IMMIGRANT UNEMPLOYMENT: THE AUSTRALIAN EXPERIENCE* by Paul W. Miller and Leanne M. Neo Department of Economics The University of Western Australia * This research was supported by a grant from the Australian Research Council. We are grateful to two anonymous referees for a number of helpful comments. 1
I. INTRODUCTION Immigration has been a major contributor to workforce growth in Australia. In 1991, twenty six percent of the workforce were immigrants, and a further thirteen percent were the Australian-born children of immigrants. A large proportion of these immigrants are of British and Continental European origins, although among recent waves of immigrants, Asian immigrants are more important 1. This change in the source countries of Australia s immigrants reflects changes in Australian immigration policy 2, together with supply and demand factors in the global economy. One of the considerations that will be important in this regard is the potential for economic success in Australia. While the net influence of this in the migration decision depends on the relative degrees of economic well-being in the country of origin and the country of destination, a first step to understanding its significance is to examine the economic well-being of immigrants in the country of destination. A prime determinant of economic wellbeing is whether the individual has a job. Immigrant employment outcomes are therefore the focus of this study. Immigrant employment outcomes in Australia were the subject of a series of studies based on data collected in the 1980s. These studies revealed that immigrants experienced relatively high rates of unemployment in the first few years after they arrive in Australia, but that this disadvantage was dissipated with increases in duration of residence for most groups. Since these studies were conducted however, the Australian labour market has deteriorated, and the mix of immigrants has continued to shift towards the Asian region. Establishing whether the findings from the earlier research carry across to the labour market of the 1990s is therefore a 1 According to the 1991 Australian Census data (ABS, 1991), immigrants from the main Englishspeaking countries and other European countries accounted for around 72 percent of the overseas born in 1991. However, there has been an increase in immigration from South East and North East Asia in the recent decade. 2 In 1973, a policy of non-discrimination on the grounds of race, colour or nationality in the selection of immigrants was adopted. 2
research priority. In addition, advances in the research methodology employed in previous studies are now possible, and a feature of the current study is that the unemployment rate differential between the Australian born and immigrant groups is decomposed into components which, on the basis of the model proposed, can be viewed as justified and unjustified. This paper is structured as follows. Following the introductory section, Section 2 uses data from the Australian Bureau of Statistics (ABS) publication The Labour Force to present statistical facts relating to immigrant relative unemployment experience in Australia since the 1980s. It shows that the unemployment gaps between immigrants and the Australian born have increased in recent years. Tabular evidence of the relative unemployment rate disadvantage of sub-groups of the immigrant population (defined by age and country of origin) is also presented. The section concludes with a review of the Australian research on the labour market position of immigrants relative to Australian-born persons. Section 3 contains a brief description of a model of unemployment. The factors that are believed to influence labour market performance, and the data source for the empirical analysis, namely the 1991 Australian Census of Population and Housing, are outlined. Section 4 presents the results of a logit analysis of unemployment and relates the major findings to the evidence reported in previous studies. The differences in the average predicted unemployment rate of the Australian born and immigrant groups are decomposed into a component that is attributable to the different characteristics of the birthplace groups (e.g., different mean levels of education), and a part that is, from the perspective of the model proposed, due simply to being of foreign origin. In Section 5 the results of the study are used as the basis for reviewing various policy scenarios. 3
II. IMMIGRANTS RELATIVE UNEMPLOYMENT EXPERIENCE: THE BASIC FACTS Tables 1, 2 and 3 present data from The Labour Force on the unemployment situation of immigrants and the Australian born. Table 1 shows the trend of unemployment experienced by both the overseas born and the Australian born from August 1980 to August 1996. These data indicate that the overseas born have consistently had unemployment levels between 0.4 and 2.2 percentage points higher than the Australian-born throughout the 1980s and 1990, with the gap rising to around 3 percentage points from 1991 and falling to 1.7 percentage points in 1996. Both male and female immigrants are characterised by an unemployment rate disadvantage and by a worsening of this disadvantage in recent years. Table 1: Unemployment Rates of Australian-born and Overseas-born Persons Aged 15 and over, August 1980 to August 1996. Males Females Total Year Australian Born Overseas Born Australian Born Overseas Born Australian Born Overseas Born 1980 4.8 5.5 7.4 7.5 5.8 6.2 1981 4.5 5.2 6.9 7.8 5.4 6.1 1982 6.1 7.0 7.1 8.6 6.4 7.6 1983 9.3 11.5 9.3 11.7 9.3 11.5 1984 8.1 10.2 7.9 9.7 8.0 10.0 1985 7.5 8.8 7.8 8.7 7.6 8.8 1986 7.5 8.3 8.2 9.0 7.8 8.6 1987 7.3 8.3 8.0 9.5 7.5 8.7 1988 6.4 7.0 7.0 8.1 6.7 7.4 1989 5.3 5.7 5.9 7.2 5.5 6.2 1990 6.7 7.4 6.8 8.1 6.7 7.7 1991 9.2 12.7 8.1 10.3 8.7 11.7 1992 10.7 13.1 8.9 11.6 9.9 12.5 1993 10.7 13.5 9.0 12.4 10.0 13.1 1994 8.7 11.5 8.1 11.4 8.5 11.4 1995 7.8 10.7 7.0 9.3 7.4 10.1 1996 8.5 10.0 7.5 9.5 8.1 9.8 Source: ABS, The Labour Force, 1980-1996 (August) 4
It is widely known that unemployment outcomes vary by characteristics such as age, educational attainment and country of origin, among others. If the variations in unemployment rates with these characteristics differ between the Australian born and the Overseas born, the relative unemployment rate disadvantage experienced by some groups of immigrants may depart considerably from the average picture portrayed in Table 1. Tables 2 and 3 present information that addresses this issue. In Table 2 unemployment rates at August 1996 are crossclassified by birthplace and age. Table 2: Unemployment Rates Cross-classified by Birthplace, Age and Sex, August 1996 Males Females Total Age Australian Born Overseas Born Australian Born Overseas Born Australian Born Overseas Born 15-19 20.5 18.5 18.6 18.3 19.6 18.4 20-24 12.4 12.6 10.6 15.7 11.6 14.0 25-34 7.8 10.7 6.6 9.7 7.3 10.3 35-44 5.7 8.1 5.2 8.0 5.5 8.1 45-54 5.2 9.0 4.5 8.6 4.9 8.9 55 and over 6.3 12.1 2.9 6.7 5.1 10.6 15-64 8.6 10.1 7.6 9.5 8.2 9.8 Source: ABS, The Labour Force, August 1996 The Table 2 data indicate that the overseas born experience higher unemployment rates than the Australian born for both males and females in each age group except the teenagers. For example, among the 35-44 year age group the unemployment rate for Australian-born males is 5.7 percent compared to 8.1 percent for overseas-born males. For the same age group the unemployment rates of Australian-born and overseas-born females are 5.2 percent and 8.0 percent respectively. A further factor that appears to be important when considering labour market success is country of origin. Table 3 shows that unemployment rates are lower for immigrants from 5
English-speaking countries than for immigrants from non-english-speaking countries. Indeed, immigrants from English-speaking countries experience an unemployment rate marginally lower than the Australian born (7.9 percent versus 8.5 percent for males and 6.6 percent versus 7.5 percent for females). Table 3: Unemployment Rates of Immigrants and the Australian Born Aged 15 and over, August 1996 Birthplace Males Females Total Australia 8.5 7.5 8.1 ESB 7.9 6.6 7.4 NESB 11.5 11.7 11.6 Total Overseas Born 10.0 9.5 9.8 Source: ABS, The Labour Force, August 1996 Note: ESB = Born in main English-speaking countries NESB = Born in non-english-speaking countries The unemployment rate disadvantage experienced by immigrants has been examined in a number of studies which have been based on the 1981 and 1986 Censuses, and survey data collected in 1983 and 1987. A number of techniques are employed in the studies (tabular, ordinary least squares, logit and probit), and there is considerable variation in the types of influences on labour market outcomes taken into account. For example, Miller s study (1986a) emphasises the significant impact of pre-immigration knowledge about job opportunities in Australia on immigrant labour market outcomes. In comparison, Beggs and Chapman (1988) focus on post-immigration factors, such as investment in Australian qualifications and the accumulation of labour market experience in Australia, as determinants of immigrants relative labour market position. There is also considerable variation in the age groups studied. Chapman and Miller (1985) and Inglis and Stromback (1986), for example, focus on the labour 6
market outcomes of 15-64 year old males and females, while Miller (1986b) and Wooden (1991) focus on the labour market success of 15-30 and 18-64 year old males and females respectively. Finally, the studies differ in terms of the groups that are the primary focus of the analysis. Jones (1992), for example, conducts analyses of unemployment for four main ancestry groups, namely the Anglo-Celts, Dutch, Italian and Chinese. In contrast, other major studies generally define the group in focus by gender status. For instance, Beggs and Chapman (1988) focus on males only, while other major studies (Brooks and Volker, 1985; Miller, 1986a; Inglis and Stromback, 1986; Chapman and Miller, 1985; Miller, 1986b; Wooden, 1991 and Chiswick and Miller, 1992) look at both males and females. A summary of immigrant unemployment studies is presented in Table 4. The main finding from these studies is that immigrants experience higher unemployment rates than comparable Australian-born persons, even when other determinants of unemployment, such as educational attainment, age and marital status, are held constant. This finding carries over to both males and females and to each age group, and there appear to be four main reasons for this. First, human capital skills acquired overseas may not be fully transferable to the Australian labour market. Second, immigrants may have less information about job opportunities in the Australian labour market than the Australian born. In addition, employers may face difficulty in assessing the productivity of the foreign born which will result in poorer employment prospects. Third, immigrants may have poor English language skills and, as a result, face inferior labour market prospects. Where this is a major contributor to immigrant s relatively high unemployment rate, the immigrant unemployment rate disadvantage should fall with duration of residence (see Wooden, 1991). Fourth, immigrants may face discrimination from employers or fellow employees, a factor that will also increase the likelihood of unemployment (Chapman and Miller, 1985). 7
The labour market disadvantage experienced by immigrants is expected to diminish with increases in their duration of residence in Australia. The empirical evidence summarised in Table 4 is consistent with this expectation. It needs to be noted, however, that while the studies agree that the unemployment rate of recent arrivals exceeds that of longer-term settlers, there is some disagreement over the nature of the unemployment rate-duration of residence relationship. For example, Miller s (1986a) analysis of 1981 Census data indicates a U-shaped relationship between duration of residence and unemployment rates, while Brooks and Volker (1985) and Inglis and Stromback (1986) suggest that as period of residence increases, the unemployment probabilities of immigrants approach that of Australian-born persons in a monotonically decreasing form. A major feature reported in many studies is that unemployment rates for immigrants from English-speaking countries are lower than for those from non-english-speaking countries, irrespective of gender. Among immigrants from non-english-speaking countries, those arriving under the humanitarian migration category and those from the Mediterranean region experience the highest unemployment rates (Miller, 1986a and b; Wooden, 1991; Jones 1992; Chiswick and Miller, 1992). Another feature of the studies of immigrant labour market outcomes to which attention can be drawn is that most studies report that educational attainment is strongly associated with reduced unemployment rates. However, Miller s analyses of 1981 Census data (1986a) and the First-Institute Manpower Program Survey data (1986b) show that additional years of education are associated with smaller unemployment rate reductions among immigrants, with the smallest unemployment rate reductions being experienced by immigrants from non-english-speaking countries (see also Jones, 1992). This suggests that human capital acquired in non-englishspeaking countries is less internationally transferable than that acquired in English-speaking 8
countries. The different links between educational attainment and unemployment rate reductions for the native born and foreign born result in additional years of education being associated with an increase in the relative unemployment rate among immigrants (see Beggs and Chapman, 1988). This phenomenon emphasises the difficulty faced by Australian employers in assessing the value of overseas education. A number of other studies have reported on the unemployment rate effects associated with post-arrival qualifications, though there is no agreement in this respect. Jones (1992) suggests that foreign qualifications provide weaker protection against unemployment than qualifications acquired after arrival, irrespective of ethnicity. However, the study by Inglis and Stromback (1986), based on earlier data, indicates that the impact of a qualification acquired abroad on the unemployment rates does not differ from that of a qualification acquired in Australia. Finally, there are some differences between the findings reported for men and women. Of major importance is that most studies conclude that the human capital effect is not significant among female immigrants. For example, Chiswick and Miller (1992) report that qualifications only increase the labour force participation rate among female immigrants, whereas they tend to increase labour force participation and reduce the unemployment rates among male immigrants. A further illustration is given by Inglis and Stromback (1986) who report that the impact of English language proficiency (another form of human capital) on the unemployment outcome is insignificant among female immigrants which contrasts to a negative and significant effect on the male immigrant predicted unemployment probability. In summary, this review has shown that the immigrant unemployment rate disadvantage varies by country of origin, and is particularly pronounced for immigrants from 9
non-english-speaking countries. Educational attainment, qualifications, duration of residence in Australia and English-language proficiency are shown to generally affect labour market outcomes in Australia, though the effects differ by country of origin and by gender. Disaggregation of statistical analyses by these characteristics is therefore an important lesson drawn from this overview. As noted in the introduction, establishing whether the Table 4 findings carry over to the 1990s, and refining the analysis to distinguish the justified and unjustified components of the immigrant unemployment rate differential, are the research priorities addressed in this study. III. A Model of Unemployment The first part of this section discusses the model of unemployment using the concept of the tendency to be unemployed which provides a framework that enables unemployment to be related to individual characteristics which are believed to influence labour market performance. These individual characteristics and their relationships with the unemployment status of individuals are subsequently discussed. The section concludes with a discussion of the data source for the empirical analysis. An individual s tendency to be unemployed is viewed as being influenced by personal characteristics. Hence we may write I i * = X i 'β, where I i * is the index of the tendency to be unemployed which is assumed to be a linear function of a vector of the characteristics of individual i, denoted by X i. β is a vector of unknown weights of the vector X, which will be estimated. 10
The relationship between the values of I* and the state of unemployment U is assumed to be monotonic. Whether or not an individual is unemployed therefore depends on a comparison of I*, which reflects an individual s particular circumstances, with a critical value of the index I. The determination of employment/unemployment status is then given as: if I* > I, the individual is unemployed, otherwise the individual is employed 3. It is important to note that the tendency to be unemployed index I* is unobservable. All that we observe is a binary indicator variable U which takes the value of one if the individual is unemployed. Hence, the case where U takes the value of one corresponds to I* > I, and the case where U takes the value of zero corresponds to the employment outcome with I* I. To link the observable indicator of unemployment status (U i ) to the characteristics of the individual (X i ), we may write the conditional probability of being unemployed as: Prob(U i /X i ) = Prob (I i * > I ) = F(X i 'β) (1) where F denotes a cumulative distribution function. The dichotomous choice between employment and unemployment can be examined by several techniques, including linear probability, logit and probit methods of estimation. The logit model is employed in the analyses presented below. The vector of characteristics X expected to influence unemployment outcomes includes specific migrant variables (e.g., birthplace and period of residence in Australia) and other characteristics applicable to both the Australian born and immigrants (e.g., age, sex, marital status, educational attainment, possession of qualifications, number of children etc.). A 3 The employment/unemployment status is conditioned upon labour market entry. Some studies have analysed the distribution of the population across the states of employment, unemployment and nonparticipation. Study of the division between unemployment and employment among labour force participants offers more direct evidence on the effects of demand-side behaviour (see Brooks and Volker, 1985). 11
brief discussion of some of the relationships expected between unemployment and the main explanatory variables follows. Birthplace Birthplace is expected to influence unemployment status as it will capture elements of discrimination and aspects of the non-recognition of foreign qualifications that can not be readily measured by other means. For example, if overseas educational qualifications, skills and experience are not fully recognised in Australia, and information on the country in which the particular skills were acquired is not available, these influences may be proxied by a variable for country of birth. A priori, it is expected that the foreign born will have higher unemployment rates than the native born, and the extent of this disadvantage will vary with the economic and social distance of the country of origin from Australia. Period of Residence Period of residence is thought to be an important determinant of immigrant unemployment for the following reasons: Firstly, as period of residence in Australia lengthens, immigrants can be expected to acquire Australian skills more suited to Australian employers' requirements. Some of these skills may not be adequately measured by variables such as educational attainment and language proficiency and so may be indexed only by duration of residence. Secondly, immigrants who have resided in Australia for a reasonable length of time would have gained more information about Australian job opportunities. Thirdly, recent immigrants may face employer or fellow employee discrimination in the workforce and this is expected to diminish with increased duration of residence. In summary, immigrants are expected to gradually adjust to the circumstances of the Australian labour market, and therefore there should be an inverse relationship between period of residence and unemployment. 12
Educational Attainment An inverse relationship between educational attainment and unemployment is expected. One reason for this expectation is that levels of education may enhance an individual's productivity, thereby making him/her more employable (Miller, 1989). Another reason is that employers may use education as a screening device to identify the ablest individuals in the pool of job applicants. Age Human capital, which includes general skills and specific human capital learned on the job, should be positively related to an individual s age. Among young individuals, particularly teenagers, a low level of specific human capital may increase their probability of being laid off and also increase their incentive to quit; since employers face low lay-off costs and young workers are less constrained by family commitments and financial responsibilities than primeaged workers. This suggests that the younger age group may be characterised by high turnover which is an important contributory factor to a relatively high unemployment rate. On the other hand, the prime age male group, who are more likely to have acquired a high level of specific human capital and have strong family commitments, will generally have a lower turnover rate. A strong inverse relationship between age and the probability of unemployment should therefore be a characteristic of the data. Marital Status The impact of marital status is likely to vary by gender. From the perspective of supply considerations, the greater family responsibilities of married males are expected to increase their incentive to work, while on the demand side, employers are more likely to employ married males as they are held to have greater work commitments, be more reliable and potentially more productive. Amongst females, the relationship between being married and the 13
unemployment rate is difficult to determine a priori. On balance, married women s lower degree of labour force attachment, relative to single women, may result in a lower measured unemployment rate. English Language Skills The ability to communicate in English will be necessary in the majority of jobs outside of non- English linguistic enclaves. Immigrants with poor English skills will have access to fewer jobs and therefore are expected to have relatively high unemployment rates. However, immigrants who speak a language other than English at home, and who report that they are proficient in English, could be at an employment advantage compared to the Australian born. This advantage could arise where the immigrants have access to employment in firms where only English is spoken and also in firms where the immigrant s mother tongue is spoken. As the latter market is expected to be small in Australia, any employment advantage to immigrants that arises from this source should be small. A number of other variables will be included in the vector X, for example the number of children in the family, spouse s employment status and locality. Their effects will not be considered in detail here. Relevant comments will be provided in the discussion of empirical results. While there are other variables that might influence employment status (e.g., whether the immigrant is a refugee, employment history), information on these was not collected in the Australian Census, and it is therefore not possible to quantify their effects on the unemployment outcome. This restriction on the model specification should be kept in mind when interpreting the results. The empirical analyses presented below are based on data from the 1991 Australian Census of Population and Housing Sample File (Australian Bureau of Statistics, 1994), which 14
is a one percent sample of households in occupied private dwellings and a one percent sample of persons in non-private dwellings. The sample comprises 39,810 males and 28,694 females 4 who were aged between 15 and 64 years old, in the labour force, residents of Australia, and not in school or at educational institutes at the time of the Census. A general description of the variables employed in the empirical work is given in Appendix A. IV. EMPIRICAL ANALYSIS OF THE LABOUR MARKET OUTCOMES OF THE AUSTRALIAN BORN AND IMMIGRANTS Tables 5 and 6 list estimates from the logit models of unemployment for males and females respectively. Four equations are listed in each Table. The first of these is for the total sample (i.e., Australian born and foreign born). The second lists estimates obtained when the Australian born are studied separately, while the third and fourth columns contain results for when the foreign born from English-speaking countries and the foreign born from non-englishspeaking countries are studied separately. The coefficients give the effect on the log-odds [ln U/(1-U)] of the variables listed in the left hand column. A positive coefficient indicates a higher probability of unemployment; a negative coefficients indicates a lower probability of unemployment. Most of the variables are shown to be statistically significant determinants of unemployment outcomes. Of particular relevance to this study is the birthplace variables in the column (i) results. These show that unemployment rates are higher among immigrants from English-speaking countries than the Australian born, and even higher still among immigrants 4 In arriving at this number of observations all records in the Census sample file with inadequately described information on variables included in the estimating equation are omitted from consideration. 15
from non-english-speaking countries. As the models estimated hold constant the observable skills (e.g., education, qualifications, age, language etc.) of the labour force participants, these positive birthplace effects would reflect the presence of unobservable characteristics that are correlated with birthplace (e.g., motivation, ability) or discrimination in labour market outcomes on the basis of birthplace. The latter explanation is consistent with the usual definition of multivariate regression coefficients as the effect of the variable in question when all other factors in the model of unemployment are held constant. It is apparent from the column (i) results that, for males, regional factors are important only among the Australian born, with residents of both non-metropolitan areas and ACT/Tasmania having higher unemployment rates than residents of metropolitan areas. Such a finding might be expected where, upon arrival, migrants settle in regions of low unemployment for their set of observable and unobservable characteristics. Among females, the ACT/Tasmania variable is insignificant for all birthplace groups while the non-metropolitan variable is positive and significant (at the 10 percent level of better) for all birthplace groups. The different results for males and females with respect to the location variable would be consistent with many female immigrants being tied movers. Years of education are an important factor leading to lower unemployment rates for all groups, though it is noted that the estimated coefficients are much smaller for immigrants from non-english-speaking countries. This finding is consistent with a major theme from the studies reviewed in Table 4. Similarly, possession of vocational qualifications is associated with unemployment rate reductions among males, though the impact in this regard is less among immigrants from non-english-speaking countries. Among females, however, it is only among the Australian born that possession of qualifications is associated with lower unemployment. These findings are consistent with previous studies (see Table 4). It would appear therefore that 16
the much greater debate over the last decade in Australia about the non-recognition of skills acquired in non-english-speaking countries has done little to ease the problems faced by immigrants in this regard. Unemployment rates decrease with age up to around 50-55 years for Australian-born males and females, and 40 years for male and female immigrants from non-english-speaking countries. Among immigrants from English-speaking countries, the age effects peak at 40 years for males and 50 years for females. This U-shaped pattern most likely reflects the fact that many of the skills possessed by older cohorts are no longer relevant in the current labour market, a factor that would increase their likelihood of unemployment. This line of argument is also advanced in the labour economics literature for the lower earnings among older employed individuals compared to those of middle age. Compared with those who are not married, those who are married and have a spouse who is not unemployed have lower unemployment rates. Where the spouse is unemployed, however, unemployment rates are much higher. Unemployment rates also tend to increase with the number of dependent children. As these models are estimated on a sample of labour force participants, this effect of children should be interpreted as deriving from limitations on the types of work that can be undertaken and from welfare system effects. While it has been noted above that the ceteris paribus unemployment rates for the foreign born are higher than for the Australian born, the unemployment rates of the foreign born decrease at a decreasing rate with duration of residence in Australia. This result is consistent with the findings from the studies reviewed in Table 4. The maximum reduction in this regard occurs after about 22 years of residence in Australia, which is slightly greater than the mean level of residency. The increases in unemployment beyond that level of residency 17
presumably reflect cohort effects, with the groups that arrived in Australia during the 1960s having skills that are less relevant to the current Australian labour market than the immigrants that arrived in the 1970s and the 1980s. Finally, language skills are an important determinant of unemployment, but only among immigrants from non-english-speaking countries. This finding is consistent with Wooden s (1991) conjecture concerning the source of the unemployment rate disadvantage experienced by refugees in Australia. The general patterns described above are quite robust across the specifications. However, while giving insights into the significance or otherwise of particular variables as determinants of unemployment, the estimates are not informative of the quantitative impact of the variable on unemployment. To this end one needs to compute partial effects. In the case of continuous variables (education, age, number of children, duration of residence) this is quite straight-forward, with the partial effect being U/ X = β$ ( U )( 1 U ), where β $ k is the estimated coefficient attached to a particular variable (X k ). For binary variables, however, the computation of partial derivatives is inappropriate, and it is conventional to base discussion of estimated impacts on differences between the predicted probabilities for different categories, for example, individuals residing in non-metropolitan areas and for the benchmark group of individuals from metropolitan areas 5. k k Relevant partial effects are presented in Table 7. From these calculations it is seen that an additional year of education results in a reduction in the predicted unemployment rate of over two percentage points for Australian born males and for male immigrants from English- 5 The predicted probabilities are calculated as follows: Partition the regressors into two sets, the first relating to the J variables under immediate consideration (e.g., location) and the second relating to the K remaining variables. Then the predictions that form the basis of the discussion are calculated as: U $ = 1/{ 1+ exp( β$ X + β $ )}, where X k denotes the mean of the k th regressor. The βs $ are j k k j k= 1 K estimated coefficients and j denotes a particular element of the set of J variables under consideration (e.g., resident of a non-metropolitan area). 18
speaking countries. The comparable unemployment rate reductions among females are in the order of 1.6 to 1.8 percentage points. For both male and female immigrants from non-englishspeaking countries the unemployment rate reductions associated with extra years of schooling are quite low: 0.6 of one year for males and 0.4 of one year for females. Unemployment rates decline with age, although the unemployment rate reductions are only of the order of one-half a percentage point per year when evaluated at age of 30 years 6. A similar reduction in the predicted unemployment rates results from increases in the period of residency in Australia, when evaluated at 10 years of residency in Australia. Examining the unemployment rate effects associated with the dichotomous variables, it is apparent that there are considerable ceteris paribus unemployment rate effects associated with birthplace. Hence, in the column (i) partial effects for the total sample, male immigrants from English-speaking countries are revealed as having predicted unemployment rates eight percentage points higher than their Australian-born counterparts, while male immigrants from non-english-speaking countries are shown to be at a nine percentage point unemployment rate disadvantage. In the female labour market, the respective unemployment rate disadvantages of the foreign born are 11 and 17 percentage points for immigrants from English-speaking and non-english-speaking countries respectively. English language deficiencies are also associated with higher rates of unemployment. Thus, compared to monolingual English speakers, those who speak a language other than English at home and self-report their English-speaking skills as good have unemployment rates two percentage points higher. Those who speak a language other than English at home and self-report their English-speaking skills as poor have an unemployment rate disadvantage 6 As a quadratic specification is used for the age variable, the partial effect of this variable will itself be a function of age. This function is evaluated at age 30 for the purposes of the discussion. A similar issue arises with regard to the period of residence variable. In this case the period of residence effects are evaluated at 10 years of residency in Australia. 19
of 4.4 percentage points in the male labour market, and 6.7 percentage points in the female labour market. As expected, the unemployment rate effects associated with English-language deficiencies are evident only among immigrants from non-english-speaking countries. Moreover, when the analysis is restricted to this group (column (iv)), the unemployment rate effects are more pronounced than when the analysis is based on the total sample of immigrants (column (i)). Table 7: Estimated Partial Effects from Unemployment Rate Models 1. Males Foreign Born Variable Total Sample Australian Born ESB NESB Education -1.68-2.30-2.43-0.60 Age # -0.37-0.35-0.39-0.49 Number of Children 1.16 0.83 1.35 2.26 Duration of Residence ## -0.38 (a) -0.32-0.33 Non-Metropolitan 1.40 1.24 0.04* 1.36* ACT / Tasmania 1.47 1.38-0.46* 2.61* English-speaking 7.81 (a) (a) (a) birthplace Non-English-speaking 9.22 (a) (a) (a) birthplace Good English 2.14 (a) -2.04* 4.93 Poor English 4.40 (a) -3.43* 12.01 Married -10.11-9.36-11.17-13.01 Spouse Unemployed if 32.41 28.07 34.16 42.25 Married 2. Females Education -1.30-1.85-1.59-0.39 Age # -0.46-0.49-0.27-0.32 Number of Children 1.02 1.13-0.09* 0.89 Duration of Residence ## -0.72 (a) -0.47-0.78 Non-Metropolitan 2.75 2.24 4.71 4.93 ACT / Tasmania 0.62* 0.58* -3.59* 3.94* English-speaking 10.59 (a) (a) (a) birthplace Non-English-speaking 16.99 (a) (a) (a) birthplace Good English 2.07 (a) 9.29 7.43 Poor English 6.67 (a) -11.79* 22.02 Married -8.58-7.63-12.16-18.77 Spouse Unemployed if Married 35.14 29.79 40.91 40.18 Notes: # = evaluated at 30 years; ## = evaluated at 10 years; * = estimated impact insignificant at 10% level; (a) = variable not relevant. 20
Finally, Table 7 indicates that marital status and the employment status of the spouse exercise important influences on the unemployment outcome. Individuals who fall into the married, spouse present category have unemployment rates considerably lower than the other marital states provided that the spouse is not unemployed. Where the spouse is unemployed, however, the predicted unemployment rates are much higher: by over 30 percentage points in some cases 7. Some of the differences discussed above would be associated with immigrants having a higher incidence of unemployment than the Australian born (e.g., the returns to education for immigrants from non-english-speaking countries) while others could be associated with immigrants having a lower incidence of unemployment (e.g., the locality effects). An overall assessment of the unemployment differences between birthplace groups can be obtained using the method proposed by Farber (1990). Under Farber s method, the aim is to decompose the difference in the average predicted unemployment rates for two groups into a component that is attributable to the different characteristics of the birthplace groups (e.g., different mean levels of education), and a part that is to be viewed as an impact associated simply with being foreign born. For the purpose of this decomposition, the average predicted probability of unemployment is defined as: [ Xij βi ] n i PX ( i' $ 1 βi) = 1 1+ exp( ' $ ) ni j= 1 (2) 7 It is possible that the unemployment status of the spouse is endogenous in this model. Consequently, the spouse s unemployment rate variable was instrumented and the model re-estimated. This has a very minor impact on all coefficients other than that for the variable that was instrumented. The estimated partial effect of the spouse s unemployment rate variable in this instrumental variable model was doubled in the male unemployment rate model and increased by 50 percent in the model for female unemployment rates. However, given the small representation in the spouse unemployed category, the decompositions presented below are reasonably insensitive to choice of method of estimation. 21
where i refers to the birthplace group (Australian born, foreign born from English-speaking countries, foreign born from non-english-speaking countries) and j refers to the members of the specific group. Group i has a total of n i members. Using Farber s methodology, the difference in average predicted probability between the Australian born (a) and the foreign born (f) can be categorised into two parts as follows: [ ] ] PX ( ' β $ ) PX ( ' β $ ) = PX ( ' β $ ) PX ( ' β $ ) + PX ( ' β $ ) PX ( ' β $ ) (3) f f a a f f a f a f a a The first term in square brackets on the right-hand side of this equation is the part of the unemployment rate differential that arises because of differences in the marketable characteristics of the two birthplace groups. For example, if the mean level of education of the Australian born exceeds that of the foreign born, then one would expect that this would result in the mean unemployment rate of the Australian born being relatively low. Such an impact would be captured by this first term and it is the component of the unemployment rate differential that is conventionally labelled as justified. [ The second term in square brackets on the right-hand side of equation (3) is the component of the unemployment rate differential between the two groups that arises because the marketable endowments of the foreign born are rewarded at rates different to those of the Australian born in the labour market. This component of the unemployment rate differential is conventionally labelled unexplained or unjustified in the literature, and is often termed the discriminatory component of the unemployment rate differential. Where the discrimination terminology is adopted, the qualification that the effect could represent unmeasured factors that are correlated with birthplace should be noted. 22
The results from this decomposition are presented in Table 8 8. There are two key features of these results. First, for each comparison immigrants are shown to have endowments that should, ceteris paribus, lead to an unemployment rate lower than that of the Australian born. In the case of males the unemployment rate advantage would be 4.4 percentage points among immigrants from English-speaking countries, and three percentage points among immigrants from non-english-speaking countries. Among females the unemployment rate advantages of immigrants that should arise due to their favourable marketable characteristics are 5.9 and 7.3 percentage points for those from English-speaking and non-english-speaking countries respectively. These advantages derive from immigrants concentration in the relatively low unemployment metropolitan areas, a greater mean age and higher mean educational attainment. The second feature of the results is that each birthplace group s labour market performance appears to be characterised by a substantial unexplained unemployment rate disadvantage. For immigrants from English-speaking countries, the unexplained unemployment rate component approximately offsets the advantage the immigrants would otherwise have gained from their favourable endowments. For immigrants from non-english-speaking countries, however, the unexplained component is sufficiently strong that it more than offsets the advantage associated with their favourable endowments. The unexplained component for immigrants from non-english-speaking countries is valued at almost seven percentage points in the case of males and almost 14 percentage points among females. 8 The decomposition outlined in equation (3) evaluates the characteristics of the Australian born β f ( X a ) using the migrant coefficients ( $ ). An alternative decomposition would be to evaluate the migrants characteristics ( X f ) using the coefficients of the Australian born ( $ ). To minimise the potential for bias, the average of the two decompositions is presented in Table 8. β a 23
Table 8: Decomposition of Unemployment Rate Differential Between the Australian born and Immigrants Comparison Australian-born males and: male immigrants from English-speaking countries male immigrants from non- English-speaking countries Unemployment Rate Differential Justified Unjustified -0.02-4.4 4.38 3.83-2.98 6.81 Australian-born females and: -0.84-5.92 5.08 female immigrants from English-speaking countries female immigrants from non- English-speaking countries 6.46-7.32 13.78 The estimates of the unjustified component of immigrants unemployment rate situation displayed in Table 8 are approximately three percentage points lower than estimates based on the more restrictive dummy variable approach in the unemployment rate model (see Table 7). These calculations were repeated using estimates of the unemployment rate model obtained using ordinary least squares. The basic patterns evident in Table 8 carry across to these alternative decompositions, indicating the major findings are not sensitive to choice of model. One aspect of the Table 8 results which may be surprising is the size of the unexplained/unjustified component in the unemployment rate differential between the Australian born and immigrants from English-speaking countries, particularly for males where it is comparable to the effect among immigrants from non-english-speaking countries. A number of studies have used the dummy variable approach to assess the ceteris paribus unemployment rate disadvantage of immigrants. Brooks and Volker (1985) distinguish immigrants from four birthplace regions in their study of the 1981 Census: Anglo-Saxon, Northern Europe, Southern Europe and Other. They report that the ranking of ceteris paribus 24
unemployment rates (from lowest to highest) is Northern Europe, Southern Europe, Anglo- Saxon, Other for males and Southern Europe, Anglo-Saxon, Northern Europe and Other for females. Inglis and Stromback (1986) report similar ceteris paribus unemployment rates for immigrants from UK/Eire and Europe. These results are consistent with the Table 8 finding of a sizeable unjustified component in the unemployment rate of immigrants from Englishspeaking countries. V. CONCLUSION This study shows that between 1980 and 1996 both male and female immigrants were at an unemployment rate disadvantage compared to the Australian born. Moreover, it is demonstrated that this disadvantage has worsened in recent years. Detailed study of the unemployment rate experience of birthplace groups based on the 1991 Census of Population and Housing was therefore undertaken in order to advance understanding of the nature of immigrants unemployment rate disadvantage in the Australian labour market. The analyses reveal that the principal factors that influence employment success in the Australian labour market are educational attainment, age, qualifications and, among the foreign born, duration of residence in Australia and English language proficiency. The returns to education, in terms of employment prospects, are not as large for immigrants from non-english-speaking countries as they are for the Australian born and for immigrants from English-speaking countries. Additionally, the employment enhancing effect of post-school qualifications is smaller among the foreign born, particularly for females. The magnitudes of the period of residency effect on labour market performance are similar for male immigrants from both English-speaking and non-english-speaking countries, while in the 25
female labour market the impact of duration of residence is much greater for immigrants from non-english-speaking countries. It was shown that the unemployment rate differential between immigrants and the foreign born can be decomposed into two components. The first of these is due to differences in the marketable characteristics of the foreign born and the Australian born. This is generally viewed as a justified component of the unemployment rate differential. The second component is due to differences in the way that the marketable characteristics of males and females translate into lower unemployment rates for the various birthplace groups. This component is unexplained by our model and is conventionally labelled as a discriminatory component of the unemployment rate differential. The decomposition presented shows that if immigrants marketable characteristics were rewarded in the labour market in the same way that the Australian born s characteristics are rewarded, then immigrants would experience unemployment rates that are considerably lower than those of the Australian born. The unemployment rate differentials actually observed in the labour market appear to arise because this potential advantage is offset by a large unjustified unemployment rate disadvantage. As a result, the unemployment rates of immigrants from English-speaking countries are only slightly lower than those of the Australian born, while the unemployment rates of immigrants from non- English-speaking countries are much greater than those of the Australian born. Many of the findings for the Australian labour market reported in this paper are similar to those reported by studies of other major immigrant-receiving countries. For example, Chiswick, Cohen and Zach (1997) report that educational attainment has a smaller negative effect on unemployment outcomes for immigrants than for the native born in the U.S. Chiswick, Cohen and Zach also report that in the U.S. labour market the ceteris paribus unemployment rate for immigrants from Europe and Canada is similar to that experienced by 26
Asian immigrants. Contrary to the finding reported in this paper, however, Chiswick, Cohen and Zach report that the unemployment rate disadvantage experienced by recent immigrants is short-lived in the U.S., and the duration of residence effects are negligible after about five years in the U.S. The duration of residence effects in the Australian labour market are more pronounced and persist for a longer period after arrival. This difference may be associated with the U.S. labour market being more flexible. 9 Reducing the unemployment rate disadvantage of immigrants may prove difficult in the short term. It is not simply a matter of improving language skills or levels of human capital. Rather, the crucial factors appear to be improving the transferability of overseas human capital skills and minimising discriminatory practices in the workplace. With respect to the former problem, skill adaptation courses and competency based skill standards might be expected to assist, although the heightened awareness of these issues in Australia in the past decade does not appear to have improved matters. A strengthening of affirmative action programs for ethnic minorities may be required if the Australian labour market is still characterised by residual elements of discrimination. 9 In the study of wage determination in the U.S. and Australia, strong duration of residence effects are usually reported for the U.S. but duration of residence effects in Australia are quite modest (see Chiswick and Miller, 1985). The links between this pattern of wage effects with duration of residence and the unemployment effects requires further analysis. 27