Electronic copy available at: http://ssrn.com/abstract=1930678

Welfare Policy and Labour Supply of Immigrants in Australia

Vincent Law

Crawford School Research Paper No. 9

Electronic copy available at: http://ssrn.com/abstract=1930678

1

Welfare Policy and Labour Supply of Immigrants in Australia Vincent Law1, Australian National University

Abstract This paper investigates the impact of social security payments on the labour supply of recent immigrants to Australia after the policy change. This research uses the first wave of two sets of Longitudinal Survey of Immigrants to Australia (LSIA) data as treatment group and Survey of Income and Housing (SIH) as control group to analyse the short-term immigrant labour market outcomes before and after the policy change. Employing difference-in-differences estimators and propensity score matching procedures, this study suggests that welfare reform caused a substantial increase in the employment of immigrants. This might imply that restricted access to welfare produced a higher proportion of new immigrants who more actively looked for jobs. JEL: O15, J22

1. Introduction

It is well known that transfer programs negatively affect work incentives for low income

earners. As a result, these adverse effects of labour supply may substantially raise the cost of

improving the living standards of the poor. Labour supply theory hypothesizes that an

income-support policy change would cause a higher proportion of welfare recipients to search

more actively for work. Therefore, the motivation for welfare reform has been the low rates

of employment and participation in the labour market among welfare recipients. Several

OECD countries, such as the US, UK and Germany, revised their social assistance programs

to encourage participation and employment of welfare recipients.

In March 1997 the Australian government introduced a welfare reform to immigration

policies: all immigrants (except humanitarian migrants) had previously been denied access to

1 Many colleagues have made valuable comments on this paper in my presentations at the Australian National University. Particular thanks go to Alison Booth, Deborah Cobb-Clark, Bruce Chapman, David Ribah, Elliot Fan, Chikako Yamauchi and my discussant at the Labour Econometrics Workshop, Jim Vere. I bear full responsibility for any remaining errors. Correspondence: Crawford School of Economics and Government, Australian National University, Canberra, ACT 0200, Australia. Email: u4177364@anu.edu.au

2

welfare payments and Austudy2 during the first six months after arrival in Australia, but in

1997 this period was extended to two years. The waiting period is applied to Disability Wage

Supplement, Carer Pension, Mobility Allowance, the Seniors Health Card and all benefits.

Additionally, the Australian immigration acceptance process adopted new selection criteria to

emphasise the productivity-related characteristics in the immigrant selection process through

a stricter points test. The intention was also to reduce the overall size of the immigrant intake.

Moreover, the higher points test requirement adopted by the Australian government from July

1999 pertaining to human capital endowment (i.e. age, skills and English ability) clearly

shows the Australian government’s increased emphasis on skill-related characteristics in

order to intensify its economic focus.

Following the policy changes, the new immigrants dramatically outperformed those who

came before the policy change in the Australian labour market (Richardson, Robertson and

Ilsley 2001; Cobb-Clark 2003). This is true for all visa categories except business skills,

employer nomination3 and humanitarian.4 The improvement is also true for both genders six

months after arrival time. Figure 1 demonstrates the difference in labour force status of

primary applicants (PA) and migrating spouses (MU)5 for all migrants between two cohorts.

The employment rate rose from 33 per cent in cohort 1 to 50 percent in cohort 2 and the non-

participation rate fell from 46 per cent to 41 per cent. This left a question for researchers to

pose: why have cohort 2 migrants done much better than cohort 1?

2 Austudy provides financial support for tertiary education students. 3 Immigrants for whom there was little change. 4 Immigrants who make up a small proportion of the intake. 5 The PA is the person upon whom the approval to immigrate was based. The groups of persons who migrate as part of the PA visa application are known as the migrating unit (MU).

3

Figure 1 Labour Force Status of Primary Applicants and Migrating Spouses, Cohort 1 and 2 (per cent)

Source: Richardson et al. (2001) Note: Cohort 1 immigrants arrived in the two-year period September 1993 to August 1995. Cohort 2 immigrants arrived approximately between November 1999 and February 2002.

There are a number of reasons earlier identified by Richardson et al. (2001) contributing to

the better labour outcomes of cohort 2. First, they had higher levels of English proficiency

and educational qualifications, and fewer people with low levels of each of these attributes

due to the higher selection criteria imposed by the Australian government. Second, migrants

may have had more incentive to search for jobs because of denial of access to social welfare

payments for longer periods. Third, there was an overall improvement in the state of the

Australian labour market. Last, cohort 2 members might possess attributes that make them

more employable in ways that are not captured in the observable characteristics.

Richardson et al. (2001) investigated the reasons for the improved labour outcomes of new

migrants in the second cohort by using a standard pooled Logit model. However, given the

0

10

20

30

40

50

60

Employed Unemployed Not in the labour force

Per c

ent

Cohort 1 Cohort 2

4

cross-sectional nature of the data, the accuracy of the results cannot be guaranteed due to the

possibility of unobserved individual heterogeneity. This problem could be particularly severe

when the unobserved heterogeneity in individual human capital characteristics in the two

cohorts is correlated with the returns attainable from related observed characteristics. This is

actually an omitted variables bias problem—the magnitude and direction of which may differ

between cohorts (Borjas and Tienda 1985; Beggs and Chapman 1988; Cobb-Clark 2003).

Such heterogeneity most likely, or to a greater extent, arises from the immigration policy

changes that skilled-based (i.e. independent and skilled Australian-linked) categories are

subject to tighter selection criteria. In contrast, other visa categories (e.g. Family Stream) are

only affected by delayed access to social security payments: they have not faced more

restrictive selection criteria. Thus Family Stream category immigrants should have much less

heterogeneity caused by policy changes than any other category.6 In addition, those authors

failed to separate income policy effects from macroeconomic effects.

Most of the studies in this area have focused on Unemployment Insurance (UI) programs.

Garcia (1991) estimates a participation Probit model for 946 wives taken from the 1987 UK

Cohort Study of Unemployed and concludes that extending the duration of UI entitlements

would increase participation of wives of unemployed men by eight per cent. Boeters et al.

(2006) uses a computable general equilibrium model that incorporates a discrete choice

model of labour supply to simulate the effects of social assistance reform proposals in

Germany. Their simulations indicate that major cuts in welfare payments are necessary to

produce substantial employment effects. However, Bingley and Walker (2001) estimate a

model of labour supply for married women allowing for endogenous unemployment

durations of husbands and wives to evaluate the effects of the Job Seeker’s Allowance

6 The humanitarian category is not affected by policy change, and not considered further in this study.

5

programme, which has reduced the duration of UI entitlement from 12 to six months. Their

results suggest that the labour supply disincentives from the welfare system facing women

married to men who remain unemployed are made significantly worse by the reform.

Ashenfelter, Ashmore and Deschenes (2005) use randomised trials to measure whether

stricter enforcement and verification of work search behaviour alone decreases

unemployment claims and benefits paid in the U.S. UI program. Their results show that the

failure to search actively for work has been a cause of overpayment in the UI system.

Empirical literature on this issue is still scarce in Australia, and Richardson et al. (2001)’s

research suffers from methodological problems, such as the heterogeneity between the

cohorts. Although Cobb-Clark (2003) adopted the decomposition method to deal with the

heterogeneity problem, potential second-order effects7 could complicate the results. Her

decomposition method is based on the assumption that the selection policy change directly

affects labour market outcomes by influencing observed human capital endowments, and that

the income support policy change directly affects returns to human capital. However, three in

four individuals captured in the second Longitudinal Survey of Immigrants to Australia

(LSIA) cohort report that they knew the ban before migration. The change in welfare policy

would be expected to result in those not confident of their ability to be economically

independent increasingly choosing to stay in their country of origin. Therefore, income

support policy may, through an endogenous immigration process, affect both human capital

characteristics and the returns to human capital.

The goal of this research is to examine the impact of the 1997 change in income-support

policy on the labour market performance of recent immigrants to Australia. The approach is

7 See Cobb-Clark (2003) for definition.

6

distinctive from previous research in two ways. First, while previous research has

encompassed all migrants, this study isolates the Family Stream category from others to

avoid the contamination of the selection criteria effects imposed on independent and skilled

Australian-linked categories. Thus, not only the containment of second-order

effectsselection criteria effects can be avoided, but also the effects of restricted access to

welfare can be identified from labour market improvements effects. Moreover, the Family

Stream category immigrants are most likely to rely upon social security benefits. Second,

using LSIA as a treatment group and Survey of Income and Housing (SIH) as a control

group, a difference-in-differences estimator is employed to separate welfare effects from

macroeconomic effects.

The study is organised as follows. The detailed overview of the LSIA and SIH data is

presented in section 2. Section 3 considers the difference between the human capital

endowments of the two cohorts. Then section 4 looks at the impact of welfare policy change

on labour supply of immigrants in Australia, followed by propensity score matching

procedures in section 5. Finally, the empirical results are discussed and conclusions derived.

2. Data

2.1 LSIA(Treatmentgroup)

The LSIA consists of two entry cohorts. The first cohort (LSIA1) arrived in the two year

period September 1993 to August 1995. In LSIA1, immigrants were interviewed three times.

The first wave of interviews commenced in March 1994 (approximately six months after

arrival). The second wave of interviews commenced one year later. The third wave of

interviews commenced in March 1997. Each wave of interviews was spread over a two-year

period. There were 3124 primary applicants included in the LSIA2 sample. LSIA2 has two

7

waves. The first interview was conducted approximately five or six months after arrival from

March 2000 to February 2001, and the second interview commenced a year later.

Non-humanitarian immigrants can be classified into two groups: 1) Family Stream, strictly

based on family relationships, and 2) all other categories based on potential labour market

outcomes, including Independents, Employer Nomination Scheme (ENS), and Business

Skills. Independent stream assesses those who pass the point test and do not have a family

relationship. ENS stream refers to those who have pre-arranged employment with an

Australian employer. Business Skills Stream applies to those who meet certain capital

requirements and wish to settle in Australia and develop new or existing businesses. The

Concessional Family category assesses individuals on the basis of both their family

connections and their skills.

The LSIA has two advantages over other datasets for policy analysis. One advantage is that

the policy change occurred just between two cohorts. This is an excellent opportunity to

examine the effect of policy change on immigrants. The other advantage is that the LSIA

contains rich information about immigrants to Australia, which allows us to investigate

immigrants’ behaviour from different perspectives.

For the purpose of dealing with the endogeneity of the migration process, this study only

focuses on labour outcomes of Family Stream visa category PAs aged 15 or above. This

reduces the sample size to 2195 and 1550 for LSIA1 and LSIA2 respectively (i.e. about 52

per cent and 45 per cent of the totals).

8

2.2 SIH(Controlgroup)

For the control group, a sample of non-immigrants is required. The most suitable survey,

because it was undertaken in the same period, was the 1994–95 and 2000–01 Survey of

Income and Housing Costs (SIH). The 1994–95 survey was conducted throughout Australia

from July 1994 to June 1995, and the 2000–01 survey from July 2000 to June 2001. The

1994–95 SIH consists of unidentified individual statistical records containing data on 13,827

persons belonging to 8,675 income units in private dwellings including State and capital

city/rest of State identifiers (ACT and NT combined with no part of State). The 2000–01 SIH

includes unidentified individual statistical records containing data on 13,193 persons

belonging to 8,410 income units from 6,786 households in private dwellings.

The surveys contain State, capital city/balance of state and section of state identifiers;

Australian Socio-Economic Indexes for Areas (SEIFA) deciles; income unit type;

relationship in family and household; age, gender, social marital status, birthplace and year of

arrival in Australia; labour force details (some items for up to eight sequential months)

including occupation, industry and hours worked; qualifications, student status and other

details of education; details of mortgages and loans; housing costs, type of tenure and

landlord, number of bedrooms and type of dwelling structure; taxes and details of weekly and

annual income by fine level source of income for persons, and income units and weekly and

annual income for households.

After pooling data from the LSIA as treatment group and SIH as control group, several

potential problems have been identified. First, LSIA and SIH are two different surveys with

different survey design and methodology, thus they might not be appropriate for comparison.

9

Propensity score matching is therefore employed, to some extent, to make the two surveys

similar. Second, since the definition of employment used in the two surveys differs, they are

not directly comparable. In SIH, the employed are persons aged 15 years and over who,

during the week before the interview worked one hour or more for pay, profit, commission or

payment in kind in a job or business, or on a farm (includes employees, employers and own

account workers), or worked one hour or more, without pay, in a family business or on a

family farm, or had a job, business or farm but were not at work because of holidays, sickness

or other reason.

Unlike SIH, LSIA simply asks respondents whether or not they work in a job, business or

farm at the time of interview. If an employed respondent is on holiday or sick, he or she

would not be counted as employed at the time of interview. Hence, this might underestimate

the number of employed. Usage of the difference-in-difference estimator only examines the

difference within the survey (hence the same definition), and for that reason one would

expect a consistent result from the difference-in-difference estimation. Third, several

variables that might affect outcome variables are excluded because either the definition of the

variables differs between two surveys or variables are available in one survey but not in the

other one. This can be a potential problem if these missing variables are time variant and

influential to the outcome variables, such as English ability or wages.

10

3. Descriptive Analysis

3.1 LabourMarketStatusTable 1 Family Stream Labour Market Outcomes of LSIA1 and LSIA2 by Gender [Ratio and t-value]

Participation Employment to population

Unemployment conditional on participation

Total Cohort 1 0.48 0.29 0.41 Cohort 2 0.50 0.39 0.20 Difference (C2-C1) 0.02 [1.38] 0.10* [6.45] -0.21* [-9.16] Men Cohort 1 0.73 0.44 0.39 Cohort 2 0.72 0.55 0.21 Difference (C2-C1) -0.01 [-0.47] 0.11* [3.70] -0.18* [-6.20] Women Cohort 1 0.33 0.19 0.42 Cohort 2 0.37 0.29 0.18 Difference (C2-C1) 0.04* [1.95] 0.10* [5.32] -0.24* [-7.10] Note: * means that the difference of ratio between the cohorts is statistically significant at 5 per cent level.

The participation rate is the percentage of individuals (over the age of 15) who are currently

employed or are actively seeking employment in the LSIA sample at the time of the survey.

The employment-to-population ratio is the proportion of individuals who are employed in the

sample at the time of the survey. The use of the employment-to-population ratio provides two

benefits. 1) It is the single most useful figure for assessing the labour market success of a

group. 2) A simple estimate of the probability of unemployment 8 is not particularly

instructive about immigrants’ labour outcomes. For example, a simple estimate does not

distinguish between people who are not employed because they are not looking for work, and

people who are unemployed despite seeking work. The probability of unemployment

conditional on participation is defined as the percentage of total people in the sample who are

unemployed and are looking for a paid job at the time of the survey.

There is a modest increase in the participation rate from 48 per cent (cohort 1) to 50 per cent

(cohort 2) for the Family Stream. From the table, men from cohort 2 experience a slightly

8 This comment refers to measures of official unemployment only and is not relevant to those who are marginally attached.

11

lower participation rate. On the other hand, women from cohort 2 have a slightly higher

participation rate than women from cohort 1.

Rising employment and falling unemployment accompanied this increase in the participation

rate for cohort 2. In contrast to participation, the change in employment to population ratio is

substantial. It has risen from 29 per cent for cohort 1 to 39 per cent for cohort 2 for the

Family Stream. The magnitude of improvement in employment (39 per cent) was substantial

for cohort 2 members six months after migration, and was similar for both genders.

The magnitude of change in unemployment rate is even more pronounced. The

unemployment rate of cohort 2 immigrants was less than half that of cohort 1 immigrants at

the same stage of the settlement process. If this result is distinguished by gender, females

underwent a more impressive drop in unemployment rate (42 per cent to 18 per cent) than

males (39 per cent to 21 per cent).

Much of the higher rate of employment for cohort 2 is mirrored by a fall in unemployment.

Over the period 1994 to 2000, the overall unemployment rate in Australia fell from 9.7 per

cent to 6.6 per cent.9 However, the reduction in unemployment among recently arrived

Family Stream migrants (from 41 per cent to 20 per cent) far exceeded this average native

experience. It is to be expected that when the labour market is improving, the impact on new

entrants to the labour force will be greater than the impact on the workforce as a whole.

However, as mentioned in the previous section, the different definitions of unemployment

used by LSIA can complicate this comparison.

9 ABS catalogue no. 6291, table 91.

12

3.2 HumanCapitalEndowments

A descriptive analysis of the human capital endowment of the Family Stream can give an

explicit insight into the changes in “quality” between the two cohorts in terms of both

absolute and relative values. Inspection of Table 2 indicates that there is not much difference

in demographic characteristics, except for 1) marital status where 82 per cent of migrants

from cohort 1 are married compared with 68 per cent of those from cohort 2; 2) country of

origin, where 19 per cent coming from South Asia falls to four per cent in cohort 2, and 10

per cent coming from North and East Asia grows to 35 per cent. However, the sample size is

fairly small for both categories.

Table 2 Family Stream Human Capital Endowment, LSIA1 and LSIA2a (Means and Standard

Deviations) Total Cohort 1 Cohort 2 Mean Std. Dev Mean Std. Dev Percent Change Weeks Since Migration 18 5.1 21 5.1 18*** Demographic Female (%) 62 0.5 61 0.5 -2*** Age 31 10.7 32 11.2 1*** Married (%) 82 0.4 68 0.5 -18*** LM Experience Before Migration (%) (%) Employed before migration 68 0.5 66 0.5 -3*** Unemployed before migration 3 0.2 3 0.2 -8 Non-Participant before migration 30 0.5 32 0.5 8*** No Work Experience 30 0.5 33 0.5 9*** WE less than 1 year 10 0.3 15 0.4 39*** WE between 1 and 2 years 15 0.4 12 0.3 -19*** WE between 2 and 5 years 19 0.4 19 0.4 -4 WE greater than 5 years 25 0.4 22 0.4 -13*** Professional before migration 42 0.5 47 0.5 11*** Skilled before migration 46 0.5 40 0.5 -12*** Unskilled before migration 11 0.3 13 0.3 11*** Prior Visit to Australia 47 0.5 47 0.5 -1 English Ability (%) (%) Native speaker 17 0.4 23 0.4 36*** English well 37 0.5 38 0.5 1*** English not well 46 0.5 39 0.5 -14*** Education (%) (%) Higher degree 3 0.2 6 0.2 90*** Post graduate diploma 4 0.2 4 0.2 -1 Bachelor degree or equivalent 17 0.4 19 0.4 12*** Tech/prof qual diploma/certificate 23 0.4 22 0.4 -6***

13

Trade 5 0.2 5 0.2 -1 12 or more years of schooling 23 0.4 21 0.4 -8*** Schooling below year 12 25 0.4 23 0.4 -7*** Country of Birth (%) (%) Oceania 4 0.2 5 0.2 17*** UK & Ireland 4 0.2 7 0.3 70*** Southern Europe 7 0.3 4 0.2 -49*** Western Europe 5 0.2 5 0.2 -6 Northern Europe 3 0.2 4 0.2 56*** South & Eastern Europe 12 0.4 10 0.3 -19*** North Africa & Mid East 17 0.4 8 0.3 -55*** Southern Asia 19 0.4 4 0.2 -79*** North & East Asia 10 0.3 35 0.5 233*** Southern & Central Asia 6 0.2 3 0.2 -52*** Northern America 3 0.2 6 0.2 114*** South America 7 0.2 5 0.2 -26*** Central America & Caribbean 1 0.1 2 0.1 93*** Other Africa 5 0.2 4 0.2 -20*** State of Residence (%) (%) NSW 34 0.5 41 0.5 21*** Victoria 20 0.4 24 0.4 18*** Queensland 18 0.4 9 0.3 -50*** South Australia 8 0.3 5 0.2 -38*** Western Australia 9 0.3 10 0.3 4*** Tasmania 3 0.2 3 0.2 4 Northern Territory 5 0.2 2 0.1 -60*** ACT 10 0.3 6 0.2 -40*** Note: a Sample size is 2079 and 1403 for LSIA 1 and 2 respectively. b *** denotes the level of significance at 1%.

The labour market status of migrants before arriving in Australia only changes a little

between the two cohorts. Members from cohort 2 experienced two months more

unemployment in their home country than those from cohort 1 on average. In terms of labour

market experience before migration, the Australian government took in a greater number of

less-experienced Family Stream migrants in 1999–00 than 1995–96. More professional and

unskilled migrants had been taken in cohort 2 than cohort 1. Alternatively, the above

phenomena are due to changes in applicants’ preferred places to migrate to.

Miller (1986) indicates that obtaining information about Australian job opportunities from

relatives or the Australian government prior to migration allows migrants to significantly

reduce their predicted probability of unemployment. Therefore, “Prior visit to Australia” is an

14

important factor in searching for a job. However, there is no change in the incidence of prior

visits between the two cohorts. English proficiency often plays an important role in the

successful labour market outcome for migrants. In response to the language consideration, the

Australian immigration acceptance process has adopted a stricter English proficiency

requirement to emphasise the productivity-related characteristics in the immigrant selection

process since July 1999 (Law 2008). As a result of the stricter English requirement,

applications from English speaking background (ESB) countries were more likely to be

successful.

Therefore, English ability of migrants is much improved in cohort 2 partially by taking in

more native English speakers and fewer poor English speakers. Family Stream immigrants

were not directly affected by the policy change, but as PAs’ spouses, they were also more

likely to be native English speakers.10 Thus, one would expect more migrants in cohort 2 to

be from ESB countries, such as the UK, Ireland and North America and fewer from non-

English speaking background (NESB) countries, such as North Africa, the Middle East, and

Southern and Central Asia. But the significant increase in the number of North and East

Asian immigrants in cohort 2 was unexpected. From the aspect of education, members of

cohort 2 are better educated although the difference in education between two cohorts is not

large in absolute value. This could be due to the fact that Family Stream migrants’ family

members faced stricter selection criteria and they might have similar educational attainments

as their family members. If the effect of better educated cohort members is large, it may

contaminate the identification strategy.

10 This change may contaminate the strategy to separate the income-policy effects from selection criteria effects. However, the change in the number of Native English speakers is arguably not very large between the two cohorts (six percentage points).

15

Table 3 Family Stream Male Human Capital Endowment, LSIA1 and LSIA2, (Means and Standard Deviations)

Total Cohort 1 Cohort 2 Mean Std. Dev Mean Std. Dev Percent Change Weeks since migration 18 5.4 22 5.1 18*** Demographic Age 32 10.6 32 11.2 2 Married (%) 82 0.4 66 0.5 -19*** LM Experience Before Migration (%) (%) Employed before migration 78 0.4 75 0.4 -4*** Unemployed before migration 3 0.2 3 0.2 -24 Non-Participant before migration 18 0.4 22 0.4 21*** No Work Experience 19 0.4 24 0.4 22*** WE less than 1 year 10 0.3 14 0.4 41*** WE between 1 and 2 years 16 0.4 14 0.3 -17*** WE between 2 and 5 years 22 0.4 22 0.4 -2 WE greater than 5 years 32 0.5 26 0.4 -17*** Professional before migration 41 0.5 50 0.5 21*** Skilled before migration 43 0.5 38 0.5 -11*** Unskilled before migration 15 0.4 12 0.3 -19*** Prior Visit to Australia 51 0.5 50 0.5 -1*** English Ability (%) (%) Native speaker 20 0.4 28 0.4 39*** English well 37 0.5 36 0.5 -2*** English not well 43 0.5 36 0.5 -16*** Education (%) (%) Higher degree 4 0.2 7 0.2 73*** Post graduate diploma 4 0.2 3 0.2 -6*** Bachelor degree or equivalent 16 0.4 18 0.4 16*** Tech/prof qual diploma/certificate 23 0.4 21 0.4 -9*** Trade 8 0.3 8 0.3 -6 12 or more years of schooling 22 0.4 22 0.4 1 Schooling below year 12 23 0.4 20 0.4 -12*** Country of Birth (%) (%) Oceania 4 0.2 5 0.2 22*** UK & Ireland 7 0.3 8 0.3 25*** Southern Europe 11 0.3 6 0.2 -46*** Western Europe 6 0.2 6 0.2 -4 Northern Europe 2 0.1 5 0.2 120*** South & Eastern Europe 9 0.4 7 0.3 -25*** North Africa & Mid East 21 0.4 8 0.3 -60*** Southern Asia 13 0.3 3 0.2 -75*** North & East Asia 8 0.3 28 0.4 265*** Southern & Central Asia 5 0.2 4 0.2 -20*** Northern America 3 0.2 9 0.3 188*** South America 7 0.3 4 0.2 -49*** Central America & Caribbean 1 0.1 3 0.2 131*** Other Africa 5 0.2 5 0.2 0

16

State of Residence (%) (%) NSW 36 0.5 36 0.5 -2 Victoria 24 0.4 29 0.5 21*** Queensland 19 0.4 11 0.3 -45*** South Australia 10 0.3 5 0.2 -49*** Western Australia 12 0.3 10 0.3 -22*** Tasmania 3 0.2 3 0.2 -2 Northern Territory 5 0.2 1 0.1 -82*** ACT 10 0.3 6 0.2 -39*** a *** denotes the level of significance at 1%. Table 4 Family Stream Female Human Capital Endowment, LSIA1 and LSIA2, (Means and Standard

Deviations) Total Cohort 1 Cohort 2 Mean Std. Dev Mean Std. Dev Percent Change Weeks since migration 18 4.9 21 5.1 17*** Demographic Age 31 10.8 31 11.1 0 Married (%) 83 0.4 69 0.5 -17*** LM Experience Before Migration (%) (%) Employed before migration 61 0.5 60 0.5 -3*** Unemployed before migration 3 0.2 3 0.2 6 Non-Participant before migration 36 0.5 38 0.5 4*** No Work Experience 37 0.5 39 0.5 5*** WE less than 1 year 11 0.3 15 0.4 39*** WE between 1 and 2 years 14 0.3 11 0.3 -21*** WE between 2 and 5 years 18 0.4 16 0.4 -6*** WE greater than 5 years 21 0.4 19 0.4 -9*** Professional before migration 44 0.5 45 0.5 4*** Skilled before migration 47 0.5 41 0.5 -13*** Unskilled before migration 9 0.3 13 0.3 52*** Prior Visit to Australia (%) 45 0.5 45 0.5 0 English Ability (%) (%) Native speaker 15 0.4 20 0.4 32*** English well 37 0.5 38 0.5 3*** English not well 48 0.5 42 0.5 -13*** Education (%) (%) Higher degree 3 0.2 6 0.2 102*** Post graduate diploma 4 0.2 4 0.2 2 Bachelor degree or equivalent 18 0.4 20 0.4 9*** Tech/prof qual diploma/certificate 23 0.4 22 0.4 -4*** Trade 2 0.2 2 0.2 6 12 or more years of schooling 24 0.4 21 0.4 -13*** Schooling below year 12 25 0.4 24 0.4 -4***

17

Country of Birth (%) (%) Oceania 4 0.2 4 0.2 13 UK & Ireland 2 0.1 6 0.2 150*** Southern Europe 5 0.2 2 0.2 -53*** Western Europe 4 0.2 4 0.2 -10 Northern Europe 3 0.2 4 0.2 27*** South & Eastern Europe 15 0.4 12 0.3 -17*** North Africa & Mid East 14 0.3 7 0.3 -49*** Southern Asia 22 0.4 4 0.2 -80*** North & East Asia 12 0.3 39 0.5 223*** Southern & Central Asia 7 0.3 2 0.2 -65*** Northern America 3 0.2 5 0.2 63*** South America 6 0.2 6 0.2 -10 Central America & Caribbean 1 0.1 1 0.1 51 Other Africa 5 0.2 3 0.2 -33*** State of Residence (%) (%) NSW 32 0.5 45 0.5 38*** Victoria 18 0.4 21 0.4 15*** Queensland 18 0.4 8 0.3 -53*** South Australia 7 0.3 5 0.2 -29*** Western Australia 8 0.3 10 0.3 29*** Tasmania 2 0.1 2 0.2 9 Northern Territory 5 0.2 3 0.2 -47*** ACT 10 0.3 6 0.2 -39***

a *** denotes the level of significance at 1%.

Both males and females follow a similar pattern as above, except for three differences. 1)

There are more female unskilled migrants but less male unskilled accepted in 1999–00. 2)

More male migrants came from Northern Europe than female in cohort 2. In contrast, female

migrants came from the UK and Ireland. 3) Male migrants of cohort 2 seem to favour

Victoria for residence only, but female migrants prefer to stay in NSW.

In the descriptive analysis, there seems to be no significant change in human capital

endowment between the two cohorts as a whole. Even though there is a noteworthy

percentage change, it is not large in the absolute value. Hence, one would expect a constant

quality across the two cohorts for the Family Stream, but this should be further examined

with empirical analysis in the following section.

18

4. Econometric Evidence

4.1 EmpiricalFramework

The labour supply change of immigrants who are subject to the welfare reform is compared

to that of natives who are not, under the assumption that in the absence of the welfare reform

they would have had the same change in labour supply. The estimation strategy in use is the

difference-in-differences estimator. The idea behind it is simple and appealing. By assuming

that the control groups of natives underwent a similar change in the unobservable factors, the

mean difference over time in the control group’s outcome is meant to provide a

counterfactual for the treatment group. This indicates what would have happened to the

treatment group, had there not been the welfare reform for migrants. It can therefore be used

to eliminate the effects of all unobservable factors and identify the effect of the welfare

reform for migrants on their labour supply. The following table of labour supply measures of

each group before and after the welfare policy reform motivates the difference-in-differences

estimator.

Table 5 Difference-in-Differences Rationale Before 1997 Welfare Reform After 1997 Welfare Reform

Control Group Pcb Pca

Treatment Group Ptb Pta

where denotes labour market outcomes (e.g. participation etc) of an individual in the

labour market. The subscript ‘t’ denotes treatment, subscript ‘c’ denotes control, and ‘b’ and

‘a’ denote ‘before’ and ‘after’ respectively. The changes in labour supply by immigrants are

(Pta -Ptb). Part of this change is due to the welfare reform, and part is due to extraneous

factors such as overall labour market improvement. The assumption is that the changes in

natives’ labour supply decisions are only affected by the same extraneous factors as

immigrants, given by (Pca -Pcb). This is the “common trend” assumption that underlies the

19

difference-in-differences estimator. Thus an estimate of the effect of income support policy

reform on the labour supply of immigrants (the treatment group) is (Pta -Ptb)-(Pca -Pcb).

With regard to the difference-in-differences estimator, there are two concerns related to group

identification. One is the satisfaction of the “common trend” assumption. That is, there

should be no demand or supply shocks to only one group between 1994-95 and 2000-01. If

there was one, the difference-in-differences estimates will be biased. For example, the

selection criteria for skilled migrants were getting stricter over that period. This could

improve the “quality” of new migrants, and hence lead to possibly better labour outcomes for

them. In contrast, Family Stream immigrants were not subject to the selection criteria.

Therefore, the analysis has less chance of being biased in this way than studies that include

immigrants from all visa categories.

The other concern is that the treatment and the control groups may differ in time trends of

either observable or unobservable characteristics. Any bias due to differential characteristics

between the treatment and control groups can be reduced in a regression-adjusted difference-

in-differences approach by controlling for relevant factors. Moreover, conditioning on

additional explanatory variables through a regression adjustment will produce more efficient

estimates than a simple difference-in-differences estimator (Meyer 1995). Thus the simple

difference-in-differences estimator is augmented with regression-adjusted difference-in-

differences estimators.

Using the estimating subsample of 30,755 individuals, a Probit model of the probability of

being in work is estimated, given by:

20

(1)

where is the latent labour market status of individual ‘i’ that equals to one if an individual

is employed, ‘i’ denotes the individual, is a vector of regressors for individual ‘i’

including gender, age, marital status, education and country of birth. is a vector of

coefficients. is a dummy equal to one for individuals in the treatment group, and zero

otherwise. Any time-invariant differences in labour skills and supply preferences across the

treatment and the control group are reflected in the coefficient . This coefficient is

expected to be negative as employment is generally lower for immigrants than for natives due

to the lack of country-specific skills or poor language skills. The variable , is a dummy that

is equal to one for the post-welfare reform period. Its coefficient ( ) is expected to be

positive since employment is generally increasing over time. If is positive, it suggests that

immigrants increased their labour supply relative to natives after the welfare reform. This

would suggest that the restriction to welfare access resulted in a higher proportion of LSIA2

immigrants more actively looking for jobs.

When the employment to population ratio is conditional on participation, it is analogous to

the above setting.

4.2 EmpiricalResults

Marginal effects from the Probit model are reported in Table 6. All standard errors are robust

to heteroscedasticity. Estimation of probability of an individual being employed suffices for

the purpose of the research because this indicator manifests the effects of welfare change.

21

4.2.1 ProbabilityofBeingEmployed

4.2.1.1 SimpleDifference­in­DifferencesEstimator

Table 6 Family Stream Immigrant’s Probability of Being Employed, Simple Difference-in-Differences Estimation.

Variable Marginal Effect Std. Error Post-reform Period (2000-01) 0.009 [0.006] Treatment Group (Immigrants) -0.317 [0.010] Difference-in-Differences 0.097 [0.017] Table 6 reports the simple difference-in-differences estimates of the effects of welfare reform

in 1997 on the probability of being employed of Family Stream immigrants. Both and

have the expected signs and are statistically significant, though the former is not

economically significant. The difference-in-differences estimate of the welfare reform impact

on labour force participation is 0.097, statistically significant at five per cent level.

4.2.1.2 Regression­AdjustedDifference­in­DifferencesEstimator

Table 7 Family Stream Immigrants’ Probability of Being Employed, Regression-adjusted Difference-in-Differences Estimation.

Variable Marginal Effect Std. Error Post-reform Period (2000-01) 0.011 [0.007] Treatment Group (Immigrants) -0.385 [0.011] Difference-in-Differences 0.128 [0.018] Demographic Female -0.221 [0.006] Married 0.053 [0.008] Agea age20-24 0.204 [0.013] age25-29 0.233 [0.012] age30-34 0.198 [0.013] age35-39 0.205 [0.013] age40-44 0.255 [0.013] age45-49 0.233 [0.013] age50-54 0.157 [0.015] age55-59 -0.029 [0.017] age60-64 -0.237 [0.017] age65 or above -0.555 [0.008] Educationa Bachelor degree or above 0.221 [0.009] Tech/prof qualification 0.141 [0.007] Country of Birth English-Speaking Country 0.132 [0.008] Note: a The omitted categories are age15-19, school and Non-English-Speaking country .

22

Table 7 shows the regression-adjusted difference-in-differences estimates of the effects of the

1997 welfare reform. This improves on the simple difference-in-differences estimates by

reducing any bias due to differential characteristics between the treatment and control groups.

The results show that cohort 2 migrant’s employment increased by almost 13 percentage

points due to the reform in the welfare system. The coefficient on the interaction of time and

treatment is positive and significant at a one per cent significance level. As expected, women

and the less educated are less likely to be employed over that period. The treatment dummy is

significantly negative. The change in “probability of being employed” across cohorts is 10

percentage points, but insignificant.

Table 8 reports the regression-adjusted difference-in-differences estimates of the welfare

effects from different education groups. The data show that immigrants with university

degrees have been affected most by the income-support policy change. The coefficient of

difference-in-differences for university group is nearly three percentage points higher than

other educational categories. This seems to be contrary to intuition. However, it may be a

result of having lower reservation wages due to the more restrictive welfare policy11. It might

also be the reflection of the fact that change in the selection criteria affects Family Stream

immigrants through the effect on family members.

As shown in Table 9, once the probability of being employed is conditional on being in the

labour force, the magnitude for immigrants with university degrees drops the most. This

implies that the more restrictive an income support policy is, the more likely it is that

immigrants with higher education will join the labour force from the non-participation pool.

11 This explanation would be more relevant to higher educated immigrants than lower educated ones because usually the qualifications of the more educated are more likely to be underutilised.

23

Other estimates are consistent with the employment to population ratio in pattern, although

are expected to be smaller in magnitudes.

Table 8 Family Stream Immigrants’ Probability of Being Employed, Regression-adjusted Difference-in-

Differences Estimation by Education.

Variable School University Tech/Prof. Qualification

Post-reform Period (2000-01)

0.013 [0.009]

-0.003 [0.018]

0.008 [0.013]

Treatment Group (Immigrants)

-0.294 [0.014]

-0.45 [0.029]

-0.425 [0.023]

Difference-in-Differences

0.104 [0.030]

0.135 [0.023]

0.109 [0.028]

Demographic

Female -0.211 [0.008]

-0.165 [0.015]

-0.239 [0.012]

Married 0.064 [0.010]

-0.023 [0.017]

0.061 [0.014]

Agea

age20-24 0.192

[0.016]

-0.873 [0.006]

0.098 [0.043]

age25-29 0.243

[0.017]

-0.95 [0.003]

0.088 [0.044]

age30-34 0.186

[0.018]

-0.943 [0.004]

0.073 [0.045]

age35-39 0.221

[0.018]

-0.922 [0.004]

0.069 [0.046]

age40-44 0.27

[0.018]

-0.911 [0.005]

0.099 [0.044]

age45-49 0.252

[0.018]

-0.893 [0.005]

0.059 [0.047]

age50-54 0.153

[0.020]

-0.861 [0.006]

0.01 [0.050]

age55-59 -0.041 [0.020]

-0.829 [0.006]

-0.2 [0.057]

age60-64 -0.211 [0.017]

-0.817 [0.006]

-0.398 [0.051]

age65 or above -0.476 [0.008]

-0.879 [0.005]

-0.719 [0.018]

Country of Birth English-Speaking Country

0.128 [0.011]

0.164 [0.018]

0.094 [0.015]

Observations 17741 4270 8717 Note:

24

a The omitted categories are age15-19 and Non-English-Speaking country.

Table 9 Family Stream Immigrants’ Probability of Being Employed, Regression-adjusted Difference-in-Differences Estimation by Education, Conditional on Participation.

Variable School University Tech/Prof. Qualification

Post-reform Period (2000-01)

0.026 [0.007]

0.005 [0.010]

0.018 [0.007]

Treatment Group (Immigrants)

-0.252 [0.027]

-0.252 [0.034]

-0.253 [0.031]

Difference-in-Differences

0.078 [0.008]

0.038 [0.008]

0.034 [0.009]

Demographic

Female 0.019 [0.006]

-0.004 [0.008]

-0.004 [0.006]

Married 0.052 [0.008]

0.028 [0.010]

0.058 [0.009]

Agea

age20-24 0.032

[0.009]

-0.978 [0.005]

0.02 [0.018]

age25-29 0.052

[0.008]

-0.983 [0.011]

0.029 [0.017]

age30-34 0.05

[0.009]

-0.984 [0.010]

0.023 [0.018]

age35-39 0.053

[0.009]

-0.986 [0.006]

0.03 [0.017]

age40-44 0.07

[0.008]

-0.983 [0.009]

0.021 [0.019]

age45-49 0.078

[0.008]

-0.981 [0.008]

0.024 [0.018]

age50-54 0.063

[0.009]

-0.977 [0.005]

0.023 [0.018]

age55-59 0.046

[0.011]

-0.959 [0.011]

-0.007 [0.027]

age60-64 0.08

[0.009]

-0.954 [0.006]

-0.02 [0.033]

age65 or above 0.089

[0.009]

-0.953 [0.004]

-

Country of Birth English-Speaking Country

0.058 [0.010]

0.065 [0.011]

0.022 [0.009]

Observations 9460 3325 6167 Note: a The omitted categories are age15-19 and Non-English-Speaking country.

25

5 Sensitivity Analysis

It is ideal to perform “placebo” tests that require two cohorts before the policy change and

another two cohorts after the policy change. Unfortunately, there are no data available to

perform this test at the time of writing this chapter. Even if there is one arranged in the future,

it would be ten years after the policy change. Things would have changed during the decade

that might affect both the unemployment and participation of those migrants.

Since the control group and treatment group are taken from two different surveys, the two

surveys might not necessarily be comparable for the estimation. Hence, in order to ensure the

SIH (control) group is similar to the LISA (treated) group, a propensity score matching

(PSM) procedure is used to reweight the sample before taking the difference-in-differences

again. This problem can be addressed better by using matching because it can minimise the

unobserved heterogeneity as well as reduce bias. The PSM provides a convenient way to

match on a large number of observed characteristics.

Bryson, Dorsett and Purdon (2002) states that the PSM attempts to provide unbiased

estimation of treatment-effects, and so allow differences of response to be attributed to

differences of treatments. Specifically, the motivation to use PSM stems from the needs for

analysing causal effects of treatment from observational data and for reducing selection

biases in non-experimental settings. Two features are commonly observed in such settings:

first, few units in the comparison group are comparable to the treatment units; and second,

selecting a subset of comparison units similar to the treatment unit is often difficult because

units must be compared across a high-dimensional set of pre-treatment characteristics.

26

A standard PSM typically assumes a common support, that is, the range of propensities to be

treated is the same for treated and control cases, even if the density functions have quite

different shapes. But if the two groups do not have substantial overlap, then an estimation

bias may be introduced: namely, if only the worst cases from the untreated “comparison”

group are compared to only the best cases from the treatment group, the result may be

regression toward the mean which may make the comparison group look better or worse than

reality.

PSM estimates each individual's propensity to receive a binary treatment as a function of

observables and matches individuals with similar propensities, usually obtained from logistic

regression to create a counterfactual group. In essence, propensity score is the probability of

taking treatment given a vector of observed variables.

(2)

denotes the probability of being observed in the treatment ( ) given the set of

characteristics ( ). If individuals with the same propensity score are divided into treated

and non-treated groups, the groups will be approximately balanced on the variables

predicting the propensity score. Rosenbaum and Rubin (1983) state that should the balancing

occur, the conditional independence assumption must satisfy, that is:

(3)

Among those with the same predicted probability of treatment , those who get treated

and not treated differ only on their error term in the propensity score equation. But this error

27

term is approximately independent of the X’s. The treatment assignment D is

independent of P, given the strata created by X’s. Blundell and Costa Dias (2000) suggest the

estimator for the repeated cross-sections data case as follows:

(4)

where ‘t’ and ‘c’ denote the treatment group and control group and similarly ‘b’ and ‘a’

denote ‘before’ and ‘after’ the policy change. is the weight placed on observation j when

being compared to treatment observation i, and N indicates the number of observations in the

treatment group. For the PSM estimator, two different weighting functions are used: 1) the

five nearest neighbours12 with replacement13 and 2) the biweight kernel function.14 The use of

nearest neighbours is a common practice, but kernel matching usually has smaller variance

than other matching procedures.

In the propensity score matching, one concern is the common support: ,

that is, the overlap condition for persons with the same X value are allowed to have a positive

probability of being in treated and control groups. In the matching, observations are excluded

12 Randomly order the participants and non-participants, then the method selects the five comparison units whose propensity scores are closest to the treated unit in question. More formally, the five nearest neighbours are determined by the distance measured as the absolute value of the difference between the ratio of the odds in the propensity score. An average is taken over the five values. 13 If replacement is allowed, an untreated individual can be used more than once as a match. Hence, the average quality of matching will increase and the bias will decrease. This is of particular interest with data where the propensity distribution is very different in the treatment and the control group. As Figure 2 shows, the distribution between control and treatment group is very different. Matching with replacement makes it better with the data. However, matching with replacement involves a trade-off between bias and variance (Caliendo and Kopeinig 2005). 14 The kernel function is the biweight with a bandwidth: where

and Nc is the size of the control group. The variance and the interquartile range are computed for the distance between the propensity scores of the treatment and the matching group. See Pagan and Ullah (1999) for more details on the properties of kernels and the choice of bandwidths.

28

if an independent variable falls outside the common support range across the control and

treatment groups. After matching, no observation is excluded from the sample. Thus, the

results from difference-in-differences estimation would be the same before and after

propensity score matching.

The other concern is the density of the coverage in the propensity score. The issue is that the

small group of matching observations are used for a substantial fraction of the treatment

group (Doiron 2004). Figure 2 shows that the bulk of the matching group is located in the

same range of the score as the treated observations. Moreover, there are far more matching

observations than the treated ones.

Figure 2 The Common Support Area from Five Nearest Neighbour Matching15

15 The numbers of those in the green area (“Treated: Off support) are essentially too small to be seen from Figure 2.

29

A measure of good PSM is the covariate imbalance testing before and after matching. The

equality of means for all covariates that were statistically significant before the matching

becomes insignificant after the matching. This means that the balancing of covariates occurs

and the conditional independence assumption satisfies. Furthermore, the matching achieved

bias reduction in most variables. The detailed results are available on request.

Table 10 Propensity Score Matching with Different Trimming 5-Nearest Neighbours Kernel Bandwidth=.01 No trimming 0.13 0.13 [0.02] [0.02] 5% trimming 0.12 0.12 (156 off support) [0.02] [0.02] 10%trimming 0.12 0.12 (329 off support) [0.02] [0.02]

Both five nearest neighbour and kernel matching give similar results, however, the five

nearest neighbour matching seems to perform better than kernel matching in terms of balance

testing before and after each matching. Five per cent trimming is also performed for both

matching estimators, and again the results from those estimators are almost the same.

After performing the propensity score matching on control and treatment group, the results

are similar to previous difference-in-differences estimation.

6 Discussion and Conclusion

Unlike previous research that has encompassed all migrants, this chapter only includes the

Family Stream migrants to separate the income-support policy effects from selection criteria

effects. Using difference-in-differences estimation to filter out macroeconomic effects, the

results show that the policy change in 1997 could be the major contribution to the substantial

improvement of immigrants’ labour market outcomes. The “quality” of Family Stream

30

migrants is almost constant between the two cohorts, as is observed from the descriptive

analysis. This implies that the denial of access to social welfare forced migrants from cohort

2 to more actively search for jobs, however there was a slight impact of overall labour market

improvement on immigrants’ labour outcomes.

Employment per se may be a poor indicator of labour market success if the immigrant

concerned is in a job that does not make full use of his or her skills and abilities (e.g. are

those immigrants without job security payment forced to take bad jobs?). It is, therefore,

necessary to also observe the indicators that reflect the “quality” of that job, such as earnings

and occupational status. There is evidence shown in Law (2008) that more migrants from

cohort 2 were competing for low-skilled jobs with natives. Moreover, Junankar and

Mahuteau (2005 and 2008) show that policy change had a negative impact with respect to

holding a good job.

Further research should shed light on endogeneity which might not be completely addressed

by the use of Family Stream migrants because their family members are likely to have faced

the change in migration criteria, and the migrants and their family members may share

common unobserved characteristics such as industriousness, ability, and diligence (for

example, assortative mating suggests similar traits among couples). If it is possible to identify

Family Stream migrants whose family migrated based on the pre-reform selection criteria,

and those whose family migrated based on the post-reform criteria, the former could be used

only as an alternative treatment group which is less subject to the endogeneity through the

household/family-level unobserved heterogeneity. Even with this alternative treatment group,

it is possible that welfare reform discouraged the application by potential migrants who had a

family member already in Australia.

31

Overall, this study provided a new way to examine the response of immigrants to the welfare

change in their labour market performance. The results suggested that the restriction to

welfare access caused immigrants to more actively look for jobs.

References

Ashenfelter, O., Ashmore, D. and Deschenes, O. 2005, "Do Unemployment Insurance

Recipients Actively Seek Work? Evidence from randomized trials in four U.S. States," Journal of Econometrics, Elsevier, vol. 125(1-2), pages 53-75.

Beggs, J. J. and Chapman, B. J. 1988, “Immigrant Wage Adjustment In Australia: Cross-Section and Time-Series Estimates”, Economic Record, vol. 64, September, pp. 161-7.

Bingley, P. and Walker, I. 2001, "Household Unemployment and the Labor Supply of Married Women," Economica, London School of Economics and Political Science, vol. 68(270), pp.157-85, May.

Borjas, G. J. and Tienda, M. 1985, eds. “Hispanic in the U.S. Economy”, New York, Academic Press.

Blundell, R. and Costa Dias, M. 2000, “Evaluation Methods for Non-Experimental Data”, Fiscal Studies, Vol. 21, 427-68

Boeters, S., Gurtzgen N. And Reinhold S. 2006, “Reforming Social Welfare in Germany: An Applied General Equilibrium Analysis”, German Economic Review, Vol.7 (4), pp.363-88

Bryson, A., R. Dorsett, and Purdon, S., 2002, “The Use of Propensity Score Matching in the Evaluation of Labour Market Policies", Working Paper No. 4, Department for Work and Pensions.

Caliendo, M. and Kopeinig, S. 2005, "Some Practical Guidance for the Implementation of Propensity Score Matching," IZA Discussion Papers 1588, Institute for the Study of Labor (IZA)

Cobb-Clark, D.A. 2003, “Public Policy and the Labor Market Adjustment of New Immigrants to Australia”, Journal of Population Economics, Vol. 16(4), November, 2003, pp. 655-81

Doiron, D. 2004, “Welfare Reform and the Labour Supply of Lone Parents in Australia: A Natural Experiment Approach”, Economic Record, Vol. 80, No. 249, June, 2004, pp.157-76

Garcia, J. 1991, “A Participation Model with Non-Convex Budget Sets: The Case of Wives of the Unemployed in Great Britain”, Applied Economics, 23, pp.1401-16

Junankar, P.N. and Mahuteau, S., 2005 ‘Do Migrants Get Good Jobs: New Migrant Settlement in Australia’, Economic Record, 81:S34-46

Law V., 2008, “English Proficiency and Migrant’s Labour Supply”, 2008, Working Paper, Australian National University

“LSIA User Document”, 2002, DIMIA, Canberra: AGPS. Mahuteau, S. and Junankar, P.N., 2008 ‘Do Migrants Get Good Jobs in Australia? The Role

of Ethnic Networks in Job Search’, Economic Record, 81:S34-46

32

Meyer, B. D. 1995, “Natural and Quasi-Experiments in Economics,” Journal of Business and Economics Statistics, 13(2), 151-161.

Miller, P. W. 1986, “Immigrant Unemployment in the First Year of Australian Labour Market Activity” The Economic Record, March, pp. 82-7.

Pagan, A. and Ullah, A. 1999, “Nonparametric Econometrics”, Cambridge University Press, Cambridge.

Richardson, S., Robertson, F. and Ilsley, D. 2001, “The Labour Force Experiences of New Migrants” AGPS, Canberra.

Rosenbaum, P. R., and Rubin, D. B. 1983, “The Central Role of the Propensity Score in Observational Studies for Causal Effects”, Biometrika, 70, pp. 41– 55.