5  Regression results on actual work experience by age 21

We know from Tables 4 and 5 that Maori in the CHDS accumulate less work experience, on average, compared to their non-Maori counterparts by age 21. This is despite the fact that Maori by this age have spent more time than non-Maori outside of education and training. In this section, we quantify the statistical relationship between potential and actual work experience, while controlling for personal and family background factors that are known from the previous section to vary significantly between Maori and non-Maori in many cases. To what extent do these differences in personal and family background characteristics influence the accumulation of work experience?

5.1  Short and long regressions: Two-way split in youth ethnicity

Results of the regression analysis using the two-way split in ethnicity are displayed in Table 10. Two specifications are used. The ‘short regression’ includes the few independent variables that are typically available in most cross-sectional data sets. These regressors include the individual’s years out of education and training, formal qualifications, ethnicity, gender and the number of children born to the respondent. The second ‘long regression’ includes various measures of personal and family backgrounds that are generally found only in panel data sets like the CHDS. They include the variables reported in Tables 6 and 7.

The dependent variable is the actual years of work experience accumulated by youth between the ages of 16 and 21. The mean of the dependent variable is 2.249 years across the sample of 973 individuals. The R² statistics are 0.299 and 0.346 for the short and long regressions, respectively. The addition of the 14 independent variables on family backgrounds in the long regression raises the explanatory power of the model by 15.7%. Using an F test, we can reject the null hypothesis that the coefficients on these family background variables are simultaneously equal to zero at better than a 1% level. Yet, just over one-third of the variation in work experience can be explained by this long regression.

** Significantly different from zero at 1% level.

* Significantly different from zero at 10% level.

Notes: These data are taken from the 973 respondents in the CHDS who provided valid information for the purposes of this study. The dependent variable is the effective years of work experience accumulated by the individual between the ages of 16 and 21. Youth are defined as ‘Maori’ in these regressions if they identify Maori as at least one of their ethnicities at age 21, and had at least one parental figure claiming Maori ethnicity by age 14 of the CHDS child. All other youth are considered to be ‘non-Maori’. Standard errors are in parentheses.

Table 4 showed that Maori, on average, had 0.288 fewer years of work experience by age 21 relative to non-Maori. If the other independent variables accounted for none of the ethnic differences in accumulated work experience between Maori and non-Maori, the estimated coefficient on Maori in these regressions would be –0.288. This estimated parameter is slightly smaller in absolute value in the short regression (-0.247), and significantly different from zero at better than a 2.5% level.

To perform the Neumark decomposition we need to re-estimate the short regression excluding the Maori indicator variable as a regressor. These regression results are not reported, but the R² statistic declines only slightly from 0.2989 to 0.2953 when we drop ethnicity from the short regression.[18] In other words, there is only a 1.2% loss in explanatory power when ethnicity is dropped from an experience regression that already includes years not enrolled in education or training, gender, gender interacted with children born to the individual, and the person’s formal qualifications. Almost all of the explained variation in actual work experience across individuals can be accounted for by measured personal characteristics and circumstances other than ethnicity. Adding the Maori indicator variable does not substantially enhance the predictive performance of this regression model.

The findings from these experience regressions can be summarised by equation (4), which was developed in the Section 3 of this report. This equation is re-produced below.

(9)     EXP_NM - EXP_M = [(POTEXP_NM - POTEXP_M)b + (X′_NM - X′_M)c] +

[(b_NM - b)POTEXP_NM + (b - b_M)POTEXP_M + (c_NM - c)X′_NM + (b - b_M)X′_M]

The left-hand side of the expression is the observed mean gap in actual work experience between non-Maori and Maori. The first term in the square brackets on the right-hand side is the explained component of the overall gap in work experience. This can be broken down into parts that can be related to ethnic differences in potential work experience and all other covariates in the regression. The second term in the square brackets is the unexplained component of the overall gap in work experience. Using estimated coefficients from an unreported ‘pooled’ regression, the estimated values for these terms can be substituted into this expression.[19]

(10)     Observed Gap = [(ΔPOTEXP)b + (ΔX′)c] + [Unexplained] = [Explained] + [Unexplained]

0.288 = [-0.238 + 0.298] + [0.228] = [0.060] + [0.228]

The results from this short regression can be interpreted in the following way. We know that non-Maori accumulated an average of 0.288 more years of work experience than Maori by age 21. However, this is the net effect of two offsetting forces. Maori have, on average, more years of potential work experience than non-Maori by age 21. The estimated coefficient on years not enrolled in education or training is positive (0.420), and significantly different from zero at better than a 1% level in the regression without ethnicity.[20] Table 4 shows that the average potential work experience of Maori exceeds that of non-Maori by 0.566 years. Multiplying this difference in means by the estimated coefficient on this variable suggests that the additional time spent outside of education and training by Maori would increase their actual work experience by slightly less than one-quarter of a year (0.566 • 0.420 = 0.238). This is the first number in the squared brackets in equation (10). In other words, if non-Maori spent the same time as Maori away from education and training, the ethnic gap in work experience would widen to more than one-half of a year (0.288 + 0.238 = 0.526).

Yet, we also know from Tables 4 and 6 that Maori have, on average, more children and fewer qualifications than non-Maori. If non-Maori had these same personal characteristics as Maori, the ethnic gap in work experience would narrow by 0.298 years. If both ethnic groups had the same amount of time away from education and training and the same observable factors in other areas, then the gap in experience would narrow by only 0.060 years. This says that about one-fifth (20.8%) of the observed gap in experience between non-Maori and Maori can be explained by the covariates in this short regression (0.060/0.288). Nearly four-fifths (79.2%) of this gap cannot be accounted for with this specification (0.228/0.288).

Suppose we restrict the CHDS sample to those without formal school or tertiary qualifications. By age 21, unqualified Maori (n=28) had accumulated 0.606 fewer years of work experience than unqualified non-Maori (n=132). This gap in experience between these ethnic subgroups is substantially larger than the gap for the overall ethnic groups of 0.288. This is consistent with the earlier results that showed that the gap in work experience between Maori and non-Maori was largest at the lower tails of these respective distributions (Figure 1). The covariates in the regressions reported above account for 0.060 and 0.166 years of the ethnic gaps in work between all and unqualified 21 year-olds, respectively. The results from our short regression are able to explain a larger proportion of the ethnic gap in work experience among unqualified youth (27.4%) compared to all youth (20.8%). Yet, our conclusion is that the majority of ethnic differences in the early accumulation in work experience cannot be explained by the covariates in this short regression.

We can re-estimate the short regression for the subsample of CHDS youth without formal qualifications (n=160). Of course, the regressors on school and tertiary qualifications are excluded from this estimation, because of an absence of variation in these regressors. These regression results are not reported, but the R² statistic declines from 0.2356 to 0.2263 when we drop ethnicity from the short regression. This means that there is a 3.9% loss in explanatory power when ethnicity is dropped from an experience regression that already includes these other covariates. Most of the explained variation in actual work experience across unqualified youth can be accounted for by measured personal characteristics and circumstances other than ethnicity. Yet, the Maori indicator variable does more to enhance the predictive performance of this regression model among unqualified youth than it did earlier among all youth.

The second set of empirical results reported in Table 10 relate to the long regression, which adds information on the youth’s personal and family background characteristics to the previous set of independent variables. It is worth noting that the estimated coefficients on the educational qualifications of parents are generally negative, and significantly different from zero in three of the six cases. Once other factors have been held constant, the education of the parents directly reduces the accumulation of work experience by the child. The proportion of years living in a single-parented family and the number of children in the family have estimated effects on work experience that are not statistically different from zero.

The proportion of years in which the family received social welfare benefits while the CHDS child was between the ages of 1 and 14 has a negative impact on his or her accumulation of work experience. This estimated coefficient is statistically significant at better than a 10% level. The mean real income of the family in which the child resided between ages 1 and 14 has a positive effect on subsequent work experience, and is significant at a 1% level. To get an idea of the magnitude of this income effect, the overall gap in experience between Maori and non-Maori is equivalent to an increase in real mean family income of Maori of $36,000 ((0.288/0.080) • $10,000). This is more than four-times the actual gap in mean family income between non-Maori and Maori ($8,824 taken from Table 6).

Measured conduct problems and criminal convictions have no statistical impact on the accumulation of work experience. Yet, the proportion of years between the ages of 12 and 16 that the child was truant, suspended or expelled from school and the proportion of years between 18 and 21 that the youth showed evidence of alcohol/drug abuse or dependence both have negative effects on work experience that are significantly different from zero at better than 10% levels.

Once all of the factors in the long regression have been held constant, the estimated coefficient on Maori is now –0.169, and only significantly different from zero at a 12.5% level. This means that the estimated coefficient on Maori has fallen in absolute value by nearly one-third in moving from the short to the long regression, and is no longer statistically significant at conventional test levels. The estimated coefficient on years not enrolled in education or training is positive and significant at a 1% level, but has increased slightly in magnitude from 0.418 in the short regression to 0.436 in the long regression.

Again, to perform the Neumark decomposition we need to re-estimate the short regression excluding the Maori indicator variable as a regressor. These regression results are not reported, but the R² statistic declines only slightly from 0.3464 to 0.3448 when we drop ethnicity from the short regression. In other words, there is only a 0.5% loss in explanatory power when ethnicity is dropped from an experience regression that already includes these other personal and family background characteristics. Adding the Maori indicator variable does not substantially enhance the predictive performance of this regression model.

We can break the actual gap in work experience between non-Maori and Maori into its explained and unexplained components using the estimated coefficients from an unreported pooled regression excluding the Maori indicator variable as an explanatory variable. These results are summarised in the following equation:

(11)     Observed Gap = [(ΔPOTEXP)b + (ΔX′)c] + [Unexplained] = [Explained] + [Unexplained]

0.288 = [-0.249 + 0.389] + [0.148] = [0.140] + [0.148]

Again, non-Maori accumulated an average of 0.288 more years of work experience than Maori by age 21. This is the effect of two offsetting forces. Maori have, on average, more years of potential work experience than non-Maori. The estimated coefficient on years not enrolled in education or training is positive (0.439), and significant at a 1% level in the regression without ethnicity. Multiplying this estimated coefficient by the observed difference in potential work experience between the ethnic groups of 0.566 years says that this effect would increase the ethnic gap in work experience by slightly less than one-quarter of a year (0.566 • 0.439 = 0.249). This effect alone would raise the ethnic gap in work experience to more than one-half of a year (0.288 + 0.249 = 0.537).

All of the other observed differences in personal characteristics and family backgrounds between non-Maori and Maori in this long regression would narrow the gap in work experience by 0.389 years. If both ethnic groups had the same amount of time away from education and training and the same observable factors in other areas, then the gap in work experience would narrow by 0.140 years. This says that slightly less than one-half (48.6%) of the observed gap in experience between non-Maori and Maori can be explained by the covariates in this long regression (0.140/0.288). Slightly more than one-half (51.4%) of this gap cannot be explained by this specification (0.148/0.288).

Suppose we again restrict the CHDS sample to those without formal school or tertiary qualifications. Recall that by age 21, unqualified Maori (n=28) had accumulated 0.606 fewer years of work experience than unqualified non-Maori (n=128). These covariates account for 0.140 and 0.385 years of the ethnic gaps in work between all and unqualified 21 year-olds, respectively. The results from our long regression are able to explain a larger proportion of the ethnic gap in work experience among unqualified youth (63.5%) compared to all youth (48.6%). Family background is particularly important in explaining the relatively poor work experience histories of Maori without school or tertiary qualifications.

We re-estimate the long regression for the subsample of CHDS youth without formal qualifications (n=160). These regression results are not reported, but the R² statistic declines from 0.3677 to only 0.3676 when we drop ethnicity from the long regression. This means that there is almost no loss in explanatory power when ethnicity is dropped from this specification. The variation in actual work experience across unqualified individuals that was previously explained by ethnicity is now captured by measured family background characteristics.