PDF file of entire paper: Print-quality (358K)

K07017; Online publication date 30 May 2008
Received 7 August 2006; accepted 19 March 2008
Kōtuitui: New Zealand Journal of Social Sciences Online, 2007, Vol. 3: 1–13
1177–083X/08/0301–0001 © The Royal Society of New Zealand 2008

Kōtuitui

New Zealand Journal of Social Sciences Online

---

Explaining Maori under-achievement in standardised reading tests: the role of social and individual characteristics

Kelly J. Lock

John K. Gibson^*

Motu Economic and Public Policy Research
PO Box 24390
Wellington 6142, New Zealand

^*Waikato Management School
The University of Waikato
Private Bag 3105
Hamilton 3240, New Zealand

Abstract  In New Zealand, Maori students have lower educational attainment levels than their Pakeha counterparts, which is a major contributor to the earnings inequality between these two groups. This study attempts to identify determinants of the gap in educational attainment using data on reading test scores for a sample of 3031 15-year-old Maori and Pakeha students. Using a decomposition method that is commonly applied to income inequality analyses, the determinants of the reading literacy test score gap between Maori and Pakeha students are identified. The main contributors to this gap are family factors, student’s opinions, and school factors (including decile). An understanding of the determinants of the test score gap provides insights into how effective policy makers can be at reducing this inequality.

Keywords  education; Maori; student achievement; test scores

INTRODUCTION

In New Zealand, Maori have lower incomes than New Zealand European (Pakeha) and experience less success in the labour market (Maani 2004). While theory suggests that this difference is likely to be explained by differing characteristics between the two groups and/or discrimination, most empirical analysis suggests that these differences reflect the lower accumulation of favourable labour market characteristics by Maori (Gibson 1998). About three-quarters of the income gap that can be explained by observable characteristics is due to the lower educational attainment of Maori (Maani 2004). Hence, questions about Maori-Pakeha income gaps that underpin many social policy debates can be largely recast as questions about differences in educational attainment.

  Moreover, beyond their role in explaining income gaps, educational gaps are important in their own right due to the benefits associated with education for both the individual and society. These benefits have been extensively reviewed in Haveman & Wolfe (1984) and Wolfe & Zuvekas (1997), as well as in the New Zealand context by Johnston (2004). For example, higher levels of education enhance an individual’s ability to cope with changes in the economy. So as economies become more technologically focused, individuals with low education levels will become more disadvantaged (Chan et al. 2003). Higher education is also associated with higher life expectancy as educated individuals are less likely to undertake unhealthy activities such as smoking (Kenkel 1991).

  Population sub-groups often have different levels of educational attainment, which can lead to different standards of living due to the lifetime benefits associated with education. A number of studies have found different levels of educational achievement and attainment between ethnic groups around the world. These studies have been especially prolific in the United States, comparing black, white, and Hispanic students, where black and Hispanic students are consistently found to have lower educational performance (e.g., Hedges & Nowell 1999; Perna 2000; Cameron & Heckman 2001; Orr 2003).

  Similar differences occur in New Zealand regarding educational participation and achievement. Maori, on average, spend less time in the education system and do less well while they are there than Pakeha (Else 1997). Maori have lower participation rates, both in early childhood education centres (Ministry of Education 2005a) and in upper secondary school once they have reached the minimum leaving age (Maré 1995). While at school, Maori students are also failing to achieve the same level of qualification success as their Pakeha counterparts in both secondary (Ministry of Education 2005a) and tertiary education (Ministry of Education 2005b). The size of the gap in educational attainment for Maori and Pakeha children has fluctuated over the past 25 years but, despite the gap falling (Else 1997), it has not closed completely (Ministry of Education 2005a).

  Internationally, a number of different approaches have been used to assess differential academic achievements by population sub-groups. These include IQ tests (e.g., McKay et al. 2003), drop-out rates (e.g., Aloise-Young & Chavez 2002), college enrolment (e.g., Perna 2000), and test scores (e.g., Orr 2003; Bali & Alvarez 2004). In addition to looking at the size of the difference between groups, studies have also investigated how the gaps have changed over time (e.g., Hedges & Nowell 1999) and factors that contribute to these differences (e.g., Brooks-Gunn et al. 1996).

  Although potentially more difficult to acquire, differences in test scores are likely to provide a more precise measure of an individual’s educational achievement than many of the other measures available. For example, differential drop-out rates may reflect more binding financial constraints for some groups rather than performance in school per se , while test scores allow the individual’s academic ability to be directly targeted. Test scores have been used in a number of studies carried out internationally which have consistently found ethnic differences (Bali & Alvarez 2004).

  However, while there is substantial evidence that Maori are underperforming in the New Zealand education system, there is little evidence based on standardised test scores. Maani & Kalb (2007) examined the effect of family income, student ability, and school peer effects on test scores. They found that when family economic factors, childhood IQ, and peer factors are controlled for, the educational performance gap between Maori and Pakeha disappears. Apart from the study by Maani and Kalb, the only other work that uses test scores to assess Maori-Pakeha educational differences has been preliminary work using the PISA dataset (Sturrock & May 2002), which is an international standardised test that is described below.^[1] This work found a significant test score difference between students from the two ethnic groups, with Pakeha students, on average, performing above the OECD average while Maori students were, on average, below OECD average. The reasons for this difference were not examined.

  The current study uses test scores to identify educational differences between Maori and Pakeha students using regression analysis. The factors that contribute to this difference are then assessed. This assessment allows the raw test score gap to be decomposed into two parts: that which is due to difference in characteristics (e.g., different levels of family wealth) and a remainder that is unexplained except that it must (by assumption) reflect different returns to the characteristics. The importance of the various observable characteristics in explaining test scores is then discussed in terms of public policy, with a special focus on the feasibility of public interventions to close the gap in educational achievement.

PISA 2000 DATASET

The Programme for International Student Assessment (PISA) was commissioned by the Organisation for Economic Co-operation and Development (OECD) to provide a dataset that allows internationally comparable measures of student performance near the completion of secondary education (Adams & Wu 2002).^[2] The study assesses the ability of students to apply their knowledge of reading, mathematics and science to real world situations, referred to as the individual’s reading, mathematical and scientific literacy. Subsequent to its first cycle in 2000, the PISA data have been collected every 3 years using a new cohort of students and focusing on a different subject area, starting with reading literacy in 2000. Since the data used in this study come from the 2000 PISA cycle, we focus on reading literacy (defined as “the ability to understand, use and reflect on written tests to achieve one’s goals, to develop one’s knowledge and potential, and to participate effectively in society” (Sturrock & May 2002). As an individual’s ability to comprehend written language is a major determinant of future success, this provides a valuable, policy relevant measure and allows investigation of one of the most important components of the school curriculum (Bedard & Ferrall 2003).

  In addition to the standardised test, all individuals completed a questionnaire covering a wide variety of topics including details on their family, socio-economic status, and attitudes. The principals of the schools where participating students were enrolled also completed a questionnaire, which included questions on conditions in and features of their school. Summary statistics were also created within the dataset from information contained within the student’s and principal’s questionnaires, and test scores were standardised internationally to a mean of 500 and standard deviation of 100. From the information contained in the questionnaires, summary statistics, and the test scores, the PISA study provides a comprehensive dataset for assessing the determinants of educational differences.^[3]

  In New Zealand, nearly 3700 secondary students from over 150 schools took part (Ministry of Education 2002). These were a good representation of the total population (Sturrock & May 2002). In the questionnaire, students were able to identify multiple ethnicities but were assigned a single ethnicity based on the Statistics New Zealand prioritised allocation of multiple ethnic affiliations into a single ethnic group. This meant that for this analysis, individuals were deemed Maori if they selected Maori as any of their ethnic groups and New Zealand European (Pakeha) if they selected Pakeha and no other ethnic group. For this study, only Maori and Pakeha individuals are included, generating a sample of just over 3000 individuals.

METHOD

We use two related methods to identify the factors that are associated with the difference in average reading literacy test scores between the Maori and Pakeha students. First, regression analysis is used with a binary indicator variable for ethnicity included along with a wide range of variables measuring other characteristics and the attitudes of students. This indicator variable will capture the effects of ethnicity on the test scores after controlling for the other characteristics and attitudes.

  Although regression with an indicator variable is a popular method of looking for differences between groups after controlling for other factors, it does have some limitations. Importantly, it constrains all the other variables, except the ethnic group identifier, to have the same effect on both Maori and Pakeha students. For example, the effect of school decile is forced to have the same effect on test scores for Maori and Pakeha students, with the indicator variable for ethnicity forced to capture the differences in test scores for Maori and Pakeha students who attend schools in the same decile level (and have the same values for the other characteristics). This conditional comparison may be too restrictive. First, it is possible that the effect of decile (or other characteristic) on test scores differs between Maori and Pakeha students. Second, part of the cause of the test score difference may be due to differences in decile (or other characteristic) but this difference is disguised in a conditional comparison which compares statistically identical individuals from different ethnic groups. Therefore, a regression decomposition is used to see how much of the difference in test scores could be attributed to differences between Maori and Pakeha students in their observed characteristics versus differences in the returns to these characteristics.^[4] This involves running separate regressions for Maori and Pakeha students and hence provides a more flexible analysis by allowing the coefficients on all variables to differ for Maori and Pakeha students. Consequently, we can identify the portion of the difference in test scores due to differences in observable characteristics and the portion that is unexplainable (Borjas 1996).

  Although the basic idea of a regression decomposition is straightforward, there are differences amongst various methods according to the parameter vector used to weight the difference in average characteristics between Pakeha and Maori students, ( ). The approach used here is the most general, of using a parameter vector β* from a regression on both samples pooled (Neumark 1988). The gap between the reading literacy test scores for Pakeha students, L ^P and Maori students, L ^M at the mean can then be expressed as:

  (1)

  The last term in Equation (1) reflects the part of the gap in reading test scores explained by differences in the average characteristics of the two groups of students. The first two terms reflect unexplained differences due to unequal coefficient vectors estimated on the Pakeha sample, β^P and the Maori sample, β^M. These unequal coefficient vectors imply different rates at which student characteristics are translated into test scores.^[5]

RESULTS

Unadjusted test score gap

There is a large, statistically significant difference between the average reading literacy test scores of Maori and Pakeha students (Sturrock & May 2002). Pakeha students, on average, are performing well above the OECD average of 500. Specifically, they achieved a test score average of 554 points, which is one-tenth of a standard deviation higher than the highest country average in the PISA 2000 cycle.^[6] However, in contrast, Maori students, on average, are performing under the OECD average at a level approximately equivalent to Hungary.^[7]

Regression model

Results from the regression analysis with a binary indicator variable for ethnicity are presented in the first column of Table 1. (Definitions of all of the variables used in this study, along with average values for Maori and Pakeha students, can be found in Appendix.) The regression specification is able to explain 42% of the variation in test scores (with an F statistic of 44.83 ( P = 0.0000)).^[8]

  After controlling for all of the observable factors included in the regression model, the test score difference between Maori and Pakeha drops from 72 to 18 test score points. Despite controlling for a number of other variables, the effect of ethnicity remains statistically significant. Thus, even if a Maori student had the same observable characteristics as a Pakeha student for each of these 23 variables, their test score would still be 18 test score points (or approximately 0.2 standard deviations) lower, on average, than a statistically identical Pakeha student.

  Some of the variables in the regression specification may be collinear and thus the interpretation of coefficients on individual variables is difficult. To investigate the effect of multicollinearity, the effect of excluding different variables, such as the socio-economic variables, was examined. When these variables were removed from the regression analysis, the change in the results was minimal, other than a fall in the explanatory power of the model. There may also be reverse causality between the enjoyment of reading variable and the student’s reading literacy test score. Once again, removing this variable from the analysis did not lead to significant changes in the results.

  To ensure that multicollinearity is not a problem, we used the Belsey et al. (1980) detection statistic. This statistic is based on the three factors that contribute to imprecise coefficient estimates: a large error variance and small eigenvalues combined with a large fraction of the variance of two or more variables that is associated with a particularly small eigenvalue. Specifically, if the X matrix from the regression model y = X b + m is scaled to unit length, Belsey et al. (1980) suggest that collinearity problems may occur when the “condition index”, which is the square root of the ratio of the largest eigenvalue to the i th eigenvalue, exceeds 30. In addition, for coefficient estimates to be degraded, two or more coefficients must have more than half of their variance associated with a “small” eigenvalue (i.e., one where the condition index exceeds 30). Our regression model has a largest “condition index” of 28 and therefore we do not need to invoke the second condition and instead can proceed with the estimation under the assumption that coefficient estimates are not degraded by multicollinearity.

  The regression coefficients suggest some interesting results, many of which support previous work. For example, male students had a lower test score than female students with otherwise identical characteristics, as has been found in a number of studies including Sammons (1995) and in the New Zealand context by Fergusson & Horwood (1997) and Maani & Kalb (2007). Other variables though were more surprising, such as the negative coefficient on the proportion of teachers employed at the student’s school with a Bachelor of Teaching. Some of the most important variables for explaining a student’s test score are the language spoken at home, the student’s school year, and the proportion of English teachers at the school with a degree in teaching.

  To further aid the ease of understanding the findings given the number of variables included, related explanatory variables were grouped (as shown in the bold titles in Table 1). For example, the number of students in English classes is likely to be related to the number of students in the school and the number of students in the Maths class. Thus, these variables were all placed in the same group, “School Factors”, along with other potentially related variables. The aggregated, standardised coefficients for each of the groups can be seen in Table 2. Higher absolute coefficients indicate a greater impact on test scores.

  The student’s characteristics and family factors are the most significant in determining the student’s test score. It is unsurprising that student characteristics are the most important variables for determining a student’s tests score as this group includes variables often linked to under-achievement—being male, being Maori, and being in a lower grade at school. Socio-economic factors are also often associated with educational under-achievement and these are found in the group with the second largest contribution to student’s test scores.

Decomposition

While the regression analysis identifies variables that are able to contribute to test scores for the population as a whole, it does not provide insight into what is attributing to the test score difference between the two groups. It fails to do so because its counterfactual is for a statistically equivalent Pakeha student. If the reading literacy test score difference between Maori and Pakeha students was completely explainable by differences in the level of these characteristics, then we would expect the binary ethnicity variable to become insignificant in the regression, as was found by Maani & Kalb (2007). But, in the current study, this variable remains highly significant ( P = 0.000). Thus, there may be differences in the effect that each of the characteristics has on test scores for Maori and Pakeha students. Through the decomposition analysis, the test score gap is able to be examined more closely to determine what factors are contributing to the difference in test scores between Maori and Pakeha.

  The results from the separate regressions for Maori and Pakeha subsamples carried out as part of the decomposition analysis can be seen in the second and third columns of Table 1. The decomposition analysis identified the proportion of the test score gap that is able to be explained by each of the variables in the model and was able to explain 67.0% of the test score gap between the two groups.

  By using the decomposition analysis, we are able to control for the differences in characteristics between the Maori and Pakeha students. For example, in the sample, 40.5% of Pakeha students are in a high decile school compared with only 17.5% of Maori students. In the initial regression analysis, the effect of this difference will be disguised. However, based on the decomposition analysis, it is possible to see that the total effect of differences in school decile contributes to 11.3% of the total test score gap between Maori and Pakeha students. Overall, the student’s enjoyment of reading, school decile, and perceived socio-economic index contributed the most (12.3, 11.3 and 11.1% of the test score gap, respectively) to the test score gap between these two groups.^[9]

  The effects of each of the variables in the decomposition were combined as previously into broad groups. The decomposition results by group are presented in Table 3. Family factors were able to explain the largest proportion of the test score gap followed by student’s opinions and school factors.

  There is also a large amount of the gap that was not able to be explained despite the large number of variables that were available in this dataset. This unexplained fraction indicates that there are unobservable variables that are not able to be included in the analysis and/or that the ethnic groups had different rates of return to characteristics (i.e., different β’s). For example, it is apparent from Table 1 that the penalty for being male (in terms of reading test scores) is larger for Maori, with a b-coefficient of –24.5 compared to only –16.5 for Pakeha. On the other hand, Maori students appear to benefit more from attending a high decile school, where reading scores are 42.1 points higher, holding other factors constant, while for Pakeha students the gain is only 9.8 points (and is statistically insignificant).

DISCUSSION

The results of this study clearly show that Maori students, on average, are doing less well in the education system than their Pakeha counterparts. Significant differences between Maori and Pakeha students in their PISA 2000 reading literacy test scores were identified in this study, even when the effect of other variables such as socio-economic status and school factors are taken into account. This is in contrast to previous New Zealand studies where the effect of ethnicity disappears once other variables are introduced to control for socio-economic background (e.g., Fergusson et al. 1991; Maani & Kalb 2007).

  In the New Zealand studies cited above, variables containing retrospective information were found to be significant in explaining academic performance and ethnicity was no longer significant once other variable were controlled for. For example, early childhood income was found to be a highly significant determinant of secondary school achievement by Maani & Kalb (2007). The PISA dataset used in the current study does not provide information previously collected on individuals and does not show a closure of the test score gap between the two ethnic groups once other variables are controlled for. It could be that some of the unexplained variation remaining in this analysis is the effect of factors such as early childhood parental income that we are unable to observe in these data.

  There may also be systematic reasons why Maori are under-achieving in the PISA dataset. First, there has been a long debate about whether standardised tests produce valid results for all ethnic groups involved, both internationally (e.g., Helms 1992), and in New Zealand (e.g., McCreanor 1988). If the test that was undertaken does not discriminate reading ability as accurately for Maori as for other ethnic groups, then we might expect to see a mean test score difference. However, since education success is often measured through tests, by employers and higher level education providers, policy makers may still need to be concerned about the under-achievement of Maori in these sorts of standardised tests. Second, Maori students in Maori immersion classes may do better in a literacy test that is undertaken in Maori rather than English. This would lead to lower results than are achieved by students taught in English. Unfortunately, we are unable to control for this.

  The test score gap is also likely to be because of systematic differences between the ethnic groups. As can be seen in Appendix 1, for many of the variables, there are big differences between the average values for the two groups. These are likely to play a big role in contributing to the test score difference. To be able to address the observed ethnic difference in the regression analysis, we need to be able to identify why the Maori students are, on average, under-performing compared to their Pakeha counterparts. By using the model specification from the regression analysis, we can get some insight into the contributing factors.

  A number of the significant variables found in this study have been found to significantly contribute to test score gaps in previous regression-based work. For example, a New Zealand study carried out by Fergusson et al. (1991) found that differences in family social background, such as socio-economic status, were able to explain a large proportion of the test score difference between Maori and Pakeha students as was found in the current study. Many of the variables identified in this study have also been found to be important in determining test score differences in studies of other ethnic groups. For example, an index of wealth is available in the PISA dataset, which is based on the ownership of various items such as cellphones. This was much higher for Pakeha than Maori, and the regression analysis suggests that higher wealth levels contribute to a larger test score. Work carried out in the United States comparing black and white students also found that higher levels of wealth are associated with better educational outcomes (Orr 2003).

  The comprehensive nature of the PISA dataset allowed a wide range of variables to be included in the analysis. This meant that variables that are not commonly used in this type of study were available. For example, PISA provided a variable that indicated the level of enjoyment that the individual got from reading. The lower reported enjoyment of reading experienced by Maori students was found to contribute to their lower test scores. Also, although both Maori and Pakeha spent more time on the computer than the OECD average, Maori spent more time on computers than Pakeha. This difference significantly contributes to the test score difference between these two groups, with our regression analysis indicating that increased time on computers actually lowers reading test scores. But the variable used in this study did not differentiate between different computer uses, and previous work finds conflicting effects of computer use on educational success. Thus, even with this very rich dataset, we do not know everything we need to in terms of how characteristics are used by students to improve, or hinder, their academic success.

  The results from the decomposition carried out in this study support the findings of previous work on other ethnic groups (e.g., blacks and whites in the United States) using test scores. Differences in family factors, followed by student’s opinions and school factors, were major contributors to the test score gap. Many of the variables contained within these groups were expected to be important based on previous work. For example, the student’s enjoyment of reading played a significant role being able to explain 12.3% of the test score gap, the largest amount explained by a single variable. Differences in socio-economic status is one of the most cited reasons for differences in educational attainment between two ethnic groups (e.g., Bradley & Taylor 2004). The current work supports this, suggesting that the lower socio-economic index experienced by Maori, on average, is one of the most significant contributors to their lower reading literacy test scores.

  The results of this work suggest that it is unlikely that the test score gap will be able to be completely closed through public policy changes. A large proportion (approximately one-third) of the test score gap was unable to be explained by the variables in the model and thus reflects differences in unobservable variables or differing rates of returns to the characteristics for each ethnic group. Moreover, not all the variables that were able to be included in the model can be influenced by policy. For example, although policy makers may be able to influence the number of students in each mathematics class, they are unlikely to be able to influence the student’s own perceived ability in mathematics. So, even if it was possible to give Maori and Pakeha students identical characteristics for all variables controllable by public policy, a somewhat reduced test score gap of approximately 63% of the original test score gap (analysis not presented here) between the two groups would still persist. This finding replicates previous international work (e.g., Todd & Wolpin 2004) which concludes that only a low percentage of the test score gap between ethnic groups is be able to be influenced by public policy.

  The inability of policy interventions to close the gap in educational achievement is likely to cause problems for New Zealand in the future. Maori are becoming a larger proportion of the New Zealand workforce due to their more youthful population structure (Statistics New Zealand 2003). Thus, if their lack of educational success is not able to be addressed, the future New Zealand workforce is likely to be less educated and consequently less productive. The educational difference between Maori and Pakeha has also been suggested as a major contributor to the differences in income inequalities observed in the labour market (Maani 2004). It is important to be able to address the differences in educational success so the income inequalities between Maori and Pakeha can be alleviated.

ACKNOWLEDGMENTS

Thank you to the staff of Motu Economic and Public Policy Research and University of Canterbury for their suggestions and advice on this paper. All remaining errors in this paper are solely those of the authors.

REFERENCES

Adams R, Wu M 2002. PISA 2000 technical report. Paris, OECD. 322 p.

Aloise-Young PA, Chavez EL 2002. Not all school dropouts are the same: ethnic differences in the relation between reason for leaving school and adolescent substance use. Psychology in the Schools 39(5): 539–547.

Bali VA, Alvarez RM 2004. The race gap in student achievement scores: longitudinal evidence from a racially diverse school district. The Policy Studies Journal 32(3): 393–415.

Bedard K, Ferrall C 2003. Wage and test score dispersion: some international evidence. Economics of Education Review 22: 31–43.

Belsey D, Kuh E, Welsch R 1980. Regression diagnostics. New York, John Wiley and Sons. 292 p.

Borjas GJ 1996. Labor economics. New York, McGraw-Hill. 165 p.

Bradley S, Taylor J 2004. Ethnicity, educational attainment and the transition from school. The Manchester School 72(3): 317–346.

Brooks-Gunn J, Klebanov PK, Duncan GJ 1996. Ethnic differences in children’s intelligence test scores: role of economic deprivation, home environment, and maternal characteristics. Child Development 67: 396–408.

Cameron SV, Heckman JJ 2001. The dynamics of educational attainment for Black, Hispanic, and White males. Journal of Political Economy 109(3): 455–499.

Chan C-K, Kato K, Davenport EC, Guven K 2003. Analysis of subsequent educational decisions of high school dropouts and their life outcomes. Paper presented at the 84th Annual Meeting of the American Educational Research Association (AERA), Accountability for Educational Quality: Shared Responsibility, Chicago, April 21–25, 2003.

Else A 1997. Maori participation and performance in education. Wellington, New Zealand, Ministry of Education.

Fergusson DM, Horwood LJ 1997. Gender differences in educational achievement in a New Zealand birth cohort. New Zealand Journal of Educational Studies 32(1): 83–96.

Fergusson DM, Lloyd M, Horwood LJ 1991. Family ethnicity, social background and scholastic achievement—an eleven year longitudinal study. New Zealand Journal of Educational Studies 26(1): 49–63.

Gibson JK 1998. Ethnicity and schooling in New Zealand: an economic analysis using a survey of twins. Wellington, Institute of Policy Studies.

Gibson JK, Scobie GM 2004. Wealth and ethnicity: evidence from the household savings survey. Paper presented to the 7th Labour Econometrics Workshop, University of Auckland.

Haveman RH, Wolfe B 1984. Schooling and economic well-being: the role of nonmarket effects. The Journal of Human Resources 19(3): 377–407.

Hedges LV, Nowell A 1999. Changes in the black-white gap in achievement test scores. Sociology of Education 72(2): 111–135.

Helms JE 1992. Why is there no study of cultural equivalence in standardized cognitive ability testing? American Psychologist 47(9): 1083–1101.

Johnston G 2004. Healthy, wealthy and wise? A review of the wider benefits of education. Treasury Working Paper 04/04.

Juhong B, Maloney T 2006. Ethnicity and academic success at university. New Zealand Economic Papers 40(1): 181–213.

Kenkel DS 1991. Health behavior, health knowledge, and schooling. Journal of Political Economy 99(2): 287–305.

Maani SA 2004. Why have Maori relative income levels deteriorated over time? The Economic Record 80(248): 101–124.

Maani SA, Kalb G 2007. Academic performance, childhood economic resources and the choice to leave school at age 16. Economics of Education Review 26: 361–374.

McCreanor T 1988. The bias in TOSCA. In: Olsen M ed. Mental testing in New Zealand: critical and oppositional perspectives. Dunedin, University of Otago Press. 270 p.

McKay PF, Doverspike D, Bowen-Hill D, McKay QD 2003. The effects of demographic variables and stereotype threat on black/white differences in cognitive ability test performance. Journal of Business and Psychology 18(1): 1–14.

Maré D 1995. Examining labour market outcomes for Māori. Labour Market Bulletin 1995(1): 116–123.

Ministry of Education 2002. PISA 2000. Overview of selected New Zealand findings. Wellington, Ministry of Education.

Ministry of Education 2005a. Māori in early childhood education and schools. Wellington, Ministry of Education.

Ministry of Education 2005b. Māori in tertiary education: a picture of the trends. Wellington, Ministry of Education.

Neumark D 1988. Employers’ discriminatory behaviour and the estimation of wage discrimination. Journal of Human Resources 23(2): 279–295.

Oaxaca R 1973. Male-female wage differentials in urban labor markets. International Economic Review 14(3): 693–709.

Oaxaca RL, Ransom MR 1994. On discrimination and the decomposition of wage differentials. Journal of Econometrics 61: 5–21.

Orr AJ 2003. Black-white differences in achievement: the importance of wealth. Sociology of Education 76: 281–304.

Perna LW 2000. Racial and ethnic group differences in college enrolment decisions. New Directions for Institutional Research 107: 65–83.

Sammons P 1995. Gender, ethnic and socio-economic differences in attainment and progress: a longitudinal analysis of student achievement over 9 years. British Educational Research Journal 21(4): 465–486.

Statistics New Zealand 2003. National Maori population projections (2001 (base) – 2021). Media Release 15 May 2003.

Sturrock F, May S 2002. PISA 2000: the New Zealand context. Wellington, Ministry of Education. 153 p.

Todd PE, Wolpin KI 2004. The production of cognitive achievement in children: home, school and racial test score gaps. Working paper, University of Pennsylvania, Department of Economics.

Wolfe B, Zuvekas S 1997. Nonmarket outcome of schooling. International Journal of Education Research 27(6): 491–502.

---

Table 1  Regression and decomposition results (including P values) predicting test scores.

				Decomposition regressions
	Regression			Pakeha			Maori
	β	P value		β	P value		β	P value
Student characteristics
Maori	–17.83***	0.000
Male	–17.67***	0.000		–16.53***	0.000		–  24.45**	0.013
School year – 11	51.51***	0.000		54.27***	0.000		37.93**	0.037
School year – 12	104.8***	0.000		102.4***	0.000		128.3***	0.000
Family factors
Mother’s perceived  socio-economic index	0.461***	0.000		0.537***	0.000		0.084	0.790
Student’s perceived  socio-economic index	1.167***	0.000		1.187***	0.000		1.013***	0.000
Family wealth	3.17	0.160		2.007	0.410		8.402	0.170
Cultural communication  with parents	3.962*	0.056		5.680**	0.011		–4.975	0.390
Language mainly spoken at  home – Maori	–59.88***	0.004		–96.4	0.160		–51.12**	0.029
Language mainly spoken at  home – Other	–44.25***	0.001		–44.57***	0.001		0	0
School factors
School decile – medium	–3.625	0.530		–8.339	0.220		6.64	0.580
School decile – high	17.02***	0.005		9.797	0.160		42.14***	0.003
School size	–0.0108**	0.016		–0.00910*	0.064		–0.0184	0.110
Proportion of teachers  with teaching degree	–44.78***	0.000		–36.29***	0.004		–72.70***	0.008
Courses for gifted students  at school	–6.763*	0.067		–5.506	0.170		–15.01	0.120
Poor quality school resources	–4.214**	0.049		–4.089*	0.087		–3.885	0.460
Number of students in  English class	1.619***	0.000		1.473***	0.004		1.862*	0.072
Number of students in  Maths class	1.369***	0.001		1.401***	0.002		1.439	0.160
Minutes spent in Science  class each week	0.0645***	0.002		0.0594***	0.009		0.103*	0.056
Student activities
Cultural activities	3.456*	0.096		2.731	0.230		5.043	0.350
Computer use	–12.22***	0.000		–12.01***	0.000		–11.15***	0.009
Student opinions
Sense of belonging at school	–3.600**	0.050		–2.561	0.210		–6.978	0.110
Use of competitive learning	11.84***	0.000		12.09***	0.000		9.568*	0.064
Interest in Maths	–1.732	0.370		–1.655	0.430		–1.942	0.720
Enjoyment of reading	23.59***	0.000		24.05***	0.000		20.00***	0.000
Constant	418.4***	0.000		391.9***	0.000		438.2***	0.000
Observations	1543			1285			258
R squared	0.42			0.42			0.39

Source: Authors’ calculations from PISA 2000 data for New Zealand Maori and Pakeha students, OECD. *** Significant at 0.01; ** Significant at 0.05; * Significant at 0.1

---

Table 2 Aggregated, standardised coefficients from regression analysis.

	Absolute coefficients	P value
Student characteristics	191.81***	0.000
Family factors	112.89***	0.000
School factors	79.47***	0.000
Student activities	15.67***	0.000
Student opinions	40.77***	0.000

Source: Authors’ calculations from PISA 2000 data for New Zealand Maori and Pakeha students, OECD. *** Significant at 0.01; ** Significant at 0.05; * Significant at 0.1

---

Table 3  The contribution of variable groups to the reading literacy test score gap between Maori and Pakeha students.

	Absolute value	Percentage
Student characteristics	1.38	3.19
Family factors	10.61	24.54
School factors	6.29	14.56
Student activities	3.97	9.19
Student opinions	6.69	15.48
Unexplained differences	14.28	33.04
Total	43.22	100.00

Source: Authors’ calculations from PISA 2000 data for New Zealand Maori and Pakeha students, OECD.

---

Appendix Definitions and average values for Pakeha and Maori subpopulations for all variables used in the analysis.

	Definition	Mean values
	Pakeha	Maori
Student characteristics
Maori	Binary = 0 for Pakeha; 1 for Maori
Male	Binary = 0 for female; 1 for male	0.509	0.484
School year	Student Grade – Year 10, 11 or 12	10.990	10.945
Family factors
Mother’s perceived socio-economic index	Index based on occupation and hours	47.689	41.141
Student’s perceived socio-economic index	Index based on expected occupation	56.826	52.804
Family wealth	Index based on home resources	0.376	–0.071
Cultural communication with parents	Index based on frequency of cultural conversations	0.071	0.018
Language mainly spoken at home	1 = English; 2 = Maori; 3 = Other	1.041	0.048
School factors
School decile	1 = deciles 1–3; 2 = deciles 4–7; 3 = deciles 8–10	2.308	1.844
School size	Number of students in school	956.629	819.385
Proportion of teachers with  teaching degree	Proportion of English teachers in school with Bachelor of Teaching	0.727	0.722
Courses for gifted students at school	Binary = 0 if have courses; 1 = if don’t	0.424	0.430
Poor quality school resources	Index of whether Principal believes learning in school hindered by resources	–0.133	–0.005
Number of students in English class	Usual number of students in English class	25.162	24.104
Number of students in Maths class	Usual number of students in Maths class	24.901	23.369
Minutes spent in Science class each week	Time spent in Science class each week	257.411	239.974
Student activities
Cultural activities	Index of participation in cultural activities	–0.085	–0.106
Computer use	Index of time spent on computer	0.145	0.374
Student opinions
Sense of belonging at school	Index based on student’s opinions about different ways of belonging at school	–0.046	0.141
Use of competitive learning	Index of the frequency of student’s use of competitive learning strategies	0.272	0.134
Interest in Maths	Index based on student’s attitude towards Maths	0.004	0.066
Enjoyment of reading	Index based on student’s attitude towards reading	0.027	–0.195

Source: PISA 2000 data for New Zealand Maori and Pakeha students, OECD.

^[1]A recent study by Juhong & Maloney (2006) investigated the effect of ethnicity on tertiary academic performance using student’s Grade Point Average (GPA) scores. Unfortunately, due to data limitations, they were only able to investigate the effect of age, gender, school qualifications, bursary scores, secondary school decile ranking, and ethnicity on the GPA. Additional socio-economic variables were not available, limiting the number of test scores determinants able to be investigated.

^[2]This study was originally carried out in 2000 in 32 different countries including 28 OECD countries and involves students ranging in age from 15 years and 3 months to 16 years and 2 months. Tests were sat in the language of instruction to prevent any countries from being disadvantaged.

^[3]For more information on the PISA dataset see Adams & Wu (2002).

^[4]This method is typically called the Oaxaca decomposition, after Oaxaca (1973) who first proposed it. Discussion of this and other decompositions can be found in Oaxaca & Ransom (1994).

^[5]See Gibson & Scobie (2004) for further details on the theory behind this approach.

^[6]This was achieved by Finland with an average of 546 points.

^[7]The average Maori test score was 482 points compared with Hungary with 480.

^[8]Due to missing data, the sample size dropped from the initial 3031 individuals to 1543 individuals. But a sensitivity analysis using imputed data was carried out on an alternative model specification which found that this drop in sample size is unlikely to alter the findings.

^[9]Complete decomposition results for the individual variables are available from the author.

---

This year's abstracts | Journal home page | All abstracts | Publishing home page

PDF file of entire paper: Print-quality (358K)

K07017; Online publication date 30 May 2008
Received 7 August 2006; accepted 19 March 2008
Kōtuitui: New Zealand Journal of Social Sciences Online, 2007, Vol. 3: 1–13
1177–083X/08/0301–0001 © The Royal Society of New Zealand 2008