2  Recent literature on ethnic disparities in labour market outcomes in New Zealand

There have been numerous empirical studies on labour market disparities between Maori and non-Maori in New Zealand. Only some of the more recent studies are reviewed in detail here.

Winkelmann and Winkelmann (1997) used unit record data from the 1981, 1986 and 1991 Population Censuses to examine the reasons behind the systematic differences in current labour force states between Maori and non-Maori. They choose four mutually exclusive and exhaustive states: full-time employment, part-time employment, unemployment and the residual category of being out of the labour force. The authors employed a multinomial logit approach to estimate the effects of age, school and tertiary qualifications, marital and parental status, the local unemployment rate and urban or rural geographic location on these labour force states. The authors found that these relatively few demographic characteristics could explain a substantial proportion of the differences in labour market outcomes between Maori and non-Maori. Between one-half and two-thirds of the ethnic differences in full-time employment rates for males, for example, could be explained by these personal characteristics and geographic factors. These same independent variables could account for nearly all of the ethnic differences in full-time employment rates for females. In particular, the authors found that poor labour market outcomes of Maori are significantly linked to their relatively poorer educational qualifications. They also found that the marginal effects of formal qualifications on the probability of being in full-time employment (ie, this particular ‘rate of return’ on education) were consistently larger for Maori compared to non-Maori.

Chapple and Rea (1998) used aggregate data from the Household Labour Force Survey (HLFS) to examine trends in labour market disparities between Maori and non-Maori between 1985 and 1998. Several indices of relative disparity were used in conjunction with various labour force states. The authors concluded that little additional insight is gained from more complex indices relative to simple aggregate employment propensities for the ethnic groups. They found that Maori labour market status deteriorated relative to that of non-Maori between 1985 and 1998. However, all of this worsening in the relative position of Maori occurred between 1985 and 1992. The relative labour market performance of Maori actually improved between 1992 and 1996. The authors made no attempt, unlike Winkelmann and Winkelmann, to estimate the extent to which these ethnic gaps in employment propensities and unemployment rates could be ‘explained’ by differences in demographic characteristics and regional location. No regression analysis was included in their study.

Alexander, Gene and Jaforullah (2001) were able access unit record data in 1997, 1998 and 1999 from the Income Survey (a supplement to the HLFS in the June quarters starting in 1997). They distinguished between individuals who reported Maori as their only ethnicity (referred to as ‘sole Maori’ in the present study), and those who reported Maori as only one of their ethnicities (referred to as ‘mixed Maori’ in this report). The authors note in their descriptive statistics that in terms of usual hourly earnings “… Maori males in most qualification groups actually outperform European males” (p.8). Although they make no attempt to summarise the overall ethnic difference in wages across the various qualification groups and years, these simple statistics suggest that Maori might have higher wages than non-Maori if only education is held constant.

The authors find in their regression analysis that once age, household type, region, gender, qualifications and occupation are held constant, there is no consistent evidence across the three years that Maori workers receive lower wages than European workers. The average estimated coefficient on the sole Maori indicator variable in these hourly earnings regression is –0.0353. This implies that sole Maori workers, holding other measured factors constant, receive wages that average 3.5% less than their non-Maori counterparts. Yet, these estimated coefficients are only clearly statistically different from zero in one of the three years (1997). The average estimated coefficient on the mixed Maori indicator variable is –0.0131. This implies that mixed Maori workers, holding other measured factors constant, receive wages that average 1.3% less than non-Maori counterparts. These estimated coefficients are never statistically different from zero in any of the three years.

Alexander et al. make a valid and potentially important observation that this finding does not imply that observationally equivalent Maori and non-Maori face the same wages in the labour market. Sample selection bias may be an issue here. Unobservable determinants of wages may be correlated with unobservable determinants of employment propensities. As a result, Maori who are working may receive much higher wages than those faced by Maori who are not working. The same may be less true for non-Maori. Only by correcting for this sample selection process can we determine whether or not observationally equivalent Maori and non-Maori face systematically different wages in the labour market.

The results reported by Alexander et al. suggest that sample selection bias is a crucial factor in comparing the wages facing Maori and non-Maori. Their wage regressions, corrected for this sample selection bias, show negative and significant effects of being Sole or Mixed Maori. The average estimated coefficient on sole Maori is –0.121, and significantly different from zero in all three regressions at better than a 1% level. This implies that sole Maori, holding other measured factors constant, face wages that average 11.4% less than non-Maori (an ethnic gap more than three and one-quarter times larger than the one previously estimated without the inverse Mills ratio). The average estimated coefficient on mixed Maori is –0.045, and significantly different from zero in two of the three regressions at better than a 1% level. This implies that mixed Maori, holding other measured factors constant, face wages that average 4.4% less than non-Maori (an ethnic gap, again, more than three and one-quarter times larger than the one previously estimated without the inverse Mills ratio).

Little weight should be placed on the regression results of Alexander et al. that correct for sample selection bias. It is well established in the sample selection literature that ‘pseudo instruments’ are needed to appropriately ‘identify’ the inverse Mills ratio (or ‘lambda term’) in regressions that correct for a sample selection process. In this case, at least one variable must be included in the probit equation on the employment propensity that is subsequently excluded from the wage equation. Even though it is technically possible to avoid perfect collinearity without this pseudo instrument, because nonlinearities will ‘identify’ this lambda term, most authors refuse to use Heckman’s technique (1979) without valid instruments (eg, see the discussion on this topic in recent articles by Fitzgerald, Gottschalk and Moffitt (1998a and 1998b). Yet, this is exactly what Alexander et al. do.[2] This is particularly problematic when the functional form of the regression (log-linear in this situation) is chosen only for convenience, and not due to an underlying theoretical model.

At the very least, the authors need to explore why their estimated effects of ethnicity on wages vary so dramatically between the ‘uncorrected’ and ‘corrected’ wage regressions. This is particularly important in this situation where earlier descriptive statistics suggest that male Maori workers receive higher hourly earnings than male non-Maori workers with the same qualifications. The sample selection issue is a potentially important factor in this study, but the authors provide little convincing evidence that this issue has been handled in an appropriate manner.