3  An analytical framework

The purpose of this study is to test the hypothesis that observationally equivalent Maori experience significantly different early labour market histories when compared to their non-Maori counterparts. The first issue that arises is how these ‘early labour market histories’ will be quantified. There are obviously many dimensions along which these employment experiences can be measured. Two variables are chosen for this analysis: the actual number of years of work experience between the 16^th and 21^st birthdays, and the hourly earnings of those working at age 21. Both are key elements of the early labour market transition period.

It is well known that Maori have fewer formal school and tertiary qualifications than non-Maori (eg, see Winkelmann and Winkelmann (1997) and Chapple and Rea (1998)). Relatively little is known about ethnic differences in human capital accumulation through the initial work experience of youth. Yet, these early experiences may be critical to the subsequent labour market success or failure of these individuals. It is also an area where discrimination may arise. Although it may be difficult for employers to discriminate on the basis of pay and promotion, unequal access to employment may be more difficult to detect and prosecute.

To gain a more comprehensive picture of any ethnic inequalities that might be present in the labour market, it is important to canvass a wide array of outcomes. Once we have investigated the possibility of systematically lower rates of work experience among Maori, we can then ask whether hourly earnings vary between Maori and non-Maori with the same qualifications, work experience and other relevant factors. We might find that ethnic differences exist in the accumulation of work experience, and not in hourly earnings. Or we might find that there are no ethnic differences in terms of acquiring work experience. Yet, conditional on this work experience and other factors, Maori might face lower wage rates than non-Maori. The key is that these issues need to be explored simultaneously. Accurate measures of work experience histories are needed to appropriately estimate a standard wage regression, and account for all individual differences in human capital.

3.1  Estimating the determinants of work experience

Consider the following specification for the work experience regression.

(1)     EXP_i= βPOTEXP_i + δMAORI_i + X′_iγ + u_i

The dependent variable EXP_i is a measure of the actual years of work experience accumulated by the i^th individual between his or her 16^th and 21^st birthdays. One of key independent variables in this equation is a measure of the years of ‘potential’ work experience POPEXP_i. This is the amount time that the individual was not enrolled in education or training programmes over this five-year interval. If the coefficient β on this regressor is equal to one, then every year away from human capital accumulation through education and training leads to an additional year of ‘on-the-job’ human capital accumulation through work experience. It is expected that β will be positive, but less than one. Potential work experience generally ‘facilitates’ the accumulation of actual work experience. On average, only a portion of each year of potential work experience is converted into a year of actual work experience. Because of the ways in which actual and potential years of work experience are constructed, it is at least possible for β>1. The discussion in the following few paragraphs describes how these variables are constructed from data available in the CHDS. The key is that there is no ‘mechanical’ relationship between potential and actual work experience, and the latter can exceed the former.

Retrospective data on the activities of youth are available from the interviews at ages 18 and 21. At age 18 we know the job tenure (in months) for youth employed at the time of the survey. We also know their weekly hours worked at age 18, and the number of other jobs held between their 16^th and 18^th birthdays. At age 21 we know the work status of youth during the three-month intervals between their 18^th and 21^st birthdays. We know whether or not they were working full-time (30 or more hours per week) or part-time (fewer than 30 hours per week) during each of these 12 quarters. All of this information on work histories is independent of the enrolment in school or tertiary education of these youth over this period.

These data were used to construct a measure of the actual work experience of youth over the five-year period between their 16^th and 21^st birthdays. Each individual was assigned three months of work experience if he or she reported working full-time during a quarter, and one and one-half months work experience if he or she reported working part-time between the 18^th and 21^st birthdays. These amounts were aggregated over the twelve quarters. Each individual was assigned two years of work experience if he or she reported working full-time at age 18 in a job held for at least two years. One year of work experience was assigned if the youth was working part-time at age 18 in a job held for at least two years. These measures of full-time or part-time work experience between the 16^th and 18^th birthdays were prorated for individuals with job tenure of less than 2 years at age 18. In addition, these youth were assumed to have received two months of work experience for every additional job held between the ages of 16 and 18. Individuals were not allowed to accumulate more than two years of actual work experience between their 16^th and 18^th birthdays.

Although arbitrary decisions were made on how these data would be used to construct an overall measure of actual work experience between the 16^th and 21^st birthdays, these retrospective data at these two interviews should provide a fairly accurate picture of these work histories. This measure of actual work experience can range from a minimum of zero to a maximum of five years.

‘Potential’ work experience was slightly easier to construct. This is the amount of time between the 16^th and 21^st birthdays that youth were not enrolled in either education or training. This variable was also constructed from retrospective data taken from the interviews at ages 18 and 21. The CHDS contains an estimate of the age when the individual left school if this event occurred before the 18^th birthday. It also contains quarterly summaries of their activities between their 18^th and 21^st birthdays, which indicate whether or not they were enrolled in school or tertiary education (full-time or part-time), and whether or not they were “… attending (an) educational training course (not at University or Polytechnic)” (Question B.9, CHDS 21-Year Interview). Each individual was assigned three months of potential work experience if he or she reported no education or training, and one and one-half months of potential work experience if he or she reported part-time education or training during a quarter. Youth were assigned zero months of potential work experience if they were involved in full-time education or training during a quarter.

The idea is that potential work experience ‘facilitates’ the accumulation of actual work experience between the 16^th and 21^st birthdays. Yet, actual work experience isn’t necessarily limited by our measure of potential work experiences (ie, time spent in education or training and time spent in work are not mutually exclusive). Actual work experience can exceed potential work experience if individuals are working while studying or training. We expect, however, that actual work experience will generally be less than potential work experience for youth because of time spent in unemployment or being out of the labour force.

Once relevant factors (including potential work experience) are held constant, our hypothesis is that Maori will accumulate less work experience than non-Maori by age 21. The coefficient on the variable indicating Maori ethnicity is hypothesised to be negative (δ<0). One of the key issues in this study is the way in which ethnicity might be measured. This issue is explored in detail in Section 4.

A vector of other independent variables X_i is included in equation (1). Alternative compositions of this vector are used. Under the ‘short regression’, this vector includes an indicator variable for gender, a series of indicator variables for the formal school, tertiary and vocational qualifications obtained, and gender interacted with the number of children born to the respondent by age 21.[3] These are typical covariates that would generally be available through other data sources (eg, the Population Census or the Household Labour Force Survey).

Under the ‘long regression’, this vector also includes personal and family background characteristics that generally are not available through other data sources. The following section provides a detailed discussion of these additional independent variables. The reason for these alternative definitions of the X_i vector is quite straightforward. If systematic differences between Maori and non-Maori are evident in the short regression specification, it is at least possible that these ethnic differences might be the result of specification error. In other words, relevant measures of family background factors that influence the accumulation of work experience by youth were erroneously excluded from this original specification. Indications of unequal outcomes between Maori and non-Maori in the accumulation of work experience found in other data sets might be related to the unavailability of detailed information on individual family backgrounds. In other words, the type of data available in the CHDS may be critical if we are to claim ‘observational equivalence’, and to attribute systematic differences to ethnicity alone.[4]

An alternative approach to the inclusion of ethnicity as an indicator variable is to separately estimate the work experience regressions for Maori and non-Maori.

(2)    EXP_i= β_MPOTEXP_i + X¢_iγ_M + u_i

EXP_i= β_NMPOTEXP_i + X¢_iγ_NM + v_i

By essentially interacting all of the covariates with ethnicity, we can determine the extent to which the incremental effects of the independent variables might vary between the ethnic groups. For example, the effects of potential work experience on actual work experience may be quite different between Maori and non-Maori (ie, β_M ≠ β_NM).

Oaxaca (1973) used separate regressions from two subsamples like these to decompose overall mean differences in wage rates into components that could and could not be explained by mean differences in observable factors between the groups. The problem that Oaxaca faced is that the coefficients from either regression model in (2) could be used to ‘weight’ these relative differences in productivity characteristics. Although the coefficients from the ‘primary group’ (eg, males in a gender decomposition or non-Maori in an ethnic decomposition) might be selected because these individuals are not expected to suffer from discrimination in the labour market, there is argument for using a weighted average of these estimated coefficients.

Neumark (1988) suggests that the appropriate coefficients to be used in this decomposition should be based on the counterfactual of an absence of labour market discrimination. In this case, this entails an experience regression dropping ethnicity as an explanatory variable. We adopt this same approach in this study.

We begin by taking the means (represented in bold type) of the sample regression functions in equation (2):

(3)     EXP_M = b_MPOTEXP_M + X′_Mc_M

EXP_NM = b_NMPOTEXP_NM + X′_NMc_NM

where Greek letters are replaced by their English equivalents to indicate that these are estimated parameters. The means of the residuals are equal to zero in both cases through Ordinary Least-Squares estimation.

Now introduce a set of estimated coefficients without subscripts, indicating that they come from a pooled regression. Adding and subtracting cross-products and performing some algebraic manipulation, we get the following expression.

(4)    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 terms in the first set of square brackets on the right-hand side of this equation represent the part of the observed gap in average work experience between non-Maori and Maori that can be explained by differences in potential work experience and other independent variables. The terms in the second set of square brackets represent the part of the work experience gap that cannot be explained by differences in measured characteristics.