dummy) to see if the coefficients fall over time. Both of these alternative strategies (across all eight panel combinations) yield results consistent (but weaker) with those obtained with consecutive cross-sections. Finally, we enact some alterations to our specifications that are meant to determine whether there are endogeneity issues that may generate inconsistent and biased estimators (notably by using shift-shares instruments). All these alterations suggest that endogeneity issues do not modify our results. We also use complementary evidence from votes in Congress to document the shift in political behavior from Democrats and Republicans over time.

We argue that combining these pieces of evidence supports our narrative. The Civil War reconfigured political incentives in ways that created bipartisan support for anti-immigration policies in general and anti-Chinese policies in particular. As such, national-level nativist policies could be adopted even though nativist hostility to migrants had been present with few policy consequences for more than four decades.

2 Nativist backlash, the civil war and policy actions

Immigration to America picked up noticeably during the 1840s and 1850s (Cohn, 2000; Davis et al., 1972; Margo, 2000). In states like New York and Massachusetts, an almost immediate nativist backlash was directed at Irish and German immigrants, while states such as California directed the backlash at the Chinese (Fong & Markham, 1991, 2002; Fisher & Fisher, 2001). Generally, this took the form of sporadic outbreaks of mob violence and harassment against immigrants. However, translating this backlash into policy was much more difficult. Actions prior to 1880 were confined to the state level and were of limited scope and effect. Only in the 1870s did federal legislation become politically feasible, and only in 1882 was a strong piece of anti-immigration legislation-the Chinese Exclusion Act-adopted.

The backlash of the 1840s and 1850s is quite illustrative in terms of understanding the lag between the backlash and policy action. This period is marked by the rise of the Know-Nothing Party in the 1850s (Alsan et al., 2020), which, by 1855, was able to control virtually every New England state and secure large delegations in Congress (Cohn, 2000, pp. 361–362, 374). Unsurprisingly, such electoral successes led to the adoption of state-level anti-immigration policies. However, such laws had only minor impacts.³ The most extreme example was that of Massachusetts. The state deported foreigners who sought help from almshouses (Hirota, 2016, p. 1). This was mostly aimed at Irish immigrants. However, only 715 of the 33,436 Irish passengers who arrived in Massachusetts between 1837 and 1845 were deported back to Ireland (Hirota, 2016, p. 215) although larger numbers were deported to other American states. In the 1850 s, Massachusetts adopted even stricter deportation criteria so that, from 1851 to 1863, some 15,438 persons were deported back to England, Canada, or other US states (Hirota, 2016, p. 216–217). Yet, this pales in comparison with population increases recorded in the census. The average annual increase in the foreign-born population in Massachusetts between 1850 (Census Office, 1853, p. 53) and 1860 (Census Office, 1864, p. 226) was 9608 people, whereas the average annual number of deportations was 1287 (roughly 13%). Moreover, Massachusetts was an outlier in that it was probably the most aggressive state.⁴ Other states adopted quite mild policies that failed to come close to matching the nativist policies of Massachusetts.⁵ With milder (or no) policies in other states, immigrants could (and did) change their ultimate destination when migrating (Cohn, 2000, p. 377). Numerous states, notably those of the Midwest, saw immigration as a way to increase their populations and thus their influence in Congress (Hirota, 2016, p. 186). As such, they were unlikely to enact any anti-immigration policies. Without federal restrictions on immigration-through changes to the Naturalization Act of 1790-the ability of states to enact strong restrictions on immigration was limited, and so was their capacity to curb the actual numbers.⁶

One reason for the lag in anti-immigration policy action by federal politicians is that states like California were not initially tremendously important (only 4 out of 296 electoral college votes in 1856). However, that is not sufficient since states such as New York and Massachusetts-with large anti-immigrant sentiment-weighed quite heavily. The other necessary detail relates to the fact that the Democrats had a unified voting bloc in the slave-holding southern states. The non-slave states can be broken into two groups: coastal (e.g., New York, Massachusetts) and non-coastal (Iowa, Wisconsin). The latter group exhibited weaker anti-immigration sentiment, notably because immigration allowed them to grow demographically (which secured more votes in the Electoral College and seats in Congress). Putting too much weight on immigration topics could divide the northern states and allow Democrats to sneak in (as was the case in Illinois and New Jersey in the presidential election of 1856).⁷ This helps to explain why the Know-Nothing Party disintegrated rapidly soon after its electoral successes of the 1850 s, despite the popularity of its anti-immigration positions within the North.

The Civil War altered this dynamic entirely.⁸ First, the Civil War’s burden solidified the Republican coalition in the North as veterans and their families were solidly opposed to the Democratic party. This made it hard for the Democrats to break through in states such as Illinois, Pennsylvania, and Ohio, which had sent some of the largest contingents of troops for the Union cause (Dyer, 1908, p. 11–13).⁹. Second, in the 1868 election, there were still some states (Texas, Mississippi, Virginia) in the South that could not participate in the election. Third, the politics of the Reconstruction era meant that, until after the election of 1876, many pro-Reconstruction Republicans were in power in many Southern states (Egerton, 2014). With their electoral base weakened for most elections until 1880 and with a more monolithic pro-Republican voting bloc in the North, the Democrats needed to find ways to compete in presidential elections by catering to marginal states outside the South.

The issue that made them competitive in key states was the advancement of anti-immigration policies (Seo, 2011). This can be seen in the fact that the Democrats frequently nominated politicians from New York-a state that was a hotbed of anti-immigration feeling and which frequently supported Democrats during the post-war years-for president.¹⁰ However, New York was far from sufficient. Inroads into the midwestern and western states were necessary. Pushing for anti-Chinese policies was one key way the Democrats found to extend to other states, such as California. One historian argues that Southern Democrats explicitly picked the Chinese question to create a West-South alliance “dedicated to white supremacy and defeat of Northeastern radicalism” (Tichenor, 2009, p. 104). The Democratic press invested considerable efforts at raising the salience of racial issues tied with the Chinese.¹¹ Early efforts included media tropes against Chinese laundries (Bernstein, 1999), restaurants (Chin & Ormonde, 2017), women (notably in relation to prostitution) (Matsubara, 2003) and the effect of Chinese workers on the wages of White workers (Locklear, 1960; Peng, 2021).

Initially, Republicans “did not want to stray too far from the party’s records on civil rights and equal citizenship” (Seo, 2011, p. 826). As a result, Chinese immigrants did “have their allies in the West" (Hall, 2013, p. 42) during the 1860s and early 1870s. However, Republicans were also “well aware of the need to court crucial swing voters in the West if they hoped to keep the highly rewarding presidential office under their control” (Seo, 2011, p. 826). As “white political opinion solidified against the Chinese", Republicans felt that the issue threatened to split the party and allow Democrats to win (Hall, 2013, p. 42).¹² As such, they gradually shifted towards anti-immigration positions in general and-in the case of California-anti-Chinese positions in particular. Republicans participated in the 1876 California Special Committee on Chinese Immigration, which provided strong support for anti-Chinese policies (Matsubara, 2003, fn. 6). In that same year, California Republicans successfully pushed for the adoption of an anti-Chinese plank (Torres-Spelliscy, 2021, pp. 1234-35).

By the early 1880 s, anti-immigration policies had achieved bipartisan support. From there, it became increasingly possible for federal-level policies to be enacted and implemented such that nativist feelings could finally begin to translate into policy actions.

3 Empirical strategy and data

Our argument above essentially predicts that the Democrats moved first on anti-immigration policies and that, upon realizing the electoral threat, the Republicans followed later. The problem with this argument is that white immigrants could vote, which limited the gains from advancing nativist rhetoric. This is why we focus on California.

First, well before the Civil War, the state had a particularly strong nativist backlash against Asian immigrants (who were labeled as Chinese).¹³ This entailed that there was a strong demand for policies targeted against the Chinese community. As the Chinese were an out-group, attempts to tie political opponents with them were a means to sway voters (Glaeser, 2005). As we pointed out above, it was initially the Democratic press that supplied hate-creating stories against the Chinese during the postbellum period. The objective was to tie Republicans with the out-group (Hall, 2013).¹⁴ Republicans had little incentive to generate counter-stories to fight back since the vast majority of the population held such views. As a result, Republicans rapidly shifted to producing hate-creating stories against the Chinese. The only distinction was that they tried to tie these stories to anti-Democratic rhetoric.

Second, the state was also increasingly important in terms of electoral college votes. By the 1868 election, in which Republicans won by a razor-thin margin of 520 votes (0.48% of the votes), it had become a swing state. In fact, its swing state status could have determined the presidential outcome in 1876. In that year, it handed its six electoral college votes over a 1.8% margin to the Republican candidate, Rutherford B. Hayes, who had won the election with 185 electoral college votes against 184 for his opponent. California switching sides would have cost Hayes the victory.

Third, the state fits the pre-war narrative described in Sect. 2. Peng (2021) and Kanazawa (2005) note that anti-Chinese feelings did not translate into policy actions for many years. A part of the lag, Peng (2021) argues, is explained by changes in economic circumstances.¹⁵ Another part of the lag is due to the fact that many state-level policies, such as an 1858 exclusion act, were ruled unconstitutional (Kanazawa, 2005, p. 787).¹⁶ As such, California experienced the same lag between the backlash and effective policy responses as elsewhere in the country.

Fourth, and most importantly, Asian immigrants could not vote. Under the Naturalization Act of 1790, only free white men were eligible for citizenship and suffrage. This meant that politicians who seized the opportunity to capitalize early on the anti-Chinese feeling would not face a major penalty in other states. In states such as New York, large cohorts of Catholic immigrants from Quebec, Ireland, and Southern Europe had caused nativist backlashes in the 1840s and 1880s (Hirota, 2014, 2020). However, this backlash was self-limiting because white immigrants could acquire citizenship and vote. As such, more extreme proposals were too risky for viable candidates for office to endorse (Hirota, 2020). California’s unique setting excluded that counterweight. In other words, the cost of supplying discriminatory policy proposals to voters was limited (Glaeser, 2005). Moreover, it limited the ability of Asian immigrants to engage in political entrepreneurship (Hirota, 2020, 12).¹⁷

California is thus an ideal case for the study of the shift due to the Civil War that permitted the adoption of federal-level policies aimed at curbing immigration, a shift that began with the Chinese Exclusion Act of 1882.

We look at presidential election results in Californian counties between 1868 and 1880 using the following econometric specification:

$$\begin{aligned} \begin{aligned} |margin|_i&= {Chinese/White}_{i}^{'} {\delta } + {x}_{i}^{'}{\beta } + \epsilon _i \\ \end{aligned} \end{aligned}$$

(1)

where the absolute value of the margin between Republicans and Democrats is our measure of electoral competitiveness. Using the absolute value allows us to disregard which party wins in a given county and concentrate exclusively on how competitive a county is, as lower values indicate tighter elections. The county-level results are taken from Burnham (1955). Our coefficient of interest is $\delta$, which measures the effect of the ratio of the Chinese to the White population in a county i on the competitiveness in that county. The ratio is built using the data from the Census Office (1883a, p. 382). Our assumption is that counties with a large Chinese population relative to the white population yielded closer contests as they were key to shifting the state’s allegiance in presidential elections. The vector $\textbf{x}_i$ contains relevant control variables that we use: the log of population, the log of manufacturing wages, and either the log of manufacturing output per capita or the ratio of the manufacturing to farming sectors. These variables are meant to control for other socio-economic determinants of competitiveness and are constructed from volumes 2 and 3 of the Census of 1880 and volume 3 from the census of 1870 (Office, 1872, 1883b, c). The descriptive statistics are below in Table 1.¹⁸

The presidential elections occurred in 1868, 1872, 1876 and 1880. Only one of these years was the election in the same year as the census data we have. This has one important consequence in terms of our methodology. We have to match the dependent variable for the elections with the closest census year. For example, the 1868 and 1872 election results will rely on values from the 1870 census, while the 1876 and 1880 results will rely on values from the 1880 census. This means we cannot employ a panel setup because there is more time variation in election years than in census years. As such, we rely on separate ordinary least squares specifications (i.e., one for each election—which we refer to as consecutive cross-sections).

Table 1 Descriptive statistics

Before we proceed, we should also point out a key prediction of our narrative in Sect. 2: we expect the effect of the Chinese-to-white ratio (measured by the coefficient $\delta$) to fade over time.¹⁹ This is because the Democrats in 1868 were the first to adopt anti-Chinese policy. The Republicans followed after. Prior to the 1870s, the Republicans tended to be more divided on anti-Chinese policies (Seo, 2011). This was true even in state-level politics as early as the 1850 s (Stanley, 1973, pp. 82–83, 85). By the 1860s, Republicans still opposed some anti-Chinese policies at both the state and federal levels (Stanley, 1973, p. 213). For example, in 1862, the control of the state legislature shifted to the hands of the Republicans. A joint committee of the legislature was appointed to investigate the Chinese issue and made a favorable report because of the tax revenue the Chinese generated (Coolidge, 1909, p. 62). Moreover, Leland Stanford, who was the eighth governor of California and a Republican, stated that tax bills targeting the Chinese were stringent and oppressive in 1862 (Coolidge, 1909, p. 72). Republican Convention also dealt with the Chinese issue in an evasive way (Eaves, 1910, p. 128). This was in line with the post-war political ascendancy of the Radical Republicans who had pushed for the Civil Rights Act of 1866 and the 14th Amendment to the United States Constitution in 1868 (Fisher and Fisher, 2001).

In contrast, Democrats were more aggressive on the Chinese issue. In 1860, the Democratic State Central Committee of California stated that “it is the imperative duty of our state legislature to exercise all its constitutional power to prevent the further immigration of coolies or Chinese of our state” (Davis, 1893, p. 122). Henry Huntly Haight who was elected governor in 1867 with a strong majority, had consistently promoted strong anti-Chinese policies (Stanley, 1973, p. 214). Specifically, Haight called upon Congress to protect the western coastal states from the Chinese and advocate the legislature to keep the Chinese out (Sandmeyer, 1991, p. 46). According to Stanley (1973), the late 1860s-marked by a near Democratic victory in the state during the presidential election of 1868 and an 11-point victory by Haight in the 1867 gubernatorial election-constitute the reversal point for Republicans. At both the federal and state levels, Republicans began adopting increasingly anti-Chinese policies. Their policy planks rapidly changed to adopt anti-Chinese proposals, so much so that by 1871, “racial issues no longer distinguished Republicans from Democrats” in state politics (Stanley, 1973, p. 215). Indeed, after being defeated in 1867 against Haight, who had warned that Chinese suffrage was imminent due to the Republicans’ commitment to Reconstruction, the Republicans shifted their message entirely. They sought the endorsement of the anti-Chinese association of the Industrial Reformers (with more than 10,000 members) by stating in its platform that the employment of the Chinese was “offensive to the exalted idea of the dignity of labor and detrimental to the prosperity of the American laboring classes" (Aarim-Heriot & Daniels, 2003, p. 163). They also stated that ‘Chinese laborers were incapable of assimilation with our race" (Aarim-Heriot & Daniels, 2003, p. 163). This allowed Republican Newton Booth to win the 1871 election against the anti-Chinese Haigt. By 1877, the same could be said to apply to federal politics as both parties adopted highly similar policies (Fisher & Fisher, 2001, p. 59).²⁰ Because the Democrats were faster to jump on the anti-Chinese policy bandwagon, we expect that larger Chinese-to-White ratios would matter more in the elections before Republicans began adopting more anti-Chinese policies. As Republicans became more anti-Chinese, we expect the effect on electoral competitiveness to wane.

The results obtained using specification (1) will be useful in that regard, but they are not sufficient. By using the absolute margin of victory regardless of party, we are assuming that the Democrats were becoming more competitive and the Republican response produced the same coefficient. This is an imperfect approach. Our discussion above suggests that the Democrats had a first mover advantage that withered away as Republicans copied them. The setup in (1) does not permit us to capture this. As such, we employ the first of multiple robustness checks (the others are discussed below) by using a second specification centered on the Republican response to the Democrats’ anti-Chinese pivot:

$$\begin{aligned} \begin{aligned} GOPSwing_i&= {Chinese/White}_{i}^{'} {\delta } + {x}_{i}^{'}{\beta } + \epsilon _i \\ \end{aligned} \end{aligned}$$

(2)

where GOPSwing is the swing against the Republican party relative to the 1868 election.²¹ The 1868 election serves as a base as it was the first year in which Republicans realized the danger of not adopting anti-Chinese policy. We should expect the swing to fall gradually (and become insignificant) leading up to the 1880 election, at which point the Republicans should no longer suffer a penalty associated with Chinese residents. This setup will rely on the same variables and a similar design, with the closest census to the election acting as the provider of control variables.

We need to make four final econometric notes. Each of these four notes are tied to a different robustness check that we enact (bringing the total to five with the specification change using swings). First, our i are the different counties. This is usual in the literature, as the most granular electoral results are at the county level. However, our model assumes that there are no spillovers. This is a limitation even in a period when news and politics were highly localized. As such, in a robustness check, we will present results where we use the Chinese-to-White ratio of neighboring counties as a control. Second, we will add the elections of 1856 and 1860 (which will be matched with the census of 1860) and the elections of 1884 (which will be matched to the census of 1880) and 1888 (which will be matched with the census of 1890). Our argument predicts that the Chinese-to-White ratio should be irrelevant before the Civil War (i.e., in the elections of 1856 and 1860) and irrelevant after the adoption of the Chinese Exclusion Act in 1882 (i.e., during the elections of 1884 and 1888). Testing with these elections provides a validity check.

Third, one might wonder why we did not employ a more usual panel setup where the counties are followed from election year to election year. This is because, as we indicated above, we have more time variation in the election year data than in the control variables. We only have three censuses from 1860 to 1880 that provide the independent variables and six elections that provide the dependent variable.²² This means that we cannot easily do a panel (i.e., for loss of variation between elections). However, there are eight possible panel combinations that could be done based on having a single census year with a single election. Table 2 below highlights eight possible panel combinations. We will use these eight different panel combinations to see if our baseline results are conceptually confirmed. This will be done by two different approaches that rely on interaction terms between the Chinese/White ratio and different time-related variables. Fourth, one might be concerned that we have endogeneity problems. To deal with these concerns, we will exploit some of the features of the panel setup and also make use of shift-share instruments (Card, 2001; Jaeger et al., 2018).

Table 2 Possible panel combinations

4 Results and robustness checks

4.1 Main results

The results from the regressions confirm our intuitions to some extent: the ratio of Chinese to White people in a county makes that county more electorally competitive in a presidential election. This applies to all elections.

First, we ran bivariate regressions, which we depict in Fig. 1 below using binned scatter plots. As can be observed there, the relationship between the absolute vote margin is negative, and the fit appears strong starting at the 1868 election. While the relation remains negative throughout the period, the fit deteriorates mildly.

Fig. 1

Bivariate Effect of Chinese/White Ratio on Absolute Victory Margin Using Binned Scatter Plots

Second, we ran multivariate regressions in OLS with each of the different elections. The results are depicted in Table 3 below. Each pair of columns represents an election with two different specifications. The difference in each pair is that one column relies on the log of manufacturing output (capturing the marginal product of labor in manufacturing) and the other measures how important the agricultural sector was relative to the manufacturing sector (with the assumption that the Chinese were disproportionately engaged in manufacturing). As can be seen, all columns report that the Chinese-to-White ratio reduces the absolute value of the victory margin in a county. The effect is always significant above the 10% level and is often significant above the 5% level. The coefficients are quite large. As the ratio is expressed in decimal points (i.e., one Chinese to ten Whites is 0.1), our coefficients imply that one extra Chinese per 100 whites reduced the margin of victory by between 0.135 and 0.352% points. To provide context to these proportions, the election of 1876 could have tipped to the Democrats had they been able to get less than 1% of California’s republican voters to switch sides.

Most importantly, the evolution of the coefficient for the ratio of Chinese to White is also consistent with our explanation and echoes the deteriorating fit in Fig. 1. Its importance diminishes monotonically from election to election. This can be seen in Fig. 2. The line labeled “Effect (1)” refers to the coefficients for the Chinese to White ratio from the odd-numbered columns of Table 3 while “Effect (2)” shows the even-numbered columns of Table 3. As we pointed out above, we expect that the Democrats initially get significant returns from adopting anti-Chinese policies. However, as Republicans realized the electoral cost of appearing pro-Chinese in California, they adopted increasingly hostile proposals. This blurred the difference between them on this key topic such that it minimized the electoral advantage conferred by endorsing anti-Chinese policies.

More importantly, the results depicted in Fig. 2 suggest that bipartisan support for anti-Chinese policies in the late 1870 s-which is well documented by historians and sociologists such as Fong and Markham (2002) and Tichenor (2009)-was born out of political need; more precisely, the need to negate the electoral advantage of the early mover (i.e., the Democrats) on anti-Chinese policies.

Table 3 Ordinary least squares results of absolute electoral margin

Fig. 2

Effect of Chinese/White Ratio on Absolute Victory Margin

4.1.1 Changing baseline specification to vote swings

However, our use of absolute margin assumes that the effect of adopting anti-Chinese policies was identical for Democrats and Republicans. This is why we try a second specification in Table 4 in which we employ the vote share swing relative to 1868 as the dependent variable. This choice of specification is because the Republican pivot on Chinese policy begins after 1868 (whereas the Democrats started earlier). Had the Republicans preserved their initial policy, we should expect the swing they suffered in the 1868 election to be permanent. However, as they shifted positions, the swing relative to 1868 should gradually fall. This is what we observe. Relative to 1868, Republicans in 1872 saw their vote share fall as the Chinese population grew relative to the White population. That reduction was modestly smaller in 1876 and no longer significant in 1880, implying that the Republicans no longer suffered from being perceived as being less anti-Chinese than the Democrats. The difference from the 1876 to 1880 election is particularly telling. Indeed, had California switched fewer than 1500 votes to the Democrats, Republicans would have lost the Presidential election by 179 to 190 votes in the Electoral College. It was crucial for Republicans to switch positions on Chinese policy so as to prevent Democrats from solidifying their inroads.²³

These results explain well the fact that the 1880 election was fought between two presidential candidates who defended significant restrictions on Chinese immigration. A sense of inevitability surrounds the Chinese Exclusion Act of 1882, as neither party appeared willing to take any other stance than massive restrictions on Chinese migration. As such, the decades-long period of federal inaction in terms of immigration policy in the face of a clear nativist backlash came to an end.

Table 4 Ordinary least squares results of republican vote swing relative to 1868 presidential election

4.2 Robustness checks and validity

4.2.1 Spillovers

There are six reasonable rebuttals regarding our results. The first rebuttal is that our unit of observation—the county—may be affected by the presence of Chinese workers in neighboring counties. Simply put, there might be spillovers. As such, we alter our results from Tables 3 and 4 to include a control for the average Chinese/White ratio in neighboring counties. In Table 5 below, the resulting coefficients on the Chinese/White ratio can be observed. For brevity, we report only these coefficients. As can be seen in Table 5, including controls for potential spillovers strengthens our results. Similar effects of the ratio on the absolute margin are present for the 1868 and 1872 presidential elections (see top panel in figure). However, the effects for the 1876 and 1880 elections are smaller and no longer significant at the 10% level as in Table 3. For the swing relative to the 1868 election, a similar shift appears as, by 1876 rather than 1880, Republicans no longer seem to suffer any electoral penalty. This suggests that they successfully adjusted their stance to negate the Democrats’ initial advantage on the topic.

Table 5 Effect of replicating our results while controlling for the Chinese/White ratio in neighboring counties

4.2.2 Was it also true before the civil war?

The second rebuttal is that our findings do not necessarily entail a shift due to the Civil War. The electoral tendencies we observe could have pre-dated the War. More importantly, our narrative is that the Civil War altered political incentives. This means that prior to the Civil War, we should see null effects of the Chinese/White ratio on the absolute margin. Finding such null effects would suggest that our narrative is coherent with the empirical evidence. As such, we replicate the results from Table 3 using the 1856 and 1860 elections (the last two pre-Civil War elections).

As can be seen from Table 6,²⁴ the results are noticeably different from those of the postwar period. First of all, they were not significant in the 1860 presidential election. Secondly, they are positive—a greater Chinese population made a sub-district less competitive prior to the Civil War. This is true for both elections, but the effect is statistically significant only for the 1856 election. This difference in coefficients and significance relative to the postwar period provides a greater sense of the plausibility of our claim that the Civil War caused the shift in political incentives that eventually led to the Chinese Exclusion Act.

Table 6 Ordinary least squares results of absolute electoral margin prior to the civil war (1856 and 1860 elections)

4.2.3 Was the effect gone after the Chinese exclusion act?

The third rebuttal is that our results for 1880 (i.e., no effects of the ratio on electoral competition) should extend after the adoption of the Chinese Exclusion Act in 1882. It appears that its enactment did eliminate the topic’s relevance in Californian presidential politics (Lee, 2003). The state reverted to being a generally safe Republican hold. It was carried by Republicans in every election until 1908 except the 1892 election (when Grover Cleveland squeezed a razor-thin win) by margins varying from 2.82% points to 34.9% points. Those margins were always far larger than the national margin for the Republicans.²⁵ The loss of the wedge issue appears to have disfavored Democrats. This can also be observed when we repeat the exercise from Sect. 4.1. The coefficients on Chinese/White ratios for the 1884 and 1888 elections show no statistically significant association, and the t-stats are far smaller than they were for the 1880 election.

There is also some qualitative evidence in that respect that the topic lost importance after the Act’s enactment. There were only two pieces of federal debate on Chinese immigration after 1882. The first was the Scott Act of 1888 which acted as a supplement to the 1882 Exclusion Act. It prohibited Chinese laborers from returning to the United States once they traveled abroad. The Geary Act of 1892 (or the Chinese Exclusion Act of 1892) was passed to extend the 1882 Exclusion Act for another ten years and added more restrictions on Chinese residents in the United States. The law required all Chinese residents of the United States to carry a resident permit that functioned as an internal passport (Lee, 2003, p. 43). Failure to carry the permit would result in deportation or a year of hard labor, though the law was not effectively enforced due to the lack of funding from Congress. Both acts were adopted with little debate and great bipartisan support (Table 7).

Table 7 Ordinary least squares results of absolute electoral margin after enactment of the chinese exclusion Act of 1882 (1884 and 1888 elections)

4.2.4 Shifting to panel combinations

The fourth rebuttal is tied to whether our results are driven by the choice of using consecutive cross-sections and comparing them rather than using panels. Unfortunately, a single panel setup is unavailable due to the fact that we have a single census year providing independent variables for multiple election years (i.e., the dependent variable). This would eliminate much variation. However, as we indicated in Sect. 3, a roundabout method allows us to use a modest panel setup as a robustness check. Our use of earlier elections for 1856 and 1860 (in Sect. 4.2.2.) allows us to create eight different panel combinations with a single census year being used for a single election year. For example, the elections of 1856, 1868, and 1880 can be used in a panel where they respectively use the census years of 1860, 1870 and 1880 for the independent variables. This would be one combination. Another combination would be a panel with the 1860, 1872, and 1876 elections, which respectively use the 1860, 1870, and 1880 censuses for the independent variables. These eight panel combinations allow us to see if we can obtain consistent results with the baseline ones. One virtue of this, which will become apparent in the next subsection, is that we can also make use of fixed effects.

Table 8 below shows the results from this approach.²⁶ To economize on space, the table only presents the key variables of interest. Each column uses a separate combination. The top part attempts to see if there was a post-Civil War effect of the Chinese/White ratio on electoral competitiveness. It adds a dummy variable for whether an election year took place after the war. That dummy variable is then interacted with the Chinese/White ratio. If our argument is correct, the coefficient on the interaction term should be significant and negative. The top panel confirms that this is the case which suggests that our results are consistent. Unfortunately, this approach treats all postbellum election years as the same, which means that we cannot evaluate whether the coefficient drifted to zero as both parties adopted the same policies on Chinese exclusion.

This is where the bottom part of Table 8 acts as a complement. Each of the combinations now includes an interaction term between the Chinese/White ratio and the specific election year in the form of a year-fixed effect estimator.²⁷ For illustration, take the first column: the combination that uses the elections of 1856, 1868, and 1876. The election of 1856 acts as a reference for the year-fixed effects. The row that reads “Year t+1 X Chinese/White Ratio" gives us the interaction effect for the year-fixed effect for 1868 and the Chinese/White ratio.²⁸ The row that reads “Year t+2 X Chinese/White Ratio" gives us the interaction effect for the year-fixed effect for 1876 and the Chinese/White ratio. Our argument predicts that coefficients in the row “Year t+1 X Chinese/White Ratio" should be negative and significant. In contrast, coefficients in the row “Year t+2 X Chinese/White Ratio" should be smaller than those in the row above (and maybe even be insignificant). Both rows across all columns suggest that this is the case. This confirms the idea that the first-mover advantage of Democrats was gradually eroded between 1868 and 1880.

Table 8 Results from eight different panel combinations

4.2.5 Endogeneity concerns

The fifth rebuttal is that there might be endogeneity issues. As studies involving spatial variation in the immigrant population often come with endogeneity bias (Card, 2001; Jaeger et al., 2018), this is a reasonable concern. In our case, we could expect some endogeneity resulting from the interplay between where Chinese immigrants decided to go and the level of hostility (proxied by the absolute electoral victory margin in a county i) from the local non-Chinese population. That interplay could go in both directions. Chinese immigrants could have avoided areas with greater racial animus. They could have also gone where there was greater animus because this is where there were already Chinese immigrants.

We have three different answers with regard to the potential endogeneity issue. The first is to assert that we think the issue is limited because we are using the Chinese-to-White ratio rather than the size of the Chinese population itself. This is of relevance because Chinese migrants probably did not choose their county of residence in California based on the ratio of Chinese to White. It is more likely that the absolute size of the Chinese population in a county i was a strong element motivating location choice. This can be best seen with the counties of San Francisco and Trinity. In 1880, San Francisco’s Chinese population of 21,790 represented 29% of California’s Chinese population, and the Chinese-to-White ratio stood at 0.05. In 1860, these numbers stood at 2,719, 8% and 0.10. In contrast, Trinity County saw its Chinese-to-White ratio surge from 0.48 to 0.70 between 1860 and 1880. However, its Chinese population increased by only 313 individuals while its share of the Chinese population fell from 4.8% to 2.6%. This suggests that Chinese migrants post-1860 were selecting places where there was a large Chinese population rather than places with a large Chinese-to-White ratio. This, we believe, minimizes the issue of endogeneity, but it does not fully rule it out.

The second element of the answer has already been provided above when we used the eight different panel combinations and included area-fixed effects. Generally, area-fixed effects are not a bulletproof solution to endogeneity concerns, but they have the ability to mitigate the problem to some degree (Marvell & Moody, 1996; Bailey, 2020). As our results above showed, shifting to the different possible panel combinations shows the same conceptual results as our baseline specification with consecutive cross-sections. The inclusion of county-fixed effects does not alter our findings.

The third element of the answer consists in using the conventional shift-share instruments in immigration literature.²⁹ Our shift-share instrument is defined as follows

$$\begin{aligned} S_t=\frac{C_{i,t^0}}{C_{t^0}}\frac{\Delta C_t}{L_{i,t-1}} \end{aligned}$$

(3)

where $\frac{C_{i,t^0}}{C_{t^0}}$ is the share of all Chinese immigrants in county i at a reference date $t^0$ that predates t, $\Delta C_t$ is the change in the Chinese population at time t, and $L_{i,t-1}$ is the population of county i in the previous period. The expected Chinese population, $S_t$, is therefore a weighted average of the state-level Chinese population change (i.e., the shift) where the weights depend on the distribution of the Chinese population at $t^0$ (i.e., the share). We set the year 1860 as $t^0$, and each share depends on a share of the state-level.³⁰ We construct shift-share instruments for the 1870 and 1880 censuses by taking the change from 1860 to either 1870 or 1880. There is, however, one downside to this strategy. To create the shift-share, we have to use the 1860 counties and there were fewer counties opened in 1860 than in 1870 or 1880. As such, the use of an instrument cuts the sample size. This will affect statistical significance tests. Table 9 shows us the results without the first stage.³¹ All coefficients are further away from zero than in Table 3. For example, the coefficients for the 1880 election were $-$0.139 and $-$0.135 with OLS (see Table 3). With the instrumental variable approach, they stand at $-$0.172 and $-$0.181.³² However, the coefficients in 1880 and 1876 are not statistically significant with the instrumental variable approach. They are significant and negative in the 1872 and 1868 elections. More importantly, the coefficients get smaller as we move from election to election. As such, Tables 3 and 9 tell highly similar stories regardless of the estimation method.

Taken individually, we do not find these three replies to concerns of potential endogeneity to be satisfying. However, taken together, we believe that they convincingly reassure us that endogeneity issues are not causing us to produce inconsistent estimators.

Table 9 Instrumental variable regression result (Second stage only)

4.2.6 Complementary evidence for validity

The sixth and final counter is that if our argument is correct, there should be some complementary evidence. Fortunately, the evidence for the effects of the Civil War and the timing of changes in electoral strategy (i.e., Democrats being first movers and obtaining initially large returns that eroded as Republicans shifted as well) can be complemented by roll call votes in the House of Representatives. We identified three post-Civil War votes that qualify for our purposes: (a) an 1869 motion “To Suspend The Rules And Consider A Resolution Stating That In Passing The Resolution For The Fifteenth Amendment To The Constitution Of The U. S. This House Never Intended That Chinese Or Mongolians Shall Become Voters"; (b) an 1879 Bill restricting Chinese immigration³³; and (c) the Chinese Exclusion Act itself. The breakdown of roll call votes is illustrated in Table 10 below, where the Yes and Nays are broken down by part. The 1869 motion was soundly defeated on party lines, and Republicans overwhelmingly voted nay. However, by 1879, things had changed. First, more Democrats were in the House, which increased the odds of passage. Second, close to half of the Republican votes voted yes. The entire California delegation (which was half Republican and half Democrat) voted yes.³⁴ By the time of the vote on the Chinese Exclusion Act of 1882, the vast majority of Republicans had become favorable to immigration restriction.

Table 10 Party breakdown of roll call votes on immigration-related topics after the civil war

5 Conclusion

In this paper, we used the state of California to explain why there were many decades separating the onset of a nativist backlash against immigrants and the adoption of federal immigration control policies. We argued that pre-war politics came with incentives that made politicians unwilling to engage in national-level policies to deal with immigration. The Civil War altered those incentives by pushing Democrats to attempt to create a West-South alliance to secure presidential victories. California was seen as a key state to capture, and adopting anti-Chinese policy proposals was the way to do so (especially as Chinese immigrants could not participate in elections). As Republicans saw the inroads made by Democrats in California and realized that it could cost them the presidency, they began adopting very similar anti-Chinese policies.

Using county-level results, we found that counties with large Chinese populations were more electorally competitive, all else being equal. Moreover, and consistent with the fact that the Democrats were the first movers, the effect diminished over time as Republicans began to adopt anti-Chinese proposals. These results provide evidence for the narrative we propose whereby the Civil War altered incentives in ways that made the adoption of immigration control policies possible. Our results hold up to multiple robustness and validity checks.

Our results and arguments should be seen as complementary to those already advanced by historians and economic historians rather than as substitutes. For example, Hirota (2016) argues that the few northeastern states, such as New York and Massachusetts, that had strong nativist constituencies developed bureaucracies that both advocated for federal intervention and later implemented federal policies. Our explanation complements his in the sense that politicians saw greater returns to listening to immigration control bureaucrats from New York or Massachusetts and also hiring them to implement policy. Others emphasize the importance of nativist feelings. Our explanation complements theirs by providing a channel by which they were converted into federal policy.

This complementarity is important. Given the rising importance of immigration policy discussions across the Western world, this complementarity-even for historical cases such as the 1882 Chinese Exclusion Act-can provide key insights into understanding the determinants of policy changes today.