Since the focus of this analysis is on differences in access to HD between Maori and European offenders, the present analysis has been based on the reduced sample of 6774 offenders, excluding approximately 830 offenders of other ethnicities.

The analysis was conducted in three stages. In the first stage the associations between ethnicity and the measures of access to HD were tested for statistical significance using the chi squared test of independence. In each case a measure of the strength of association was based on the relative risk (RR) of access to HD amongst European offenders compared to Maori offenders.

In the second stage of the analysis a series of logistic regression models was fitted to the data to adjust the observed associations between ethnicity and access to HD for confounding by other factors that were also related to the likelihood of access to HD (personal characteristics, offence severity, offending history, etc). The general adjustment model fitted for each outcome was of the form:

Logit (Y) = B0 + B1 Ethnicity + ∑BjZj

where Y was the measure of access to HD (at sentencing or at the parole hearing stage) and logit(Y) was the log odds of access to HD; Ethnicity was a dichotomous measure reflecting whether the offender was Maori or European; and Zj were the set of covariate factors. The analysis of granting HD at the parole hearing stage was based on the full sample of offenders rather than the reduced sample of those who had been granted leave to apply for HD at sentencing, but with leave to apply at sentencing included as an additional predictor in the model. This was done to avoid issues of selection bias resulting from the processes by which leave to apply for HD at sentencing was given.

To aid in the interpretation of the regression models and to reduce the number of variables in the analysis a number of factors in the database were combined to produce composite score measures reflecting different aspects of the offender’s background. Composite scores were created by first grouping the covariate factors into conceptually similar domains; a series of regression models was then fitted in which each outcome was regressed separately on the variables in each domain; finally composite regression scores were then constructed from linear composites of the variables in each domain weighted by the corresponding regression coefficients from the fitted models. Composite scores were created for each of two domains: (a) Offence seriousness: this was a weighted combination of measures reflecting most serious offence type, sentence seriousness and sentence length. (b) Offending history: this was a weighted combination of measures reflecting the number of previous custodial sentences, the history of sexual and violent offending, gang membership, previous breach of sentence convictions, and previous sentences involving community work or supervision. It should be noted that while the composite score approach was adopted to aid simplicity of presentation, a range of alternative adjustment procedures all produced essentially the same conclusions to those reported here.

From the fitted regression models estimates of the adjusted rates of access to HD for each outcome were calculated using the method described by Lee (1981). The adjusted rates provide an indication of the size of any remaining ethnic differences in access to HD after adjustment for covariates, and can be interpreted as the hypothetical rates of access to HD that would have been observed had the distribution of covariates factors been the same for Maori and European offenders. The strength of the adjusted association was further summarized by the adjusted relative risk of access to HD for European compared to Maori offenders, calculated from the adjusted rates.

Finally, to examine whether the same general explanatory model was applicable for Maori and Europeans, the regression models above were extended to fit a series of nested models to test whether the regression coefficients for each covariate factor were the same for Maori and European offenders. If the regression models for Maori and European offenders were noticeably different then this would suggest that the processes influencing access to HD in each group might also be different, which in turn would make it more difficult to examine the issue of bias in the HD process.