Friday, July 9, 2010


By Ashley S. Timmer - Jeffrey G. Williamson

Contrary to conventional wisdom, the doors did not suddenly slam shut on American immigrants when Congress passed the Emergency Quota Act of May 1921. Rather, the United States started imposing restrictions a half century earlier. Argentina, Australia, Brazil, and Canadá enacted similar measures, although the anti-immigration policy drift often took the form of an enormous drop in (or even the disappearance of) large immigrant subsidies. Contrary to conventional wisdom, there wasn't simply one big regime switch around World War I. What explains immigration policy between 1860-1930? This paper identifies the fundamentáis that underlay the formation of immigration policy, distinguishes between the impact of these long run fundamentáis and short run timing, and clarifies the difference between market and non-market forces. The key bottom line is this: Over the long haul, immigrant countries tried to maintain the relative economic position of unskilled labor, compared with skilled labor, landowner and industrialist. The moráis for the present are obvious.

After the 1880s, there was a gradual closing of New World doors to immigrants. The words that matter here are gradual and New World.  Contrary to what American history textbooks may suggest, the doors did not suddenly and without warning slam shut on American immigrants when the United States Congress overrode President Wilson's veto of the immigrant literacy test in February 1917 or when it passed the Emergency Quota Act of May 1921.  Not too long after the Civil War, a half-century prior to the Literacy Act, the United States started imposing restrictions on what had been free immigration.   The United States was hardly alone.   Argentina, Australia, Brazil, and Canadá enacted similar measures, although the timing was different, and the policies often took the form of an enormous drop in (or even disappearance of) large subsidies for immigration rather than of outright exclusión.   In short, there was considerable variance in immigration policy across these five countries and over the half century.   Contrary to the conventional wisdom, there was not simply one big regime switch around the time of World War I.
What was true of immigration policy was also trae of trade policy.   Globalization proceeded in fíts and starts after 1846 when Britain repealed the Corn Laws and started a liberal trend towards free trade.   In the late-nineteenth century, enormous declines in international transport costs precipitated the first great globalization boom.   It took the form of mass migrations, a trade boom, and international capital flows at (relative) levéis never reached before or since, and it helped induce economic convergence (Williamson 1996a). The liberal trend did not last long, however.   There was a globalization backlash.   Tariffs started to rise on the European continent.   Immigration and trade restrictions emerged in the New World.   With the end of World War I, the world economy plunged into a dark age of
de-globalization and policy antagonism towards the mobility of goods and factors.   The long interwar period of darkness was followed by a liberal renaissance.   Globalization, convergence, and more liberal policy have been on the rise ever since 1945, especially since the early 1970s.
What explains a change in immigration policy?  The answer ought to be consistent with the correlations invoked in the previous paragraph, but a number of candidates have been nominated:   increasing racism, xenophobia, widening ethnicity gaps between previous and current immigrants, more immigrants, lower-quality immigrants, the threat of even lower-quality immigrants, crowded-out native unskilled workers, rising inequality, greater awareness of that inequality by the powerful, and greater voting power in the hands of those hurt most~the working poor.  There have, however, been few attempts to introduce these underlying, long-run fundamentáis into explicit models of policy formation.   We have already discussed the exceptions to this generalization in the survey by Timmer and Williamson (1995), and we start this paper by stripping that survey down to what we think is relevant to our empirical attack here on a newly-constructed panel data set.
The goal of this abridged survey is to identify the fundamentáis that might underlie changes in immigration policy, to distinguish between the impact of these long-run fundamentáis and the determinants of short-run timing, to clarify the differences between market and non-market influences, and to look at the tools others have used to assess these influences.   The literature on immigration policy is not nearly as mature as the parallel literature on trade policy, but we hope to exploit a few connections between the two.
The Theory of Political Economy
The "new" political economy attempts to make positive predictions about the interaction between political institutions and the marketplace.   This interaction has its roots in two schools of thought.   The social choice school, following Kenneth Arrow (1963), looks for equilibrium policies that are relatively free of any particular institutional framework.   The school analyzes various rules for aggregating the preferences of individuáis into a social welfare function.   An elegant median-voter result under majority rule has emerged.   Political economy models that use a majority-voting framework have proliferated, especially those confronting politically-driven business cycles and growth.1
The public choice school, founded by James Buchanan (Buchanan and Tullock 1962), explicitly addresses the nature of political institutions, political agents, and multidimensional policy goals.   The basic premise of the public choice school is that politicians do exist and they rationally maximize their own utility functions.   These utility functions include preferences for money and power, and thus they might not reflect the preferences of the majority.   The public choice school recognizes that few decisions are made by majority vote of the electorate.   Rather, politicians and bureaucrats control the agenda and often make decisions unilaterally.   Those who are in a position to give political agents the power and money they want are able to influence policy in their favor.   These have come to be called "pressure groups."   In general, pressure group models define a policy equilibrium where the political agent maximizes his utility, subject to the constraints imposed by the various interest groups and perhaps the public at large.2
1 See Persson and Tabellini (1994) for a survey.
2 Pressure group theory has developed in tándem with the theory of "rent-seeking," as first elaborated by
Tullock (1967) and formalized by Krueger (1974). Pressure group models are an ideal way to think about rent-
seeking; conversely, potential rents are an ideal way to define an interest group.
Each of these approaches to political economy has benefíts and shortcomings. Median-voter models require a single policy dimensión, and single-peaked preferences along that dimensión, in order to guarantee an equilibrium.   When policies are redistributive, it cannot be guaranteed that preferences are single-peaked.   The pressure-group model requires more detail about the parties involved, and it therefore lacks the elegant simplicity of the median-voter model.   Models with their origins in the public choice school seem more promising as explanations for policy differences across countries in the pre-1930s, and even over time.   After all, these countries were undergoing political liberalism and more inclusive suffrage, albeit some much faster than others.   Yet, to introduce political institutions is to créate difficulties in making comparisions across countries.   For example, Canadá chose to give the power to adjust immigration quotas to bureaucrats, whereas the United States set quotas which could be changed only by legislative vote (Green 1995).   Canada's policy was to set quotas on the basis of individual industries, substantially reducing the free-riding problem for those lobbies.   In the United States, legislators set quotas by country of origin, so that no one industry stood to gain much by lobbying for a quota change.
In what follows, we take an eclectic view of this debate.   Our interest is to identify which market and non-market forces mattered, rather than to explore the mechanisms by which they were translated into policy.   Indeed, we are much more interested in community attitudes towards those forces than in the actual impact of the policies implemented.
Immigration Policy: Searching for Hypotheses
There is a general consensus in the literature that immigration policy has always been sensitive to labor market conditions.3   At the same time, immigration flows themselves have always been sensitive to wage differentials between countries and unemployment rates.   For example, Claudia Goldin (1994) notes that in the United States in the late 1890s, during a time of economic recession and high unemployment, there was a new push for immigration restrictions.   At that time, however, the rate of immigration slowed dramatically, reaching a low in 1897, the same year that the first vote on immigration restriction was taken in the House of Representatives.   Similarly, Australian inflows dropped sharply in the recession of the 1890s when attitudes towards immigrant subsidies hardened (Pope and Withers 1994).   These concurrences would seem to suggest that the Ímpetus to restrict immigration was far more sensitive to labor market conditions than to immigration levéis.
On the other hand, the ethnic composition of immigrants is clearly a factor in the politics of restriction.   Australia maintained a strict policy aimed at keeping the country one of British and Irish descent, and certainly not "yellow" (Pope and Withers 1994).   The United States completely banned immigrants from China in 1882 and immigrants from all of Asia in 1917 (Green 1995).   In the United States, increasing demands for restriction in the 1880s and 1900s paralleled an increase in the relative numbers of immigrants from southern, central, and eastem Europe, the so-called "new" immigrants.   The world labor market was by 1890 almost completely segmented into what economists today would cali "North" and "South" (Lewis 1978; Taylor 1994; Hartón and Williamson 1994).   Because of this segmentation, it is unclear whether these policy goals were a result of racism and
3 After World War II, a focus on human rights developed; most Western countries changed their immigration policies to provide for special consideration of political and economic refugees. Prior to the 1930s, such classifications did not exist.
xenophobia or whether ethnic origin merely served to signal, however imperfectly, the human capital content or "quality" of the immigrants (Foreman-Peck 1992).
Three Models of Immigration Policy
As James Foreman-Peck (1992) notes, two questions for any model of policy formation are:   Who gains and who loses?   Who decides the policy?   There is a clear consensus regarding the first question.   Wage earners—unskilled workers in particular—lose with immigration, as the labor pool swells and wages sag.   Owners of other factors of production—land, capital, and perhaps even skills—gain from the more abundant unskilled labor supply that makes these other factors more productive.   Having said this, two caveats deserve stress.   While most attempts to measure the impact of immigration on wages have found that wages were downwardly sensitive to immigration (Williamson 1974; Taylor and Williamson 1997; Green 1994; Goldin 1994; Hatton and Williamson 1995; Williamson 1996a), one study, of Australia, found that wages actually increased with immigration, if only marginally (Pope and Withers 1994).   This perverse result can emerge if immigrants augment labor demand enough to offset their impact on increased labor supply (for example, by working previously unsettled land or by inducing an accumulation response). If labor demand keeps pace with labor supply, it looks as though native labor is not hurt by immigration.   The problem for us and the voters, however, is to distinguish between labor demand conditions that are dependent on the immigrants and those that are not.  Under conditions of sagging wages, policy might still be used to keep out immigrants even if their presence had nothing to do with the deteriorating labor-market conditions.
The issue of unemployment has not really been examined in the context of immigration policy.   Suppose wages are sticky downwards and unrelated to the size of the unemployment pool, perhaps for efficiency reasons or "fairness."   Immigration in this case
will not have any effect on wages, but it will add to the unemployed, all the more so if new immigrants are last hired and fírst fired (Hatton and Williamson 1995).  No one benefits. Capitalists do not gain by a fall in wages, ñor do the unemployed gain.   In addition, unemployed workers tend to express their discontent by strikes and street violence. Eventually, both sides might imite in favor of immigration restrictions.   Goldin (1994) suggests that this aligning of interests is exactly what happened in the United States during the 1890s.
On the other hand, "guestworker" effects should minimize the impact of an economic downturn on native unemployment, as recent (but now jobless) immigrants return home—that is, immigrants do voluntarily what policy aims to do.4  While this argument was used to justify postwar European guestworker policies, and while it was certainly present in the United States in the 1890s, the "guestworker" effect failed to work with any quantitative muscle during that critical decade (Hatton and Williamson 1995).
These caveats aside, most discussions of the politics of immigration assume that the interests of capital and labor are divided.   Foreman-Peck (1992) argües that land ownership might have mattered too, especially in the late-nineteenth century when agriculture was still a big sector.   Foreman-Peck takes the following approach.   Assume that the individuáis receive their incomes primarily from one source:   wage earnings, capital income, or land rents.   Depending on the franchise, the government maximizes the following weighted objective function:
4 Immigrants do it even better, of course. A policy of immigrant exclusión can do no better than reducing the net inflow to zero. Voluntary return migration can drive up emigration rates to levéis high enough to make net inflows negative.
,        V = aw1 + Pit + (1 - o - P)r s.t.      Q =/{Lv^ZK)
where L, and L2 are, respectively, native and immigrant labor.   Note that the return to immigrant labor, w2, is excluded from the objective function.   Output is subject to the constraints of the production function, and the critical question is this:   are immigrant and native labor complements or substitutes?   Estimating a trans-log production function, Foreman-Peck concludes that they were substitutes in the late-nineteenth century U.S. economy.   Thus, the larger the weight on labor interests, a, the more restrictive the immigration policy.   The reverse is true as the political system attaches larger weights on capital or land.
Foreman-Peck allows for the possibility of two types of immigrant labor:   skilled and unskilled.   It might be that skilled immigrant labor was a complement to domestic labor, whereas unskilled immigrant labor was a substitute.   We would then expect to see a policy that encouraged immigration of skilled workers and discouraged unskilled ones. Foreman-Peck argües that this concern, and not any racism or xenophobia, was responsible for policies in the Americas that restricted Asian immigrants and for South África' s policy toward African immigrants.
Although Foreman-Peck does not implement a formal empirical test, his discussion of Argentina, Britain, South África, and the United States indicates that some of the facts are consistent with his theory.   For example, landed interests were largely in control of Argentina's policy, and the government offered generous immigration subsidies to attract farm laborers from the Mediterranean Basin.   In contrast, the United States had a more universal franchise, rejected subsidies, and gradually closed the door as the frontier itself was closed (by 1890, or so said the Census Commissioner at that time).
Goldin (1994) takes a different approach.   Following a long tradition in American historiography that has focused on "sectional interests," Goldin looks at regional splits and rural-urban differences in a way consistent with a median-voter model.   Although she does not model the relationship formally, she assumes that individual Senators and Representatives pursue policies that favor their constituents, in proportion to the numbers represented by each urban, rural, and regional interest group.   The passage of the literacy test, which was first attempted in 1897 and was finally successful in 1917, seems to have been the result of two (often opposing) forces:   demographic changes, and changes of heart. The changes of heart were many.   Goldin suggests that capitalists were for the first time aligned with labor in opposing immigration during the recessionary years of the 1890s (for reasons we have already conjectured).   Later, capital would shift back to its pro-immigration stance, but the South would shift to an anti-immigration stance, probably a change of heart motivated by the urge to protect its relative population share, since few immigrants ended up in the South.   Finally, the northern Midwest, fairly pro-immigration in the 1890s, would undergo an anti-immigration switch following World War I.  Goldin argües that this was mostly a change of heart by older immigrant groups, pushed to patriotism by the war.
Where does demographic change enter the story?  As the South and northern Midwest were shifting to anti-immigration positions, cities were becoming increasingly pro-immigration.   Goldin finds that the probability that a legislator would vote for immigration restrictions was strongly negatively related to the proportion of foreign-born in the district and was al so negatively related to the level of urbanization.   This relationship suggests that efforts to reunify families were operating in the cities, or, alternatively, that there were only small differences in the ethnic make-up of resident stocks and new immigrant flows.   In
any case, cities were on the rise, and thus pro-immigration urban interests increasingly made themselves heard.
More important than either of these non-market forces, however, was the impact of increasing immigration on wages and the subsequent effect on votes.   Especially after the turn of the century, Goldin fínds a signifícant negative impact of immigration on wages, a result consistent with other historical studies.   The change in real wages is, in turn, a significant explanatory variable in accounting for the Congressional vote to override the presidential veto of the literacy test in 1915.   The higher the growth in wages, the less likely was the representative to vote for an override (and thus for restriction).   At the same time, the higher the proportional increase in the foreign-born population, the more likely was the congressman to vote for an override (and thus for restriction).   But large, established communities of immigrants had the reverse effect.   Once a district attained a foreign-born population of about one-third of the total in the district, there was almost no chance that the representative would support restrictions on immigration.
The two important elements of Goldin's model for our cross-country study are the role of changing wages on policy and the evidence that confirms that immigrants infiuenced wages.   All that is really required, however, is that politicians and their constituents believed that immigration retarded wage advance.   It appears that they did.
William Shughart, Robert Tollison, and Mwangi Kimenyi (1986) offer a model of the shifting degrees of enforcement of immigration restrictions over business cycles.   The model would apply readily to changes in policy as well.   The basic premise is that politicians try to maximize votes by catering to different interest groups.   Workers want high wages, and they pressure politicians to enforce immigration restrictions.   Capitalists and landowners want lower wages in order to raise rents and profits, and they try to reduce enforcement.   As the economy goes through business cycles, the ideal policy mix shifts,
resulting in changes in the degree of enforcement against immigration.   The authors test their model empirically on data from the United States from 1900 to 1982, using two alternative measures of the degree of enforcement.   The explanatory variables include three alternative measures of economic well-being:   unemployment, real wages, and real GNP. The authors' results are supportive of the model.   Even taking into account official changes in immigration policy, the size of the enforcement budget, and the party in the White House, the degree of enforcement is significantly, and negatively, related to real GNP. Unemployment and the real wage were also significant explanators, but not so consistently as real GNP.   Had Shughart, et al., looked at U.S. policy towards indentured labor contracts prior to 1900, they would have seen the same correlation:   harsh policy during slumps; soft policy during booms.
The models of immigration policy built by Foreman-Peck, Goldin, and Shughart, et al., are the only ones that offer empirical support for their theories.5  These models focus on the absolute gains and losses associated with some given immigration policy, however.   What about relative gains and losses?   What about income distribution?
The Politics of Income Distribution
Economists have recently awakened to the fact that migration can créate more inequality in the receiving country and, perhaps, lessen inequality in the sending country. In only a short time, the empirical literature on this issue has grown to enormous proportions, perhaps because the consequences of immigration have been given renewed emphasis on the American political scene.   The debate began over the impact of
s Jess Benhabib (forthcoming) takes the median-voter approach, allowing individuáis to earn both labor and capital income in the spirit of the growth model of Alesina and Rodrik (1994); voters determine the amount of capital that immigrants must bring with them in order to be admitted. The model is an attempt to look at the dynamics of policy implications, and it gets very complicated. Perhaps for that reason, Benhabib does not test the model empirically.
immigration in the United States (Borjas 1994), expanded to consider European immigration (Freeman 1995), and spilled over into emigration from developing countries (Wood 1994).  The distributional impact of migration has even been confirmed for the late-nineteenth century—that is, inequality increased in the receiving countries, and inequality decreased in the sending countries (Williamson 1996b).
Rational and farsighted voters will consider the impact of immigration on future economic growth.   A labor-scarce country will do better to allow immigration than to allow the export of capital, thus becoming larger rather than smaller (Cheng and Wong 1990). But if immigration induces greater inequality, and inequality in turn inhibits economic growth (a rejection of the more traditional Smithian trade-off), additional immigrants might not foster faster growth.   If a country lets its median voter become too poor, that individual might vote for distortionary redistributive policies that can slow growth.
What are the facts?   Does inequality speed up or retard growth?6  While economists do not yet have a clear answer, citizens might vote for restriction on immigration simply because they dislike the increased inequality in their social environment and the lower living standards of their unskilled neighbors.   Alternatively, changes in income distribution might tip the balance of political power among competing interest groups, leading to policy change   (Timmer, 1996).
6   Perotti (1996) attempts to answer this empirical question.
Links to the Literature on Trade Policy
The literature on the political economy of migration policy is recent and small, whereas the related literature on the political economy of trade policy is mature and large. Trade theorists have been puzzled for a long time by the widespread use of protectionist policies when free trade is usually welfare-maximizing.   By adding politicians, interest groups, and distributional matters to the theory, models of "endogenous tariffs" have flourished.
As Wong (1983) has noted, trade and immigration policies have an historical symmetry.   Trade policy might seek to protect wages in labor-intensive industries by restricting imports of goods made with cheap labor.   Immigration policy might seek to protect wages by restricting growth in the labor pool.   While free trade can be a partial substitute for free migration, we expect open trade policy and open immigration policy to go hand in hand.7
Who are the interest groups in trade theory?   In the short run, when factors are assumed to be relatively immobile, protection of a given industry (like textiles or steel) benefits both capital and labor in that sector.   As prices rise, the marginal valué product of all factor inputs rises, raising wages and profits.   In the long run, when capital and labor have had time to reallocate, protection helps the scarce factor (labor in countries of net immigration) as long as import-competing industries use relatively more of the scarce factor.   Most models of trade policy take the short-run approach, focusing on the pressure from specific industries, although some of the empirical tests focus on the long-run importance of factor endowments.
Heckscher, Ohlin, Stolper, and Samuelson explored this substitution long ago (Flam and Flanders 1991; Stolper and Samuelson 1941).
Magee, Brock, and Young (1989) present some empirical evidence for the United States from 1900 to 1988 that confronts the long-run issues.   As capital per worker rose after World War II, tariff levéis fell, but any long-run change in the terms of trade was not significant.   To complement these fíndings, the authors run regressions to test short-run predictions of their model.   They propose that the median voter opposes protectionist policies but is unlikely to vote unless "roused out of his normal stupor" by high inflation. On the other hand, rising unemployment weakens the voter's hostility toward protection. Their empirical results tend to support these hypotheses.   Changes in tariffs are signifícantly and positively related to changes in unemployment and negatively related to changes in the inflation rate.8
Marvel and Ray (1983) also provide empirical support for the pressure-group approach.   They look at the ability of United States industries to resist tariff reductions in the Kennedy Round of GATT negotiations.   Among other things, they hypothesize that industries with higher concentration ratios (and therefore less of a free-rider problem) will be able to fight off tariff reductions, as will industries whose outputs are consumer products rather than industrial inputs, since consumer groups are theoretically the weaker political actors compared with industry groups.   They find that both variables are significant (and of the correct sign) in predicting tariff levéis after the Kennedy Round.
We have offered only a small window on a very large literature, but it should be large enough to see the obvious parallels between the literature on endogenous tariffs and that on immigration policy.   The concepts and paradigms are similar.   The important point is that trade policy can undo what immigration policy does.   Thus, we expect consistency between them—unless, of course, immigration policy is driven entirely by non-economic
Unfortunately, U.S. tariffs are also signifícantly lower under a Democratic administration, adversely affecting workers. Because Democrats are the pro-labor party in the Magee, et al, model, this result does not fit with their theory.
concerns.   Policies towards the mobility of international capital can also undo the effect of immigration policy, but political-economy models have yet to look at these general-equilibrium effects.
A Menú of Hypotheses
This brief review of the literature offers several promising hypotheses, which we organize here around a set of explanatory variables.9
First, immigration policy might respond to either the quantity or the quality of immigration, or both.   The size of the immigrant flow as a share of the native labor forcé is one obvious candidate, although the experience of the 1890s has already suggested that net labor market conditions might have mattered far more.   The quality of the immigrants is another candidate, measured in comparison with the native labor forcé.   The vast majority of the immigrants carne from and entered unskilled jobs.   Some had good health, high levéis of literacy, numeracy, on-the-job training, and considerable exposure to work discipline.   Other immigrants did not.  Quality and quantity are highly correlated prior to World War I.  The switch of emigrant source from higher-wage to lower-wage áreas of Europe correlated with the rise in immigration rates.   It is likely that these two effects reinforced each other in their impact on policy.   A variable that combines the rising quantity of immigration with the falling quality might do better than the two measures of quantity and quality in competition with each other.
Second, immigration policy might respond to labor market conditions.   This likely possibility can be sharpened by distinguishing between short-run and long-run influences. Unemployment, wage growth, or more aggregative indicators should serve to isolate the role of business cycles, trade crises, world price shocks, and other short-run events that
9 The details of and sources for the variables themselves can be found in Appendix A.
might influence the timing of immigration policy.   In addition, the use of lagged dependent variables should help identify just how slowly policy responds even to long-run fundamentáis.
We expect the fundamentáis to be captured by real unskilled wages~a measure of absolute performance~or the behavior of unskilled wages relative to incomes of the average citizen—a measure of relative performance.   The latter offers a measure of inequality that gauges the unskilled worker's economic performance against that of the average, and it is a measure that the politician and the voters could most easily see and understand.   Neither measure asserts that immigration was the key forcé driving the living standards of the working poor in the New World.   Both require only that the politicians and voters believed that immigration was a powerful influence on the living standards of the working poor. Whether it was the absolute or the relative performance that mattered is an empirical issue.
Third, immigration policy must have been influenced by other policies.   In particular, did governments pursue consistently liberal or restrictive policies regarding migration and commodity trade?  Or did they pursue policies, intentionally or not, that offset each other?  If the goals of such policies were to differentially affect the economic condition of certain groups, we would expect to see consistency across the policy dimensions.
Fourth, what non-market forces remain after controlling for these market forces? After controlling for immigrant quality, did racism have an independent influence?   Did differences in ethnicity matter?  Did the population have less sympathy for free immigration if new immigrants were not of their own ethnic origin?  Informed by increasingly strident social reformers, did voters and politicians become more aware of inequality, thus sparking a change in political response to market events?  Did the political response to market events change as the working poor found their political power increasing?
Quantifying Immigration Policy
The fírst step towards answering these questions is to assess the immigration policies themselves.   Despite universal openness to immigration in the 1860s, the doors to the New World were effectively closed by 1930.   The policy evolution varied widely over those seven decades.   The United States exhibited a steady drift away from free immigration. Brazil remained open much longer, suddenly slamming the door shut in the 1920s.   Canadá reversed the trend more than once over the period.
We have designed a policy index in order to assess the various hypotheses that might have driven a change in policy—among them, wages of unskilled workers, trends in inequality, the size or quality of immigration flows, the state of the macroeconomy, and ethnic concerns.   The index uses a scale of +5 to -5, with a positive score denoting a pro-immigration policy.   The methods used to construct the index can be found in Appendix B.
Policies can be gauged by their intent or their impact.   Although most models of policy formation do not make the distinction, we recognize that policy has two functions: fírst, to signal to groups that their interests are being tended to; and, second, to change the status quo.   The fírst function does not necessarily depend on the actual change in immigration that followed a change in policy.   It is clear that political agents were trying to affect the flow of immigrants and to respond to their constituencies, despite oñen ineffective results.10  The goal here is to capture this sentiment, or political signal.   Still, we shall try to ignore political rhetoric or other noncommittal attempts to win support of various interest groups.
Immigration policy often seems to have responded to conditions that were no longer current, and in many cases policy was not effective in changing the patterns of immigration.
Empirical Tests
Our questions address long-run relationships.   Although we look at annual data for the individual countries, most of our analysis uses panel data consisting of five-year averages of the variables.   The panel includes observations for Argentina, Australia, Brazil, Canadá, and the United States from 1860 to 1930.   Five-year averages serve to minimize year-to-year variance and the measurement error inherent in the annual data.   Unlike Goldin (1994), who explored the timing of policy changes in the United States, we are looking for the underlying fundamentáis dictating policy choices.   We hope to determine if there was a similar reaction to rising immigrant flows in all countries in the New World and to sort out whether policy was driven in the long run by market conditions rather than ethnic concerns.
The next step was to use annual time-series data to look for patterns in individual countries. The results generally confirmed those found in the panel data. However, the relative impact of each explanatory variable oñen differs across countries.
The country-specific equations allow us to quantify the importance of market and/or non-market conditions in driving changes in policy.   Foliowing the literature on economic growth, we do "policy accounting."   Each country had at least one critical period for immigration policy on which we focus our attention.   Did Brazil cióse its doors to immigrants in the 1920s because of rising inequality?   Was the literacy test in the United States really the result of changes in the ethnic composition of the immigrants?
Panel Data
The regressions using panel data reveal some consistent effects across time and across countries (see Table 1).  The most consistent result is that policy is slow to change. Even with five-year averages, the lagged dependent variable (policy, our index of immigration policy) is highly significant and has a large valué.   Nonetheless, even highly
path-dependent policy is influenced by other factors.   After sorting through all the variables in the equations, the most pervasive result is that policy is sensitive to the labor-market performance of native and immigrant unskilled workers, both in relation to others in the country and in relation to expected would-be immigrants.''
Political Effects.   Surprisingly, the political environment was insignificant for policy change.   There is no evidence that different political institutions and franchises affect the degree and direction of policy change in any systematic way.   The two measures of political openness—DEMOC, the ten-point index of democratic characteristics constructed by Gurr (1991), and parcom, Gurr's measure of competitivenes of political participation~are both significant in explaining the level of the policy index, but they are not important in explaining a policy shift once fixed effects are used.12  The nature of the political system is not unimportant in explaining variation in immigration policy across countries, but it explains little of the changes in policy over time.   Perhaps the variables did not prove useful because there was no significant political change during the time period in question.
Immigration Effects.   Two variables measure the labor market effects of immigration.   imwage, the wages of unskilled workers in the source countries, is a proxy for the quality, or human capital content, of the immigrants.   threat, the measure of the threat to native labor, includes both quantity and quality effects (see Appendix A).   Both variables are consistently significant, but the coefficient for imwage has the wrong sign.
The other two measures of immigration~the percentage of the population that is foreign-born (forpop) and the difference in ethnic composition between the current immigration flow and the foreign-born population (GAP)--seem to have no bearing on
All the of methodology and sources of the variables are in Appendices A and B.
Only the regressions using fixed effects are reported here because the economic data are indexed to 1900. These data cannot be used to explain levéis, only the change over time.

policy. Contrary to the conventional wisdom, we do not fínd that immigration restrictions stemmed from a rising gap between the ethnic source of "oíd" and "new" immigrants, ñor from expanding immigrant ghettoes.
Macroeconomic Effects.   Measures of current macroeconomic conditions—real wage growth (WGRR), growth in real GDP per capita (YPCGGR), and unemployment (UNEMP2)~are also of little help in accounting for long-run policy changes.   Each consistently records the wrong sign, and none is significant.   Despite a literature that supports their influence on short-run timing, there is no evidence here that these factors contributed to policy formation in the long run.   By averaging the data into five-year periods, all the year-to-year cyclical behavior of these variables has been removed, eliminating the possibility for these short-run effects to emerge.   The only effects captured would be more fundamental shifts in long-run economic performance.
Real Wage Effects.   Real wages (wager) do not show the empirical importance they are often assigned in the literature.   In the absence of measures of relative income (equation 11), real wages have the wrong impact on policy.   High wages are associated with restrictive policy, low wages with open policy.   This perverse sign is not the result of confused causality, since causality tests strongly reject the view that tighter policy was effective in raising wages.13 When relative wages (wtoy, the ratio of unskilled wages to per capita gdp) are included in the equation, real wages have no significance at all (equations 1-4).
Trade Policy Effects.   The measure of openness to trade, xmtoy, seems to conflict with the measure of the human capital contení of immigrants, IMWAGE.   When both are included in the regression, neither has a significant impact (equation 6).  But when either
Argentina is an exception to these causality-test results. These results do not imply that immigration had no impact on wages. Indeed, we know that it did (Taylor and Williamson 1997; Hatton and Williamson 1997). Rather, it implies that policy changes did not have a large enough impact on immigration to matter.
one is alone in the regression, they are significant.   However, unlike imwage, openness to trade behaves as one would expect:   it is positively correlated with openness to immigration.   This result follows from the premise that trade and immigration were substitutes in the New World economies (Collins et al, 1997).   Restricting immigration to shore up wages would be ineffective if the cheap labor were imported in the form of goods rather than the people themselves.   Conversely, raising tariffs to restrict trade would be ineffective if free immigration were allowed.   Thus, immigration policy and trade policy should have moved together in these labor-scarce economies.
Relative Income Effects.   Aside from the lagged dependent variable, the most consistently significant variable is wtoy.   Recall that wtoy is the ratio of the unskilled wage to income per capita, or of income near the bottom of the distribution to income of the average.   Immigrants tended to be unskilled and to receive lower wages.   Thus, wtoy captures both the relative position of recent immigrants and those most threatened by immigration—the unskilled native born.  Regardless of what else is included in the equation, including wage growth, real wages, and immigrant wages relative to native wages, this measure of unskilled labor's relative position stands up as a positive factor influencing policy.   High wages paid to the unskilled, relative to average income, correlate well with more open immigration policies.14
The significance of the successful variables is fairly stable across specifications. The exceptions are in equations 9-11, where the threat variable is decomposed into its component parts.   The immigration rate (imrate) seems to be driving the results and causing severe multicollinearity problems.   It is conventional wisdom that immigration rates correlate highly with immigration policy.   However, in other (unreported) specifications of
This effect is likely to be biased downwards since open immigration policy implies more immigrants, driving down unskilled wages and thus lowering WTOY.
the regression equation, the immigration rate did not have the hypothesized effect.   Relative immigrant wages (imwrel) are highly signifícant, but have the wrong sign.   Furthermore, real wages also have the wrong sign.   Although equations 9-11 offer a good fit to the data, they make little theoretical sense.   They are therefore purged from the analysis.
The pattern that has emerged from the New World economies taken together is that the size and quality of the immigrant flows did not have any predictable impact on long-run policy change, except when the flows posed a threat to native labor.   It was not a pattern of racism or xenophobia.   Rather, labor market effects mattered, especially income inequality.
Individual Country Results
The regressions using annual time series for individual countries produce interesting results.   Table 2 presents our favorite specifications.15  In general, the measure of relative income (wtoy) continúes to be the common source of change in immigration policy (here called polism).   The exception is Australia, where the coefficient is significant but of the opposite sign.   However, empirical results for Australia discussed above suggest that immigration seems to raise wages rather than depress them.   If so, Australian policy should have responded to rising inequality by opening the gates to immigration!   We place little weight on this argument.16
Australia and Argentina seem to have been better than other New World countries in coordinating trade and immigration policy. For them, openness to trade was significant and positively related to pro-immigrant policy.   The data from these two countries are probably
15 In these regressions, we have exponentially smoothed the dependent variable.
16 It turns out that Australia is unusual all around. Indeed, it was the only country in which immigration
policy responded to current economic conditions in an expected way: opening the doors when growth was good
and shutting the doors when unemployment was high. The other countries in the sample showed no correlation
between current economic conditions and policy.

driving our results from the panel data.   The relationship was not significant for the other countries.
The various measures of the immigrant flow do not hold up well for the individual countries.   We see a significant effect of the human capital proxy at work in Canadá, but only weakly present in Australia and not at all in the other countries.   The other measures of immigrant flows seem to be responding to policy, rather than the other way around. Some regressions in Table 3 illustrate these endogeneity problems.   In the United States, for example, the positive relationship between gap and polism suggests that either the American public was xenophilic or United States policy was effective in selecting flows of immigrants that more closely matched the current ethnic make-up of the country.   Causality tests suggest the latter was the case, although not with a high degree of confidence.
The threat variable exhibits similar behavior for Australia, Brazil, and the United States.   Relatively open policy was correlated with high rates of immigration or low relative immigrant wages, or both.  The Brazilian data allow us to reject the idea that this correlation stemmed from a successful policy in reducing the threat, although we could not reject this idea for the United States and Australia.   In the pooled regressions, the use of immigrant flows from five years prior minimized these causality problems.   There, the relationship was negative; so there is no need to worry about overestimating the impact of the threat to native labor.
How Big are the Effects?
Using the results from Table 2, the next question is how much each variable contributed to closing the doors to immigrants.   For each country, we identified a period of major change towards more restrictivé immigration policy.   How much of the change was

due to changes in inequality, trade policy, or immigrant flows?17
Between 1865 and 1885, the immigration policy index for the United States dropped by 2 points (Table 4).18 Roughly two-thirds of that drop can be attributed to falling relative incomes of the unskilled.   On the other hand, Goldin (1994) is not wrong when she attributes the passage of the literacy test to other (non-market) factors.   The 2.5-point drop from 1885 to 1917 was due only in small part to rising inequality.   Furthermore, the residual is very large during this period, confirming the views of American historians who stress non-market forces.
Canadá offers the strongest argument that markets mattered.   During the Prairie Boom from 1899 to 1919, the policy index dropped 6 points.   Nearly 50 percent of this drop can be attributed to rising inequality o ver those two decades, and another 33 percent to increased threat from wages abroad.   We estímate that 10 percent of the shift was due to falling human capital content of the immigrants.   The twentieth-century response to this drop in quality was a change in Canadian policy to allow for very effective manipulation of immigrant "quality."
In Argentina, the change in openness to traded goods more than explains the change in openness to immigrants.   That is, given the estimated relationship between trade and immigration policy in Argentina, the collapse in trade flows accounts for more than the observed retreat from a pro-immigration policy.   Other factors supported this retreat, including increasing inequality.   What remains a puzzle are the offseting variables (the residual) that kept immigration policy from becoming even more restrictive.
We do not measure the impact of past performance on the lagged dependent variable, although we do multiply through the changes in the explanatory variables as they play out slowly within the period.
18  Appendix E details the methodology of the calculations.

When the Brazilian door slammed shut in the 1920s, about one-fourth of the 6.5-point drop in the policy index was due to rising inequality.   About one-third was due to the drop in real wages.   Although the residual is large, market forces still account for 60 percent of the policy switch.
For Australia, the estimated equations do not explain nearly as much of the change during the late 1920s.   The residual is almost 70 percent, which is similar to the results for the United States from 1885 through World War I.
These results point to long-term influences driving immigration policy that are very different from the short-term influences about which so much has been written.   There is no compelling evidence that xenophobia or racism was at work in these economies, once underlying economic variables are given their due.   Real wages of the unskilled did not matter all that much by themselves.   Ñor did economic growth or unemployment matter. Over the long haul, these countries tried to maintain the relative economic position of the unskilled worker, compared with the skilled worker or the industrialist.   Labor became relatively more abundant when immigrants poured in, and governments sought to stop the relative decline in their wages and in the wages of the native unskilled with whom they competed.   The greater the perceived threat to these wages from more immigrants or lower-quality immigrants, the more restrictive policy became.   Governments responded in the same way to rising land valúes in relation to unskilled labor, suggesting that Foreman-Peck was right to argüe that land ownership mattered, too.19
The results here may offer some predictions for the outcome of the contemporary debates about immigration.   The parallels are clear.   Inequality has been on the rise in the
19  The results for these regressions are not reported here.
OECD economies since the early 1970s, especially the gap between unskilled and skilled workers, just as it was in the New World economies in the late-nineteenth century.   We should therefore not be surprised by the renewed interest, both in the United States and Europe, in reducing the flow of immigration.   Labor-scarce economies have been sensitive in the past to trends of greater inequality in their midst, using restrictive immigration policy to offset those trends.   If the story repeats itself, policies will become increasingly anti-immigrant in the future, at least as long as unskilled workers lag behind other economic groups.