The Labour Market Adjustment of Immigrants in New Zealand Report
5. Main Results
5.1 Regression model
We extend the descriptive evidence by estimating regressions models of the relationship between labour market outcomes, whether an individual is an immigrant, if so, how long they have lived in New Zealand, and other characteristics. These models take the following form:
Yit=ΒImmit+ƒ(YrsNZit)+δXit+αt+εit (1)
where i indexes individuals and t indexes time, Yit is an indicator variable for whether an individual is employed, their log real hourly wage (if employed and responding to the NZIS), their real annual income (if responding to the NZIS) or the average log real wage for New Zealand-born in their 2-digit occupation (if employed). Immit is an indicator variable for whether an individual is an immigrant to New Zealand, YrsNZit is the number of years that an individual has lived in New Zealand (set to zero if they are New Zealand-born),[15] Xit are other control variables to allow for differences between immigrants and the New Zealand-born, such as human capital, that are related to differences in outcomes, αt are time fixed effects which control for aggregate changes in employment, wages and incomes over time and eit is a mean zero idiosyncratic error term.
We extend upon previous papers in the international literature by allowing outcomes for immigrants to change with years spent in New Zealand in a semi-parametric manner that makes no assumptions about how labour market outcomes evolve as more host country experience is acquired.[16] We do this by including a series of indicator variables for all observed magnitudes of years in New Zealand (zero to fifty-eight years). In all cases, we also estimate separate OLS regressions stratified by gender to allow for different assimilation profiles for male and female immigrants. We rely on an OLS regression for each outcome even though employment is a discrete outcome, because this approach is more amenable to semi-parametrically estimating the impact of years spent in New Zealand.
5.2 Regression specifications
We begin by estimating five specifications of equation (1) that include progressively more control variables (Xit). In the first specification, we include the baseline variables in equation (1) and no additional control variables. The impact of years in New Zealand on average outcomes for immigrants relative to the New Zealand-born is illustrated by the solid line in each panel of Figure 5. As in Figure 1, the upper three panels in this figure display the results for men and the lower three panels display the results for women. The first column illustrates how employment rates for immigrants relative to the New Zealand-born differ with time spent in New Zealand. The second column illustrates the same results for log real wages and the third column for real annual income. In each case, we apply a smoothing algorithm to reduce the volatility of the estimates. Specifically, we use an Epanechnikov kernel with a 3-year bandwidth. In other words, each point on the graph in Figure 3 is a weighted average of five adjacent coefficients for neighbouring years spent in New Zealand, with declining weights.[17] We also graph only up to 35 years in New Zealand since the remaining coefficients out to 58 years in New Zealand are typically extremely imprecisely estimated and based only on specific immigrant cohorts.[18]
Figure 5: Regression Adjusted Outcomes for Immigrants by Gender and Years in New Zealand - Different Specifications
The first important thing to notice when examining these results is that the assimilation profile is almost never quadratic, which is a restriction that is commonly imposed in this literature. Thus, allowing for a semi-parametric profile reveals meaningful differences in evaluating the performance of immigrants as they spend more time in New Zealand. In particular, for employment rates for both men and women, and for wage and annual incomes for women, the improvement is relatively steep through until around 10-20 years, after which the gradient is essentially flat.
Each graph in Figure 5 contains four more profiles in addition to the bold 'no controls' line. These relate to the different regression specifications with progressively fuller sets of covariates added. The first extension is to control for differences in human capital between immigrants and the New Zealand-born. Specifically, we include a quadratic in age, indicator variables for whether an individual has low school qualifications (primary proficiency examination, school certificate or other school qualifications), has high school qualifications (sixth-form, higher school leaving certificate, or university bursary), or has foreign school qualifications (with a default category of no qualifications), an indicator variable for whether an individual has post-school vocational qualifications and an indicator variable for whether they have a university degree. The impact of ageing and qualifications on labour market outcomes is assumed here to be the same for immigrants and the New Zealand-born. We later examine whether the returns to qualifications are, in fact, different for immigrants and the New Zealand-born, but it is not possible to allow age effects to differ and at the same time identify the impact of years spent in New Zealand since these both increase at the same rate.[19]
Given that immigrants to New Zealand are generally more qualified than the New Zealand-born, we expect that adding these control variables will shift the profiles for immigrants in a downward direction (i.e. they will look relatively less successful than the New Zealand-born). The results from this specification are presented as long-dashed lines in Figure 5. As expected, the relative outcomes for immigrants look slightly less favourable when we standardise for age and qualification differences. The impact is most pronounced for the log wage outcome, and for men's incomes, both of which are strongly related to age and education. We present the coefficients for the control variables (Xit) included in this model (as well as the remaining specifications) in Table 3 (employment rates), Table 4 (log real wage rates) and Table 5 (real annual income).
Table 3: OLS regression of employment rates by gender
Table 4: OLS regression of log real hourly wage by gender
Table 5: OLS regression of annual real income by gender
In the third specification, we include additional controls for whether the individual is married, widowed/divorced/separated (with a default category of never married), their family type (couple with no children, couple with children, single with children or non-family, which is the default), an indicator variable for whether they live in an urban area, and a series of indicator variables for geographic location (one of twelve local government regions). As shown in Table 1, many of these characteristics differ between immigrants and the New Zealand-born and are likely to be associated with differential success in the labour market. These results are presented as intermittently long-dashed lines. The impact on the estimated relative outcomes for immigrants is largest for wages and for incomes, although for all of the graphs, controlling for these household and location characteristics makes immigrant outcomes look worse. This reflects the more advantageous household and location characteristics of immigrants. Once we control for these advantages and compare similar immigrants and New Zealand-born adults, the immigrant disadvantage appears greater.
Our results up to this point assume that outcomes are the same for all immigrants conditional on their human capital and other observables characteristics. However, it is quite likely that the unobserved quality of immigrants varies over time due to changes in immigration policy and the relative attractiveness of migrating to New Zealand. In the fourth specification, we add controls for the arrival cohort to which a particular immigrant belongs. Specifically, we include five indicator variables for whether an immigrant arrived in 1958-67; 1967-78; 1978-87; 1988-97; and 1998-2007.[20] A sixth indicator variable for arriving prior to 1958 is dropped from the model. The included variables are not defined as typical 0/1 variables, but instead using the deviation contrast where an indicator variable is coded as 0 if the individual did not arrive in that cohort, and 1 if they did arrive in that cohort (as is the typical way these variables are coded), but all included indicator variables are coded as -1 when the individual arrived prior to 1958 (ie. in the omitted category).
When this coding scheme is used the estimated coefficients sum to zero over the full set of categories (including the category that is dropped from the model, ie. whether an immigrant arrived prior to 1958) and are interpreted as the difference in the outcome for an immigrant in a particular cohort versus an immigrant from the average cohort (as opposed to versus the outcome for immigrants in the omitted category). The coefficient for the omitted category can be calculated as minus the sum of the estimated coefficients. This approach is used for all immigrant specific variables included in the regression model (in particular in the fifth specification), because this allows β, the coefficient on the Immit indicator variable, to retain its interpretation as the difference between the average New Zealander and the average immigrant, conditional on other characteristics. On the other hand, if the traditional approach for defining indicator variables was used, this coefficient would instead be interpreted as the difference between the average New Zealander and the average immigrant in the omitted cohort (here, the pre-1958 cohort).
The results from this regression specification are presented as dotted lines in each panel of Figure 5 and again in Tables 3-5. With one exception, controlling for unobserved cohort effects leads to a flattening of the slope of the adaptation profiles. Some of the apparent improvement in relative outcomes for immigrants as they spend more years in New Zealand can be attributed to differences in unobserved cohort characteristics. As first found in Borjas (1985) for the United States, more recent immigrant cohorts to New Zealand generally have less favourable unobservable characteristics. Thus, for any given cohort, there is less improvement with years spent in New Zealand. The one exception is employment rates for males. In this case, adjusting for cohort effects leads to a steeper profile, implying that recent cohorts have unobservable attributes that make them more likely to be employed, although the differences are small. We speculate that this may be related to immigration policy settings, which over time have given increased priority to residence applicants having a job offer.
In the fifth and final specification, we include additional controls for differences in immigrant characteristics. This controls for compositional differences in the immigrant population that are related to how long individuals have lived in New Zealand. In other words, it accounts for the fact that some immigrant groups have generally been less successful in the New Zealand labour market and have been in New Zealand for more or less time than the average immigrant. In particular, we control for whether an immigrant arrived in New Zealand prior to age 18 and thus likely received some education in New Zealand, and whether an immigrant is from Australia (the omitted category), the United Kingdom, Asia, the Pacific Islands or elsewhere (coded Other). As in the prior specification, these are all defined using the deviation contrast with the coefficients on each category adding to zero. In the case of the indicator for whether an immigrant arrived in New Zealand prior to age 18, where there are only two categories, the impact of arriving prior to 18 compared to arriving at 18 or greater can be calculated as 2 times the reported coefficient (recall that the coefficient on the omitted category is just minus the sum of the other coefficients and that all coefficients are interpreted as the difference versus an immigrant with the average likelihood of arriving prior to 18).
The results from this regression specification are presented as intermittently dashed and dotted lines in each panel of Figure 5 and again in Tables 3-5. In most cases, the profiles are similar to those obtained in the previous specification which controlled for immigrant cohort fixed effects. This suggests that the cohort fixed effects generally capture the same information as is contained in the region of birth and age at arrival measures. For men's wages and incomes, the additional controls lead to a further flattening of the years-in-New Zealand profile, reflecting that even within 10-year arrival cohorts, some of the apparent improvement in wages is a result of more recent arrivals having less favourable region-of-birth characteristics.
5.3 Summary of main results
We believe that the extended regression model presented in the fifth specification provides the most robust comparison of outcomes between immigrants and New Zealanders since it allows for both differences in human capital and socio-demographic characteristics between immigrants and the New Zealand-born and allows for differences in outcomes for diverse groups of immigrants. It therefore comes closest to tracing the adaptation path followed by an individual migrant.
In Figure 6, we again present the results from the final regression specification, but now also graph 95 percent confidence intervals for our estimates. The confidence intervals are calculated as twice the standard error on the weighted mean of neighbouring coefficients. Again, the upper three panels in this figure display the results for men and the lower three panels display the results for women. The first column illustrates how employment rates for immigrants relative to the New Zealand-born differ with time spent in New Zealand. The second column illustrates the same results for log real wages and the third column for real annual income.
Figure 6: Regression Adjusted Outcomes for Immigrants by Gender and Years in New Zealand - Main Estimates
For both employment rates and annual incomes, there is evidence of a statistically significant improvement in relative outcomes over the first 10 years in New Zealand, and a stabilisation after that at levels at or slightly below that of comparable New Zealanders. However, both male and female migrants have wage rates that are generally below those of comparable New Zealanders. The confidence intervals are relatively wide, so that for immigrant men, we cannot reject the absence of any post-arrival improvements. For immigrant women, the only statistically significant improvement is for the comparison of entry wages and wages after 15 years.
5.4 The role of occupational choice
Using this same framework, we now consider the role that occupational choice plays in explaining differences in outcomes between immigrants and the New Zealand-born. As with the wage outcome, occupational rank is defined only for people who are employed. First, in the first column of Figure 7 and in Table 6, we present the results from estimating the five specifications of regression model (1) where the outcome variable is defined as occupational rank, as measured by the average log real wage for the New Zealand-born in each 2-digit occupation. In the second column of Figure 7, we present the results from the fifth specification including confidence intervals as in Figure 6.
Figure 7: Regression Adjusted 2-Digit Occupational Distribution of Immigrants by Gender and Years in New Zealand
Table 6: OLS regression of occupation classified by average wages by gender
The solid line shows relative occupational rank without any covariate controls. Immigrant men have occupational rank that is consistently above that of the average New Zealand-born worker, while immigrant women have occupational rank that is generally similar to that of New Zealand-born women. As was the case for the other labour market outcomes, controlling for age, qualification, household type and location serve to reduce the estimated relative outcomes of immigrants. The more advantageous characteristics of immigrants account for some of their better raw outcomes, especially for more recent migrants. Adjusting for unobserved cohort characteristics has minimal impact on the profile, but as for the wage outcomes, controlling for region of birth leads to a further flattening of the occupational rank profile. In particular, even within decadal arrival cohorts, migrants who have been in New Zealand for more than 25 years have region-of-birth and age-at-arrival characteristics associated with high occupational rank.
Overall, controlling for the full set of individual and household characteristics makes the relative occupational rank of immigrants look less favourable. For both men and women, immigrants with less than 15 to 20 years in New Zealand have significantly lower occupational rank than comparable New Zealand-born workers. Improvements are evident for both men and women, although the confidence intervals are reasonable large. For men, the improvement of occupational rank is barely significant between their first few years and 20 years after arrival. For women, there is a significant improvement within the first 15 years after arrival.
Note that the only way that immigrants can improve their occupational rank is by changing two-digit occupation. The results imply that some occupational upgrading does occur for immigrants as part of their adaptation to the New Zealand labour market. In order to gauge the contribution of occupational upgrading to estimated wage profiles, we estimate the full-model specification for the wage outcome, but include also a set of 2-digit occupational dummy variables. The resulting wage profile shows the pattern of wage adaptation that occurs within occupations. i.e. excluding the contribution of the occupational upgrading that was shown in Figure 7. The first column in Figure 8 again presents the results for log real wages as estimated in the fifth regression specification (the second column of Figure 6), Then, in the second column, we present the equivalent results when occupational fixed effects are added to the model. The profiles are visually very similar, and not statistically distinguishable, implying that occupational upgrading is not a significant contributor to estimated wage adaptation.
Figure 8: Regression Adjusted Hourly Wages for Immigrants by Gender and Years in New Zealand - Controling for Occupation
In Figure 9, we repeat this exercise but examine relative differences in annual income. For women, we again find that occupational upgrading is not a significant contributor to estimated income adaptation. However, for men, we see that, controlling for differences in occupation, the income gap for migrants in New Zealand for less than 5 years is 25 percent smaller (7,500 vs 10,000) and consequently the annual income - years in New Zealand adaptation gradient is now entirely flat. This indicates that the relative increase in income for male migrants during the first 10 years in New Zealand occurs because these migrants are switching into higher paid occupations in terms of annual income.
Figure 9: Regression Adjusted Annual Income for Immigrants by Gender and Years in New Zealand - Controling for Occupation
5.5 The importance of different returns to human capital
We next extend our regression model by examining whether the relationship between qualifications and labour market outcomes differs for migrants and the New Zealand-born, and the role that this plays in explaining differences in outcomes between the two groups. This is a flexible way of allowing for the possibility that the value of the human capital held by immigrants with the same qualifications as New Zealanders is less because of the imperfect transferability of skills gained overseas or because of poorer complementary skills, such as English language ability. In Figure 10, we present results that compare the impact of years in New Zealand on each of the four outcomes derived in our main model (ie. the fifth specification in Figure 5) to results from a similar model that, in addition, allows the return to qualifications to differ for New Zealanders and immigrants. This is done by interacting each of the qualification control variables with an indicator variable for whether an individual is an immigrant and again with an indicator variable for whether they arrived at less than age 18. This allows for different returns to qualifications for these two immigrant groups.
Figure 10: Regression Adjusted Outcomes for Immigrants by Gender and Years in New Zealand - Returns to Qualifications Differ for Immigrants
In the underlying regressions, there is no statistical difference in the returns to qualifications between the New Zealand-born and immigrants who arrived in New Zealand before the age of 18. Immigrants arriving at later ages earn more of a premium from vocational qualifications than the New Zealand-born do, in terms of wages, incomes, and occupational rank, perhaps reflecting the particular mix of vocational qualifications held. University-qualified immigrants receive less of an income premium from their qualifications than do New Zealand-born graduates, and immigrant males also receive less benefit in terms of occupational rank. Overall, unlike what Friedberg (2000) finds for the United States, foreign-earned qualifications appear to be fairly portable to the New Zealand labour market.[21] Consistent with this, the results in Figure 10 show that the estimated assimilation profiles from models that allow for group-specific qualification premia are very similar to those that constrain qualification premia to be the same across all groups. Thus, in the New Zealand context, differences in returns to qualification make a limited contribution to the estimated patterns of immigrant adaptation.
5.6 Heterogeneity across immigrants
In this last sub-section, we examine how the process of labour market assimilation varies for immigrants with different educational qualifications, those born in different regions, and those who arrived in New Zealand at different ages. While one weakness of the NZIS for examining immigrant outcomes is that detailed country of birth information is unavailable, we are still able to classify migrants as being born in one of five regions between which there are large differences in immigrant characteristics and outcomes.
We first examine models that stratify by educational qualifications. Specifically, we divide the sample into four groups, individuals with no qualifications, those with school qualifications, those with post-school vocational qualifications, and those with university degrees. We estimate the fifth specification of regression model (1) for each of these groups. The results are presented in Figure 11 (employment, wages and income for men), Figure 12 (employment, wages and income for women), and Figure 13 (occupational choice for both genders).
Figure 11: Regression Adjusted Outcomes for Male Immigrants by Qualifications and Years in New Zealand
Figure 12: Regression Adjusted Outcomes for Female Immigrants by Qualifications and Years in New Zealand
Figure 13: Regression Adjusted 2-Digit Occupation by Gender, Qualifications and Years in New Zealand
There is an entry-level disadvantage in employment rates for immigrant men who have university qualifications, and also for those who lack qualifications. Subsequent improvements in the relative employment rates for university-qualified men see them reach parity with their New Zealand-born counterparts within about 10 years. For those without qualifications, the process of catching up is slower, taking around 20 years. The patterns are slightly different for immigrant women. The entry disadvantage of immigrant women without qualifications is relatively small, and not statistically significant. For other qualification groups, immigrant women enter with a relative disadvantage that is eliminated after about 10 years.
Conditional on being employed, university qualified immigrant men, and immigrant women with vocational or university qualifications are the only groups to experience a significant wage disadvantage at the point of entry. Even then, the differences from the New Zealand-born are only just significant due in part to imprecisely estimated effects. Surprisingly, school-qualified immigrants appear to lose ground in terms of relative wage rates after about 20 years in New Zealand.
Immigrant women of all qualification levels have annual incomes that are similar to those of their New Zealand-born counterparts. In contrast, immigrant men have incomes that are at or below the level of comparable New Zealand-born men. University qualified immigrant men experience low initial incomes that approach New Zealand-born levels after about 15 years. It takes considerably longer for unqualified immigrant men to catch up to the New Zealand-born, and for those with vocational qualifications, there is no evidence of catching up. Relative annual incomes of unqualified immigrant men are initially low, and remain low for at least 20 to 25 years. In contrast, immigrant women without qualifications experience no significant income gap.
As shown in Figure 13, convergence of occupational rank is strongest for employed immigrants with vocational or university qualifications, and for unqualified immigrant women. However, the size of effects is not strong, and with the exception of a long period of relatively low occupational rank for immigrant men with vocational qualifications, is mostly statistically insignificant. As with wages, there is some evidence that school-qualified immigrants lose ground in occupational rank after 15 to 20 years compared with their New Zealand-born comparators.
We next examine models that stratify by immigrant region of birth. Specifically, we divide the sample into the five region-of-birth groups used throughout the analysis. Since this is a characteristic that is defined only for immigrants, in each case we compare outcomes for immigrants from a particular region of birth to outcomes for the full sample of the New Zealand-born, conditional on the variables included in the regression model.[22] Again, we estimate the fifth specification of regression model (1) for each of these groups. The results are presented in Figure 14 (employment, wages and income for men), Figure 15 (employment, wages and income for women), and Figure 16 (occupational choice for both genders).
Figure 14: Regression Adjusted Outcomes for Male Immigrants by Region of Birth and Years in New Zealand
Figure 15: Regression Adjusted Outcomes for Female Immigrants by Region of Birth and Years in New Zealand
Figure 16: Regression Adjusted 2-Digit Occupation by Gender, Region of Birth and Years in New Zealand
There are two common and striking patterns across all four outcome variables. First, the pattern of entry disadvantage followed by subsequent relative improvement is primarily a feature of adaptation for immigrants from Asian countries and to a lesser extent to the group of 'other' countries. Second, immigrant men from Pacific Island countries have consistently worse outcomes than the New Zealand-born, with no evidence of convergence.[23] This contrasts with the findings of Poot (1993) who shows income convergence for Pacific immigrants in particular occupations using 1986 Census data. A lack of convergence is also evident for the occupational rank of immigrant women from Pacific countries, but not for their other outcomes. For Australian and United Kingdom immigrants, there is little evidence that they have outcomes any different from those of comparable New Zealanders.
Finally, we examine models that stratify by whether an immigrant arrived in New Zealand prior to turning 18. These results are presented in Figure 17 (employment, wages and income) and Figure 18 (occupational choice). As in the previous analysis, since this is a characteristic that is defined only for migrants, in each case we compare outcomes for immigrants from one of the two age-at-arrival groups to outcomes for the full sample of the New Zealand-born, conditional on the variables included in the regression model. Because we only include people in the sample when they are 25 and older, no individuals have arrived in New Zealand prior to turning 18 and had been in New Zealand for less than 7 years. The coefficients for 8 and 9 years in New Zealand for this particular group are also estimated over a very small sample (i.e. only individuals that arrived at age 16 and 17 in 1988-1989) and the resulting coefficients were extremely imprecisely estimated, thus we start the graphs for this group at 10 years in New Zealand.
Figure 17: Regression Adjusted Outcomes for Immigrants by Gender, Age at Arrival and Years in New Zealand
Figure 18: Regression Adjusted 2-Digit Occupation by Gender, Age at Arrival and Years in New Zealand
Immigrants who arrived before they turned 18 have outcomes that are indistinguishable from those of comparable New Zealand-born people, with the possible exception of immigrant women, who appear to lose ground relative to their New Zealand-born counterparts after 20 to 25 years in New Zealand. In contrast, those who arrived at older ages experience poor initial employment rates and incomes that converge towards those of the New Zealand-born. For males, the convergence is only partial but for female immigrants, is complete within 15 years. Relative wages are also lower for immigrant who arrived later in life, although not always significantly so for men, and the wage gap is still evident after they have spent 35 years in New Zealand. Occupational rank also remains relatively low for immigrant men and women who arrive after age 18, for at least 30 years after arrival.
Footnotes
[15] Setting years since arrival to zero for the New Zealand-born has no impact on the results because a separate indicator variable is included for whether an individual is an immigrant (ie. this variable can be set to any number for the New Zealand-born without impacting the results).
[16] Clark and Lindley (2009) also take a semi-parametric approach to estimating immigrant labour market assimilation using local linear regression models. Given that years since arrival is a discrete variable, our approach is preferable since local regression techniques are designed to be applied to continuous variables.
[17] The coefficient at years=0 which indicated the initial difference in outcomes between migrants and the New Zealand-born is not averaged.
[18] With only 11 years of data, all points in the assimilation profiles are, in fact, identified by the variation in outcomes across 11 annual entry cohorts of new migrants. Thus, it is not possible to separately identify the role that long-run changes in immigration policy have had on say initial labour market outcomes. However, with further assumptions, it would be potentially possible to identify the impact of business cycles on initial labour market outcomes. One important advantage of the semi-parametric approach used here is that long-run changes in cohort quality will not bias our results for differences in initial labour market outcomes and early assimilation (ie. because we have no functional form assumption, the observations that are used to identify say changes in outcomes from 20 to 30 years in New Zealand have no influence on the results for changes in outcomes from 0 to 10 years in New Zealand). This is not the case when parametric models are estimated.
[19] In our current regression model, it would actually be possible to allow for different age effects for immigrants and the New Zealand-born because we are restricting the age effects to be quadratic, but it is difficult to justify this given the arbitrary nature of the restriction on the age effects.
[20] The choice of ten-year cohorts and the particular year cutoffs used to assign the cohorts is entirely arbitrary. It is not possible to jointly identify single year cohort effects and semi-parametrically estimate the impact of years in New Zealand since these will perfectly co-vary. However, we have tested whether our main results are robust to using either five-year or two-year entry cohort effects. Making this change has little qualitative impact, but it does decrease the precision of our estimates. Thus, we have decided to continue using ten-year cohorts.
[21] This results is consistent with the fact that skilled migrants to NZ typically need to have their qualifications ‘recognised’ as being identical to their NZ equivalents in order for them to count in the points system.
[22] Comparing immigrants from each region of birth to the full sample of the New Zealand-born allows for a simple comparison of the outcomes for one group of immigrants to those for another group. For Asian and Pacific Island immigrants, an alternative would be to compare their outcomes to only New Zealand-born individuals with Asian or Pacific Island ethnicity. This approach implicitly assumes that there is something about being Asian or a Pacific Islander that leads to different labour market outcomes in New Zealand and that we should be controlling for this when examining outcomes for immigrants from this ethnic group. We find this reasoning unsatisfactory; however, there is scope for a worthwhile empirical study to jointly consider the impact of ethnicity and immigration status on labour market outcomes.
[23] As discussed in the previous footnote, these results are consistent with both there being pathways that lead to poor labour market outcomes for Pacific Islanders in New Zealand, in general, and there being pathways specific to immigrants from the Pacific Islands.
