The widening socio-economic gap in UK higher education.

Galindo-Rueda, Fernando ; Marcenaro-Gutierrez, Oscar ; Vignoles, Anna 等

This paper provides up-to-date empirical evidence on the socio-economic gap in higher education (HE) participation, for the period spanning the introduction of tuition fees. We assess whether the gap has widened and ask whether the socioeconomic gap emerges on entry into university or much earlier in the education system. We do this in two ways. Firstly we consider the likelihood of going to university for school leavers in poor neighbourhoods and analyse changes in this likelihood over time. Secondly, we use more detailed individual level data to model the determinants of HE participation, focusing on changes in the relationship between family background and HE participation over time. We find that the growth in HE participation amongst poorer students has been remarkably high, mainly because it was starting from such a low base. However, the gap between rich and poor, in terms of HE participation, has widened during the 1990s. Children from poor neighbourhoods have become relatively less likely to participate in HE since 1994/5, as compared to children from richer neighbourhoods. This trend started before the introduction of tuition fees. Much of the class difference in HE participation seems to reflect inequalities at earlier stages of the education system.

I. Introduction

In 1998, up-front tuition fees were introduced for degree courses in the UK. Although poorer students are exempt from such fees, or at least pay lower amounts, many commentators predicted that fees would cause a fall-off in the number of students entering HE. Furthermore, student maintenance grants were reduced, and then abolished (1) the following year (1999). Critics argued that this de facto increase in the costs of attending university, would reduce the number of applicants to HE (2) and that students from the poorest groups in society would be most likely to be put off, assuming that poorer students tend to be more debt and risk averse, and credit constrained (Callender, 2003).

On the other hand, many economists have argued that, quite apart from the efficiency and equity arguments in favour of the introduction of tuition fees for HE (see for example Barr, 2002; Barr and Crawford, 1998; Dolton, Greenaway and Vignoles, 1997; Greenaway and Haynes, 2003), the huge wage gains from a degree, combined with a relatively low tuition fee, would be unlikely to put students off going to university. Crudely put, given estimates of the wage gain from having a degree of just under 30 per cent, as compared to having just A levels (Dearden et al., 2002; Sianesi, 2003), it would perhaps be surprising if a 1000 [pounds sterling] per annum tuition fee were to dissuade large numbers from attaining graduate status. Furthermore, the poorest students are exempt from fees anyway.

In any case, it has been argued that the problem of access to HE is in fact not rooted in the HE sector itself. It may be that the gap in educational achievement between poorer and richer students occurs prior to entry into university, i.e. poorer students may be much less likely to get good GCSEs and A-levels in the first place. If this is the case, it is conceivable that tuition fees might actually help to reduce inequality in the system. An argument can be made that if the access problem comes before entry into HE, diverting taxpayers' money away from HE towards earlier stages of our education system could potentially broaden access co higher education.

This paper provides up-to-date empirical evidence on the socio-economic gap in HE participation, for the period spanning the introduction of tuition fees. We assess whether the gap has widened and ask whether the socio-economic gap emerges on entry, into university or much earlier in the education system. We do this in two ways. Firstly we consider the likelihood of going to university for children from poor neighbourhoods and analyse changes in this likelihood over time. Secondly, we use more detailed individual level data to model the determinants of HE participation, focusing on changes in the relationship between family background and HE participation over time.

The paper relates to a burgeoning theoretical literature on educational inequality (Benabou, 1996; Fernandez and Rogerson, 1996). It is also motivated by recent empirical evidence on educational inequality in the UK (3) which suggests that, between the late 1970s and the early 1990s, parental income became a more important determinant of whether an individual went on to higher education. Blanden and Machin (2003) have investigated in some detail the relationship between parental income and higher education participation. They conclude that the expansion of the education system in the 1970s, through to the 1990s, was associated with a widening of the gap in HE participation between rich and poor children. Glennerster (2001) also found evidence of a strengthening of the relationship between social class and the HE participation rate in the 1990s. His data end however in 1998, which of course is when tuition fees were introduced. We build on this literature but focus specifically on the period spanning the introduction of tuition fees in 1998. We also examine the relationship between social class and HE participation, (4) as well as parental income.

The next section gives more detail about the recent changes to higher education policy in the UK. Section 3 discusses the trends in UK higher education participation. In section 4, we discuss our methodology and data. Some neighbourhood evidence on changes in HE participation is considered in section 5. In section 6 we estimate our individual level models of the probability of studying for a degree. Finally, section 7 concludes.

2. Higher education policy in the UK

In recent decades, a number of countries have introduced tuition fees for higher education on efficiency and equity grounds (Woodhall, 2002). In the UK, tuition fees were implemented in 1998. Prior to that time, students did not pay for their higher education courses and there was a means tested grant. Tuition fees are currently payable up front and stand at 1,125 [pounds sterling] per annum. Students whose parents earn more than around 30,000 [pounds sterling] per annum are liable for the full amount. Some exemption is given for students whose parents earn between approximately 21,000 [pounds sterling] and 30,000 [pounds sterling] per annum and students whose parents earn less than 21,000 [pounds sterling] per annum are exempt from fees. Student loans have also replaced the grant system. However, it is worth stressing that the real value of student grants was being eroded well before their abolition in the late 1990s.

In 2004, the UK parliament narrowly passed legislation to make further changes to the funding of higher education. Variable tuition fees have been proposed, i.e. fees that vary both by course and by institution. But perhaps the most important feature of the White Paper proposals, from the perspective of widening access, is that fees will be repaid after graduation via an income contingent loan system, and grants will be restored to low-income students. The empirical evidence presented here, which focuses on the period spanning the introduction of up-front tuition fees, will not therefore necessarily apply to this new income contingent loan scheme due to be introduced in 2005.

3. Trends in higher education participation

The introduction of tuition fees does not appear to have had a major impact on aggregate HE participation. (5) Charts 1 and 2 show the simple trend of first-degree students (all years including new entrants and full and part-time students), (6) by gender and over time, for England and Wales, and Scotland respectively. (7) A slight stagnation of the upward trend in student numbers is evident in all countries, following the introduction of tuition fees. However, the trend then resumes its upward path.

[GRAPHICS OMITTED]

Access to HE in the UK has always been predominantly limited to those from higher socio-economic groups. Certainly if one looks at the very top and bottom of the socio-economic scale, the situation is dire. More than three quarters of individuals from professional backgrounds study for a degree compared to just 15 per cent of those from unskilled backgrounds. Moreover, this inequality in the HE system has persisted over the past forty years.

An analysis of the DfES age participation index (8) (table 1) suggests a rise in participation by all socio-economic groups during the 1990s and a marginal widening of the gap in participation rates between richer and poorer students. The gap in the participation by the highest (A-C1) and lowest (C2-E) social groups is around 30 percentage points in 1996, prior to the introduction of fees, and 31 percentage points in 2001. These raw data do show some changes around the time of the introduction of tuition fees. There was a noticeable fall in participation in 1997 and 1998 (the former, possibly, in anticipation of the fees), before participation started rising again in 1999.

4. Methodology and data

From a methodological perspective, determining the causal impact of tuition fees on the demand for HE is problematic. Up-front tuition fees were introduced universally across England and Wales (and simultaneously student maintenance grants abolished). There was no 'experiment' to determine the impact of fees, for example by introducing tuition fees in some areas but not others. (9) Simply looking at student numbers before and after the introduction of tuition fees is likely to be informative but quite problematic, given that there has been a secular rise in the number of HE entrants over the past 30 years. We use two sources of data to address this issue in a largely descriptive way.

4.1 Higher Education Statistics Agency data

The time series data we use come from the Higher Education Statistics Agency, and cover the period 1994-2001. The data set includes limited information (gender, ethnicity, university, degree subject, home postcode, etc.) on all students in HE. We focus specifically on 18-24 year olds enrolled (either part time or full time) in HE. As we are most interested in the impact of tuition fees on the participation of poorer students, and as the HESA data set does not contain information on the income or social class of the student's parents, we use the students' home postcodes as indicators of their socio-economic status. These postcodes are almost always the postcodes of the students' parents and thus reflect the neighbourhoods in which the people grew up.

Specifically, we have merged CACI Paycheck household income data into the HESA database to construct a postcode level dataset, on the basis of each student's home postcode. CACI data are derived from a commercially produced data set, based on over 4 million households. (10) This data set can provide us with an estimate of the income distribution of each postcode sector. (11) Furthermore, for each postcode sector, we obtained from the 2001 Census the proportion of heads of household (age 35 years or older) who classify themselves as being in socio-economic group I or II (professional or intermediate) and socio-economic group III (supervisory, clerical, junior managerial or administrative). This gives us separate indicators of the socio-economic profile of each neighbourhood.

Obviously the number of young people actually in higher education in each postcode will vary according to the total population of young people in each postcode. Obtaining annual estimates of the number of young people in each postcode proved impossible. We therefore have to rely on various population estimates from one point in time, namely the 2001 Census. This ignores any year-on-year population changes in each postcode. Two main population estimates were used. Firstly, we used the number of 18-24 year olds living in a particular postcode. However, neighbourhoods located near Colleges and Universities will tend to have a large 18-24 year old population, even if most of these students are not from that particular postcode originally. We therefore also used Census data on the number of 10-15 year olds living in each postcode as an alternative estimate of the number of young people who originated from that postcode and who potentially could go on to HE. Nonetheless, the analysis is necessarily limited by the fact that we cannot know exactly the number of young people who could potentially enter HE, and how this changed from year to year during the period.

Another issue is that our indicators of the socioeconomic composition of each postcode are not time varying. For example, we only have income data for each postcode for two years (1996 and 1999). We use the more recent 1999 data and have to assume some stability in the income distribution across different postcodes over the 8-year period. Comparisons of the 1996 and 1999 CACI data suggest that this assumption is reasonable. We will, however, have introduced measurement error into both an explanatory variable (the income level of the neighbourhood) and the dependent variable. (12) The mean neighbourhood household income level is 21.89 (i.e. 21,890 [pounds sterling] per annum in 1999 prices) with a standard deviation of 5.6 (5,600 [pounds sterling]). The mean proportion of heads of household that are in groups I/II is 20 per cent (s.d. 0.089) and in group III is 30 per cent (s.d. 0.054).

4.2. Youth Cohort Study data

The second data set we use for our micro-analysis is the Youth Cohort Study (YCS), which is a series of longitudinal surveys conducted by the Department for Education and Skills. The surveys are of a particular academic year group or 'cohort', and are carried out contacting cohort members by post three times, at yearly intervals, when they were 16/17, 17/18 and 18/ 19. Respondents are first surveyed in the year after they are eligible to leave compulsory schooling. They are then followed up, generally over a two-year period. (13) The data collected include information about the economic status of the young person, and in particular whether they have entered higher education by age 18/ 19, as well as their educational background, qualifications, family background and other socioeconomic indicators. The survey is nationally representative (England and Wales) and the sample size of each cohort is around 20,000.

To date ten cohorts of young people have been surveyed. We concentrate on Cohort 7, consisting of individuals who were aged 18 in 1996, and Cohort 9, who were aged 18 in 2000. This spans the period during which tuition fees were introduced.

The YCS has been used extensively as a resource to analyse educational outcomes and subsequent transitions into the labour market (Dolton et al., 2001; Payne et al., 1996, Payne, 2001; Rice, 1999). It is not without its faults however. It lacks measures of parental income, and thus we focus on the impact of parental social class (14) as our main proxy indicator for family background (along with parental education). Attrition is the major problem in the YCS, and there has been extensive academic research on this issue (Lynn, 1996). Cohort 9 started out with an initial target sample size of 22,500. In the first survey at age 16/17, the response rate was 65 per cent. A similar response rate was also achieved in the 17-year-old and 18-year-old surveys. This means that the 18-year-old sample constitutes only 28 per cent of the initial sample (6,304 young people). Furthermore, the attrition rate is considerably higher in the Cohort 9 age 18 survey, than in the Cohort 7 age 18 survey, leading to a smaller sample size for Cohort 9. This is an issue we return to later when we present the descriptive statistics for the YCS samples and discuss our results. Here we simply note that the data are re-weighted to allow for attrition and to bring them in line with population estimates, and our results are robust to re-weighting.

5. Neighbourhood income levels and changes in HE participation

We turn now to our postcode analyses. We want to explain why more young people from one postcode enter HE, as compared to other postcodes, and how this has changed over the period. In particular we want to assess whether more pupils from richer postcodes are enrolled in HE and whether the relationship between postcode income and HE participation has changed over time.

Table 2 presents various regression models that seek to explain differences in HE participation across postcodes and time. The dependent variable is the natural logarithm of the number of 18-24 year olds (15) enrolled in HE. Here we only report the coefficients on postcode income measures, since our focus is on the nature of the relationship between neighbourhood income and the number of students going into HE. (16) However, the model also controls for other variables depending on the specification, as we now discuss.

All of the specifications include time dummies (base case is 1994/1995), allowing for national trends in HE participation over time. In Column 1 the only other control is the population estimate of the number of 18-24 year olds living in each postcode. Thus we allow for the fact that postcodes with more young people are likely to have more of them enrolled in HE. Column 2 includes the same variables but using a fixed effects formulation, thus focusing on changes within postcodes over time. This should remove differences in HE participation across postcodes that are down to characteristics of the postcode that we do not observe and are not explicitly included in our model.

Column 3 then includes interactions between the number of 18-24 year olds in the postcode and each year variable. Thus this specification tests whether 'larger' postcodes, i.e. those with more young people, experience a different trend in HE participation over time, as compared to smaller postcodes. In other words, it is an attempt to go some way to overcome the problem we mentioned earlier, namely that we do not have annual estimates of the population in each postcode.

Column 4 then adds terms indicating the qualification level in each postcode, i.e. the proportion in each postcode with a level 4 or 5 qualification (degree or above). This is to determine whether any positive relationship between neighbourhood income and HE participation is simply down to the fact that richer postcodes have higher proportions of more educated individuals who may be more likely to encourage their children to go on to HE.

Lastly, to check the robustness of the results, Column 5 uses an alternative measure of population, namely the number of 10-15 year olds in each postcode.

The basic OLS model in Column 1 suggests higher income postcodes have a higher number of young people enrolled in HE, as one would expect. What are of greater interest however, from a policy perspective, are the interactions between the income and year variables. Have richer postcodes experienced a more rapid increase in the number of young people going into HE, as compared to poorer neighbourhoods? These interaction terms are generally insignificant in this OLS model. In Column 2 however, once one allows for other differences across postcodes by using a fixed effects model, the interaction terms become always positive and significant. There is an increase in the coefficients by and large through to 1999, with some reduction thereafter. This suggests that for the early part of the period, richer neighbourhoods experienced a somewhat more rapid increase in the number of 18-24 year olds enrolled in HE over the period.

This same pattern is maintained in the other specifications, i.e. even when one allows for interactions between the number of young people in a postcode and the time dummies, and the qualification levels in each postcode. Using alternative measures of the population of young people (the number of 10-15 year olds in each postcode) does reduce the magnitude of the coefficients somewhat but still suggests that postcode income level was more positively associated with HE participation in 1996 through to 2000, as compared to 1994. We also undertook a similar analysis using a different measure of the socio-economic profile of the neighbourhood, namely the proportion of heads of household who classify themselves as being in socio-economic group I or II (professional or intermediate) and the proportion in socio-economic group III These results gave qualitatively similar results. (17)

This can be shown more clearly in chart 3. For this figure, we calculated the ratio of the number of students enrolled in HE to the number of 18-24 year olds in that postcode. We then show the trend in this ratio for postcodes from the top, middle and bottom quintiles of the postcode household income distribution. The gap in the ratio for rich, middle and poor postcodes is normalised to zero for 1994. Chart 3 then clearly shows the widening gap in HE participation between postcodes from the top household income quintile and those from the middle or bottom quintiles. This widening of the gap predates the introduction of fees and appears to be part of a longer-term trend dating back at least to the early 1990s. A separate analysis using the same methodology was used to analyse separately the number of students enrolled in 'new' and 'old' universities (18) in charts 4 and 5. A similar pattern is observed; however the gap between rich and poor postcodes is greater for the old universities. This would suggest that individuals from poorer neighbourhoods are not only less likely to go into HE in the first place but also make different choices of institution, an issue that merits further research.

[GRAPHICS OMITTED]

We undertook a number of further robustness checks. Firstly, we used different measures of postcode income. For example, we used the ranking of the postcode in the income distribution, which should abstract from any changes in the shape of the income distribution over this time period. We also used income quintiles to indicate the income level of each postcode, as in chart 3. Both these methods yielded qualitatively similar results.

The timing of these trends suggests that causes other than tuition fees may well be responsible. The continuing decline in the real value of student grants might be one possible culprit. However, in this analysis we cannot talk about causality. For example, we do not control for students' prior attainment (at area level) and thus we cannot be sure whether we are observing increasing socio-economic inequality on entry into HE or the results of increasing inequality emerging far earlier in the education system. Therefore, we now turn to our micro-analysis of the YCS data in an attempt to identify more precisely the determinants of HE participation over the period in question.

6. Micro-analysis of the determinants of HE participation

Descriptive statistics for the YCS sample are shown in table 3. The first columns describe the samples participating in higher education and not participating in HE for the cohort aged 18 in 1996. Of those aged 18 in 1996, 32.8 per cent were in higher education. The second set of columns provides the same information for the cohort aged 18 in 2000, of which 40 per cent were participating in higher education.

Even over this relatively short period of time, there have been some changes in the characteristics of those participating in higher education. Of those participating in HE in 1996, some 29 per cent were from a professional, managerial or technical background. By 2000, this had risen to 38 per cent. Over the period there was a decline in the proportion of HE students from skilled manual, semi-skilled and unskilled backgrounds. Some of the changes observed are very substantial. Eight per cent of those participating in HE in 1996 had a father with a degree: by 2000 this had risen to 12 per cent, and a similar trend is observed for the proportion of students whose mother had a degree. Whilst 55 per cent of those participating in HE in 1996 had 3 or more A levels, this proportion had risen to 72 per cent for those in HE in 2000. Of course this latter trend may reflect rising A level achievement across the board, as much as a change in the composition of the HE student body. We suspect however, that the higher attrition rate for Cohort 9 may also explain some of the changes we observe. We discuss how we deal with this problem below, when we present the results of our modelling.

Table 4 shows the marginal effects from a probit model, where the dependent variable took a value of one if the person was in higher education at age 18 and zero if he or she was not. (19) Individuals not participating in HE could be in various states, either in or out of the labour market, or studying for lower level qualifications. The model is estimated separately for the two cohorts. Specification 1 shows the impact on HE participation of gender, socio-economic background, ethnicity, parental education and school type. (20) The purpose of this model is to measure the maximum possible impact from an individual's socio-economic background on the likelihood of HE participation and we therefore do not include the individual's level of prior academic achievement. In this model, family background may act directly on the decision to enter HE at 18 or indirectly via lower academic achievement prior to HE. It is nonetheless an important policy question, as to whether family background impacts on pupils' educational outcomes earlier in the education system or only on entry into HE. (21) We test this argument in specification 2 by including measures of the student's prior academic achievement, namely the number of A levels held.

Our results from table 4 show quite clearly that a student's socio-economic background has a large impact on the probability of participating in HE when one does not control for prior educational achievement. Thus, in column 1, the results suggest that students from a professional/managerial or technical background are almost 3 percentage points more likely to be participating in HE in 1996. The impact from these socio-economic background variables largely disappears once indicators of the students' achievements at A level are included, although the coefficient on the 'other' category remains highly negatively significant. In other words, in 1996, for a given level of achievement at age 18, being from the upper socio-economic groups does not appear to have an additional impact on the decision to go to university.

By 2000, however, the situation appears to have changed somewhat. Not only is the impact of coming from a professional/managerial or technical background larger (12 percentage points) but it remains sizeable (6 percentage points) and significant even when A level measures are included in the model. A similar pattern is observed for students with non-manual backgrounds.

At the other end of the distribution, in 1996, students with unskilled parents were not significantly less likely to attend HE compared to students from a skilled background. By 2000, students from an unskilled background were 10 percentage points less likely to attend HE, not controlling for prior attainment. Once achievement at A level was included in the model however, the negative impact of coming from an unskilled background became insignificant.

The impacts of other family background characteristics are also of interest, although we cannot comment on them all for reasons of space. (22) Noteworthy changes include a small decline in the impact of parental education on HE participation, and a reduction in the impact from attending a selective (grammar) school. Of course one could argue that school type is endogenous, reflecting pupil prior attainment and choices made by different pupils and families. In specifications without the school type variables, the impact of socio-economic background increases, as one would expect. However, the general pattern of a large increase over the period in the impact from coming from a higher socio-economic background on HE participation remains with or without these additional controls. Moving to the specifications that control for attainment at A level, it appears that the impact of qualifications attained prior to entry to HE became only marginally more important in 2000 than in 1996.

We conclude that over the period there was an increase in the impact of the social class variables and that this increased impact remained even after controlling for A level attainment and school type. There is however a potential problem with these results. As we have discussed, the sample from Cohort 7 is substantially larger by age 18 than is the case for Cohort 9, reflecting the different attrition rate across the two cohorts. Thus the rather significant changes in the coefficients (in particular the rise in the magnitude of the coefficients on the socio-economic background variables and reduction in the importance of parental education) may partly reflect attrition bias caused by students from poorer backgrounds dropping out of the sample to a greater extent in the Cohort 9 survey. There is some evidence to support this in the descriptive statistics. One possible solution to this problem is to restrict the sample to a more homogenous group of higher attaining pupils, where the attrition problem is less of an issue. We therefore re-estimate our models, restricting the sample to only those who achieved five good GCSEs, as discussed below.

6.1. GCSE comparator group

Of course a legitimate question is whether the comparator group used for the analysis in table 4 is appropriate anyway. It could be argued that we should restrict our comparator group to those who are potentially able and qualified to go on to HE. Restricting our sample to those with five or more good GCSEs would appear to be appropriate. (23) Increasingly individuals enter HE without A levels and thus limiting the sample to only those with A levels is perhaps too restrictive. However, our results remain qualitatively similar even when the comparator group comprises all individuals with one or more A levels. Table 5 presents a summary table of the results from this more restricted sample; (24) however, before discussing these we need to address the problems of potential ability bias.

One of the weaknesses of the YCS data, as compared to other British cohort data sets, is that we have no independent measures of ability. Thus it is possible to argue that our socio-economic background variables are subject to ability bias. To test this we re-estimated our specifications including proxy indicators of ability, namely GCSE mathematics and English grade, (25) as shown in specification 3 in table 5. (26) We recognise that GCSE grades are potentially endogenous variables, as for that matter are our measures of attainment at A level. Students may well not apply as much effort to their GCSE studies if they do not expect to stay on at school and take A levels. Likewise students may not take many A levels if they do not expect to go on to HE. (27) However, from a policy perspective it is useful to have an indication of when the socio-economic inequality in the education system emerges. By controlling for GCSE grade score, we can ask the question whether, for a given level of achievement at GCSE and attainment at A level, students from lower socio-economic groups are less likely to participate in HE. (28)

Table 5 confirms the pattern of results in table 4, namely that we cannot find significant social class effects in 1996 but by 2000 some coefficients on the socioeconomic status variables become positive and significant. However, when we add in our (endogenous) measures of GCSE grade performance, the impact of the social class variables becomes smaller and insignificant, even in 2000. As in table 4, the effect of attainment at A level became somewhat more important by 2000, even when the comparator group is more restricted and even in the specifications that control for GCSE grades. The GCSE grade variables themselves are highly positively significant, as one would expect. However, there is a small diminution of the effect from both mathematics and English GCSE grades over the period.

6.2. Changes over time

The previous section suggested that there has been increasing income-driven inequality in HE participation over a longer period of time, i.e. from 1994/5 onwards, and not simply after the introduction of tuition fees. This is supported by other evidence of a long-term increase in income-driven inequality in HE in the UK (Blanden and Machin, 2003; Machin and Vignoles, 2004). Blanden and Machin (2003) find that, even after controlling for attainment at A level, there was an increasing effect from parental income on HE participation from the 1970s to the 1990s. Our YCS results by contrast do not indicate an independent effect from socio-economic background on HE participation, conditional on earlier educational attainment, even by 1996. However, since our YCS models assess the impact of socio-economic background, whilst Blanden and Machin use parental income, comparisons between the two pieces of research are problematic. It is quite conceivable that the relatively small impact from parental income found by Blanden and Machin (2003) is not being picked up by our broader measures of family background.

7. Conclusions

Children from all socio-economic backgrounds are considerably more likely to go to university in 2001, as compared to 1994. In fact the growth in HE participation amongst poorer students has been remarkably high, mainly because they were starting from such a low base. Nonetheless our results suggest that poorer neighbourhoods (postcodes) saw a less rapid growth in the number of young people enrolled in HE as compared to richer neighbourhoods, particularly in the early and mid-1990s. The strength of the relationship between neighbourhood income levels and HE enrolment grew most rapidly in the early part of the period, rather than after the introduction of tuition fees. This would seem to imply that any income-driven inequality in HE is part of a longer-term trend, perhaps related to the gradual reduction in student support in HE and the big expansion of the university sector that occurred in the early 1990s. (29)

Our detailed individual level analysis suggests that in 1996, i.e. before tuition fees, there was certainly substantial social class educational inequality in HE but that it occurred largely as a result of inequalities earlier in the education system. By 2000 however, one can observe social class effects on HE participation, even after conditioning for the number of GCSEs and A levels that an individual has. This seems to suggest a widening of the social class gap in higher education itself in the period after the introduction of tuition fees. However, in models that include finer measures of educational achievement, the social class effects become smaller and insignificant. We conclude that much of the impact from social class on university attendance actually occurs well before entry into HE. Of course just because we observe inequality in attainment at earlier ages, does not mean it is not related to problems in HE. Students may look forward and anticipate barriers to participation in HE and make less effort in school as a result. This is an area that requires further research.

Appendix A. Probability of getting five or more
good GCSEs (marginal effects)

 1996 2000

Male -0.058 -0.060
 (5.07) *** (5.13) ***
Parents' Socio-economic status:
 Professional, managerial &
 tech. occ. 0.056 0.126
 (3.53) *** (7.98) ***
 Other non-manual occ. 0.023 0.085
 (1.42) (5.52) ***
 Semi-skilled occ.-manual -0.052 -0.067
 (2.66) *** (3.32) ***
 Unskilled occ. -0.082 -0.191
 (2.36) ** (5.39) ***
 Other -0.067 -0.098
 (3.45) *** (3.87) ***
Ethnicity:
 Black -0.191 -0.078
 (3.07) *** (1.59)
 Asian -0.001 0.013
 (0.05) (0.57)
 Other ethnicity -0.140 -0.031
 (1.88) * (0.64)
 Ethnicity missing flag -0.100 -0.073
 (1.65) * (1.17)
Parental education:
 Father degree 0.141 -0.018
 (4.63) *** (0.83)
 Father at least one A level 0.171 0.011
 (10.42) *** (0.54)
 Father education missing flag -0.057 -0.114
 (3.46) *** (6.86) ***
 Mother degree 0.139 0.082
 (3.45) *** (4.47) ***
 Mother at least one A level 0.132 0.132
 (7.97) *** (7.06) ***
 Mother education missing flag -0.047 -0.043
 (2.76) *** (2.61) ***
Type of school attended:
 Comprehensive age 16 -0.018 -0.034
 (1.46) (2.75) ***
 Grammar 0.337 0.244
 (10.96) *** (8.29) ***
 Secondary modern -0.153 -0.099
 (4.87) *** (2.85) ***
 Independent 0.295 0.176
 (12.84) *** (7.93) ***
-2 ([log.sub.R] - [log.sub.U]) 1345.5 *** 1033.3 ***
Observations 7985 6268

Source: Youth Cohort Survey Full Sample.

Notes: Dependent variable--value of one if 5+ Good GCSEs, zero
otherwise, probit estimation. Base case: skilled background,
white, father/mother's education less than A level, attended a
comprehensive. Absolute value of t statistics in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%.

Table 1. Age participation index (API) (%) by social class,
1991/2-2001

 Year of Entry

Class 1992 1993 1994 1995 1996

Professional (A) 71 73 78 80 82
Intermediate (B) 39 42 45 46 47
Skilled non-manual (C1) 27 29 31 31 32
Skilled manual (C2) 15 17 18 18 18
Partly skilled (D) 14 16 17 17 17
Unskilled (E) 9 11 11 12 13
A-C1 40 43 46 47 48
C2-E 14 16 17 17 18

 Year of Entry

Class 1997 1998 1999 2000 2001

Professional (A) 79 72 73 76 79
Intermediate (B) 48 45 45 48 50
Skilled non-manual (C1) 31 29 30 33 33
Skilled manual (C2) 19 18 18 19 21
Partly skilled (D) 18 17 17 19 18
Unskilled (E) 14 13 13 14 15
A-C1 48 45 45 48 50
C2-E 18 17 17 18 19

Source: Department for Education and Skills Age Participation Index
which measures the proportion of the under 21s in each social class
participating in Higher Education for the first time (i.e. young
entrants from each social class as a percentage of all young
people in each social class).

Table 2. Postcode level analysis of the number of students attending HE

 Dependent variable:
 Ln (Number of HE students)
 in each postcode

 OLS Fixed Effects

Ln (Income) 1.333 --
 (36.95) ***
Interaction Ln (Income) * year 1995 0.096 0.052
 (1.87) * (2.67) ***
Interaction Ln (Income) * year 1996 0.075 0.096
 (1.48) (4.95) ***
Interaction Ln (Income) * year 1997 0.047 0.099
 (0.93) (5.13)***
Interaction Ln (Income) * year 1998 0.029 0.082
 (0.57) (4.27) ***
Interaction Ln (Income) * year 1999 0.071 0.142
 (1.41) (7.37) ***
Interaction Ln (Income) * year 2000 0.027 0.103
 (0.53) (5.37) ***
Interaction Ln (Income) * year 2001 0.017 0.080
 (0.34) (4.18) ***

Controls:
Year dummies Yes Yes
Population aged 18-24 from Census
 (in thousands) Yes Yes
Interactions population aged
 18-24 * years
Ratio population aged older 24
 Qualification level 4/5
Interactions ratio (Population
 qualification level 4/5/
 Population aged older 24) * years
Population aged 10-15 from Census
 (in thousands)
Interactions population aged
 10-15 * years
Constant -1.046 3.465
 (9.46) *** (997.21) ***
Observations 57743 57743
R-squared 0.34 0.40
Number of postcodes -- 7382

 Dependent variable:
 Ln (Number of HE students)
 in each postcode

 Fixed Effects Fixed Effects

Ln (Income) -- --

Interaction Ln (Income) * year 1995 0.061 0.076
 (3.10) *** (2.71) ***
Interaction Ln (Income) * year 1996 0.108 0.165
 (5.60) *** (5.99) ***
Interaction Ln (Income) * year 1997 0.112 0.139
 (5.78) *** (5.04) ***
Interaction Ln (Income) * year 1998 0.089 0.081
 (4.60) *** (2.95) ***
Interaction Ln (Income) * year 1999 0.147 0.136
 (7.62) *** (4.93) ***
Interaction Ln (Income) * year 2000 0.109 0.082
 (5.65) *** (2.97) ***
Interaction Ln (Income) * year 2001 0.087 0.058
 (4.53) *** (2.10) **

Controls:
Year dummies Yes Yes
Population aged 18-24 from Census
 (in thousands) Yes Yes
Interactions population aged
 18-24 * years Yes Yes
Ratio population aged older 24
 Qualification level 4/5 Yes
Interactions ratio (Population
 qualification level 4/5/
 Population aged older 24) * years Yes
Population aged 10-15 from Census
 (in thousands)
Interactions population aged
 10-15 * years
Constant 3.465 3.465
 (997.89) *** (998.12) ***
Observations 57743 57743
R-squared 0.40 0.40
Number of postcodes 7382 7382

 Dependent
 variable: Ln
 (Number of HE
 students) in
 each postcode

 Fixed Effects

Ln (Income) --

Interaction Ln (Income) * year 1995 0.033
 (1.20)
Interaction Ln (Income) * year 1996 0.095
 (3.49) ***
Interaction Ln (Income) * year 1997 0.074
 (2.74) **
Interaction Ln (Income) * year 1998 0.051
 (1.89) *
Interaction Ln (Income) * year 1999 0.113
 (4.18) ***
Interaction Ln (Income) * year 2000 0.058
 (2.14) **
Interaction Ln (Income) * year 2001 0.029
 (1.06)

Controls:
Year dummies Yes
Population aged 18-24 from Census
 (in thousands)
Interactions population aged
 18-24 * years
Ratio population aged older 24
 Qualification level 4/5 Yes
Interactions ratio (Population
 qualification level 4/5/
 Population aged older 24) * years Yes
Population aged 10-15 from Census
 (in thousands) Yes
Interactions population aged
 10-15 * years Yes
Constant 3.465
 (998.10) ***
Observations 57743
R-squared 0.40
Number of postcodes 7382

Sources: Higher Education Statistics Agency.

Notes: Sample restricted to full and part-time students aged 18-24
domiciled in England or Wales enrolled in a first-degree course.
See main paper for other sample restrictions. Absolute values of
t statistics in parentheses. * significant at 10 per cent;
** significant at 5 per cent; *** significant at 1 per cent.

Table 3. Descriptive statistics for YCS sample

 1996

 Participating Not participating
 in HE in HE

Male 0.41 0.43
Parents' socio-economic status:
 Professional, managerial &
 technical occupations 0.29 0.20
 Other non-manual occupations 0.21 0.18
 Skilled occupations -- manual * 0.31 0.34
 Semi-skilled occupations --
 manual 0.08 0.12
 Unskilled occupations 0.02 0.03
 Other 0.08 0.13
Ethnicity:
 White * 0.92 0.92
 Black 0.01 0.01
 Asian 0.07 0.06
 Other 0.01 0.01
Parental education:
 Father degree 0.07 0.03
 Father at least one A level 0.32 0.15
 Father below one A level * 0.61 0.81
 Mother degree 0.04 0.02
 Mother at least one A level 0.31 0.15
 Mother below one A level* 0.65 0.83
Type of school attended:
 Comprehensive age 16 0.24 0.35
 Comprehensive age 18 * 0.49 0.53
 Grammar 0.08 0.03
 Secondary modern 0.02 0.04
 Independent 0.17 0.05
Highest school qualification:
 One or two A levels 0.12 0.11
 Three or more A levels 0.55 0.11
 Five or more A-C GCSEs 0.96 0.47

Observations 2343 5642
 7985

 2000

 Participating Not participating
 in HE in HE

Male 0.39 0.42
Parents' socio-economic status:
 Professional, managerial &
 technical occupations 0.38 0.22
 Other non-manual occupations 0.27 0.21
 Skilled occupations--manual * 0.23 0.33
 Semi-skilled occupations --
 manual 0.07 0.11
 Unskilled occupations 0.01 0.04
 Other 0.04 0.08
Ethnicity:
 White * 0.90 0.91
 Black 0.01 0.02
 Asian 0.08 0.06
 Other 0.01 0.02
Parental education:
 Father degree 0.12 0.09
 Father at least one A level 0.32 0.15
 Father below one A level * 0.56 0.76
 Mother degree 0.16 0.11
 Mother at least one A level 0.23 0.12
 Mother below one A level* 0.61 0.78
Type of school attended:
 Comprehensive age 16 0.24 0.33
 Comprehensive age 18 * 0.51 0.54
 Grammar 0.09 0.04
 Secondary modern 0.01 0.04
 Independent 0.15 0.06
Highest school qualification:
 One or two A levels 0.11 0.13
 Three or more A levels 0.72 0.19
 Five or more A-C GCSEs 0.98 0.57

Observations 2186 4082
 6268

Source: Youth Cohort Survey.

Note: * Base case in subsequent regressions.

Table 4. The determinants of HE participation (marginal effects)

 1996

 Specification 1 Specification 2

Male -0.008 0.000
 (0.78) (0.02)
Parents' socio-economic status:
 Professional, managerial & 0.027 0.009
 technical occupations (1.89) * (0.59)
 Other non-manual occ. 0.022 0.006
 (1.50) (0.43)
 Semi-skilled occupation - manual -0.032 -0.029
 (1.75) ** (1.51)
 Unskilled occ. -0.008 0.009
 (0.25) (0.26)
 Other -0.054 -0.048
 (3.00) *** (2.58) ***
Ethnicity:
 Black -0.044 -0.021
 (0.77) (0.34)
 Asian 0.089 0.109
 (3.86) *** (4.59) ***
 Other ethnicity -0.057 -0.078
 (0.94) (1.30)
 Ethnicity missing flag -0.106 -0.075
 (1.78) * (1.16)
Parental education:
 Father degree 0.107 0.047
 (3.76) *** (1.66) *
 Father at least one A level 0.120 0.069
 (7.94) *** (4.52) ***
 Father education missing flag -0.012 0.002
 (0.73) (0.14)
 Mother degree 0.110 0.060
 (3.05) *** (1.68) *
 Mother at least one A level 0.096 0.063
 (6.49) *** (4.16) ***
 Mother education missing flag -0.055 -0.039
 (3.43) *** (2.38) **
Type of school attended:
 Comprehensive age 16 -0.040 -0.033
 (3.40) *** (2.69) ***
 Grammar 0.251 0.151
 (9.11) *** (5.45) ***
 Secondary modern -0.117 -0.075
 (4.09) *** (2.45) **
 Independent 0.195 0.171
 (9.72) *** (8.44) ***
Highest school qualification:
 One or two A levels 0.190
 (10.61) ***
 Three or more A levels 0.488
 (35.89) ***
-2 ([log.sub.R] - [log.sub.U]) 760.5 *** 2116.81 ***
Observations 7985

 2000

 Specification 1 Specification 2

Male -0.036 -0.009
 (2.91) *** (0.70)
Parents' socio-economic status:
 Professional, managerial & 0.122 0.060
 technical occupations (6.72) *** (3.19) ***
 Other non-manual occ. 0.095 0.046
 (5.23) *** (2.44) **
 Semi-skilled occupation - manual -0.023 -0.024
 (0.96) (0.94)
 Unskilled occ. -0.101 -0.059
 (2.60) *** (1.33)
 Other -0.034 -0.039
 (1.15) (1.25)
Ethnicity:
 Black -0.062 0.011
 (1.17) (0.19)
 Asian 0.130 0.164
 (5.00) *** (5.87) ***
 Other ethnicity -0.065 -0.028
 (1.23) (0.49)
 Ethnicity missing flag 0.007 0.028
 (0.10) (0.36)
Parental education:
 Father degree 0.010 0.022
 (0.46) (0.95)
 Father at least one A level 0.089 0.063
 (4.52) *** (3.08) ***
 Father education missing flag -0.071 -0.020
 (3.90) *** (1.03)
 Mother degree 0.062 0.038
 (3.11) *** (1.84) *
 Mother at least one A level 0.073 0.010
 (3.66) *** (0.48)
 Mother education missing flag -0.062 -0.044
 (3.31) *** (2.21) **
Type of school attended:
 Comprehensive age 16 -0.038 -0.026
 (2.68) *** (1.71) *
 Grammar 0.131 0.005
 (4.74) *** (0.19)
 Secondary modern -0.166 -0.069
 (4.41) *** (1.60)
 Independent 0.155 0.113
 (6.71) *** (4.79) ***
Highest school qualification:
 One or two A levels 0.227
 (10.63) ***
 Three or more A levels 0.514
 (35.18) ***
-2 ([log.sub.R] - [log.sub.U]) 653.4 *** 1998.4
Observations 6268

Source: Youth Cohort Survey Full Sample.

Notes: Dependent variable--value of one if in HE, zero otherwise,
probit estimation. Base case: skilled background, white, father/
mother's education less than A level, attended a comprehensive with
provision up to age 18 and with no A levels (as indicated by * in
table 3). Absolute values of t statistics in parentheses * significant
at 10 per cent: ** significant at 5 per cent; *** significant at 1 per
cent.

Table 5. The determinants of HE participation
(marginal effects) - sample with 5+ good GCSEs

 1996

 Specif. 1 Specif. 2

Parents' socio-economic status:
 Professional, managerial & technical 0.013 0.004
 occ. (0.69) (0.20)
 Other non-manual occupation 0.016 0.004
 (0.77) (0.18)
 Semi-skilled occupation - manual -0.003 -0.009
 (0.11) (0.32)
 Unskilled occupation 0.044 0.054
 (0.84) (1.00)
 Other -0.035 -0.034
 (1.28) (1.19)
GCSE grades:
GCSE maths grade

GCSE math grade missing flag

GCSE English grade

GCSE English grade missing flag

Highest school qualification:
 One or two A levels 0.020
 (0.93)
 Three or more A levels 0.344
 (21.44) ***
-2 ([log.sub.R] - [log.sub.U]) 183.5 *** 703.5 ***
Observations 4883

 1996 2000

 Specif. 3 Specif. 1

Parents' socio-economic status:
 Professional, managerial & technical -0.008 0.088
 occ. (0.39) (4.06) ***
 Other non-manual occupation -0.008 0.074
 (0.38) (3.39) ***
 Semi-skilled occupation - manual -0.004 0.011
 (0.14) (0.35)
 Unskilled occupation 0.056 -0.010
 (1.00) (0.17)
 Other -0.029 0.045
 (1.01) (1.08)
GCSE grades:
GCSE maths grade 0.115
 (13.10) ***
GCSE math grade missing flag -0.138
 (0.73)
GCSE English grade 0.067
 (6.35) ***
GCSE English grade missing flag 0.014
 (0.10)
Highest school qualification:
 One or two A levels 0.026
 (1.18)
 Three or more A levels 0.254
 (14.56) ***
-2 ([log.sub.R] - [log.sub.U]) 3220.38 *** 234.2 ***
Observations

 2000

 Specif. 2 Specif. 3

Parents' socio-economic status:
 Professional, managerial & technical 0.052 0.032
 occ. (2.28) ** (1.36)
 Other non-manual occupation 0.042 0.024
 (1.81) * (1.04)
 Semi-skilled occupation - manual -0.009 -0.011
 (0.28) (0.32)
 Unskilled occupation 0.003 0.009
 (0.04) (0.14)
 Other 0.014 0.011
 (0.31) (0.25)
GCSE grades:
GCSE maths grade 0.074
 (7.97) ***
GCSE math grade missing flag 0.359
 (2.34) **
GCSE English grade 0.050
 (4.66) ***
GCSE English grade missing flag 0.293
 (2.68) ***
Highest school qualification:
 One or two A levels 0.084 0.083
 (3.39) *** (3.32) ***
 Three or more A levels 0.402 0.341
 (22.74) *** (17.83) ***
-2 ([log.sub.R] - [log.sub.U]) 838.7 *** 957.76 ***
Observations 4457

Data: Youth Cohort Survey Sample with at least 5 Good GCSEs.

Notes: Dependent variable--value of one if in HE, zero otherwise,
probit estimation. Base case: skilled background, white, father/
mother's education less than A level, attended a comprehensive and
with no A levels. All specifications also control for gender,
ethnicity, parental education and school type. Absolute values of t
statistics in parentheses * significant at 10 per cent: ** significant
at 5 per cent; *** significant at 1 per cent.

NOTES

(1) Limited student grants were re-introduced in September 2002 for Welsh students. These are means tested and are for up to 1,500 [pounds sterling] per annum.

(2) Given that the supply of HE places is constrained, it is possible for student demand to fall whilst overall student numbers do not. In effect the excess demand for HE might have been reduced by the introduction of tuition fees.

(3) Blanden and Machin (2003), Galindo-Rueda and Vignoles (forthcoming). Machin and Vignoles (2004).

(4) Erickson and Goldthorpe (1992), Saunders (1997) and Schoon et al. (2002) have examined issues relating to education and social mobility, to cite lust a few.

(5) From the Higher Education Statistics Agency.

(6) We also excluded overseas students, those who did not report a domicile postcode, students with missing data on various fields, namely previous qualifications, institution type, qualification aim and degree subject. Full details of these samples are available from the authors. Charts 1 and 2 include students of all ages. For the main model, we restrict our sample to 18-24 year old students enrolled in HE

(7) Scotland is shown separately because both the educational system and the funding regime are different. In Scotland in September 2000, annual tuition fees were replaced by a 2,000 [pounds sterling] graduate contribution repayable after graduation.

(8) Based on the DfES Age Participation Index which measures the proportion of the under 21s in each social class participating in higher education for the first time (i.e. young entrants from each social class as a percentage of all young people in each social class).

(9) Tuition fees were not introduced in Scotland but there were other changes to the funding regime in that country making inter-country comparisons to evaluate the impact of fees on HE participation problematic. However, our analysis of HESA data suggested that the trends observed for England and Wales were broadly similar for Scotland too, and thus could not be explained by tuition fees specifically.

(10) Further details, and an alternative use for these data, can be found in Gibbons (2001).

(11) We grouped the information at postcode sector level as CACI provides data for the first S digits of the postcode only.

(12) We have also dropped misreported or discontinued postcodes, constituting 3.05 per cent of the sample. Where we have missing data on postcode income levels, we use 1996 income data if available or, if that is missing too, we aggregate the data up to the 4-digit postcode level and impute data on the basis of this more aggregated grouping.

(13) Some of the early cohorts have since been followed up to age 21 and beyond.

(14) The YCS provides data on the socioeconomic group to which the individual belongs, based on his or her parents' occupation. We use the term social class for these groupings for ease of exposition.

(15) As has been previously said, we only include students domiciled in England and Wales but we do include all full-time and part-time students.

(16) Full results available on request.

(17) Results available on request.

(18) Using the Guardian newspaper classification of universities into old and new, available on request.

(19) Thus, we are measuring HE participation, not degree attainment. If drop out and degree failure vary by social class, and if these variables have been affected by the introduction of tuition fees, we may be underestimating the impact of social class on HE achievement.

(20) The base case for school type is a student attending a comprehensive with provision up to age 18.

(21) There is a large literature relating to when family background has the greatest impact on pupil attainment at school and the optimal timing of any policy interventions to improve the educational attainment of poor children (Cameron and Heckman, 2001; Carneiro et al. (2003); Currie and Thomas, 1999; Haveman and Wolfe, 1995).

(22) For example, our results are robust to the inclusion of regional fixed effects. Our sample size is not sufficient to estimate the model by region but we are aware of very recent evidence of regional differences in the determinants of education participation (Rice, 2004).

(23) Appendix A shows the results of a probit model estimating the determinants of getting five or more good GCSEs. The impact of parental socio-economic status became more important between 1996 and 2000. This suggests that success in the education system at GCSE level was also becoming more closely linked to family background (see Blanden and Machin, 2003, and Galindo-Rueda and Vignoles, forthcoming, for a further discussion of this trend). Rice (2004) has also conducted a comprehensive analysis of the determinants of staying on at age 16 using YCS data for the whole of the 1990s.

(24) Full results available on request.

(25) Coded as Grade A--5 points, Grade B--4 points, Grade C--3 points, Grade D--2 points, Grade E--I point. Otherwise zero.

(26) These variables were also added to the models on the full sample--results available on request.

(27) Since most people who take at least one A level go on to university, this measure may be somewhat Jess problematic.

(28) We are currently devising an instrument for GCSE attainment at age 16, based on the average GCSE grade of the pupils in the individual's school.

(29) A number of policy changes occurred at this time, most of which caused student numbers in HE to grow. One significant policy change was the merging of 'old' universities and polytechnics into one new category just entitled 'universities'. A disproportionate amount of the growth in HE in the early 1990s occurred in former polytechnics. The same patterns can be observed, however, whether one considers HE participation in old or new universities (evidence on this is available on request).

REFERENCES

Barr, N. (2002), "Funding higher education: policies for access and quality', House of Commons Education and Skills Committee, Post-16 student support, Session 2001-02, 24 April 2002.

Barr, N. and Crawford, I. (1998), 'The Dearing Report and the government response: a critique', Political Quarterly, 69, 1, January-March, pp. 72-84.

Benabou, R. (19%), 'Equity and efficiency in human capital investment: the local connection', Review of Economic Studies, 62, pp. 237-64.

Blanden, J. and Machin, S. (2003), "Educational inequality and the expansion of UK higher education', mimeo. London School of Economics, Centre for Economic Performance.

Callender, C. (2003), 'Student financial support in higher education: access and exclusion', in Tight, M. (ed.), Access and Exclusion: International Perspectives on Higher Education Research, London. Elsevier Science.

Cameron, S. and Heckman, J. (2001), 'The dynamics of educational attainment of black, hispanic and white males', Journal of Political Economy, 109, pp. 455-99.

Carneiro, J., Hansen, K. and Heckman, J. (2003), 'Estimating distribution of counterfactuals with an application to the returns to schooling and measurement of the effect of uncertainty on schooling choice, International Economic Review.

Currie, J. and Thomas, D. (1999), 'Early test scores, socioeconomic status and future outcomes', NBER Working Paper No. 6943, Cambridge, MA., National Bureau of Economic Research.

Dearden, L., McIntosh, S., Myck, M. and Vignoles, A. (2002), 'The returns to academic, vocational and basic skills in Britain', Bulletin of Economic Research, 54, 3, pp. 249-74.

Dolton, P., Greenaway, D. and Vignoles, A. (1997), 'Whither higher education? An economic perspective for the Dearing Committee of Inquiry', Economic Journal, 107, 442, pp.710-26.

Dolton, P.J., Makepeace, G.H., Gannon, B.M. (2001), 'The earnings and employment effects of young people's vocational training in Britain', Manchester School, 69, 4, pp.387-417.

Erickson, R. and Goldthorpe, J. (1992), 'The constant flux: a study of class mobility in industrial societies', Oxford University Press.

Fernandez, R. and Rogerson, R. (1996), 'Income distribution, communities, and the quality of public education', Quarterly Journal of Economics, 111, 1, pp. 135-64.

Galindo-Rueda, F. and Vignoles. A., "The declining relative importance of ability in predicting educational attainment', Journal of Human Resources and CEE Discussion Paper (both forthcoming).

Gibbons, S. (2001), "Paying for good neighbours? Neighbourhood deprivation and the community benefits of education', Centre for the Economics of Education, Discussion Paper No. 17. London School of Economics.

Glennerster, H. (2001), 'United Kingdom education 1997-2001', CASE Paper 50, Centre for the Analysis of Social Exclusion.

Greenaway, D. and Haynes, M. (2003), 'Funding higher education in the UK: the role of fees and loans', Economic Journal, 113, F150-167.

Haveman, R. and Wolfe, B. (1995), 'The determinants of children's attainments: a review of methods and findings', Journal of Economic Literature, 33, 4, pp, 1829-78.

Howieson, C. and Croxford, L. (1996), 'Using the YCS to analyse the outcomes of careers education and guidance', The Stationary Office, DFEE Research Series, London.

Lynn, P. (1996), England and Wales Youth Cohort Study: The Effect of Time Between Contacts, Questionnaire Length, Personalisation and Other Factors on Response to the YCS (Research Studies), London, The Stationery Office Books.

Payne, J. (2001), Patterns of Participation in Full-time Education after 16: An Analysis of the England and Wales Youth Cohort Study, London. Department for Education and Skills.

Payne, J., Cheng, Y. and Witherspoon, S. (1996). Education and Training for 16-18 Year Olds in England and Wales--Individual paths and national trends, London, Policy Studies Institute.

Machin, S. and Vignoles, A. (2004), 'Education inequality', Fiscal Studies, 25, 2, pp.107-28.

Rice, P. (1999), "The impact of local labour markets on investment in further education: evidence from the England and Wales youth cohort studies', Journal of Population Economics, 12, 2, pp. 287-31.

--(2004), 'Education and training post- 16--differences across the British regions', CEE conference, March.

Saunders, P. (1997), "Social mobility in Britain: an empirical evaluation of two competing explanations', Sociology, May.

Sianesi, B. (2003). 'Returns to education: a non-technical summary of CEE work and policy discussion', draft report for Department for Education and Skills.

Schoon, I, Bynner, J., Joshi, H., Parsons, S., Wiggins, RD., and Sacker, A. (2002). 'The influence of context, timing, and duration of risk experiences for the passage from childhood to mid-adulthood'. Child development, Sep-Oct, 73, 5, pp. 1486-1504.

Woodhall. M. (ed.) (2002), 'Paying for learning: the debate on student fees, grants and loans in international perspective', The Welsh Journal of Education, 11. 1.

Fernando Galindo-Rueda, * Oscar Marcenaro-Gutierrez ** and Anna Vignoles ***

* Centre for Economic Performance, London School of Economics and Centre for the Economics of Education. ** Centre for Economic Performance, London School of Economics and Centre for the Economics of Education. *** Institute of Education and Centre for the Economics of Education (e-mail: [email protected]).