5   Decompositions

The analyses in the previous sections have established the role of hours of work, occupation, industry and locality in explaining the positive relationship between educational qualifications and income, and thereby the income gap across ethnic groups. An important question for strategies toward reducing the income gap is the extent to which lower educational attainment, or alternatively, potential difficulties for Maori in translating qualifications to income returns can explain the income gap and its increase between 1986 and 1996. This question is addressed by using the estimates of equation 1 (specification 2 in Tables 2a-2d) to decompose the income gap into a component due to differences in the characteristics (such as qualifications) of ethnic groups, and a component due to ethnic differences in the way that these characteristics are translated into income (Oaxaca, 1973). Comparisons are with the European group:

(2)      

where ‘E’ denotes European, M Maori, (and subscript ‘P’ may be substituted for M for Part-Maori, and ‘O’ for the ‘Other’ ethnic group depending on the comparison), k the number of explanatory variables, regression coefficients, mean characteristics,

is the differential in log incomes due to the difference in intercepts, is the differential explained by differences in mean personal characteristics, and

is the income differential explained by coefficients.

The above formulation uses the coefficients from the ‘European’ equations as the weight for the effect of differences in the mean ‘characteristic’ of each ethnic group relative to the European group. This has the advantage of using the same set of base coefficients for the decompositions across ethnic groups. However, the choice of the base group for coefficients and mean characteristics can affect the results. In particular, when coefficients (or returns to characteristics) are greater for the European sample, this specification tends to estimate slightly larger ‘characteristic’ effects than when using alternative weights, in which the base groups are reversed. Results from these alternative specifications are compared in Appendix D. The weighting choice does not change any of the conclusions of the Maori, non-Maori decomposition analyses[12].

A second factor relating to the interpretation of the decomposition results when sets of binary variables are present (e.g. for education, occupation, locality, and industry) is that the intercept effects are not invariant to the choice of each base category. However, the sum of the coefficient effects (the sum of the intercept and other variable coefficient effects) is invariant (Oaxaca and Ransom, 1999). Therefore, in Table 3, the sum of ‘the coefficient effects’ is provided for comparison to the effect of ‘characteristics’.

Table 3 – Decompositions of log income differentials - employed males and females

Log income differential explained by:	1986			1996
Males
Overall difference	0.25064	0.16121	0.22514	0.35388	0.23317	0.31301
Intercept differential	0.00026	0.09320	0.09003	0.09290	0.15122	0.16687
Effect of characteristics	0.18199	0.14181	0.04331	0.23184	0.18064	0.04812
	(72.6%)	(88.0%)	(19.2%)	(65.5%)	(77.5%)	(15.4%)
Effect of coefficients (intercept+ coefficients)	0.06865	0.0194	0.18183	0.12204	0.05253	0.26489
( b0E – b0M/P/O )+	(27.4%)	(12.0%)	(80.8%)	(34.5%)	(22.5%)	(84.6%)
Females
Overall difference	0.09795	0.04198	0.00642	0.15531	0.10585	0.08972
Intercept differential	-0.04226	0.23776	-0.15329	0.08485	0.18352	0.20747
Effect of characteristics	0.09007	0.02963	-0.00510	0.11610	0.08367	0.02149
	(92.0%)	(70.6%)	(-79.5%)	(74.7%)	(79.0%)	(24.0%)
Effect of coefficients (Intercept+ coefficients)	0.00788	0.01235	0.01152	0.03921	0.02218	0.06823

Oaxaca Method, Based on the specification 2 in Tables 2a-2d. Percentage contributions to the overall difference in the log income differential are in parentheses.

Table 3 shows both the mean income differentials by ethnicity in 1986 and 1996, and the decomposition of these differentials into a component due to ethnic differences in characteristics (‘Effect of characteristics’), and a component due to ethnic differences in translating these characteristics into income (‘Effect of coefficients’). The first row for both males and females shows the generally smaller gap for the Part-Maori group, and the increase in the gap for all three ethnic groups between 1986 and 1996.

First, a major result of the decomposition analyses is that the majority proportion of the Maori-European income differential for both males and females (65.5% of the differential for males and 74.7% of the differential for females in 1996) can be explained by their higher educational qualifications and differences in other control variables. These results support the hypothesis that, among those employed, differences in education, occupation, hours of work and other work related characteristics make a stronger contribution to differences in income between Maori and European, than do differences in translating similar characteristics into income.

Table 3 also shows that intercept effects made a greater contribution in 1996 compared to 1986 in explaining Maori relative income levels for both males and females. While these intercept effects are specific to the choice of the base categories in the model (e.g. in this case, ‘no school qualifications’, clerical occupation, ‘semi-urban’ locality, single, etc.), the comparisons across ethnic groups are relevant and they highlight the earlier finding that by 1996 those in the base category of no school qualifications were relatively more disadvantaged in the labour market, and that the effect was greater for Maori males in the base category of ‘no school qualifications’.

In addition, Table 3 shows that, for both male and female Part-Maori, income returns to characteristics were higher than for Maori or Other ethnic groups, and closer to those for the European group. Conversely, Other ethnic groups had the lowest income return to characteristics. This is likely to reflect the effect of language barriers and other factors negatively influencing the labour market outcomes of immigrants in this group.

The estimations and decompositions have highlighted the importance of differences in educational attainment in explaining why relative Maori income levels have deteriorated over the period 1986 to 1996. While Maori became more qualified over the period, so did other groups. This, combined with higher returns to educational qualifications meant that Maori had relatively lower incomes in 1996, compared to 1986. The analysis shows that educational qualifications have been exerting their influence on ethnic differences in income over the decade partly through differences in occupation and hours of works. Other factors such as locality have also influenced differentials over time. For example, while European males in major urban areas had significantly higher income levels than those residing in rural or semi-urban areas, Maori males did not face advantages in major cities, but had higher income returns in semi-urban areas. Finally, decompositions have shown that differences in group characteristics rather than returns to characteristics can explain the major part of the ethnic income differentials.