J Popul Econ (2000) 13: 509±525
2000 9 99
How much did immigrant ``quality'' decline in late nineteenth century America? Timothy J. Hatton Department of Economics, University of Essex, Wivenhoe Park, Colchester CO4 3SQ, UK (Fax: 44-1206-872724; e-mail:
[email protected]) Received: 1 September 1997/Accepted: 6 June 1998
Abstract. Early twentieth century observers argued that recent American immigrants were inferior, and in particular less skilled, than the old. I estimate wage equations for 1909 allowing for di¨erent e¨ects by nationality and for di¨erent characteristics on arrival. I then apply the estimated wage di¨erentials to the immigrant composition to measure the e¨ect of changing composition on immigrant earnings. Finally I ask how immigrant earning power changed relative to that of native Americans. I conclude that immigrant ``quality'' in terms of earnings did decline due to shifting composition but these e¨ects are very small compared with those reported in studies of the post-second World War period. JEL classi®cation: J15, J31, J61, N11 Key words: United States immigration history
1. Introduction Recent research on postwar American immigration has been much concerned with the economic implications of the rise in immigrant numbers and the evident decline in the ``quality'' of immigrants as judged by their labour market performance. Economists often refer in passing to an earlier era which shared
For useful comments I am grateful to Je¨ Williamson, Bob Whaples and to participants at the Cliometrics Conference at Toronto, May 1997, at the Canadian Economic History Conference at Niagara-on-the-Lake, May 1997, and at the European Society for Population Economics at Essex, June 1997. I am also grateful to two anonymous referees for their helpful suggestions. Responsible editor: Klaus F. Zimmermann.
510
T.J. Hatton
these characteristics ± the period between 1870 and 1913. This era of mass immigration culminated in the imposition of quotas with the Immigration Acts of 1921 and 1924, an event often linked to the rise in numbers and the alleged decline in ``quality'' (see for example, Borjas 1994, p. 1668±9). How close are the parallels between these two eras of mass immigration? Until now it has been impossible to say with con®dence because, although the rise in numbers in the earlier period is clear enough, there have been no studies of immigrant ``quality'' which would a¨ord comparison with those for postwar immigration. Contemporary observers in the early twentieth century concentrated on the characteristics, such as occupation and literacy, of immigrants when they arrived in the United States. They formed their assessments of changes in immigrant quality largely on these rather than on the subsequent performance of the di¨erent immigrant groups in the American labour market. In particular they contrasted the characteristics of the so called new immigrants ± those who came from southern and eastern Europe, with the so-called old immigrants ± those from Northwest Europe. It should be noted that, as in the later period, the distinction between new and old is strictly in terms of country of origin rather than the time of arrival of individual immigrants. More recently a number of historical studies have examined the wageearning performance of immigrants compared with the native-born in order to assess the degree of immigrant assimilation. These studies have often pointed to di¨erences in labour market performance between old and new immigrants. But they have not distinguished between individual nationalities and they have not used their results to assess the labour market implications of the changing composition of the immigrant in¯ow. As a result it has not been possible to compare trends in immigrant wage earning performance in the late nineteenth century with those experienced in the last 40 years. In this paper I re-examine the wage rates of new and old immigrant groups, linking these to their characteristics on arrival. I then use the results to measure the putative decline in the ``quality'' of the immigrant in¯ow. The results indicate that immigrants' relative wages did decline but that the decline was more modest than some contemporary observers suggested and much smaller than recent studies indicate for the post-second World War period. Thus while there are similarities in qualitative terms between the two periods, the late-nineteenth century decline in ``quality'' was very modest compared with that of more recent times. 2. Immigrant characteristics on arrival The parallels between immigration in the era of free migration and that of the last 40 years are quite striking. The number of immigrants rose from about 200,000 per annum in the 1870s to 800,000 per annum in the decade before the ®rst World War ± similar to the rise between the 1950s and the 1990s. In 1910 the proportion of foreign-born in the population was 14.6% ± nearly twice its level in 1990. But, because natural increase is slower than it was a century ago, the contribution of immigration to population growth is now approaching the level of the late nineteenth century. Equally striking is the change in the composition of the immigrant in¯ow by country of origin. The share of immigrants who came from Europe fell
Immigrant ``quality'' decline
511
Fig. 1. American immigration from Europe 1873±1913
from a half to 10% between the 1950s and the 1980s. This was paralleled in the late nineteenth century by the decline in the share of those from Northwest Europe and the rise in the share from southern and eastern Europe. It might be argued that the diversi®cation in immigrant origins towards non-European sources is qualitatively di¨erent from the shift within Europe in the earlier period. But the same issue, that of the e¨ects of the shifts in composition, is at the heart of the debate in both cases. In both periods the so-called new immigrants have been perceived as ethnically di¨erent and they have been judged to be inferior in terms of labour market performance than the old immigrants and the native-born. Figure 1 illustrates the uneven rise in European immigrant numbers between 1873 and 1913 and it shows that this rise was entirely due to the growth in numbers from southern and eastern Europe. In 1873 these new immigrants were just 6% of all European immigrants, by 1895 they were in the majority and by 1913 they formed nearly 83% of all European immigrants. Although their rates of return migration were higher than that of the old immigrants, the proportion of new immigrants among the stock of (European) foreign born rose from 10% in 1890 to 44% in 1910. These trends alarmed many contemporary observers and such concerns led to the establishment of the United States Immigration Commission (the Dillingham Commission) which reported in 1911 after collecting evidence for 4 years. The Immigration Commission took a very negative view of the new immigrants, emphasising the ethnic di¨erences between them and the old immigrants. They argued that the new immigrants, because of their lack of industrial quality, undercut the wages of native American and old immigrant
512
T.J. Hatton
labour and that, unlike the old immigrants, they assimilated poorly into American society and into the American labour market. In the recommendations of the (majority) Report it was argued that: ``while the American people, as in the past, welcome the oppressed of other lands, care should be taken that immigration be such both in quality and quantity as not to make too di½cult the process of assimilation'' (Vol. 1, p. 45). Among the methods canvassed for achieving this aim were the imposition of a literacy test, the exclusion of unskilled labourers unless accompanied by their wives, and some form of quota on individual nationalities (Vol. 1, p. 47). While it is doubtful that many contemporaries read the full 41 volume Report the same ideas were summarised and presented more forcefully by one of the leading members of the Commission, Jeremiah Jenks, together with Jett Lauck. Of the new immigrants, they argued that: ``their general, as well as their industrial progress and assimilation are retarded by segregation in colonies and communities where they have very little contact with American life and small opportunity to acquire the English language'' (1926, p. 78). Their book The Immigration Problem reached a wide audience and went through six editions between 1911 and 1926. Other writers expressing similar sentiments, and in some cases drawing on the ®ndings of the Immigration Commission, also reached a wide audience (for example Commons 1920; Grant 1923). Thus, directly or indirectly, the conclusions of the Commission exerted a considerable in¯uence on the course of events. Jenks and Lauck found that in the decade 1899±1909, 9.2% of new immigrants were classi®ed as either professional or skilled as compared with 22% of old immigrants. For the same period, 35.8% of new immigrants (aged 14 and over) were illiterate when they arrived as compared with only 2.7% of old immigrants. Paul Douglas (1919) criticised Jenks and Lauck for omitting Hebrews (predominantly from eastern Europe) from their new immigrant group. But more importantly, he argued that a more reasonable comparison would between new immigrants in 1899±1909 and old immigrants in 1871± 1882 when the latter had formed the bulk of the in¯ow. Douglas showed that, in the earlier period, 12.2% of immigrants from northwestern Europe were either skilled or professional while for the later period, 17.0% of those from southern and eastern Europe were so classi®ed. If the comparison is restricted only to those declaring occupations (rather than all immigrants) then 24.6% of old immigrants in 1871±1882 were skilled or professional as compared with 18.8% of new immigrants in 1899±1909 (1919, p. 400±1). Douglas concluded that there was little evidence that the ``new'' immigrants were signi®cantly less skilled than the old. The ®ndings of Jenks and Lauck carried the clear implication that the shifting composition of immigrant origins lowered immigrant ``quality'' and Douglas's argument does nothing to undermine that point. In the absence of the shift in immigrant composition, immigrant quality as judged by these measures would have signi®cantly increased. The shifting immigrant composition eliminated, or at least retarded, the improvement in quality that would otherwise have occurred. This compositional e¨ect can be examined further in Table 1. The table compares characteristics of the actual immigrant ¯ow with ®xed weight indices using the 1893 composition based on the division into 24 countries. Rows (1) and (2) show that the proportion of immigrants skilled or professional among all immigrants reporting occupations (male and female) fell by
Immigrant ``quality'' decline
513
Table 1. Characteristics of European immigrant ¯ow 1873±1913
Percentage skilled or professional among those with occupations (1) Actual (2) 1893 weights Percentage skilled or professional/total males (1) Actual (2) 1893 weights Percentage literate (aged 14 and over) (1) Actual (2) 1893 weights
1873
1893
1913
24.1 22.1
19.6 19.6
13.7 17.5
19.6 17.7
15.6 15.6
15.3 20.4
± ±
77.4 77.4
73.0 83.8
Notes: The 24 countries included in the calculations are as follows. Old immigrant countries: Belgium, Denmark, France, Germany, Netherlands, Norway, Sweden, Switzerland, England, Ireland, Scotland and Wales. New Immigrant countries: Austria, Bulgaria/Serbia, Greece, Hungary, Italy, Poland, Portugal, Rumania, Russia, Finland, Spain and Turkey. In 1899 the Immigration Bureau changed the basis of its classi®cation from country of origin to nationality or ethnic group. Fortunately they also provided tables mapping this ®ner breakdown into the country of origin classi®cation used earlier. A matrix from the table for 1913 was used to convert the characteristics recorded in the later classi®cation on to the basis used earlier.
10.4 percentage points while the ®xed weight index fell by 4.6 percentage points. Rows (3) and (4) report the similar comparison for the ratio of those reporting skilled or professional occupations (overwhelmingly male) to all male immigrants. The actual decline of 4.3 percentage points is compared to a rise of 2.7 percentage points in the ®xed weight index. Finally, rows (5) and (6) show that the decline of 4.4 percentage points between 1893 and 1913 in the proportion of immigrants who were literate (in any language) is transformed into a 6.4 percentage point increase in the ®xed weight index. On the face of it these statistics point to the conclusion that the changing country of origin mix reduced immigrant ``quality''. But such statistics raise a number of questions which have not previously been addressed. Are these characteristics linked to subsequent labour market performance? Did the changing immigrant mix signi®cantly reduce the earning power of immigrants? How did the change in immigrant quality and earning power compare with that of native Americans? In order to answer these questions we need to examine the earnings of immigrants by country of origin. 3. Immigrant earnings and immigrant origins The usual method of assessing the labour market performance of immigrants is to estimate earnings functions for immigrants and for natives using controls such as education, experience and (for immigrants) years since arrival. The economic assimilation of immigrants is judged by the degree to which the immigrants catch up with, and perhaps even overtake, the earnings levels of the native-born. The pioneering studies for the postwar period by Chiswick (1978) and Borjas (1985) have been followed by a plethora of studies using
514
T.J. Hatton
essentially this approach, many of which are surveyed in Borjas (1994). The approach adopted here is similar in spirit to these. The available data is limited. A number of studies have been performed on earnings for individual workers from around 1890 (some of which are discussed further below) but these datasets contain too few observations on ``new'' immigrants to provide estimates for individual nationalities. In order to estimate immigrant earnings by nationality we have little choice but to use the results of the extensive enquiries of the Dillingham Commission itself. The Commission conducted a survey of over half a million employees in a variety of manufacturing and mining industries in 1909. Unfortunately the original micro data do not survive. But the published volumes of the Immigration Commission's Report contain wage rates and employee characteristics by industry for about 60 di¨erent ``races'' or nationalities including the native-born of native parent and the native-born of foreign parent. These cells of industry by ``race'' have formed the basis of studies by Higgs (1971), McGouldrick and Tannen (1977) Blau (1980) and Chiswick (1992). Those studies did not provide separate earnings estimates for individual races or nationalities (with the partial exception of Chiswick who distinguished hebrews separately) although they did distinguish between the broad groupings of old and new immigrants. The approach adopted here is somewhat di¨erent to that of the previous studies. Those studies did not control for the heterogeneity arising from di¨erent national origins within broad immigrant groups and they are potentially vulnerable to Borjas's (1985) critique of cross-sectional estimates in the presence of changes in the quality of di¨erent cohorts. If the quality of successive cohorts declined then the cross-sectional estimate of the immigrant age-earnings pro®le will be upward biased because the more recent (low quality) cohorts receive lower wages than the earlier (high quality) immigrants did at a similar stage in the life cycle. Borjas (1994, p. 1685) suggests that 90% of the cross sectional upward bias in the earnings pro®les on postwar samples can be accounted for by changes in country of origin mix. These e¨ects can therefore be captured by controlling for nationality. Furthermore, declines in cohort quality within a country of origin are not likely to be important in the era before World War I because there were no changes in the rules governing entry (and in¯uencing immigrant quality) comparable with the 1965 Amendments to the Immigration Act. I return to this point further below. In order to make estimation of nationality e¨ects feasible a simple model with few controls is adopted. This is appropriate in the present context because I wish to avoid using as explanatory variables characteristics such as literacy which could have been acquired since arrival in the United States. I constructed a new dataset from the cross-tabulations of average wages and characteristics by race or nationality in 16 separate industries for which su½cient detail is reported in the ``Immigrants in Industry'' volumes of the Immigration Commission Report (Vols 7±16). Since the variables required have fewer missing values than some of those used in previous studies, I am able to double the number of available observations ± to 409 in all. Even so, there are relatively few observations for some races or nationalities and it was necessary to aggregate some of them into larger (but ethnically similar) groups, 26 in all. This aggregation was also guided by the classi®cation of immigrants adopted after 1899 by the United States Immigration Bureau. Using these observations, I regress the log of the average weekly wage on dummy variables for each race (26 in all), for each industry (16 industries,
Immigrant ``quality'' decline
515
15 dummies) and on the share of employees in certain age groups. Thus the di¨erence between nationalities appears only as an intercept shift and the omitted group is the native-born of native parent. Also included is a variable representing the time since arrival in the United States. This variable is de®ned as [(Y/20)*S1 S2] where S1 is the share of immigrants in a cell who had been in the United States for less than 20 years and S2 is the share who had been in the United States for 20 years or more. Y is the average number of years in the United States for those who had been present for less than 20 years (based on 5-year duration categories) and it is divided by 20. Thus the variable ranges between a value of zero for newly arrived immigrants and unity for immigrants who had been present for 20 years or more. It embodies the plausible assumption that assimilation e¨ects are only important for the ®rst 20 years and it summarises the assimilation e¨ects in a single variable. On a more practical level, it also solves the problem of what to do with the open ended category ``20 years or more in the United States''. The results of the estimation appear in Table 2. Unweighted least squares was used for these regressions. For equation (1), the Cook-Weisberg test for heteroskedasticity using the number of workers in each cell as the weighting variable yielded a w2 (1) test statistic of 0.28 compared with the 5% critical value of 3.84. When weighted least squares was used the null hypothesis of no heteroskedasticity was strongly rejected. In equation (1) the variables for the shares of workers in the younger age groups give negative coe½cients re¯ecting a steeply rising age-earning pro®le in the early years and which is consistent with evidence from nineteenth century micro data (Hatton 1997). The dummy variable for the native-born of foreign parent indicates that their earnings were 6.5% higher than the native-born of native parent (the excluded group) ± a result also consistent with other evidence (Chiswick 1977). One interpretation of this result is that immigrants are positively self-selected and they pass some of their labour market characteristics on to the next generation. Since second generation immigrants do not su¨er the same assimilation disadvantage as the ®rst generation, the inherited self-selection e¨ect dominates. The coe½cient on the assimilation variable indicates that in the ®rst 20 years after arrival immigrant earnings increased by 20.6% relative to the native-born. The dummies for race or nationality provide estimates of the immigrant wage disadvantage on arrival in the United States relative to the native-born of native parent. All the coe½cients are negative and signi®cant. Among the old immigrant group the initial wage disadvantage ranges from 8.9% for the English to 24.2% for the Irish. Among the new immigrant group the initial disadvantage ranges from only 14.2% for the group of Bosnians, Dalmatians and Herzegovinians to 30.2% for the Syrians and Turks. The unweighted average for new immigrants is 23.5% compared with 15.9% for the old immigrants. Thus the new immigrants as a whole did su¨er a clear disadvantage compared with the old although there is some overlap between old and new immigrant nationalities. Two of the largest old immigrant groups, the Germans and the Irish, su¨ered an initial disadvantage of more than 20% while four of the new immigrant groups had an initial disadvantage of 20% or less. The wage di¨erential after 20 years in the United States can be calculated by adding the individual race or nationality coe½cient to the coe½cient on the assimilation variable. As an illustration, the ``fully assimilated'' wage di¨eren-
516
T.J. Hatton
Table 2. Weekly wage functions: Immigration commission data (dependent variable: log weekly wage)
Constant Share aged 18±19 Share aged 20±24 Native-born, foreign parent Foreign-born assimilation duration variable Foreign-born assimilation duration *new immigrants
(1)
(2)
(3)
2.772 (93.20) 0:649 (6.53) 0:195 (2.99) 0.065 (2.25) 0.206 (4.17)
2.663 (60.66) 1:087 (7.27) 0:118 (1.14) 0.074 (1.52) 0.204 (2.57)
2.660 (60.08) 1:076 (7.14) 0:108 (1.026) 0.074 (1.51) 0.155 (1.43) 0.098 (0.66)
0:151 (3.52) 0:157 (3.88) 0:089 (1.99) 0:158 (3.67) 0:142 (3.62) 0:205 (4.42) 0:242 (4.93) 0:127 (2.76)
0:158 (2.28) 0:226 (3.40) 0:113 (1.54) 0:138 (1.94) 0:188 (2.92) 0:221 (2.92) 0:264 (3.07) 0:152 (2.03)
0:120 (1.32) 0:198 (2.54) 0:075 (0.81) 0:158 (2.04) 0:160 (2.08) 0:179 (1.81) 0:221 (2.12) 0:111 (1.14)
0:285 (7.29) 0:192 (5.51) 0:142 (2.75) 0:236 (7.43) 0:200 (6.14) 0:281 (8.91) 0:157 (4.57) 0:219 (6.19) 0:282 (8.39) 0:249 (6.96)
0:358 (5.52) 0:199 (3.50) 0:075 (0.88) 0:223 (4.23) 0:155 (2.88) 0:292 (5.57) 0:186 (3.26) 0:212 (3.61) 0:301 (5.38) 0:274 (4.60)
0:376 (5.32) 0:222 (3.33) 0:088 (1.00) 0:232 (4.25) 0:170 (2.91) 0:303 (5.52) 0:206 (3.18) 0:231 (3.54) 0:318 (5.14) 0:291 (4.49)
Race/Nationality dummies: Old immigrant group (no of observations in parentheses) Danish/Norwegian/Swedish (32) Dutch/Flemish (11) English (16) Finnish (7) French/Belgian (14) German/Swiss (22) Irish (15) Scottish/Welsh/Scots-Irish (19) Race/Nationality dummies: New immigrant group (No of observations in parentheses) Armenian (9) Bohemian/Moravian/other Austrian (24) Bosnian/Dalmatian/Herzegovinian (4) Bulgarian/Montenegrin/Serbian (16) Croatian/Slovenian (18) Greek/Macedonian (18) Hebrew Russian/Hebrew other (20) Northern Italian (15) South Italian/other Italian (17) Lithuanian (13)
Immigrant ``quality'' decline
517
Table 2. (Continued) (1) Magyar (12) Polish (15) Portuguese/Spanish (4) Rumanian (8) Russian (14) Ruthenian (8) Slovak (12) Syrian/Turkish (14) R2 RSS No. of observations
0:259 (7.40) 0:245 (6.89) 0:193 (3.78) 0:239 (6.381) 0:237 (7.10) 0:269 (6.91) 0:238 (6.46) 0:302 (8.52) 0.839 2.432 409
(2)
(3)
0:250 (4.32) 0:261 (4.41) 0:336 (3.99) 0:202 (3.23) 0:257 (4.62) 0:310 (4.82) 0:233 (3.80) 0:371 (6.30)
0:266 (4.26) 0:280 (4.24) 0:355 (4.00) 0:212 (3.29) 0:271 (4.54) 0:325 (4.78) 0:252 (3.71) 0:385 (6.15)
0.521 7.171 409
0.522 7.162 409
Note: Industry dummies included in (1) but not in (2) and (3). The industries are: bituminous coal, iron and steel, woollen and worsted, cotton, leather boots and shoes, silk, gloves, clothing, slaughtering and meat packing, agricultural implements, cigars and tobacco, furniture, sugar re®ning, copper mining, oil re®ning and iron ore mining.
tial for the English would be 0:206 0:089 0:117 while that for the Armenians would be 0:206 0:285 0:079. On average the old immigrants had a wage advantage of 4.7% over the native-born of native parent when fully assimilated while the new immigrants still had a disadvantage of 2.9%. These ®gures are the unweighted averages for each immigrant group. If the coe½cients are weighted by the number of immigrants or the number of observations then the di¨erentials are, respectively, for old immigrants 3:7 and 4.7% and for new immigrants 3.0 and 2:7%. The unweighted di¨erence of 7.6% between the two immigrant groups is consistent with the ®ndings of previous studies based on data drawn from the Immigration Commission Report. McGouldrick and Tannen (1977, p. 731) estimated a di¨erential of 5.1% while Blau (1980, p. 27) obtained a di¨erential of 5.6% (although both used additional controls such as literacy). The estimate for Hebrews which gives a fully assimilated di¨erential of 4.9% is also consistent with Chiswick's (1992, p. 283) ®nding that Jews did considerably better than other new immigrants and that they overtook the native-born. Equations (2) and (3) in Table 2 address two possible concerns about the speci®cation of the model. Industry dummies were included in (1) order to capture di¨erences in weekly hours worked, to re¯ect compensating di¨erentials, and (more tentatively) to allow for regional wage di¨erences. Some of the industry coe½cients (not shown) are quite large, yielding a wage di¨erential, for example, of 30% in iron and steel over woollen and worsted. If immigrants achieved wage growth by moving across industries then the inclusion of industry dummies might bias the coe½cient on the assimilation variable
518
T.J. Hatton
down. As equation (2) shows, this is not the case: when the industry dummies are omitted, the coe½cient on the assimilation term is unchanged and the nationality coe½cients change only slightly. Another possible concern is that immigrants who arrived with few skills and low levels of literacy, especially the new immigrants, would achieve faster subsequent earnings growth as they made greater investments in human capital. Duleep and Regets (1997) examining immigrants arriving in 1968±80 ®nd support for this view over the alternative that immigrants with di¨erent arrival wages assimilated at similar rates In order to test this hypothesis, in equation (3) the assimilation variable is interacted with a dummy for new immigrants. The result indicates that although the interaction term takes a positive coe½cient it is not signi®cant. Further regressions were estimated which interacted the assimilation variable ®rst with ®ve regional dummies rather than two and then with a dummy for each race or nationality. In neither case could the restrictions implied by equation (2) be rejected. Thus although there is some evidence that new immigrants, or those with lower arrival wages, had faster rates of assimilation, in no case was the di¨erence in assimilation rates statistically signi®cant.1 4. Immigrant earnings and characteristics on arrival The evidence suggests that, even allowing for duration in the United States, the new immigrants fared somewhat worse than the old immigrants, although the di¨erence should not be exaggerated. How do these di¨erences in wage performance relate to the characteristics exhibited by the di¨erent immigrant groups on arrival in the United States? Were contemporary observers correct to point to measures such as the proportion with skilled occupations or the proportion who were literate as indicators of labour market quality? We can approach these questions by examining the relationship between the wage and the arrival characteristics of di¨erent nationalities, again using the Immigration Commission data. In Table 3, the individual nationality dummies used in Table 2 are replaced by variables for literacy and occupational structure for the immigrant in¯ow in 1899±1909. These characteristics are de®ned for 28 di¨erent nationalities and are taken from the Immigration Commission Report (Vol. 1, pp. 171±175). The results in Table 3 indicate that there is only a small fall in explanatory power when these variables are used instead of the dummies. Furthermore the coe½cients on the other variables are little changed although the assimilation variable gives a slightly larger coe½cient than it did before. The ®rst two regressions show that there is a strong positive and signi®cant relationship between earnings and either the level of immigrant literacy or the proportion skilled relative to all males. In both cases the `t' statistic exceeds 5. Thus contemporary observers were correct in pointing to these characteristics of the immigrant in¯ow as indicators of immigrant quality as re¯ected in subsequent labour market performance. When a variable for the ratio of farm labourers relative to all males is included with the proportion skilled or professional as in equation (3) it gives a negative sign ± also consistent with contemporary observations. Finally in equation (4) when both the proportion skilled and the literacy rate are included, both coe½cients are signi®cant although they are or smaller than when each variable appeared alone.
Immigrant ``quality'' decline
519
Table 3. Wage functions with immigrant arrival characteristics (dependent variable: log weekly wage)
Constant Share aged 18±19 Share aged 20±24 Native-born, foreign parent Foreign-born duration variable Proportion 14 and over literate Skilled or professional/total male immigrants
(1)
(2)
(3)
(4)
2.411 (61.72) 0:690 (6.79) 0:162 (2.58) 0.065 (2.10) 0.249 (7.25) 0.173 (5.08)
2.515 (84.66) 0:662 (6.59) 0:215 (3.61) 0.066 (2.13) 0.263 (8.02)
2.560 (65.40) 0:662 (6.60) 0:230 (3.84) 0.066 (2.14) 0.239 (6.76)
0.125 (5.96)
0.092 (3.30) 0:086 (1.76)
2.432 (63.36) 0:668 (6.73) 0:158 (2.60) 0.065 (2.13) 0.229 (6.76) 0.119 (3.36) 0.100 (4.54)
0.799 3.009 409
0.804 2.939 409
0.805 2.915 409
Farm labourers/total male immigrants R2 RSS No. of observations
0.809 2.856 409
Note: The characteristics used as explanatory variables are for 28 di¨erent nationalities, two more than the number of nationality dummies used in the estimates of Table 2. This arises because the Portuguese and Spanish are treated as separate groups here as are the Scottish and the Welsh.
This re¯ects the positive correlation between these two characteristics across nationalities. When further variables were added to equation (4), such as the share of males, the share of adults in the immigrant in¯ow or number of farm labourers relative to all males, these never took a signi®cant coe½cient. It might alternatively be suggested that the endowment of skills or literacy on arrival would in¯uence the rate at which the wages of certain immigrants grew. But interactions between the assimilation variable and either the share who were literate or the share who were skilled failed to produce signi®cant coe½cients. This supports the ®nding in Table 2 that di¨erences in rates of assimilation are not statistically signi®cant. How do the estimated wage di¨erentials translate into changes in the quality of the immigrant in¯ow between 1873 and 1913? The ®rst row of Table 4 weights the estimated wage di¨erentials (fully assimilated) from the dummy variables in Table 2 (equation 1) by the proportions of the country of origin in the immigrant in¯ow for 1873, 1893 and 1913. The total decline between 1873 and 1913 in the immigrant relative wage based on the 1909 estimates is less than 5%. This re¯ects both the modest di¨erential between new and old immigrants estimated for 1909 and the relative decline in immigration from some of the low wage old immigrant groups such as the Irish and the Germans. If new immigrants or those with lower initial wages had faster subsequent wage growth then the compositional e¨ect would be even smaller. Since the calculation is based on an estimate where rates of assimilation do not vary, the e¨ect can be treated as an upper bound.
520
T.J. Hatton
Table 4. Impact of immigrant composition on the relative wage (percentage di¨erence from native Americans in 1909) 1873
1893
1913
(1) Weighted 1909 wage di¨erential (Table 2, eq. 1)
4:2
1:8
0:5
Wage e¨ect of compositional change due to: (2) Proportion skilled or professional/total males (Table 3, eq. 2) (3) Proportion literate (aged 14 and over) (Table 3, eq. 1) (4) Proportion skilled and proportion literate (Table 3, eq. 4)
0:2 ± ±
0 0 0
0:6 1:7 1:6
Source: Calculated from estimates in Table 2 and Table 3
Such comparisons take no account of changes in the indicators of quality within and between di¨erent the di¨erent country of origin groups. The lower part of the table compares the wage predictions based on the variable and ®xed weight indices of immigrant quality presented in Table 1. The values reported in Table 4 are derived by applying the values of the characteristics in Table 1 and the equations in Table 3 to predict the average wage. These values measure the percentage di¨erence in the predicted wage between the actual and ®xed weight indices of characteristics. Using the ratio of skilled and professionals to total males (row 2) indicates a relative wage decline of 0.8% (0:2 0:6). Thus the e¨ect of changing skill due to the nationality composition on the relative wage is less than 1%. Row (3) suggests that the literacy e¨ects of composition on relative earning power was 1.7 percentage points between 1893 and 1913 and it was about the same for literacy and skill combined (row 4). These comparisons suggest that the e¨ects of the change in immigrant composition were relatively small ± no more than 5% based on applying estimated relative wages to the changing immigrant composition and less than this if the relative wage is related to the characteristics of the immigrant in¯ow. But these estimates are based on the assumption that immigrant wage di¨erentials remained constant relative to the native-born in 1909. There remains the question of whether, or by how much, the labour market performance of immigrants deteriorated (or improved) relative to that of nativeborn Americans. 5. Did immigrant ``quality'' decline relative to natives? Paul Douglas (1919) suggested that, judging by characteristics on arrival, the quality of old immigrants improved between the 1870s and the 1900s. But he was not able to compare their performance relative to native Americans at two points in time. Recent evidence suggests that literacy and skills grew relatively rapidly in Europe between 1870 and 1913 and that unskilled wage rates in Europe grew faster than those in the United States (Tortella 1994; Williamson 1995). If so, it is possible that this underlying relative growth in immigrant ``quality'' o¨set the e¨ects of shifting immigrant composition away from old and towards new immigrant groups. Comparing independent crosssections at two points in time would reveal whether there were appreciable cohort e¨ects of the sort suggested by Borjas (1985).
Immigrant ``quality'' decline
521
Most of the datasets available for the study of immigrant and native wage rates for periods earlier than 1909 come from State Labor Bureau surveys and date from around 1890. While they contain relatively few new immigrants, and their coverage is sometimes narrow, they do o¨er micro-level data. A number of studies have been based on such sources (Hannon 1982a,b; Eichengreen and Gemery 1986; Hanes 1996). These studies have suggested that immigrant wages often failed to increase relative to those of native Americans with years since migration in contrast to the ®ndings of studies of post-second World War data and of Tables 2 and 3 above, but this conclusion has recently been challenged. Using a functional form which allows for the steep rise in the observed age-earnings pro®le between the ages 16 to 25 and the ¯attening out thereafter, I found that the wages of immigrants rose faster than those of natives in the twenty years following their arrival in the United States (Hatton 1997). In order to make a direct comparison with the results from the Immigration Commission data, I use two State Labor Bureau datasets, one for male workers in the agricultural implement and ironworking industries in Michigan in 1890 and one covering male workers in a mixed group of manufacturing industries in California in 1892. These datasets are explored in more detail in Hatton (1997). Both contain a considerable share of foreign-born among males aged 18 and over, but not surprisingly, relatively small shares of new immigrants. The dataset for Michigan ironworking contains observations on 7,925 workers of whom 32.6% are old immigrants and only 2.4% are new immigrants. The dataset for California manufacturing contains 2320 useable observations of which 42.6% are old immigrants and only 2.7% new immigrants. It clearly is not feasible to disaggregate the new immigrants in these datasets and so I compare the wages of the broad groupings of old and new immigrants with native born (®rst and second generation). In order to make the closest possible comparison with the results from the Immigration Commission data the same age groupings were used and the same variable is de®ned to capture assimilation e¨ects (zero for a newly arrived immigrant and rising with duration to unity for an immigrant who had been present for 20 years or more). The results appear in Table 5. The restricted version of the model estimated for the grouped data from the Immigration Commission indicates that, when fully assimilated, old immigrants earned 1.6% more than the nativeborn while new immigrants earned 4.3% less ± a di¨erence of 5.9% between the two immigrant groups. Although the assimilation variable takes a slightly larger coe½cient than it did in Table 2, the estimated wage di¨erential is reasonable close to the 7.4% (weighted by observations) calculated from the nationality coe½cients in Table 2. Thus there is little evidence that failing to allow for di¨erent immigrant origins within old and new immigrant groups seriously biases the estimated coe½cients for these groups as a whole. The result for the Michigan agricultural implement and iron workers in 1890 is strikingly similar to that from the Immigration Commission data ± a remarkable result given the di¨erences in period and composition of the data. The coe½cient for the assimilation term is marginally larger and that on the dummy for new immigrants somewhat smaller. The coe½cients indicate that, when fully assimilated, old immigrants earned 4.1% more than the nativeborn and new immigrants 6.1% less than the native-born. The results for
522
T.J. Hatton
Table 5. Weekly wage functions for 1909 and the early 1890s (dependent variable: log weekly wage)
Constant Age 18±19 Age 20±24 Foreign-born duration variable Old immigrant New immigrant R2 RSS No. of observations
Immigration Commission grouped data 1909
Michigan implement and iron workers 1890
California manufacturing workers 1892
2.821 (101.67) 0:673 (6.49) 0:219 (3.54) 0.232 (5.61) 0:216 (6.72) 0:275 (12.25) 0.79 3.16 409
2.428 (495.17) 0:553 (45.84) 0:172 (20.78) 0.249 (13.67) 0:208 (15.93) 0:310 (13.20) 0.25 706.17 7925
2.788 (233.10) 0:724 (23.82) 0:183 (9.65) 0.152 (4.37) 0:134 (5.28) 0:222 (4.58) 0.22 288.84 2320
California manufacturing workers in 1892 indicate a somewhat smaller coe½cient for the assimilation term. But the coe½cients imply that when fully assimilated, the old immigrant group earned 1.8% more and new immigrants 7.0% less than the native-born. These results suggest strong similarities between the labour market performance, relative to the native-born, of new and old immigrants who arrived before 1890 and those who arrived before 1909 (over 80% of whom had arrived since 1890). There is perhaps some indication that between these dates the performance of the new immigrants improved relative to old immigrants and the native born. The gap between old and new immigrants appears to have fallen from 9 or 10% in 1890 to 6% in 1909. The results suggest that there was no deterioration and there may have been a slight improvement in immigrant labour market performance after 1890. That slight improvement, when weighted by the composition of the immigrant in¯ow at these two dates would imply an improvement of about 1% in wage earning capacity and would provide some o¨set to the deterioration of 2.3% due to changing composition (between 1893 and 1913) estimated in Table 4. 6. Then and now How do the patterns of immigrant assimilation and the trends in immigrant ``quality'' found in the late nineteenth century compare with those observed in more recent times? The cross-sectional estimates for the late nineteenth century in Table 2 indicate that, on average, immigrant wage rates grew by about 20% in the ®rst 20 years after arrival ± a ®nding con®rmed in Table 5 and elsewhere (Hatton 1997). Estimates for the 1970s and 1980s which track cohorts across di¨erent censuses yield similar results. Lalonde and Topel (1992,
Immigrant ``quality'' decline
523
p. 89) found that the earnings of male immigrants who arrived in 1965±1969 increased by 9% relative to the native born in the decade 1970±1980. Borjas (1995) found that male immigrants arriving in 1965±1969 increased their wage rates relative to the native born by 18% between 1970 and 1990. Thus the rates of economic assimilation observed in the late nineteenth and early twentieth century, although based only on cross-section estimates, seem broadly comparable with those of the more recent past.2 How do average wage levels of immigrants relative to natives compare between now and then? In 1909 the average wage for immigrants was 6.4% lower than for natives; for Michigan in 1890 it was 2.3% lower; and in California in 1892 it was 6.1% higher. By comparison, Borjas (1994, p. 1674) reports that for men aged 25±64, immigrants earned 0.9% more than natives in 1970, 9.2% less in 1980 and 15.2% less in 1990. Although immigrants seem to have done slightly better in 1909 than in 1980, it must be remembered that the surveys used here are largely for blue collar occupations in which immigrants were somewhat overrepresented. In order to more closely approximate the current immigrant ¯ow it is more useful to look at newly arrived immigrants. The results in Table 2 imply that immigrants arriving in the 5 years prior to 1909 earned on average 20.4% less on arrival than the native-born. Borjas (1994, p. 1674) ®nds that recently arrived immigrants earned 16.6% less than natives in 1970, 27.6% less in 1980 and 31% less in 1990. Thus, relative to natives, recent arrivals, at least between 1970 and 1980, seem to have been in a similar position to those in 1909. However the variation in earnings experience of recent immigrants seems to be much greater across nationalities in 1980 than it was in 1909. The standard deviation of the (log) wage of recently arrived immigrants across the 26 national groups derived from Table 2 was 0.056 as compared with 0.295 in across 41 nationalities in 1980 (calculated from Borjas, 1992, p. 33±4). Much of this di¨erence is accounted for by the gap between old and new immigrants. In 1909 the average wage gap (weighted by total immigrants) between old and new immigrant groups was 6.7%. By contrast, the average gap between European immigrants and those from Africa Asia and South America in 1980 was 30.7%. Clearly since the wage gaps among newly arrived immigrants were wider in recent times, the e¨ects of changing immigrant nationality mix are potentially more powerful. The ®gures in Table 6, taken from Borjas (1992) show the e¨ects of changing mix on years of education and the wage rate of recent immigrant arrivals. The ®rst two rows show that years of education relative to natives fell between 1955±1960 and 1975±1980 whereas in the ®xed weight index it rose ± a feature somewhat reminiscent of the ®nding for literacy between 1893 and 1913 in Table 2. But the e¨ects of changing composition on wage rates are much more marked. Rows 2 and 3 show that whereas the average wage for recent immigrants relative to natives fell by 27% (0.30±0.03), in the ®xed weight index it fell by only one percentage point. Thus the e¨ect of changing composition over 1935±1940 to 1975±1980 is ®ve times larger than the modest 5% estimated in Table 4 for 1873 to 1913. The adjusted log wage rate (adjusted for characteristics such as age and education) also shows a decline of about 25% in the actual but a decline of less than half that in the ®xed weight index. This also is consistent with the more modest composition e¨ects found in the wage rates predicted by characteristics on arrival as compared with the actuals in Table 4, but again on a vastly larger scale.
524
T.J. Hatton
Table 6. E¨ects of changing immigrant composition 1940±1980 (di¨erence from native-born males)
Years of education Actual 1955±60 weights Log wage rate Actual 1955±60 weights Adjusted log wage rate Actual 1955±60 weights
1935±40
Composition of arrivals in: 1955±60
1975±80
0.74 0:01
0.41 0.41
0:63 0.27
0:03 0:08
0:13 0:13
0:30 0:09
0:03 0.03
0:11 0:11
0:22 0:08
Source: Calculated from Borjas, 1992, p. 37. Notes: The comparisons made here are not quite the same as those made in Tables 1 and 4. Here, the cross-nationality relative wage (or relative education years) is that of the census year after arrival rather than of a ®xed year. However if, for example, the log wage di¨erential is evaluated throughout at 1980 wage rates then the immigrant wage di¨erentials would be very similar to those reported above, viz: 1935±40: 0.04; 1955±60: 0.23; 1975±80: 0.30 (Borjas 1992, p. 37).
7. Conclusion How strong are the parallels between the two mass immigration eras? With regard to the e¨ects of changing immigrant composition, the same features were present in the four decades up to 1913 that were present in the four decades up to 1980. There was a dramatic shift in the nationality composition of the immigrant in¯ow and this accounted for almost the whole of the decline in immigrant wage rates relative to natives. In both eras this decline is less marked when based only on arrival characteristics. But the magnitude of the wage decline in the late nineteenth and early twentieth century was at most only a ®fth as large as that observed since 1940. In this perspective, the rhetoric about, and the ultimate result of, the perceived decline in immigrant ``quality'' in the forty years before World War I seems out of all proportion to the magnitude of the observed e¨ects. Endnotes 1 Robert Whaples (personal communication) has demonstrated to me a strong positive association between the average heights in 17 European countries and the wage coe½cients in equation (1). This seems to provide circumstantial evidence that some component of the wage di¨erential was due to physical strength which would be correlated with height but would not erode with assimilation. 2 Estimates from a single cross-section tend to show rather faster rates of assimilation (Borjas 1985, 1994; Chiswick 1986). As Lalonde and Topel show, for 1980 the cross-sectional e¨ect of 10 additional years in the United States is 31% compared with 9% for the 1965±1970 arrival cohort observed in 1970 and 1980. It is not possible to make such comparisons in the earlier period but if the cohort e¨ects can be attributed to changing nationality composition then the e¨ect should be observable in the comparison of cross-sectional estimates with and without controls for nationality. When equation (1) in Table 2 was estimated without the nationality dummies the coe½cient on the assimilation variable was 0.16 (t-statistic 4.98). Thus there seems to be no upward bias and perhaps a slight downward bias in the earlier period.
Immigrant ``quality'' decline
525
References Blau FD (1980) Immigration and Labor Earnings in Early Twentieth Century America. Research in Population Economics 2:21±41 Borjas GJ (1985) Assimilation, Changes in Cohort Quality, and the Earnings of Immigrants. Journal of Labor Economics 3:463±489 Borjas GJ (1992) National Origin and the Skills of Immigrants in the Postwar Period. In: Borjas, GJ, Freeman, RB (eds) Immigration and the Workforce: Economic Consequences for the United States and Source Areas. University of Chicago Press, Chicago, 17±24 Borjas GJ (1994) The Economics of Immigration. Journal of Economic Literature 32:1667±1717 Borjas GJ (1995) Assimilation and Cohort Quality Revisited: What Happened in the 1980s? Journal of Labor Economics 13:201±245 Chiswick BR (1977) Sons of Immigrants: Are they at an Earnings Disadvantage? American Economic Review 67:376±380 Chiswick BR (1978) The E¨ect of Americanization on the Earnings of Foreign-Born Men. Journal of Political Economy 86:897±921 Chiswick BR (1986) Is the New Immigration Less Skilled than the Old? Journal of Labor Economics 4:168±192 Chiswick BR (1992) Jewish Immigrant Wages in America in 1909: An Analysis of the Dillingham Commission Data. Explorations in Economic History 29:274±289 Commons JR (1920) Races and Immigrants (2nd edn, reprinted 1967). Augustus Kelly, New York Douglas P (1919) Is the New Immigration more Unskilled than the Old? Journal of the American Statistical Association 16:393±403 Duleep HO, Regets MC (1997) The Decline in Immigrant Entry Earnings: Less Transferable Skills or Lower Ability? Quarterly Review of Economics and Finance 37:189±208 Eichengreen BJ, Gemery HA (1986) The Earnings of Skilled and Unskilled Immigrants at the End of the Nineteenth Century. Journal of Economic History 46:441±454 Grant M (1923) The Passing of the Great Race, 4th edn. Charles Scribner, New York Hanes C (1996) Immigrants' Relative Rate of Wage Growth in the Late Nineteenth Century. Explorations in Economic History 33:35±64 Hannon JU (1982a) Ethnic Discrimination in a Nineteenth century Mining District: Michigan Copper Mines, 1888. Explorations in Economic History 19:28±50 Hannon JU (1982b) City Size and Ethnic Discrimination: Michigan Agricultural Implements and Ironworking Industries. Journal of Economic History 42:825±845 Hatton TJ (1997) The Immigrant Assimilation Puzzle in Late Nineteenth Century America. Journal of Economic History 57:34±62 Higgs R (1971) Race, Skill and Earnings: American Immigrants in 1909. Journal of Economic History 31:420±429 Hourwich I (1922) Immigration and Labor: The Economic Aspects of European Immigration to the United States, 2nd edn. Huebsch, New York Jenks JW, Lauck WJ (1926) The Immigration Problem, 6th edn. Huebsch, New York Lalonde RJ, Topel RH (1992) The Assimilation of Immigrants in the U.S. Labor Market. In: Borjas JG, Freeman RB (eds) Immigration and the Workforce: Economic Consequences for the United States and Source Areas. University of Chicago Press, Chicago, 67±92 McGouldrick PF, Tannen MB (1977) Did American Manufacturers Discriminate Against Immigrants before 1914? Journal of Economic History 37:723±746 Tortella G (1994) Patterns of Retardation and Recovery in Southwestern Europe in the Nineteenth and Twentieth Centuries. Economic History Review 47:1±21 United States Immigration Commission (1911) Reports, 61st Congress, 3rd Session, Washington D.C., Government Printing O½ce Williamson JG (1995) The Evolution of Global Labor Markets since 1830: Background Evidence and Hypotheses. Explorations in Economic History 32:141±196