Why does Year Twelve retention differ between Australian states and territories?

Watson, Louise

Year Twelve retention rates have a number of well-known deficiencies that prevent proper comparisons of school completion between school systems. This paper compares secondary school completion rates across Australian states and territories from 1989 to 2002 and adjusts 'official' 2002 retention rates to take account of the acknowledged measurement problems. We identify a pattern of mismeasurement of national Year Twelve retention over the 1990s. We estimate that the Year Twelve retention rate was a particularly poor measure of national school completion in the early 1990s, when it appeared to peak in the official estimates. In contrast to the official figures, our adjusted measure of Year Twelve retention was no lower in the late 1990s than it had been in the early 1990s. Our findings suggest that governments should be cautious in using official Year Twelve retention rates as a measure of the performance of Australian school systems.

Keywords

educational policy

grade repetition

secondary school students

school holding power

performance indicators

school entrance age

Introduction

Retention at school to Year Twelve has traditionally been used as an indicator of school system performance in Australia. For example, the 1989 National Report on Schooling in Australia indicated that state and Commonwealth ministers for education had 'agreed to work towards a national Year Twelve retention rate of sixty-five per cent by the early 1990s' (Australian Education Council, 1991, p. 8).

The Year Twelve retention rate is measured as the number of students in Year Twelve in a given calendar year divided by the number of students who were in the first year of secondary school when that Year Twelve cohort commenced secondary school. While comparisons of retention rates across jurisdictions have always been made, official publications have repeatedly pointed to a list of factors that limit their usefulness, such as:

* population changes, including international and interstate migration

* the effect of full fee-paying overseas students at upper secondary level

* Year Twelve repetition

* the availability of part-time school study options

* the effect of alternatives to school education, most notably vocational courses available through Technical and Further Education (TAFE) institutions

* the different age-grade structures in the states (Steering Committee for the Review of Commonwealth/State Service Provision, 2000, pp. 66-7). (1)

In response to these deficiencies, governments have developed alternative measures of school participation and completion, such as 'attainment rates' by age nineteen and 'full-time participation' rates, which include participation in full-time study or work and jointly in part-time work and study (Ministerial Council on Education, Employment, Training and Youth Affairs, 2000). (2) Nevertheless, we will argue that these indicators require just as careful interpretation as differences in retention rates across jurisdictions, because the different age-grade structures in jurisdictions have similar effects on alternative participation indicators to their effect on retention rates.

This paper estimates the impact of the confounding influences identified above on retention comparisons to assess their importance. The first two factors identified above are handled through adjustments to the estimated retention rates that build in changes in the school-age population. (3) The impact of grade repetition is estimated from available Australian Bureau of Statistics (ABS) data in conjunction with regression analysis. The role of the last three factors is also assessed with the regression analysis.

The second section describes the methodology used here to analyse the problems associated with retention rates and quantify their effects and presents adjusted estimates that take account of these effects. The third section uses these results to re-assess what happened to national retention rates over the 1990s. The fourth section discusses these estimates in the context of current directions in school system performance indicators. The policy implications of our results are discussed in the conclusion.

How to adjust to retention rates

The 'official' Year Twelve retention rates produced by the ABS are estimates of the proportion of any cohort who commence secondary school and proceed to Year Twelve in the minimum possible number of years in a jurisdiction. Secondary school commences in Year Seven in New South Wales, Victoria, Tasmania and the Australian Capital Territory. It commences in Year Eight in Queensland, South Australia, Western Australian and the Northern Territory. Consequently, the calendar year in which the denominator is measured varies between jurisdictions for any Year Twelve retention figure in a calendar year.

Most of the data used in this paper are taken from the National Schools Statistics Collection (NSSC), which is published by the ABS in Schools, Australia (Cat. No. 4221.0). (4) There are two particular limitations in the NSSC data for the analysis undertaken below. First, the data do not allow identification of the prevalence of Year Twelve repetition. Second, the data relate primarily to full-time students. Since both of these factors are forms of school participation that individuals may choose and their incidence has reportedly changed over time, their mismeasurement in these data might affect the validity of the inferences drawn here. Attempts to estimate rates of repetition and part-time study from the published data are described below.

The next three sub-sections describe the various adjustments we make to the official ABS retention rate estimates anticipated in the introduction: the first describes the adjustment to account for population changes; the second how we account for Year Twelve repetition; and the third the regression results that lie behind each of the adjustments required for TAFE study, part-time study and differences in age-grade structures between jurisdictions.

Adjustments for population changes

We make three adjustments to the standard retention rate definition used by the ABS to produce estimates that are more consistent across jurisdictions and to minimise the effects of departures or additions on the cohort through either grade repetition or migration, drawing on data published by ABS in Population by Age and Sex (1985-2002).

First, for all jurisdictions, retention rates are estimated by dividing the number of Year Twelve students in calendar year t with the number of Year Eight students in calendar year t minus four (in jurisdictions with six years of secondary school, the 'official' estimate uses t minus five). This changes marginally the estimated retention rates for those jurisdictions where secondary school commences in Year Seven, but leaves unchanged the estimates for jurisdictions where it commences in Year Eight.

Second, only those Year Eight students aged twelve to fifteen years inclusive are counted in the denominator (these ages represented 99.8 per cent of Year Eight students in Australia in 1998) and only those Year Twelve students aged sixteen to nineteen in the numerator (these ages represented 98.8 per cent of Year Twelve students four years later in 2002). (5)

Third, the estimates are adjusted for changes in the population of the relevant age cohorts in jurisdictions over these years. That is, population growth in each Year Eight single year of age cohort over the years to Year Twelve is used to adjust the retention rate estimate. In effect, the denominator is increased to match population growth over the intervening years.

The effects of these adjustments on estimated Year Twelve retention rates for 2002 are shown in columns two, three and four of Table 1. The estimates in the table are cumulative: the estimates in the fourth column include those of the second and third columns.

In the official retention rate estimates, retention is highest in the Australian Capital Territory, Queensland, Victoria and Western Australia (see column one of Table 1). The first two adjustments have very little impact on the estimated retention rates, other than in Tasmania, where ten per cent of the Year Twelve cohort in 2002 was aged twenty years or older. In 1998 in Tasmania, there were no Year Eight students older than 15 years (nor were there Year Seven students older than fourteen in 1997). Hence this substantial group of students were clearly not original members of the grade cohort. The adjustment for population change has more of an impact in other jurisdictions, lowering estimated retention in most jurisdictions (the Australian Capital Territory, Queensland and Victoria especially), but raising it in Tasmania and the Northern Territory.

Overall, the three adjustments lower estimated retention by about four percentage points in Australia, but do not change the relative retention position of jurisdictions substantially.

Adjustments for Year Twelve repetition

The retention rate estimates presented in the previous sub-section excluded those Year Twelve students aged twenty years or older in July 2002. This adjustment will have removed some of the effects of Year Twelve repetition. We developed a further two-stage strategy to estimate the impact of Year Twelve repetition on our adjusted retention rates.

In the first stage, we generate an estimate of Year Twelve repetition in all jurisdictions in all years based on changes in the average ages of grade cohorts between Years Eleven and Twelve. In the second stage we make assumptions about the way we have mismeasured the 'true' Year Twelve repetition and use (instrumental variable) regression techniques in an equation designed to explain Year Twelve retention to obtain better estimates of the extent of Year Twelve repetition.

The regression parameter on the Year Twelve repetition variable in the retention equation can be used to re-scale the repetition estimates obtained through the first stage described above. These repetition estimates can then be subtracted from the retention estimates to obtain repetition-adjusted retention rates. These are shown for 2002 in column five of Table 1.

It is likely that the initial Year Twelve repetition estimates underestimate the true level, since 'true' progression from Years Eleven to Twelve may be higher for younger members of grade cohorts than older ones. Hence the reduction in jurisdictions' retention rates for Year Twelve repetition may also be too small in Table 1. Nevertheless, the patterns in the estimated Year Twelve repetition series both through time and between jurisdictions accord with what little is know about it in Australia. (6)

Nationally, we estimate that Year Twelve repetition added about one and a half percentage points to retention in 2002. This contribution varied considerably by jurisdiction, being higher in Tasmania, Western Australia and South Australia than in other jurisdictions.

Adjustments for differences in education systems

Our adjustments to account for differences in state and territory education systems are based on the parameters estimated from a regression equation designed to explain changes in retention rates in jurisdictions between 1989 and 2001. Essentially, we estimate a regression equation where the dependent variable is the adjusted retention rate estimate from which the fourth column of Table 1 is drawn. We use data from thirteen years (1989 to 2001) for each of the eight jurisdictions, so the equation is estimated using 104 observations. It is a linear equation estimated by Instrumental Variables to deal with the measurement error of the Year Twelve repetition variable already described, and the potential endogeneity of some of the other explanatory variables. (7) The explanatory variables include labour market and education system-related factors. The labour market variables are designed to capture both cyclical and structural factors likely to influence the employment opportunities for young people. The cyclical factors are captured through the change in annual state unemployment rates over the period. The structural factors are captured through measures of the annual state proportions of fifteen to nineteen year olds who work full-time, (8) drawn from data in Labour Force Australia (ABS, 1987-2002).

Differences in the education systems are captured with a series of variables. The first is the proportion of seventeen year olds in a jurisdiction studying at a TAFE college in any year, drawn from data published in Participation in Education (ABS, 1989-1999), and from the National Centre for Vocational Education Research (2000, 2001).This variable is interpreted as reflecting alternatives to school in jurisdictions, both educational and labour market alternatives, since apprentices and trainees are included in the TAFE student figures. (9) The second is an estimate of the proportion of the relevant Year Eight cohort studying part-time in Year Twelve. The third variable is the proportion of Year Eight students of any grade cohort who are Indigenous. This was highest in the Northern Territory in 1998 at about twenty-five per cent of the grade cohort, but was no more than five per cent in the other jurisdictions.

Accounting for different age-grade structures The final education system-related variable is designed to capture differences in the age-grade profiles of jurisdictions arising from their different formal school systems. Age-grade structures differ between the states and territories because their school starting-age policies vary and some states provide an initial year of primary school prior to Year One.

Differences in age-grade structures might affect retention across jurisdictions via a number of potential mechanisms. In conjunction with minimum school leaving ages, systems with twelve years in their school system structure provide fewer grades prior to Year Twelve in which individuals might leave school. Explanations of why such jurisdictions might enjoy higher school completion rates include: screening theories were individuals choose their schooling to differentiate themselves from the earliest leavers; models where the experience of being among the poorer performers in class induces non-completion by individuals; and simple stochastic models, where individuals have a chance of leaving each time they make a decision to stay or leave, so that the more times they make the decision, the more individuals will be observed to leave school. Similarly, in human capital models, more individuals might be expected to complete school with twelve-year structures than one with thirteen. In these models, individuals choose a desired level of schooling based on its relative costs and benefits. All those whose desired level is twelve years or more will complete school in systems with twelve years in their formal structure, while only those whose desired level is at least thirteen years will complete it in systems with thirteen. Since in any population the size of the former group can be no smaller than the size of the latter, school completion should be no smaller, other things equal, in systems with twelve years in their formal structures than those with thirteen.

The variable we use to capture differences in age-grade structures across Australian jurisdictions measures the proportion of twelve year olds in the Year Eight cohort. (10) Higher proportions of students aged twelve years reflect 'young' age-grade structures in jurisdictions. The proportion is highest in jurisdictions without a pre-Year One level of primary school (Queensland and, at the time of research, Western Australia). These proportions in 1998 varied from more than forty per cent in Western Australia to about one per cent in a number of other jurisdictions.

The proportions aged twelve years old in Year Eight changed substantially in two jurisdictions in the period we analyse, South Australia and Queensland, for reasons associated with school entry or early school progression policies. The proportions aged twelve in Year Eight for the cohorts reaching Year Twelve between 1989 and 2002 in these jurisdictions are shown in Figure 1, along with the proportion in Western Australia, which changed little over the relevant period. The decline in the proportion is most marked in South Australia.

[FIGURE 1 OMITTED]

The changes in South Australia resulted from the implementation of the Early Years of School policy, announced in 1984 and implemented from 1985. South Australia has a 'continuous admission' policy for five year olds (described in Trethewey, 1997). It involves regular (not less than once a term) admission of recently turned five year olds into individual schools over the school year. Prior to the policy reform, students who commenced school at the start of the school year (approximately those born from October to mid-February) typically completed just two years of schooling before entering Year Three. After the reform, they completed three years before entering Year Three. Students who entered school later in the year, from March to June, were not affected by the policy change, while some born from July to September also did an extra year of schooling.

The main effect of the policy change was that grade cohorts were older (the proportion aged twelve in Year Eight fell). The change in the composition of the cohorts meant that the reported ages of students in Year Eight went from being approximately split between twelve and thirteen years in the ABS data prior to the change to predominantly thirteen years after it. (11)

The changes in Queensland were less dramatic. In Queensland, the school entrance policy since the 1950s had allowed (but did not require) students to commence school at the start of the school year provided they turned five by the end of February. Over time this practice became less common. In 1986, the Education Department announced a new policy: that in 1987 students could commence school provided they turned five by the end of January, and from 1988 they could only commence school if they had turned five in the preceding calendar year. The 1988 cohort reached Year Eight in 1994, but as is evident from Figure 1, the proportion aged twelve in Year Eight in Queensland declined over much of the period we analyse. (12)

Regression results and further adjustments to retention The regression results (reported in the Appendix) largely confirm the role of economic factors in explaining retention (both the long-term decline in full-time youth jobs and the role of cyclical effects through changes in the unemployment rate). Like findings in Larum and Beggs (1989) and Lewis and Koshy (1999), the results suggest that the decline in full-time job opportunities for teenagers was the main determinant of increased school retention.

The results also confirm the role of education system effects on Year Twelve retention. Retention is lower the higher the proportion of seventeen year olds who study at TAFE and the higher the Indigenous share of the commencing Year Eight cohort. The results imply that Indigenous retention is about thirty per cent of that of non-Indigenous Australians. That estimate seems low compared to estimates based on national figures in Long, Frigo and Batten (1999, p. 50) of just over forty per cent. Our estimates are heavily dependent on retention patterns and the treatment of Indigenous students in the ABS collection for the Northern Territory. Changes in the treatment of Indigenous students in the Northern Territory statistics affected apparent retention rates there substantially from 1999. (13)

Retention is higher among jurisdictions with 'young' Year Eight cohorts and in jurisdictions where Year Twelve repetition is higher. There was no evidence from the regression results that growth in part-time Year Twelve student numbers had detracted from Year Twelve retention, so that variable was excluded from the results presented. However, data on part-time Year Twelve student numbers were not available for the earlier years and had to be estimated. While various assumptions were used to construct the series, some of which were 'sympathetic' towards finding a negative effect of part-time students on retention and no such effect was found, we think it best not to make too much of this particular finding until a longer time series of data is available.

Two of the education effects from the results reported in the Appendix table are incorporated into further adjusted retention rates for Australian jurisdictions in 2002 in Table 1. Like the earlier adjustments, these adjustments remove the relevant effect from the national retention estimates and those of the individual jurisdictions. The sixth column incorporates all of the other adjustments in Table 1 and corrects for the age-grade structure via the age profile of the Year Eight cohort. The comparison is made across jurisdictions as if there was no Year Twelve repetition and as if jurisdictions had no students aged twelve in Year Eight. The final column of Table 1 incorporates the Year Eight student body Indigenous share effect. (14) Again, the comparison assumes implicitly that there are no Indigenous students in jurisdictions. (15) As with the Year Twelve repetition adjustment, the estimated parameters for the age profile of the Year Eight cohort and the Indigenous share effect from the Appendix table are multiplied by the relevant values of those variables for all jurisdictions to adjust their 2002 retention estimates.

The adjustments to jurisdictions with large Indigenous shares of the Year Eight population and/or with young age-grade structures are substantial. For example, the results indicate that the young age-grade structures of Queensland and Western Australia in the data added 11.6 and 14.4 percentage points to their retention in 2002. Such estimates might seem implausibly large. This age-grade effect is identified in our results by the change in the age-grade structure that took place in South Australia described in Figure 1. The proportion aged twelve years in Year Eight fell by thirty percentage points between 1992 and 1996 in that state. Between the same years, the official retention rate fell by 24.2 percentage points--from 92.7 to 68.4 percent, compared with the national decline of 5.8 percentage points. We would attribute just over ten percentage points of the decline in relative retention in South Australia to the changes in their age-grade structures in that period, with the decline in Year Twelve repetition contributing most of the balance. Consequently, we consider that the evidence supports the existence of substantial age-grade effects on Year Twelve retention.

We are less confident about the magnitudes implied by the Indigenous share adjustment than the age-grade adjustments. The parameter estimate implies a very large penalty for jurisdictions with large Indigenous shares and its estimation may have been affected by changes in data definitions in the Northern Territory.

The adjusted, relative rankings evident in columns six and seven of Table 1 were quite stable over the period 1989 to 2002. The Australian Capital Territory, Victoria and New South Wales had consistently higher retention rates than the other states between 1989 and 2002. The rate in Queensland was consistently above that of the remaining states, while Tasmania was the only jurisdiction to exhibit a trend change (upwards) in its relative retention ranking over the period. This hierarchy is also evident in the jurisdiction effects captured in the regression results reported in the Appendix. (16)

What drove changes in 'official' national retention rates in the 1990s?

It is useful to distinguish between factors that influence the true or underlying rate of school completion and those that affect its measurement. For example, the corrections embodied in columns three to five of Table 1 involve the removal of three sources of measurement error in the retention rate--school age population growth; the addition of older students to the Year Twelve class; and Year Twelve repetition. These factors get in the way of estimating the true proportion of students who commence secondary school who proceed to Year Twelve in the shortest possible time. Moreover, just as the factors that influence true school completion may affect the path it follows over time, so too these sources of measurement error may change in their impact on retention over time.

Between 1989 and 2002 these factors made varying contributions to the 'official' retention estimate. However, their contributions all peaked in either 1992 or 1993, and their sum peaked in 1992, adding about eight and a half percentage points to retention in that year. By 1996, the trough in 'official' retention in the 1990s, the contribution of these three factors had fallen by five percentage points from their 1992 peak, close to the decline in 'official' retention over that period. The implication of this is that part of the rise to 1992 and much of the decrease after 1993 in 'official' retention reflected measurement problems rather than changes in underlying school completion rates.

This is not to say that true school completion rates were unchanged in this period. The regression equation described earlier can be used to identify those factors that influenced true school completion rates from the late 1980s. Figure 2 shows the impact of key variables from the equation on underlying school completion. The effects of these variables are shown relative to their impact in 1989. For example, the unemployment rate added about three percentage points to retention in 1991 relative to its impact on retention in 1989. The aggregate change in retention and the total predicted change based on changes in all the variables in the regression equation are also shown in Figure 2. (17)

[FIGURE 2 OMITTED]

The broad magnitude of changes in underlying retention appears to have been driven principally by the decline in available full-time jobs for teenagers. In addition, the recession, via the increase in the unemployment rate between 1989 and 1991 added about three percentage points to retention. The impetus to retention from the loss of full-time teenage jobs slowed from 1993. A further effect of a structural nature that detracted from retention after 1994 was that grade cohorts aged somewhat (the South Australian and Queensland changes described earlier).

What factors caused the observed decline in 'official' national retention of six percentage points between 1992 and 1996? The major element of this fall came from the decline in Year Twelve repetition among students aged sixteen to nineteen in Year Twelve (it contributed three percentage points to the decline in retention). Three other elements made individual contributions to the fall in retention of one to one and a half percentage points: the fall in the number of Year Twelve students aged twenty or older, which must have included a substantial proportion of students repeating Year Twelve; the decrease in the growth rate of the senior school age population; and the ageing of grade cohorts, that is, the fall in the share of the Year Eight student body aged twelve years.

Figure 3 shows what the path followed by Australian retention over the 1990s looks like once the figures are corrected to take account of the measurement problems discussed earlier in this section. It contains the 'official' estimate of retention between 1989 and 2002 and an estimate when the rate is adjusted for population changes, the contribution of older Year Twelve students and estimated Year Twelve repetition. No adjustment is made for the ageing of the grade cohort or to take out the effect of full-time job losses.

[FIGURE 3 OMITTED]

The picture from the adjusted measure is somewhat different from that provided by the 'official' retention estimate. The peak in retention during the early 1990s recession is less pronounced, the subsequent decline smaller and the adjusted retention estimates exceed those of 1992 and 1993 from 1999. That is, unlike the pattern in the 'official' estimate, the measure of adjusted retention is higher at the end of the estimation period than it was during the recession of the early 1990s.

Implications for policy and performance measurement

As emphasised in the third section, the main commonly-identified factor that stands in the way of comparisons of retention between jurisdictions is differences in the age-grade structures of jurisdictions. But alternative participation and school completion measures are also affected by differences in age-grade structures (a point acknowledged in relation to school age participation rates in Steering Committee for the Review of Commonwealth/State Service Provision, 2003, p. 3.28).

Just as jurisdictions with 'old' age-grade structures face a 'disadvantage' in relation to Year Twelve retention compared to those with younger structures, these structures also affect their relative position in relation to measures of participation by age. However, the position of relative disadvantage is reversed. That is, jurisdictions with young age-grade structures are relatively disadvantaged by comparisons of school participation among individuals of a particular age. The extent of this degree of 'disadvantage' across jurisdictions is summarised in Table 2. (18) The relative 'poor' performance of Queensland and Western Australia improves after this 'disadvantage' is taken into account.

Conclusion

Although government authorities acknowledge that a range of factors limits the usefulness of the estimates of Year Twelve retention rates produced by the Australian Bureau of Statistics, these data are still widely used as a measure of school system performance in Australia. This paper indicates that the 'official' figures can be adjusted to accommodate the influence of these factors to produce more reliable and useful measures of Year Twelve retention that can be used for comparative purposes.

When the 'official' Year Twelve retention rate for each state and territory is adjusted to take into account of these various factors, we find:

* By removing older age students from the Year Twelve cohort and the effect of population change, Year Twelve retention was about four percentage points lower than in the 'official' estimates in 2002.

* Year Twelve repetition added about one and half percentage points to national retention in 2002 with a bigger effect in some Tasmania, Western Australia and South Australia.

* Age-grade structures have a substantial impact on retention, completion and participation measures, adding over ten percentage points to the 2002 Year Twelve retention estimates for Queensland and Western Australia. Nevertheless these two states faced a penalty from those same age-grade structures in their participation measures. The penalty was of similar magnitude to the retention benefit for participation among seventeen year olds and approaching five percentage points for participation among fifteen to nineteen year olds.

* Removal of three important influences on retention in the 1990s--older age students in the Year Twelve 'cohort', the effect of population change, and Year Twelve repetition--substantially changes retention nationally. The peak in retention during the early 1990s recession is less pronounced in the adjusted estimates and the subsequent decline is smaller. The adjusted Year Twelve retention rate in the late 1990s was no lower than it had been in the early 1990s.

* The main reason Year Twelve retention was higher in 2001 than it was in 1989 was the loss of available full-time jobs for young people. This factor contributed about twelve percentage points to the increase in retention over the period analysed.

Alternative measures of school participation or completion are subject to at least some of the same criticisms as those directed to Year Twelve retention. Like retention, the alternative measures can be adjusted to remove the confounding influences. When this is done, the alternative measures reveal a similar picture to the adjusted retention estimates over the 1990s.

As Year Twelve retention rates are influenced to a significant extent by the availability of full-time jobs for teenagers (see Figure 2)--an economic condition beyond the control of schools or school systems--we should be cautious in applauding rises in 'official' Year Twelve retention rates as evidence of superior school performance or condemning falls without taking labour market conditions into account. By illustrating the extent to which labour market conditions drove national Year Twelve retention rates during the 1990s, this paper provides implicit endorsement for broadening the scope of school system performance measures.

Although labour market factors may explain the pattern followed by national Year Twelve retention over the estimation period, they are not the sole determinants of school retention. Unexplained differences in retention performance between the jurisdictions (the state effects in the regression results) may point to differences in system performance. There is a range of policies, practices and institutional arrangements that could contribute to differences in retention outcomes between jurisdictions. The impact of differences in school starting ages, the availability of part-time study options and the provision of TAFE alternatives to Year Twelve is discussed in this paper. Other state characteristics not accounted for here, such as rurality and ethnicity, may also affect retention. Further, state policies and practices that are also likely to influence Year Twelve retention rates include curriculum and assessment arrangements, modes of school organisation and the allocation of resources across school systems. Where these practices differ between jurisdictions, they are likely to influence differences in Year Twelve retention, although their effects are difficult to quantify (Vickers & Lamb, 2002).

Our findings suggest that any pessimism about school performance based on the decline in official Year Twelve retention rates since 1992 is misplaced. The peak in Year Twelve retention recorded in 1992 and the subsequent decline in the 'official' estimates was caused, in part, by measurement problems in the retention series. Our adjusted measure of Year Twelve retention was no lower in the late 1990s than in the early 1990s.

While it is possible to adjust the National Schools Statistics Collection data to produce retention estimates that are relatively comparable across jurisdictions and through time, it would be better to analyse differences between jurisdictions with data that were already comparable. For example, use of a unique student identifier in the National Schools Statistics Collection would make such comparisons more straightforward and allow issues such as the impact of grade repetition and part-time studies on retention and participation to be addressed more directly than was possible in this paper.

Appendix Retention equation regression results

 Adjusted retention rate

 Std.
 Coeff. Err. t-ratio

Year Twelve Repeats 1.577 0.44 3.55
Full-time employment share
 squared (NSW, VIC, TAS, ACT) -2.277 0.17 -13.46
Full-time employment share
 squared (QLD, SA, WA, NT) -1.248 0.24 -5.28
Proportion aged seventeen at TAFE -0.190 0.09 -2.10
Increase in the unemployment rate 0.015 0.01 2.67
Decrease in the unemployment rate 0.003 0.01 0.41
Indigenous share of
 Year Eight cohort -0.717 0.28 -2.52
Twelve year old share
 of Year Eight cohort 0.339 0.05 6.61

State effects
NSW 0.769 0.01 53.14
VIC 0.779 0.01 58.68
QLD 0.686 0.02 38.36
SA 0.653 0.01 45.47
WA 0.626 0.02 30.53
TAS 0.681 0.02 32.35
NT 0.690 0.09 7.82
ACT 0.847 0.02 56.00

Marginal effects
Year Twelve Repeats 1.577 0.44 3.55
Full-time employment share
 squared (NSW, VIC, TAS, ACT) -0.970 0.07 -13.46
Full-time employment
 share squared (QLD, SA, WA, NT) -0.580 0.11 -5.28
Proportion aged seventeen at TAFE -0.190 0.09 -2.1
Increase in the unemployment rate 0.015 0.01 2.67
Decrease in the unemployment rate 0.003 0.01 0.41
Indigenous share of
 Year Eight cohort -0.717 0.28 -2.52
Twelve year old share
 of Year Eight cohort 0.339 0.05 6.61

Observations 104
 Dependent Variable: Mean 0.666
 Std dev. 0.115
 Residual Sum of squares 0.078
 Std dev. 0.030
R-squared 0.943
F[15, 88] 97.2
Probability value 0.000
Log likelihood 226.6
Restricted (b=0) Log likelihood 77.6
Estd. Autocorrelation of e (i,t) 0.29

Acknowledgements

The authors are grateful to Tue Gorgens for his comments on an earlier draft of this paper. All errors are the responsibility of the authors.

References

Australian Bureau of Statistics (ABS) (1985-2002). Schools, Australia: Cat. 4221. O. Canberra: Australian Bureau of Statistics.

Australian Bureau of Statistics (ABS) (1985-2002). Population by Age and Sex: Cat. 3201.0. Canberra: Australian Bureau of Statistics.

Australian Bureau of Statistics (ABS) (1987-2002). Labour Force Australia: Cat. 6201.0. Canberra: Australian Bureau of Statistics.

Australian Bureau of Statistics (ABS) (1989-1999). Participation in Education: Cat. 6272. O. Canberra: Australian Bureau of Statistics.

Australian Education Council (1991). National Report on Schooling in Australia 1989. Carlton: Curriculum Corporation.

Department of Employment, Education and Training (1994). Retention and Participation 1967 to 1993. Canberra: Australian Government Publishing Service.

Hausman, J.A. (1978). Specification tests in econometrics. Econometrica, 46, 1251-71.

Lamb, S. (1996). Completing School in Australia: Trends in the 1990s (Longitudinal Surveys of Australian Youth, Research Report No 1). Melbourne: Australian Council for Educational Research.

Lamb, S. (1998). Completing school in Australia: Trends in the 1990s. Australian Journal of Education, 42(1), 5-31.

Larum, J. and Beggs J.J. (1989). What drives Australian teenage labour force participation? Australian Journal of Statistics, 31 A, 125-142.

Lewis, P.E.T. and Koshy P. (1999). Youth employment, unemployment and school completion. Australian Journal of Education, 43 (1), 42-57.

Long, M. Frigo, T. and Batten M. (1999). The School to Work Transition of Indigenous Australians: A Review of the Literature and Statistical Analysis. Canberra: Department of Education, Training and Youth Affairs. Retrieved 1 July, 2003, http://www.dest. gov.au/schools/indigenous/publications/schooltowork/.

Ministerial Council on Education, Employment, Training and Youth Affairs. National Report on Schooling in Australia, various years. Melbourne: Curriculum Corporation.

Ministerial Council on Education, Employment, Training and Youth Affairs (2001). Stepping Forward--improving pathways for all young people. Canberra: Ministerial Council on Education, Employment, Training and Youth Affairs.

Morgan, R. (1995). National Project on Repeating Year Twelve Students. Canberra: Department of Employment, Education and Training.

National Centre for Vocational Education Research (2000, 2001). Australian Vocational Education and Training Statistics: In detail. Adelaide: National Centre for Vocational Education Research.

Newey, W.K. and West, K.D. (1987). A simple, positive-definite heteroskedasticity and autocorrelation consistent covariance matrix. Econometrica, 55, 703-8.

Russell, J. (1993). Post-Compulsory Education and Training Arrangements in the Australian States and Territories: National Board of Employment, Education and Training Report. Canberra: Australian Government Publishing Service.

Steering Committee for the Review of Commonwealth/State Service Provision (2000). Report on Government Services 2000. Canberra: AusInfo.

Steering Committee for the Review of Commonwealth/State Service Provision (2003). Report on Government Services 2003. Canberra: AusInfo.

Trethewey, L (1997). Beginning school in the 1970s and 80s: the (re-) introduction of continuous admission in South Australia. New Zealand Journal of Educational Studies, 32, 175-89.

Vickers, M. and Lamb S. (2002). Why state policies matter: The influence of curriculum policy on participation in post-compulsory education and training. Australian Journal of Education, 46(2), 172-88.

Chris Ryan

Australian National University

Louise Watson

University of Canberra

Notes

(1) Other factors, such as ungraded enrolments, may also result in the mismeasurement of retention rates.

(2) Governments are also developing performance measures across a broader range of outcomes, such as literacy and numeracy; student participation and attainment; vocational education and training (VET) in schools; science and information and communications technology (Ministerial Council on Education, Employment, Training and Youth Affairs, 2000; Steering Committee for the Review of Commonwealth/State Service Provision, 2003). The Australian Bureau of Statistics (ABS) has also published a broader set of education system 'performance' indicators (ABS, 2002).

(3) We do not adjust for full fee-paying overseas students directly and the extent to which full fee-paying overseas students are captured in the population data is unclear because only permanent and long-term migrants (those intending to remain in Australia for more than 12 months) are included in the resident population estimates. No data series are available to make any direct adjustment. However, our view is that full fee-paying overseas are unlikely to affect comparisons between jurisdictions as much as the other factors identified in the Steering Committee for the Review of Commonwealth/State Service Provision list.

(4) Other data sources used in the paper include the ABS publications: Labour Force Australia (Cat. No. 6201.0); Population by Age and Sex (Cat. No. 3201.0); and Participation in Education (Cat. No. 6272.0).

(5) For 1991 and earlier years, the oldest age category in the ABS data was 'nineteen years and older'. We used the observed 1992 share of nineteen year olds among students aged 'nineteen years' or 'twenty years or older' in jurisdictions to estimate those aged nineteen years in Year Twelve before 1992.

(6) For example, we estimate Year Twelve repetition added 3.1 percentage points to retention in 1993; Department of Employment, Education and Training (1994: 12) contains an estimate that at that time Year Twelve repetition added about four percentage points to the national retention estimate in 1993. Morgan (1995) contains estimates of Year Twelve repetition in Australian jurisdictions in the early 1990s, the magnitudes of which support earlier estimates in Russell (1993). The estimates were based on figures provided by the Boards of Study. They showed that Year Twelve repetition was highest in South Australia, Tasmania and Queensland, and in those jurisdictions where there was some kind of time series, that repetition had fallen by the mid-1990s from higher rates in the early 1990s. The estimates of Year Twelve repetition derived here are lower than those contained in Morgan (1995), but match the pattern of those estimates.

(7) Standard errors are allowed to vary between jurisdictions and are 'robust' in the presence of unspecified serial correlation of one lag (Newey & West, 1987).

(8) This variable is likely to be 'endogenous' (it failed Hausman 1978 tests), so we use lagged values of the proportion as instruments for current values in estimation. The equation was estimated via instrumental variable techniques, for this and reasons detailed in the paper.

(9) The equation is estimated by instrumental variables to account for the mismeasurement of Year Twelve repetition.

(10) There are other ways of characterising differences in the age distributions of cohorts, but their use does not change the results.

(11) The proportion aged twelve in Year Eight also fell in Northern Territory and may reflect practice imported from South Australia, given the historical links between the school systems of the two jurisdictions.

(12) Other changes also took place to education systems at the upper secondary school level across jurisdictions that may have affected Year Twelve retention, such as the introduction of Certificates of Education in Victoria, South Australia, the Northern Territory and Tasmania. Lamb (1996, 1998) reported that Year Twelve repetition increased substantially in the late 1980s and early 1990s in South Australia and the Northern Territory following changes in certification requirements. Vickers and Lamb (2002) show that differences in educational structures between Australian jurisdictions explained divergent patterns of school retention in the late 1990s.

(13) The change involved reclassifying secondary-aged students in remote Aboriginal schools to an 'ungraded' category from 1995. This lowered the Indigenous share of the 1995 Year Eight cohort by ten percentage points compared with the 1994 cohort. The apparent retention rate increased by ten percentage in the Northern Territory between 1998 and 1999.

(14) The adjustments to retention are based on parameters which are estimated, and hence subject to uncertainty about their actual magnitude. The adjustments for Queensland, for example, associated with Year Twelve repetition, the age-grade structure and the Indigenous share are estimated to be 1.4, 11.6 and--3.2 percentage points respectively. The ninety-five per cent confidence bands around these adjustments are 0.6 to 2.2, 8.1 to 15.1 and -5.8 to -0.7, respectively.

(15) An alternative would have been to assign the Australian averages for students aged twelve in Year Eight and the Indigenous share to jurisdictions. This would re-scale the estimates and centre them on the national estimate from column five of Table 1, but not affect the relativities between jurisdictions.

(16) Retention to Year Twelve is not the only school participation measure of interest to policy makers. When we replicated the regression approach with three other dependent variables: school participation at age seventeen; school participation among fifteen to nineteen year olds; and Year Twelve completion, a similar hierarchy of performance across jurisdictions was apparent.

(17) The TAFE and Indigenous share variables are excluded from the figure, since they contributed little to changes in national retention over this period. The Year Twelve repetition variable is also excluded, since it affects the measurement of retention, but not changes in the underlying rate.

(18) This is calculated as the estimated parameter from the regression equation multiplied by the proportion of students aged twelve years in Year Eight.

Chris Ryan is a Research Fellow, Social Policy Evaluation, Analysis and Research Centre, Research School of Social Sciences, Australian National University, ACT 0200. Email Chris.Ryan@anu.edu.au

Louise Watson is Associate Professor and Principal Researcher, Lifelong Learning Network, Division of Communication and Education, University of Canberra, ACT 2601. Email louise.Watson@canberra.edu.au

Table 1 Adjusted Year Twelve retention rates in 2002

Column
No. (1) (2) (3) (4)

 Adjusted retention rates

 Direct adjustments

 Cohort
 Official Year 8 age Population
 estimates denominator restriction correction

NSW 69.9 69.9 69.7 67.0
Vic 80.9 80.5 79.9 75.4
Qld 81.3 81.3 80.4 76.2
SA 66.7 66.7 65.7 64.2
WA 73.7 73.7 72.5 69.2
Tas 72.6 73.1 65.7 67.5
NT 53.0 53.0 52.8 54.0
ACT 88.1 88.4 87.9 80.6
AUST 75.1 75.1 74.3 71.1

Column
No. (5) (6) (7)

 Adjusted retention rates

 Regression-based adjustments

 Year
 Twelve Age-grade indigenous
 repetition structure share

NSW 66.4 65.7 67.6
Vic 74.1 73.7 74.2
Qld 74.8 63.2 66.4
SA 62.3 60.1 61.9
WA 66.5 52.1 55.6
Tas 61.4 61.1 64.9
NT 54.0 50.6 69.7
ACT 80.6 80.3 81.3
AUST 69.6 65.3 67.4

Table 2 Age-grade structures and their effect
on age participation rates, 2002

 Age 17 participation rate

 Age-grade
 Estimated structure Adjusted
 rate adjustment rate

NSW 66.3 -0.8 67.1
Vic 76.1 -0.4 76.5
Qld 51.1 -13.8 64.9
SA 60.1 -2.6 62.7
WA 41.6 -17.1 58.7
Tas 63.1 -0.4 63.5
NT 43.9 -4.0 47.9
ACT 88.9 -0.3 89.2
AUST 62.8 -5.1 67.9

 15 to 19 years participation rate

 Age-grade
 Estimated structure Adjusted
 rate adjustment rate

NSW 50.2 -0.2 50.4
Vic 55.4 -0.1 55.5
Qld 45.9 -3.9 49.8
SA 48.8 -0.7 49.5
WA 43.2 -4.8 48.0
Tas 53.3 -0.1 53.4
NT 40.9 -1.1 42.0
ACT 61.4 -0.1 61.5
AUST 50.0 -1.4 51.4