II. The Model and the Data

A. The Model

Language skills are a form of human capital. They satisfy the three criteria that define human capital (Becker, 1964). First, language skills are created at a cost, where this cost includes not only the time of the person, but also the time of parents, teachers and others that is devoted to enhancing language skills, as well as purchased inputs, including books (for home use and school use), school rooms and other school supplies. Even early language development among young children is not without cost, as parental and other adult time in talking with and reading to young children is essential to their language development.

Second, there is a presumption, to be developed further below that language skills are productive. Language skills are presumably productive in the individual’s role as a consumer and in the role as a producer (supplier of labor services). Those deficient in language skills will find it more difficult (costly) to search for the lowest price, or the highest quality for a given price, of the goods and services they may wish to purchase. Searching for information regarding prices, quality and efficacy of a good or service requires a degree of speaking proficiency and/or literacy.

Language skills are presumably important in the labor market. The language skills may be required for the job, because of oral communication with others (e.g., consumers, suppliers, supervisors, or co-workers) or because literacy (reading, writing or both) may be needed to do the job efficiently or to do it safely.

Finally, language skills are embodied in the person. Unlike physical capital (e.g., a machine) which may be separated from its owner, a person’s language proficiency cannot be alienated from the individual.

If language skills are a form of human capital, as are schooling, on-the-job training, and health status, individuals will invest in these skills and will use them in consumption and production activities so as to maximize the person’s own economic well-being, subject to various constraints. These constraints include ones own time, resources to purchase inputs into the production of human capital, ability, home environment, school quality, etc.

The primary language in the U.S. labor market is English and it would be reasonable to hypothesize that, at least up to some point, greater proficiency in English is associated with greater productivity in the labor market, and hence in higher earnings. Someone with greater proficiency could, presumably, do a job that could be done by someone with lesser proficiency, but in addition would have job opportunities that would not be available to someone with a lower degree of proficiency.

A person who is fully proficient in English, but who also has some proficiency in another language may have an additional economic advantage. Knowledge of a second language may expand job opportunities. This could arise if that language was valuable in international trade, or in economic interactions with suppliers, consumers, or others who use that language.

There are, however, circumstances in which a person who speaks a language other than English, as well as being able to speak English, incurs an economic disadvantage. This could arise, for example, if the speaking or studying of the other language detracts from the individual’s full proficiency in English. That is, the person may be bilingual, but not fully proficient in English, and the returns, if any, from knowing the other language are more than offset by deficiencies in English. It may also arise if the bilingualism is associated with an accent or intonation, and if this serves as a basis for discrimination in the labor market. The discrimination may be pure prejudice against those with a particular accent or speech pattern, or it may be based on the false notion that this implies less education or less ability.

In addition, speaking a language other than English may, for some, reflect isolation from the mainstream American economy. This may arise from either growing up in or currently living and working in an immigrant/ethnic/indigenous group enclave. Having been raised in such an environment may result in less human capital acquired in the home as a child or in school relevant for the mainstream American economy. Those with limited English language proficiency but with knowledge of another language may optimize their labor market success by working in a partial or full language enclave in which the non-English language serves as a partial or primary medium of communication.

It is, therefore, an empirical question as to the extent to which limited English language proficiency penalizes individuals in the labor market and the extent to which full bilingualism, in the sense of being fully proficient in English but with a knowledge of another language, enhances labor market opportunities.

B. The Data

The purpose of this study is to investigate the relation between language skills and labor market outcomes. To analyze labor market consequences requires data on adults in the labor force. Immigrants would not be an appropriate study group for this purpose since, for many, some or all of their schooling and early childhood experience occurred outside the United States. Thus, the analysis is limited to persons born in the United States.

The simplest and most direct measure of labor market attainment is earnings, that is, income from wages, salaries and self-employment. The analysis of earnings will be limited to males age 25 to 64 years. The age limitation is imposed to develop more precise estimates by excluding the aged, many of whom are partly or fully retired, and youths, many of whom are in school or investing heavily in on-the-job training. Analyses of the earnings of women are more complex than that of men because of a far greater propensity on the part of women to move in and out of the labor market. At this stage, to keep the analyses straightforward, the study is limited to adult men.

The 1990 Census of Population and Housing of the United States is the most appropriate data set for the purpose of this analysis. It is the most recent large data set that provides information on a variety of characteristics of each respondent, including their earnings and information on language skills. In particular, the Census in April 1990 asked the respondents to report their wage, salary and self-employment income in 1989. The Census also asked respondents to identify the language, if any, other than English currently spoken in the home (other than just a few words). Those who reported a language other than English were also asked whether their speaking ability in English was : Very Well, Well, Not Well or Not at all (none). These latter four categories plus the "Only English" group constitute the five language categories.

These data are, of course, not ideal. There are no data on proficiency in English among those who report English is the only language spoken at home. There are no data on languages that are known but are not spoken at home. The data on English language proficiency for those who speak another language are self-reported. Finally, the Census does not ask questions on literacy. Analyses show, however, that the Census questions are very productive for research purposes and that there is a strong relation between literacy and speaking proficiency among limited English proficient individuals (Chiswick and Miller, forthcoming). However, in spite of its limitations, there is no data set that is superior to the 1990 Census of Population for an analysis of earnings and language proficiencies for a large random sample of the entire native-born population. Since relatively few in the adult native-born population speak a language other than or in addition to English at home, the very large samples for analysis in the microdata files created from the decennial census are essential for generating sufficiently large samples for those who speak a language other than English.

III Analysis

Speaking a language other than English at home is not common among adult native-born men in the United States. According to the 1990 Census of Population, 5.5 percent of men reported speaking a non-English language at home, and were to some degree bilingual as nearly all reported speaking at least some English. Of those reporting a non-English language spoken, 60 percent (primarily Hispanics) reported Spanish. Another 9 percent each reported German and French. This was followed by other European, Asian and American Indian languages, representing the languages of old immigrants, new (post-1965) immigrants and indigenous peoples.

Table 1 indicates the proportion of the racial/ethnic groups that report speaking a language other than English. While overall this is done by only 5.5 percent of the men, the range is from 3.3 percent for white non-Hispanic men to about two-thirds of all Hispanic and Mexican-origin men. About one-in-five American Indians and native-born men of Asian and Pacific Island origin speak a language other than English. While the Hispanic and Mexican-origin men nearly all report Spanish, the other groups report much greater linguistic diversity.

Table 2 reports the mean earnings in 1989 of adult (age 25 to 64) men by language skills. Those who speak only English at home have the highest earnings, over $32,300, while those who speak Spanish or other languages have lower earnings. Even those who report they speak English "very well" have lower earnings than English-only speakers, just over $31,800 for languages other than Spanish and just under $26,000 for Spanish speakers. Earnings generally decline in Table 2 with a decline in self-reported proficiency in English.

The data in Table 2 are suggestive but cannot be taken as definitive because they represent simple relationships. Other factors associated with speaking a language other than English at home may be influencing earnings. Those who speak English at home are more likely to be Hispanic, have less schooling, may live in poorer regions of the country (rural areas and the South), may work fewer weeks in the year, may be members of other disadvantaged racial and /or ethnic minorities, or may have had foreign-born parents. With the exception of parental nativity, it is possible to control statistically for each of these potentially confounding influences.

The "human capital earnings function," which relates the natural logarithm of earnings to a set of explanatory variables, is estimated statistically using ordinary least square regression techniques. The explanatory variables include years of schooling completed, potential labor market experience (age-schooling-5) and its square, the natural logarithm of weeks worked, and dichotomous variables for whether the respondent is black by race, married and living with his spouse, living in a rural area, and living in a southern state. Two types of language variables are included separately in the analysis. One is whether the respondent speaks a language other than or in addition to English at home. The second set is the degree of English language fluency, that is, four dichotomous variables for speaking English very well, well, not well, or not at all, with speaking only English as the benchmark. The analysis is computed for all adult native-born men with earnings in 1989, and then separately for groups defined by race and Hispanic origin.

The results of the estimated effect of the language variables on earnings for the seven racial/ethnic groups under study are reported in Table 3. Among native-born men age 25 to 64 in 1990 with earnings, only 5.5 percent reported speaking a language other than English. Of these 76.7 percent reported speaking English "very well", 16.3 percent reported "well", 6.6 percent "not well", and 0.5 percent "not at all". While it is difficult to understand the circumstances under which adult men born in the U.S. who have earnings would report speaking English not well or not at all, it is reassuring that these are very rare circumstances. It would indeed be quite remarkable for there to be native-born adult but not aged men in the labor market who do not speak English at all or only a few words. In part because this group is very small, it may be dominated by measurement error.

The data in the first column of the top panel in Table 3 indicate that, other things being the same, adult men who speak a language other than English at home receive 8 percent lower earnings than those who speak only English at home.

The dichotomous non-English language variable is replaced by the four categories for English language proficiency in the second panel of Table 3. Compared to those who speak only English, those who speak English "very well" have about 7.5 percent lower earnings, those who speak it only "well" have about 12 percent lower earnings, with the two very small groups with more limited English skills being associated with about 5 percent lower earnings for the "not well" and no significant effect for the "not at all" category.

The remaining columns in Table 3 look at the effect of language on earnings for adult native-born men, separately by racial/ethnic category. This is done to remove the possible confounding effects in the first column of interactions of language patterns with race/ethnicity. The race/ethnic categories are essentially mutually exclusive, except for those of Mexican origin who are a subset of Hispanics. Among adult U.S.-born men, those who speak a language other than English at home earn less than those who do not by about 4 percent among white non-Hispanic men, 10 percent among Hispanic men (9 percent for those of Mexican origin) and about 17 percent among American Indians (converting the coefficient of - 0.19 to a percent). Although the coefficients are negative, there is no significant earnings effect among the small samples of Black and Asian and Pacific Islander men who speak a non-English language.

Interesting results emerge when the English language fluency variables are entered into the analysis. Those who speak English "very well" have lower earnings than those who speak only English. The effect is largest for American Indian men (about 16 percent), followed by Hispanic (9 percent), Mexican origin (7 percent) and white non-Hispanic men (3 percent), but is not statistically significant for Black and Asian-origin men. Those whose self-reported English is only "well" have a larger deficiency for each group. The earnings disadvantage associated with this level of English proficiency is now statistically significant for Asians, but is not significant among Blacks. The "not well" and "not at all" categories become even smaller in size when divided into race/ethnic categories, and generally show negative patterns.

In summary, among adult native-born men with earnings, those who speak a language other than English at home receive lower earnings than those who speak only English at home. Even those who report that they speak English very well have lower earnings than those who speak only English at home.

Additional statistical tests were undertaken to determine why even those bilingual speakers who speak English very well earn less than monolingual English speakers. Among white non-Hispanic men only 27 percent speak Spanish at home. The earnings of the Spanish speakers do not differ significantly from those of monolingual English speakers, although the earnings differential is significant for the other languages, (4 percent lower earnings for those who speak English very well). This pattern did not change when the analysis were performed separately for the five states with a high concentration of Spanish speakers (Arizona, California, Florida, New Mexico and Texas) and all other states.

Among Hispanics, nearly all who speak a language other than English report it is Spanish. When the analysis is performed separately for the five high Spanish concentration states and all other states an important disparity arises. The labor market disadvantages from speaking Spanish at home are much larger in the high Spanish concentration states. It is striking that among adult male native-born Hispanics, other things being the same, those who report they speak English very well receive about 11 percent lower earnings in the 5 high Spanish concentration states, and 4 percent lower earnings in other states, compared to those who speak only English at home.

The results for white non-Hispanics and for Hispanics, the two numerically largest groups of bilingual speakers, suggest that ethnicity and ethnic concentrations matter. The earnings disadvantage appears to be larger among those who speak their ethnic-origin language, or do so in an area where many others speak the same language.

IV Summary and Conclusions

This study has been concerned with the effect of bilingualism and English language ability among adult (age 25 to 64) native-born men on their earnings. The data are from the very large microdata sample from the 1990 Census of Population. Among these men in the U.S., 5.5 percent report speaking a language other than English at home. Of those speaking another language, 60 percent report Spanish (primarily of Hispanic origin), followed by German and French (9 percent each) and other old immigrant, new immigrant and indigenous (American Indian) languages.

Overall and other variables the same, those who speak another language earn less than those who at home are monolingual English speakers. Earnings decline with a lower degree of proficiency in English, but even those who report they speak English very well receive lower earnings than those who speak only English. The earnings disadvantage among those who speak English very well, compared to those who speak only English at home, ranges from about 3 percent for white non-Hispanic men, to about 10 percent for Hispanic men, to about 17 percent for American Indians.

There are several reasons why those who report they are bilingual and speak English "very well" might have lower earnings than those who speak only English. One is that their level of proficiency in English might be lower than the monolingual English speakers because speaking in childhood and/or as adults this other language competes with their obtaining full proficiency in English. That is, they may have less English language proficiency than monolingual English language speakers. Another is that they may experience discrimination because of an accent or speech pattern related to their other language. Living and working in an ethnic concentration area because of their language deficiencies may also be, in part, responsible for their lower earnings.

These hypotheses are supported by the finding that among white non-Hispanic men speaking Spanish has a smaller adverse effect than speaking other languages, and that among Hispanics living in a state with a high concentration of Spanish speakers has a larger negative effect on earnings. This suggests that it is not the Spanish language per se that is associated with lower earnings, but rather speaking the ancestral language and living among others who do so.

There appears to be no statistical support for the proposition that bilingualism, as measured in this study, enhances earnings in the U.S. It does provide support for the proposition that whatever detracts from full proficiency in English has an adverse effect on earnings.

Table 1
Available by request.

Speaks Only English at Home: $ 32,315

Speaks Spanish and Speaks English:
Very Well: --- 25,987
Well: ----------- 21,664
Not Well: ----- 23,258
Not at All: ---- 16,649

Speaks Language Other Than Spanish and Speaks English:
Very Well: ---- 31,833
Well: ----------- 26,406
Not Well: ----- 27,675
Not at All: ---- (b)

(a) Earnings are wage, salary and self-employment income.
(b) Sample size is only 23 cases (unweighted).

Source: U.S. Bureau of the Census, 1990 Census of Population, Public Use Microdata
Sample, One percent sample.

Table 3
Available by request

References

Becker, Gary S. (1964) Human Capital New York: National Bureau of Economic Research.

Chiswick, Barry R. (1977) "Sons of Immigrants: Are They at an Earnings Disadvantage?" American Economic Review (February) pp. 376-380.

Chiswick, Barry R. and Paul W. Miller (1995) "The Endogencity Between Language and Earnings: International Analyses" Journal of Labor Economics (April) pp.245 - 287.

Chiswick, Barry R. and Paul W. Miller (1996) "The Languages of the United States: What is Spoken and What it Means" READ Perspectives (Fall), pp. 5 - 41.

Chiswick, Barry R. and Paul W. Miller (forthcoming) "Language Skill Definition: A Study of Legalized Aliens" International Migration Review