The medium-term impact of a conditional cash transfer programme on educational outcomes in England

ABSTRACT

This paper uses longitudinal data from England to examine the medium-term impact of a means-tested conditional cash transfer programme, Education Maintenance Allowance (EMA), on higher educational participation and attainment. Combining regression modelling with entropy balancing, this paper finds that two-year EMA recipients are more likely to participate in higher education than non-recipients. Moreover, the impact of EMA is more substantial for male students, those with higher prior academic attainment, and students whose parents have higher educational qualifications. These findings suggest that even though EMA is a costly programme, it will benefit young people over a longer time frame.

KEYWORDS:

• Educational inequality
• Education Maintenance Allowance (EMA)
• higher education
• socio-economic gaps

JEL CLASSIFICATION CODES:

• I22
• I24
• I28

1. Introduction

The socio-economic gap in post-compulsory education in the UK has been widely discussed (Otero 2007; Thomas 2005; Thompson and Simmons 2013). Although the gap seems to be closing in recent years (Crawford 2012; Higher Education Funding Council for England 2013; Iannelli 2007; Murphy, Scott-Clayton, and Wyness 2017), socio-economic differences in higher education (HE) participation remain substantial (UCAS 2021). Using the 2008 General Certificate of Secondary Education (GCSE)¹ cohort, Crawford and Greaves (2015) find that students from the highest socio-economic quintile group are approximately three times more likely to participate in higher education and seven times more likely to enrol in a selective university than those from the lowest group. Disadvantaged students are, however, more likely to go to further education rather than higher education, studying courses at National Vocational Qualification (NVQ)² Level 3 and below (Department of Education 2018).

The socio-economic gap in degree completion is even more substantial. Previous studies have shown that disadvantaged students have higher chances of dropping out of university than students from wealthier backgrounds even after controlling for personal characteristics, prior attainment and university characteristics (Crawford et al. 2016; Johnes and McNabb 2004; Quinn et al. 2005; Vignoles and Powdthavee 2009). This leads to the socio-economic gap in degree completion. Moreover, socio-economic differences also exist in degree classification as students from higher socio-economic backgrounds are more likely to be awarded a first or upper second-degree³ than those from lower socio-economic backgrounds (Crawford 2014; McNabb, Pal, and Sloane 2002; Smith and Naylor 2001).

Therefore, what motivates this work is that the long-existing socio-economic gap in education remains and continues to pose a challenge for policymakers despite a range of policy measures to close it. This study evaluates the medium-term impact of EMA on the higher education participation and attainment of young people from low-income families eight to nine years after receiving it.

The EMA programme began in pilot areas in 1999 and has been rolled out in England, Scotland, Wales, and Northern Ireland since September 2004.⁴ The scheme was a key component of the government’s education policy to encourage 16-to 19-year-olds to continue their education in specific further education courses beyond compulsory education. Eligibility for the EMA depended on household income thresholds, with the amount of allowance tiered accordingly. Students whose parental income was below £20,818 were eligible for £30 per week, those with parental income between £20,818 and £25,521 received £20 per week, and those with parental income between £25,522 and £30,810 received £10 per week.⁵ To qualify for the allowance, they were required to attend approved educational institutions and courses and to maintain regular attendance to receive the allowance. Payments were made directly to students on a weekly basis, aiming to directly support their educational needs.

At its peak, the EMA programme covered over 500 thousand students per year with an expenditure of over £500 million. Despite a notable increase in post-16 education enrolment observed in the years following its implementation, EMA was discontinued in 2011 in England⁶ as part of wider governmental austerity measuresConsequently, students are now required to stay in education until the age of 18.⁷ Although previous studies have shown that the EMA has positive impacts on participation, retention, and attainment in secondary education (see, for example, Dearden et al. 2009; Spielhofer et al. 2010), there is very limited evidence on the longer-term impact of the programme. Thus, this study conducts a retrospective empirical analysis to ask whether offering EMA is an effective way of narrowing the socio-economic gap in post-secondary education by incentivising young people from disadvantaged backgrounds to participate in higher education, and to improve their performance.

In this study, I estimate a multivariate regression model using the Next Steps study, controlling for a rich range of observed factors such as demographic characteristics, prior attainment, behaviours and attitudes, and school fixed effects to determine the influence of EMA receipt on higher education participation and achievement. Focusing on students from low-income backgrounds, including those with parental incomes slightly above the EMA threshold, I compare EMA recipients to non-recipients. Non-recipients are those who either had incomes too high to be eligible for EMA, or were eligible for EMA but did not receive it, for reasons explained in Section 3. To mitigate the effect of unobserved factors and ensure comparability between treatment and control groups, I use entropy balancing to balance characteristics of the multi-treatment and control groups, including those who received EMA for one and two years, as well as non-recipients.

In addition to the overall regression, the impact of EMA by gender is estimated in this study as existing studies have shown that male students are less likely to participate in post-compulsory education but respond better to EMA than female students (Ashworth et al. 2001; H. Chowdry, Dearden, and Emmerson 2008; Dearden et al. 2005; Middleton et al. 2005). Furthermore, this study also examines the different impacts of EMA by prior attainment and parental education. The results from the model indicate that EMA raises higher education participation among those who received the allowance for two years. The positive impact is especially strong for male students, those with higher prior academic achievement, and students whose parents have advanced educational qualifications. These results provide insights that can inform more efficient policymaking by tailoring financial support programmes to the needs and behaviours of different student demographics.

2. Relevant literature

In order to encourage children from deprived families to stay in education, a number of countries, particularly Latin American countries, have introduced conditional cash transfer (CCT) programmes (see in Table 1). Adato and Hoddinott (2010) and Rawlings and Rubio (2005) review the literature in Latin American and Caribbean (LAC) region and surmise that the Opportunities programme (previously called PROGRESA) in Mexico increases years of schooling by 0.5–0.7 years (Behrman, Sengupta, and Todd 2005; Schultz 2004; Todd and Wolpin 2003), the Red de Protecion Social (RPS) in Nicaragua boosts school attendance for all children aged 7–13 by approximately 20 percentage points (Maluccio and Flores 2005), the Families in Action (FA) in Colombia raises secondary school enrolment rates by 14 and 5.5 percentage points in urban and rural respectively (Attanasio et al. 2004). Evidence from Turkey also suggests that the Social Risk Mitigation Project (SRMP), which aims to increase school attendance rates for disadvantaged students and for secondary-school girls in particular, increases the overall enrolment rate for 14–17-years-olds (post-compulsory) by 9.9 percentage points and raises the secondary school completion rate for girls by 7.8 percentage points (Ahmed et al. 2006). In developing countries, conditional cash transfer programmes have become an effective way to boost the education participation of disadvantaged children.

Table 1. Conditional cash transfer (CCT) programmes in different countries.

Country	Programme	Target Population	Conditions (education related)	Transfer size	Principal impacts	Authors
Mexico	Opportunities programme (PROGRESA)	Low-income families with children ages 8–18; age limits raised to 20 in 2001	Enrolling primary school (grade 3 and higher) or secondary school and attending at least 85 per cent of the school days monthly and annually	For boys: varies by grades, Mex$70-225 (£5.4–16.8) per week For girls: varies by grades, Mex$70-255 (£5.4–16.8) (1998) per week	An increase of 0.5–0.7 years in years of schooling.	Behrman, Sengupta, and Todd (2005); Schultz (2004); Todd and Wolpin (2003)
Nicaragua	Red de Protecion Social (RPS)	Low-income families with children ages 6–13	Enrolling in primary school grades 1–4 and fewer than six days of unexcused school absence in a two-month cycle	School attendance transfer: C$1,440 (£70.3) per household per year School supplies transfer: C$ 275 (£13.4) per child beginning of school year (2000)	An average attendance increase of 20 percentage points for children age 7-13.	Maluccio and Flores (2005)
Colombia	Families in Action (FA)	Low-income families with children ages 7–17	At least 80 per cent school attendance in a two-month cycle	Monthly payment of 14,000 pesos (£4.2) for primary school children and 28,000 pesos (£8.4) for secondary school children (2002)	An increase in secondary school enrolment of 14 percentage points in the urban area and 5.5 percentage points in the rural area.	Attanasio et al. (2004)
Turkey	Social Risk Mitigation Project (SRMP)	Low-income families with children ages 6–17	Attending at least 80 per cent of the total school days and not to repeat the same grade more than once	Annual payment of 216YTL (£90) for boys and 264YTL (£110) for girls in primary school, and 336 YTL (£140) for boys and 468 YTL (£195) for girls in secondary school (2005)	An increase of 9.9 percentage points in overall enrolment rate for 14–17 years old, an increase of 7.8 percentage points in secondary school completion rate for girls.	Ahmed et al. (2006)
Australia	Young Allowance (AUSTUDY)	Ages 16–18, with family income and assets below the threshold	Full-time higher-education students and secondary students completing their final two years of school	Maximum weekly payment of $64.15 (£26.1) for 16-17-year-olds and $77.10 (£31.3) for 18-year-olds (1992)	An increase of 3.5 percentage points in Year 11 and 12 enrolments for poor children.	Dearden and Heath (1996)
UK	Education Maintenance Allowance programme (EMA)	Ages 16–19, with family income and assets below the threshold	Attending either a full-time further education course at a school/college, a course leading to an apprenticeship or a Foundation Learning Programme	Maximum weekly payment of £30 depending on annual household income, plus a retention bonus (2006)	An increase of 4.5 percentage points in Year 11 enrolment, an increase of 6.7 percentage points in receiving two years of education.	Dearden et al. (2009)

In high-income countries, there have also been a range of cash transfer programmes to boost enrolment rates in post-compulsory education. The Australian government launched an educational assistance scheme, known as the AUSTUDY Scheme, in 1987 to reduce the youth unemployment rate and to encourage students to stay in education past the minimum school-leaving age. AUSTUDY offers around £26 to 16–17-year-olds and £31 to 18-year-olds who participate in post-compulsory education, provided that their parental incomes are below a certain threshold. Walker, Johnson, and Osei (2001) find a significant increase in the enrolment of AUSTUDY students within the Australian 15–29 age group during the AUSTUDY period. The proportion of AUSTUDY students more than doubled, indicating an over 100% increase. In contrast, the increase for non-AUSTUDY students was less than 50%. This comparison suggests that AUSTUDY effectively reduces the educational barriers encountered by disadvantaged students. Dearden and Heath (1996) use longitudinal data to estimate the impact of AUSTUDY on secondary-school retention in Australia and find that the policy contributes to a 3.5 percentage points increase in Year 11 and Year 12 participation rates among students from lower socio-economic backgrounds. Additionally, they explored the implications of implementing a comparable programme in the UK. Although this UK programme was not yet in place at that time, hence ‘hypothetical,’ they argue that it could be advantageous over time due to the substantial returns on educational investment.

In the UK, the vast majority of past studies focus on the impact of EMA on participation, retention and achievement in Years 12 and 13 during the pilot period (from 1999 to September 2004). Spielhofer et al. (2010) interview 2,000 Year 11 pupils to explore the barriers they experience in order to stay in education at the end of compulsory schooling. They find that approximately 12 per cent of young people who received EMA stated that they would not have stayed in education if they had not received EMA. This result is consistent with the empirical findings from Middleton et al. (2005) that EMA eligibility is associated with a 5.9 percentage point increase in participation among 16-year-olds, and a 6.1 percentage point increase in participation among 17-year-olds. Based on the data from the first cohort of the EMA pilot study, Dearden et al. (2009) use propensity score matching to control for the individual and local differences between pilot and control areas. They find that eligible young people are 4.5 and 6.7 percentage points more likely to stay in post-compulsory full-time education at the age of 16 and 17, respectively. Although Spielhofer et al. (2010) point out that the ‘deadweight’ of EMA is very high- about 88 per cent said their participation decisions were not affected by the receipt, EMA is still considered to be beneficial (H. Chowdry and Emmerson 2010). It not only leads to higher wages that can offset costs in the long-term (Dearden et al. 2009) but also has a positive impact on wealth redistribution and crime reduction, which provides spillover benefits to society (Feinstein and Sabatés 2005). The impact of EMA on the retention of post-compulsory full-time education was initially evaluated by Ashworth et al. (2001). They combine one-way matching with a difference-in-difference approach and suggest that young people who receive EMA, especially those who receive full payments (£30 a week), are less likely to drop out during the academic year. Moreover, EMA raises the retention rate from Year 12 to Year 13 by 3.9 percentage points in urban areas and 6.4 percentage points in rural areas. Further, Chowdry et al. (2007) provide evidence of an impact on achievement, suggesting that the impact on Level 2 and 3 attainment rates was around 2.5 and 2.0 percentage points for females and males respectively when comparing the EMA pilot areas with the rest of England.

With regards to the evaluation of the EMA national roll-out, Aitken et al. (2007) conduct interviews with 375 16–19-year-olds and compare the EMA recipients with the non-recipients of similar characteristics. They find that the in-year retention rate for recipients is 2.3 percentage points higher than that for non-recipients, but that recipients are 0.9 percentage points less likely to achieve the learning aims of the course they were taking than the non-recipients. Overall, the success rate of the learning aims is 1.2 percentage points higher for recipients than that for non-recipients. Moreover, O’Sullivan (2011) compares the pilot and national roll-out estimates of the impact of EMA on post-compulsory education participation and suggests that estimated impacts of the national roll-out are smaller than the impact of the pilot.

Evidence regarding the impact of EMA on higher education is mixed. Comparing the pilot and control areas, Fitzsimons (2004) estimates a dynamic discrete choice model and finds that EMA has no impact on enrolment in higher education. However, this pilot result is challenged by later study after the national roll-out. Valbuena (2012) uses the first seven waves of the Next Steps study and estimates a linear probability model controlling for personal characteristics, family backgrounds, attitudes and behaviours, prior educational attainment and student’s expectations towards the university. He suggests that the recipients are 4.2 percentage points more likely than the non-recipients to enter higher education, but they are about 3.0 percentage points less likely to attend Russell Group⁸ universities. As the main variable of interest of Valbuena’s work is the socio-economic status rather than the EMA, the sample was not restricted to individuals whose family income was below the requirement of EMA. Unlike Valbuena (2012), this study will exclude individuals whose family income was too high for EMA, focusing only on pupils from low-income families. Moreover, only those who have completed an NVQ level 3 or above will be included in the sample in order to reduce the influence of dropping out of age 16–18 education, which could be correlated with both receipt of, and the impact of EMA. This study aims to add more evidence on the medium-term effect of EMA—its impact on higher education participation and achievement—to the existing literature.

3. Data and descriptives

The data used in this analysis comes from waves 1, 3–8 of Next Steps, previously known as the First Longitudinal Study of Young People in England (LSYPE1) (University College London, UCL Institute of Education, Centre for Longitudinal Studies 2022). Started in 2004, Next Steps is a large-scale and innovative panel study, which documents the lives of approximately 16,000 young people born in England in 1989–90. From 2004 to 2010, the cohort members were interviewed annually until the age of 19/20. The survey mainly focuses on young people’s educational and early labour market experiences, but also collects information on their family, health and happiness, behaviours and attitudes, and aspirations for the future. The last wave, collected at the age of 25, was conducted in 2015/16 to capture the independent adult lives of the cohort members.

Next Steps adopted a two-stage probability proportional to size (PPS) sampling procedure (Department for Education 2011a). First, schools, considered as the primary sampling units (PSUs), were sampled separately for the maintained schools, the independent schools, and pupil referral units (PRUs) to obtain the sample stratum. Maintained schools were stratified based on their deprivation levels, with deprived schools oversampled by 50%. Independent schools were stratified by the proportion of pupils obtaining five or more A*-C GCSE grades in 2003 within boarding status and gender of pupils. As for the pupil referral units (PRUs), they formed a stratum of their own. Then, within selected schools, pupils from major minority ethnic groups were oversampled to achieve 1,000 sampling units in each group. Furthermore, the sample excluded those solely educated at home, boarders and those who resided in England for education purposes only.

3.1. Sample selection

As is the nature of longitudinal surveys, sample attrition is a problematic issue in Next Steps, where the available sample size reduced substantially from 15,770 in 2004 (wave 1) to 7,707 in 2015 (wave 8). Attrition in Next Steps not only leads to a smaller sample size and lower statistical power but also could result in sample bias if the probability of dropping out of the survey is correlated with the sociodemographic characteristics of the participants (Calderwood et al. 2021). Apart from the design weights which adjust the sample composition to take account of over-sampling of specific subgroups, Next Steps also includes attrition weights as the inverse of the predicted probabilities of response. Following Calderwood et al. (2021), this analysis will use the wave 8 final weights, which combine the design weights with the attrition weights.

Young people were not eligible to receive EMA unless their household’s gross annual income was £30,810 or lower. In order to rule out the influence of income-related unobserved factors, such as family resources, ideally, one would drop the respondents whose parental income was too high and focus only on young people who are not too dissimilar to those who have received the allowance. However, the parental income data in Next Steps is banded in waves 3 and 4, and thus, the exact level of income is unknown. The cut-off point for eligibility for EMA (£30,810) is in the income group 7 (£26,000-£31,199) in the dataset. As the household income was self-reported in Next Steps, I include income groups 1–8 (up to £36,399) instead of groups 1–7 (up to £31,199), which may introduce a small bias in the reported income, but also increases the sample size, especially for the control group. Moreover, having a more comprehensive income range also allows for a control group that consist not only of young people who did not apply to EMA but also those who were marginally ineligible. It also reduces the chance that eligible individuals are not in the sample because of misreporting (Britton and Dearden 2015). In addition, I recode the missing values in control variables as a separate group using missing flags to increase sample size and to reduce bias. Table 2 shows the treatment and control groups for the analysis.

Table 2. Treatment and control groups for regression analysis.

Overall sample
Those whose family income was below £36,400
Treatment groups			Control group
One-year EMA	Two-year EMA	No information
Eligible*, applied^† and received for one year	Eligible*, applied^† and received for two years	Missing values for EMA receipt status	Not eligible* or did not apply^† for both years

Notes: *Eligible here means that young people need to be enrolled in eligible courses of further education and their family income at that time need to be below £30,810.

^† In order to receive EMA, eligible young people need to fill in an EMA application form every academic year.

3.2. Variables

The primary outcomes of interest in this study are higher education participation and degree attainment. Higher education participation is defined as a binary variable equal to one if a young person has enrolled in any HE institution by the age of 25, while degree attainment is defined as a binary variable for whether or not the young person has achieved a first degree or higher by the age of 25. Moreover, I select the NVQ as one of the outcome variables to capture the impact of EMA on the attainment of non-degree qualifications. In addition, I have also included attendance at a Russell Group university (RGU) and the attainment of a first or upper second-class degree as two additional outcome variables to provide a more comprehensive understanding of the educational impacts. The detailed definitions of all outcome variables are listed in Appendix Table A1. Table 3 shows that, in the sample, females (41.3%) are slightly more likely to participate in HE (including both degree and non-degree courses) than males (34.3%), to obtain a first degree or higher (22.5% vs 20.3%) and to achieve NVQ Level 4 or above (30.7% vs 25.4%). Among those who have completed their degree, only approximately 21% attended a Russell Group university, but about two-thirds graduated with a first or upper second-degree.

Table 3. Summary of outcomes.

Outcome Variables	Gender of Young Person			Observations
	Male (%)	Female (%)	Total (%)
HE participation	34.26	41.34	37.68	3,203
Degree completion	20.29	22.45	21.23	3,203
NVQ Level 4+	25.36	30.67	27.92	3,203
Russell Group university	26.64	16.94	21.29	576
First or upper second-class	67.20	68.51	67.91	629

Notes: Sample for regression analysis. Weighted using wave 8 weights.

Source: University College London, UCL Institute of Education, Centre for Longitudinal Studies (2021) Next Steps: Sweeps 1-8, 2004-2016. [data collection]. 16th Edition. UK Data Service. SN: 5545, http://doi.org/10.5255/UKDA-SN-5545-8.

The key variable of interest is EMA receipt status. EMA status is measured by the number of years (0, 1 or 2) a young person may have received the allowance for. Young people who have no information on EMA status are excluded from the sample. Table 4 displays the summary statistics of EMA receipt status, both overall and separately by gender. Overall, around 60.8% of young people in the sample have ever received the EMA, and most of them receive it for two years. Females are approximately 3.3 percentage points more likely than males to receive the allowance for two years and 2.8 percentage points more likely to receive it for one year.

Table 4. Summary of EMA receipt status.

EMA Status	Gender of Young Person
	Male (%)	Female (%)	Total (%)	Weighted N
Never	42.14	36.05	39.20	1,343
One year	23.46	26.21	24.79	849
Two years	34.40	37.74	36.01	1,234
Total	100	100	100	3,427

Notes: Sample for regression analysis. Weighted using wave 8 weights.

Source: University College London, UCL Institute of Education, Centre for Longitudinal Studies (2021) Next Steps: Sweeps 1-8, 2004–2016. [data collection]. 16th Edition. UK Data Service. SN: 5545, http://doi.org/10.5255/UKDA-SN-5545-8.

I would like to measure how educational attainment differs by EMA receipt status. However, the challenge lies in the possibility that recipients and non-recipients of EMA may inherently differ in ways not caused by the EMA. Such pre-existing differences could lead to a misattribution of educational outcomes to the EMA, thereby rendering any estimated effects of the EMA on educational attainment both biased and invalid. To get closer to causal estimates between EMA receipt and educational attainment, I use a rich set of measures to control for the demographic and non-demographic differences across groups. The control variables include personal characteristics, family background, prior attainment, and young person’s behaviours, attitudes and expectations. Furthermore, I also implement a reweighting strategy, detailed below, to create a more comparable treatment and control groups. Appendix Table A1 describes all variables used in the analysis.

3.3. Descriptive statistics

It is instructive to explore the unconditional relationship between educational attainments and EMA receipt status before accounting for the control variables. Table 5 shows that about 26.2% of those who received EMA for one year attended university, and 13.0% completed their degree, compared to 27.8% and 15.9% of those never received EMA. In contrast, young people who received EMA for two years are more likely to participate in higher education (56.3%) and obtain a degree (33.0%) than those who never received the allowance. However, both one-year and two-year EMA recipients are less likely to attend a Russell Group university, or graduate with a first or upper second-class degree than non-recipients. Overall, there are considerable raw gaps in educational attainments between one-year and two-year EMA recipients. It is hard to know why some recipients only receive the allowance for one year since the answers in the surveys are unclear.⁹ Dropout might be one of the main reason here as those who did not stay in full-time education would become ineligible for EMA. Another possible explanation is that those who received EMA for one-year were the less motivated pupils, who did not apply for EMA nor university in the last year of school. Moreover, pupils can only receive EMA once their family income drops below the threshold. Those who receive EMA only in Year 13 might have suffered some family financial crisis, which encouraged them to find a job rather than attend higher education. Furthermore, pupils will lose their allowance if they enrol in courses that are ineligible for EMA. Some young people who want to work after school might attend more career-focused programmes in the last year of school and become ineligible for EMA. However, it is unfortunately not possible to test any of the above hypotheses with the data.

Table 5. Educational attainments by EMA receipt status

Outcome Variables	EMA Receipt Status				Observations
	Never (%)	One year (%)	Two years (%)	Total (%)
HE participation	27.78	26.23	56.33	37.68	3,203
Degree completion	15.85	13.03	33.02	21.33	3,203
NVQ Level 4+	22.27	20.34	39.30	27.92	3,203
Russell Group university	26.99	20.24	18.42	21.29	576
First or upper second-class	70.73	65.68	66.95	67.91	629

Notes: Sample for regression analysis. Weighted using wave 8 weights.

Source: University College London, UCL Institute of Education, Centre for Longitudinal Studies (2021) Next Steps: Sweeps 1-8, 2004–2016. [data collection]. 16th Edition. UK Data Service. SN: 5545, http://doi.org/10.5255/UKDA-SN-5545-8.

Figures 1–3 presents the gender differences in the effects of EMA on educational attainments. While there are notable gender gaps in HE participation and degree attainment, the impact of EMA on these two outcome variables shows a similar pattern for both males and females. HE participation and the chance of obtaining a degree are highest for those who received EMA for two years, and those who never received the allowance are more like to attend higher education and graduate with a degree than those who received the allowance for one year. Nevertheless, both one-year and two-year male EMA recipients have a higher chance to achieve NVQ Level 4 or above than the non-recipients, whereas one-year female recipients have a lower chance than the non-recipients to achieve NVQ Level 4 or above. In general, it can be concluded from the descriptive statistics that receiving EMA for two years is associated with higher educational attainments, especially for male students.

Figure 1. HE participation by gender and EMA receipt status.

Notes: The figure displays the proportions of individuals who have ever participated in HE by age 25, based on a sample used for regression analysis. It is weighted using wave 8 weights, with a sample size of N = 3,203. The vertical lines in the figure denote the 95% confidence intervals. Source: University College London, UCL Institute of Education, Centre for Longitudinal Studies (2021) Next Steps: Sweeps 1–8, 2004–2016. [data collection]. 16th Edition. UK Data Service. SN: 5545, http://doi.org/10.5255/UKDA-SN-5545-8.

Figure 2. Degree attainment by gender and EMA receipt status.

Notes: The figure displays the proportions of individuals who have obtained a degree by age 25, based on a sample used for regression analysis. It is weighted using wave 8 weights, with a sample size of N = 3,203. The vertical lines in the figure denote the 95% confidence intervals. Source University College London, UCL Institute of Education, Centre for Longitudinal Studies (2021) Next Steps: Sweeps 1–8, 2004–2016. [data collection]. 16th Edition. UK Data Service. SN: 5545, http://doi.org/10.5255/UKDA-SN-5545-8.

Figure 3. NVQ Level 4+ achievement by gender and EMA receipt status.

Notes: The figure displays the proportions of individuals who have obtained NVQ Level 4+ by age 25, based on a sample used for regression analysis. It is weighted using wave 8 weights, with a sample size of N = 3,203. The vertical lines in the figure denote the 95% confidence intervals. Source: University College London, UCL Institute of Education, Centre for Longitudinal Studies (2021) Next Steps: Sweeps 1–8, 2004–2016. [data collection]. 16th Edition. UK Data Service. SN: 5545, http://doi.org/10.5255/UKDA-SN-5545-8.

4. Methodology

This study adopts a regression method with a multivariate reweighting approach, entropy balancing (Hainmueller 2012), to estimate the impact of EMA on educational attainments and to explore how other factors influence the observed impact. The basic model to be estimated can be written as: (1) $Y_i=\alpha +\text{β}\mathrm{EM}{A}_i+{\gamma }^{\mathrm{\prime }}{X}_i+{\eta }_s+{ϵ}_i$(1) where $Y_i$ is the educational outcome (HE, Degree, NVQ, RGU, Class) for individual i; $\mathrm{EM}{A}_i$ represents the EMA receipt status, specifically, $\mathrm{EM}{A}_i$ = 1 if respondents received EMA for one year, and $\mathrm{EM}{A}_i$ = 2 if respondents received EMA in both 2007 and 2008, and $\mathrm{EM}{A}_i$ = 0 if respondents did not receive EMA in both 2007 and 2008; $X_i$ denotes a vector of background characteristics (see Appendix Table A1 for a full list); ${\eta }_s$ is a school fixed effect; ${ϵ}_i$ is the error term; and $\alpha$, $\text{β}$ and $\gamma$ are the parameters, with $\text{β}$ indicating the size effect of EMA receipt on degree outcome. The standard errors are clustered by primary sampling units (PSUs) and sample strata.

4.1. Model specification and estimation

I estimate Equation (1) additively and sequentially to explore the potential drivers of the relationship between EMA and educational attainments. The baseline model (Model 1) includes only the variable of interest, EMA, in order to show the raw underlying gaps in educational attainments by EMA receipt status. Due to the heterogeneity between cohort members, the baseline model would fail to account for the true associations between EMA and the outcome variables and its estimates could be biased and inefficient. Thus, I then estimate the second model (Model 2) which augments the baseline model by controlling for personal characteristics and family background, including Gender, Ethnicity, Special Educational Needs (SEN), Family Income, National Statistics Socio-economic classification (NS-SEC),¹⁰ Parental Education and First Language. The third model (Model 3) adds the Key Stage¹¹ 2 and Key Stage 4 results to examine the extent to which the impact of EMA can be explained by gaps in prior attainment. Moreover, previous studies suggest that bad behaviours in school and negative attitudes towards school and post-16 education often rule young people out of further education, especially higher education (Archer and Yamashita 2003; Archer, Hollingworth, and Halsall 2007; Department for Education 2011b; Gorard, Huat See, and Davies 2012). In the fourth model (Model 4), I further include a set of behaviour and attitude indicators, including Truancy, Exclusion, Cannabis, Attitude, Post16 Intention, and HE Intention. Furthermore, the attainments of young people tend to be clustered within schools, as those in the same school share the same facilities, curriculums, teachers and teaching methods (H. Chowdry et al. 2013; Crawford and Greaves 2015; Lleras 2008). Hence, the last model (Model 5) adds school fixed effects to capture the variance between schools.

One crucial issue that needs to be considered is how to incorporate school fixed effects in the final specification. Caudill (1987) and Oksanen (1986) point out that the coefficient of the group fixed effects cannot be estimated in a logit or probit model but in a linear probability model (LPM) if every member in the group has similarities. In this case, pupils in the same schools tend to have some common traits, and there are some schools in the sample containing only one pupil.¹² Thus, a logit or probit model with school fixed effects is not an optimal choice for the final specification. This analysis uses a linear probability model with school dummies to take into account the influence of schools. Nevertheless, an ordinary least squares (OLS) model in this situation, the linear probability model, has its own problems when estimating binary outcomes (Aldrich and Nelson 1984; Horowitz and Savin 2001; Maddala 1983). Because the error term does not have constant variance and is not normally distributed, a linear probability model will not yield optimal estimates and cause problems for t-testing and F-testing. Moreover, the predicted values of the outcome variable in a linear probability model are not bounded, which can be greater than one and less than zero. Therefore, in this paper, I also report the results of logit models for specifications 1 to 4 for robustness check.

4.2. Entropy balancing

A potential threat to the validity of the estimates is that young people need to come from low-income families, and apply for EMA to receive it, which means the assignment of EMA is not random. Consequently, the EMA receipt status is very likely to be endogenous. As shown in Table A., those who received EMA are systematically different from those who did not, even though I have already restricted the sample to household incomes of less than £36,400. For example, EMA recipients are more likely to come from poorer and lower socioeconomic status (SES) backgrounds, and their parents tend to hold lower level of qualifications. Among those who received EMA, individuals who received it for two years generally perform better in KS2 and KS4 exams, and are less likely to have played truant or to be excluded from school compared to their one-year recipient counterparts. Two-year recipients also tend to have higher intentions of staying in full-time education after 16 and applying to university comparing to the other two groups. In order to avoid biases in the estimates, it is necessary to balance the characteristics of the one-year and two-year EMA recipients (treatment groups) as well as non-recipients (control group) before running regressions for each outcome variable.

Matching methods, such as propensity score matching, are widely used to evaluate the treatment effects in education studies (Alcott 2017; Dearden et al. 2009; McGuinness and Sloane 2011; Nguyen, Taylor, and Bradley 2006). However, many matching methods do not focus directly on achieving covariate balance and might be unable to balance the covariate moments in finite samples (Hainmueller 2012; Hirano, Imbens, and Ridder 2003). Thus, instead, a pre-processing technique, entropy balancing, is used to estimate the impact of EMA. According to Hainmueller (2012), entropy balancing is an entropy maximisation method, which matches the first, second, and possibly higher moments of the covariate distributions for treatment and control groups. Unlike other methods, it directly incorporates covariate balance into the weight function and keeps valuable information in the data by choosing the weights as close to the base weights as possible. As entropy balancing is applicable only to binary treatments, I first use wave 8 final weights as the base weights and balance the group of one-year EMA recipients and the control group with respect to the first, second and third moments¹³ of all control variables, separately for the regression sample of each outcome variable.¹⁴ Then I apply this approach again for two-year EMA recipients and non-recipients. Finally, I combine these two sets of weights to derive the combined entropy balancing weight for the analysis. Table A shows an example of the balanced sample, in which one-year and two EMA recipients, as well as non-recipients, have similar characteristics after the entropy balancing. Apart from the results using the entropy balancing weight, I also report the results with the inverse-probability-weighted regression adjustment (IPWRA)¹⁵ in the appendix for a robustness check, ensuring that the findings are not sensitive to the choice of statistical method.

5. Results

Overall, the estimated impact of EMA in this paper refers to the association between EMA receipt status and education attainment rather than the causal effect of EMA on attainment. As mentioned in the previous section, young people are required to submit EMA applications before they can receive the allowance. Even though we have included a number of controls in our model, there is still a possibility that some unobserved factors are correlated with both EMA receipt status and education attainment. Thus, the term, such as ‘impact’ and ‘influence’, is used in this paper to demonstrate only the statistical association.

Table 6 shows the estimated impact of EMA on higher education participation. The first and third panels of the table present the estimates without matching. Initially, before the inclusion of any control variables, EMA recipients are 16.3 percentage points more likely to participate in higher education than the non-recipients. This effect is predominately driven by the those who received EMA for two years, who are 28.5 percentage points more likely to participate in higher education than their peers who did not receive EMA. Models 2 and 3 highlight the importance of demographic factors and prior attainment in explaining the impact of EMA on higher education participation. The positive effect of receiving two-year EMA drops from 28.5 to 19.8 percentage points, whereas in impact of receiving one-year EMA on participation remains statistically insignificant in Model 3. The reduction in the estimated impact suggest that EMA receipt status is correlated with demographic factors and prior attainment, and thus, the raw difference partly proxies the influence of demographic factors and prior attainment.

Table 6. Impact of EMA on higher education participation (linear probability model).

EMA (base=Never received)		Model 1 LPM	Model 2 LPM	Model 3 LPM	Model 4 LPM	Model 5 LPM
Without reweighting	Ever received	0.163*** (0.0190)	0.191*** (0.0182)	0.113*** (0.0170)	0.0353** (0.0163)	0.0290 (0.0191)
R-squared		0.027	0.206	0.362	0.443	0.579
With entropy balancing	Ever received	0.0519* (0.0302)	0.0521** (0.0254)	0.0519** (0.0220)	0.0519*** (0.0196)	0.0315 (0.0217)
R-squared		0.003	0.251	0.405	0.494	0.654
Without reweighting	One year	−0.0155 (0.0224)	0.0439** (0.0205)	0.0125 (0.0190)	−0.0369* (0.0191)	−0.0442* (0.0227)
	Two years	0.285*** (0.0221)	0.304*** (0.0215)	0.198*** (0.0205)	0.105*** (0.0200)	0.0983*** (0.0223)
R-squared		0.084	0.247	0.382	0.455	0.588
With entropy balancing	One year	−0.00869 (0.0368)	−0.00869 (0.0284)	−0.00842 (0.0264)	−0.00853 (0.0252)	−0.0533* (0.0277)
	Two years	0.0714 (0.0471)	0.0715* (0.0372)	0.0713** (0.0294)	0.0713** (0.0280)	0.0823*** (0.0287)
R-squared		0.005	0.268	0.408	0.472	0.671
Demographic factors			✓	✓	✓	✓
Prior attainment				✓	✓	✓
Behaviours and attitudes					✓	✓
School fixed effects						✓
Observations		3,203	3,203	3,203	3,203	3,203

Notes: Cluster-robust standard errors in parentheses. *** indicates p<0.01, ** p<0.05, * p<0.1.

Source: University College London, UCL Institute of Education, Centre for Longitudinal Studies (2022) Next Steps: Linked Education Administrative Datasets (National Pupil Database), England, 2005–2009: Secure Access. [data collection]. 6th Edition. UK Data Service. SN: 7104, DOI: 10.5255/UKDA-SN-7104-6.

Model 4 shows how the impact of EMA is mediated by including a set of behaviour and attitude indicators. The impact of receiving EMA for two years deceases further to 10.5 percentage points, indicating that the decision of whether to apply for EMA and participation in higher education is greatly driven by young people’s plans for higher education. Moreover, one-year recipients are 3.7 percentage points less likely to attend higher education after accounting for behaviour and attitude indicators. Finally, the estimated impacts in Model 5, where school fixed effects are included, are similar to estimates in Model 4. It is worth noting that there are 626 schools in the sample with 3,203 observations in total. Model 4 is preferred here because the number of pupils in some schools is extremely low.

The second and fourth panels of Table 6 shows the estimated impact of EMA on higher education participation changes after applying the entropy balancing approach. Compared to the results without reweighting, the coefficients for those who have ever received EMA are lower across all models, but still statistically significant at the 1% level up to Model 4. Regarding the duration of receiving EMA, receiving EMA for one year appears to have no significant impact expect in the final specification, which aligns with the results without reweighting. However, for the two-year recipients, the impact is again positive and becomes statistically significant as the controls are gradually added into the model. In Model 4, receiving EMA for two years is associated with 7.1 percentage points higher chance to attend higher education. This result, which is one of the main results of this paper, is in line with the previous study of Valbuena (2012), who suggests that those who received EMA are 4.2 percentage points more likely than the others to participate in higher education. The reason behind this could be that long-term financial aids are more effective than short-term ones. Moreover, even though entropy balancing is used and attitudes are included in Model 4, there are still some unobserved factors such as motivation that can affect the estimates of impact. If those who received two years of EMA were inherently more motivated and thus more likely to apply for two years, the estimates could just indicate the difference in motivation rather than the impact of EMA.

The estimated impact of EMA on degree completion among all participants is presented in Table 7. Without reweighting, the results suggest that for one-year recipients, the initial observed negative impact (Model 1) becomes more pronounced and statistically significant in the later models. By Model 4, one-year EMA receipt is associated with 3.7 percentage points lower likelihood of obtaining a degree compared to those who have never received EMA. For two-year recipients, the positive impact is significant across all models, though it also diminishes as more controls are introduced, dropping from 17.2 percentage points in Model 1 to 5.3 percentage points in Model 4.

Table 7. Impact of EMA on whether obtained a first degree (linear probability model).

EMA (base=Never received)		Model 1 LPM	Model 2 LPM	Model 3 LPM	Model 4 LPM	Model 5 LPM
Without reweighting	Ever received	0.0902*** (0.0150)	0.111*** (0.0149)	0.0569*** (0.0142)	0.00881 (0.0137)	−0.00326 (0.0158)
R−squared		0.012	0.118	0.229	0.277	0.448
With entropy balancing	Ever received	0.0137 (0.0250)	0.0139 (0.0223)	0.0138 (0.0205)	0.0138 (0.0186)	0.00364 (0.0200)
R−squared		0.000	0.161	0.266	0.326	0.548
Without reweighting	One year	−0.0282* (0.0160)	0.0126 (0.0161)	−0.00816 (0.0155)	−0.0372** (0.0153)	−0.0530*** (0.0173)
	Two years	0.172*** (0.0186)	0.186*** (0.0182)	0.112*** (0.0176)	0.0533*** (0.0173)	0.0438** (0.0193)
R−squared		0.047	0.144	0.241	0.283	0.454
With entropy balancing	One year	−0.0388 (0.0319)	−0.0389 (0.0272)	−0.0387 (0.0259)	−0.0388 (0.0247)	−0.0532** (0.0232)
	Two years	0.0431 (0.0395)	0.0432 (0.0316)	0.0431* (0.0259)	0.0431* (0.0247)	0.0348 (0.0253)
R−squared		0.006	0.195	0.291	0.341	0.589
Demographic factors			✓	✓	✓	✓
Prior attainment				✓	✓	✓
Behaviours and attitudes					✓	✓
School fixed effects						✓
Observations		3,203	3,203	3,203	3,203	3,203

Notes: Cluster-robust standard errors in parentheses. *** indicates p<0.01, **p<0.05, *p<0.1.

Source: University College London, UCL Institute of Education, Centre for Longitudinal Studies (2022) Next Steps: Linked Education Administrative Datasets (National Pupil Database), England, 2005-2009: Secure Access. [data collection]. 6th Edition. UK Data Service. SN: 7104, DOI: 10.5255/UKDA-SN-7104-6.

When entropy balancing is applied, the overall effect of having ever received EMA is small and not statistically significant across all models. However, when focusing specifically on those who received EMA for two years, I find that the two-year recipients are 4.3 percentage more likely to obtain a degree than the non-recipients in the preferred model, Model 4. In contrast to the impact on higher education participation, the smaller impact observed here suggest that although providing a conditional cash transfer to deprived students during secondary school is an effective way to encourage participation, it has less impact on their attainment at a degree level. One possible explanation could be that EMA recipients who have participated in some sort of higher education tend to end up with other qualifications or even no qualification. Thus, in the next section, I will further look into this point by examine the impact of EMA on all Level 4 qualifications.

As EMA supports children from low-income families who are more likely to choose a vocational pathway after the compulsory education (Department of Education 2020), it is important to see whether EMA has an impact on both academic qualifications and vocational qualifications. The first and third panels of Table 8 present the estimated impact of EMA on NVQ Level obtained without matching. Receiving EMA for one year starts with a non-significant negative coefficient in Model 1, which becomes significant in Models 4 and 5, indicating a small negative association with the outcome. For two-year recipients, the impact is initially strong and positive at 17.0 percentage points in Model 1 and remains positive but diminishes in significance and magnitude across the subsequent models, ending at 4.2 percentage points in Model 4. However, after applying entropy balancing, the coefficients become smaller and statistically insignificant for both one- and two-year recipients. Therefore, while EMA can motivate disadvantaged students to participate in higher education, the insignificant results in this section indicate that these students are more likely to drop out or obtain a Level 3 qualification or below.

Table 8. Impact of EMA on whether obtained NVQ Level 4 or above (linear probability model).

EMA (base=Never received)		Model 1 LPM	Model 2 LPM	Model 3 LPM	Model 4 LPM	Model 5 LPM
Without reweighting	Ever received	0.0930*** (0.0175)	0.114*** (0.0177)	0.0616*** (0.0172)	0.00459 (0.0166)	−0.00257 (0.0186)
R-squared		0.010	0.110	0.197	0.248	0.434
With entropy balancing	Ever received	0.0245 (0.0275)	0.0250 (0.0243)	0.0248 (0.0223)	0.0247 (0.0204)	0.0159 (0.0221)
R-squared		0.001	0.158	0.249	0.305	0.528
Without reweighting	One year	−0.0193 (0.0200)	0.0216 (0.0200)	0.00119 (0.0194)	−0.0345* (0.0193)	−0.0421* (0.0215)
	Two years	0.170*** (0.0206)	0.185*** (0.0208)	0.113*** (0.0208)	0.0423** (0.0200)	0.0348 (0.0219)
R-squared		0.036	0.129	0.206	0.252	0.437
With entropy balancing	One year	−0.0128 (0.0343)	−0.0129 (0.0288)	−0.0127 (0.0278)	−0.0130 (0.0269)	−0.0260 (0.0272)
	Two years	0.0101 (0.0403)	0.0151 (0.0334)	0.0138 (0.0281)	0.0138 (0.0260)	0.0218 (0.0265)
R-squared		0.000	0.184	0.266	0.316	0.566
Demographic factors			✓	✓	✓	✓
Prior attainment				✓	✓	✓
Behaviours and attitudes					✓	✓
School fixed effects						✓
Observations		3,203	3,203	3,203	3,203	3,203

Notes: Cluster-robust standard errors in parentheses. *** indicates p<0.01, **p<0.05, *p<0.1.

Source: University College London, UCL Institute of Education, Centre for Longitudinal Studies (2022) Next Steps: Linked Education Administrative Datasets (National Pupil Database), England, 2005-2009: Secure Access. [data collection]. 6th Edition. UK Data Service. SN: 7104, DOI: 10.5255/UKDA-SN-7104-6.

So far, the analysis has focused on participation and completion of higher education. It is also worth knowing whether EMA has an impact on achievement in higher education because attending a high-status institution and obtaining a first or upper second-degree have been shown to be associated with higher returns in the labour market (see, for example, Hussain, McNally, and Telhaj (2009) for university quality and Walker and Zhu (2011) for degree classification). Among the university graduates, I find that receiving EMA for two years is associated with an 11.6 percentage point lower chance of obtaining a degree from a Russell Group university (see Appendix Table A7). Furthermore, receiving EMA for two years has a negative, but statistically insignificant impact, on obtaining a first or upper second-class degree (see Appendix Table A9). These results are not surprising because EMA recipients are from low-income backgrounds and they are the marginal students who are less likely than the non-recipients to attend high-status institution or obtain a first or upper second-degree in the first place. Moreover, the primary purpose of EMA is to encourage children from low-income families to stay in education rather than improve their performance. Before July 2008, young people were required to attend all learning session of their chosen programmes to receive the weekly payment, but there was no achievement requirement (Hubble 2008).¹⁶ Although Chowdry, Dearden, and Emmerson (2008) find a positive effect of EMA on average Key Stage 5 scores, the results here suggest that EMA has no positive impact on academic performance in higher education. However, the validity and robustness of the results need to be further examined as the sample size is around 600 young people.

5.5. Heterogeneous treatment effects

As there is a gender gap in educational attainment in the sample (see Table 3) and different gender might respond differently to financial aids, the impact of EMA on educational attainments by gender will be examined in this section. To investigate whether males and females react differently to the allowance, I estimate the models again using Model 4 with entropy balancing, separately for males and females. Treatment and control groups are balanced with respect to the first, second, and third moments of all control variables.

Table 9 presents the estimates of the impact of EMA on educational attainments for males and females. For both genders, receiving EMA for one year shows no impact on higher education participation, degree completion, or NVQ achievement. However, when it comes to those who received EMA for two years, the situation becomes quite different. For males, receiving EMA for two years is associated with an 11.8 percentage point increase in the likelihood of attending higher education and a 6.4 percentage point increase in the likelihood of obtaining a degree. For females who received EMA for two years, there is a 10.1 percentage point increase in the likelihood of participating in higher education compared to non-recipients, but no impact on degree completion.

Table 9. Impact of EMA on educational attainments, by gender (with entropy balancing).

		HE		Degree		NVQ
		Male	Female	Male	Female	Male	Female
EMA (base=Never received)	One year	−0.0358 (0.0392)	0.00896 (0.0310)	−0.0162 (0.0326)	−0.0503 (0.0375)	0.0250 (0.0371)	−0.0228 (0.0399)
	Two years	0.118** (0.0516)	0.101*** (0.0376)	0.0643* (0.0367)	0.0155 (0.0386)	0.0279 (0.0382)	−0.00765 (0.0401)
Observations		1,423	1,780	1,423	1,780	1,423	1,780
R-squared		0.491	0.516	0.429	0.326	0.377	0.320

Notes: Cluster-robust standard errors in parentheses. *** indicates p<0.01, ** p<0.05, * p<0.1.

Source: University College London, UCL Institute of Education, Centre for Longitudinal Studies (2022) Next Steps: Linked Education Administrative Datasets (National Pupil Database), England, 2005-2009: Secure Access. [data collection]. 6th Edition. UK Data Service. SN: 7104, DOI: 10.5255/UKDA-SN-7104-6.

These results indicate that males are more sensitive to financial incentives than females, which is consistent with the findings of Chowdry, Dearden, and Emmerson (2008) during the initial piloting of EMA. They run multivariate regressions to compare educational attainments in the pilot areas with those in the chosen control areas and find a statistically significant positive impact on Level 3 attainment (1.2 percentage points) and Key Stage 5 tariffs (5.0 points) for males, but no impact on attainments for females. Moreover, previous studies suggest that EMA has a larger impact on participation in post-16 full-time education for males, who tend to have a lower participation rate without the allowance (Ashworth et al. 2001; Dearden et al. 2005; Middleton et al. 2005). Thus, it can be inferred that the statistically significant positive estimate of the impact of EMA presented in the previous sections of this paper are mainly driven by the young men’s positive reactions to the programme.

Building on the findings of Belfield et al. (2018), which highlight the gender disparity in returns to higher education (HE), we gain insight into the differential impact on early-career earnings. Their study suggests that the average man who attended HE earns approximately 25% more than their non-HE counterparts by age 29, whereas this earnings premium is significantly higher for women, reaching around 50%. Focusing on lifetime returns, Walker and Zhu (2011) find that there are high average returns for women across all subjects, but for men, high returns are only found in Law, Economics, and Management. Therefore, the lower earnings gains from HE for men might disincentivise male students from continuing after secondary school. This discrepancy in earnings gains may indeed lead to reduced motivation among male students to pursue HE after secondary school. Given this context, the role of the EMA becomes particularly relevant. It potentially serves as an equalising force, mitigating the disincentives for male students by offsetting the less immediate financial pay-off from HE.

Apart from gender, the impact of EMA can vary based on prior academic attainment (Chowdry, Dearden, and Emmerson 2008). Table 10 presents the different impact of EMA by prior attainment. High or low achievement here is determined by whether the student achieved 5 or more A*-C grades at GCSE or equivalent.¹⁷ For students with high prior attainment, receiving EMA for two years is associated with a significantly higher likelihood of HE participation (10.0 percentage points), degree attainment (11.6 percentage points), and NVQ achievement (11.0 percentage points), all statistically significant at least at the 1% level. For students with low prior attainment, the impacts are smaller and mixed, with a significant but lower increase in HE participation (6.4 percentage points), no significant impact on degree attainment, and a significant negative impact on NVQ achievement (−4.3 percentage points).

Table 10. Impact of EMA on educational attainments, by prior attainment (with entropy balancing).

		HE		Degree		NVQ
		Male	Female	Male	Female	Male	Female
EMA (base=Never received)	One year	0.0193 (0.0472)	0.00798 (0.0309)	−0.0770 (0.0469)	0.00390 (0.0208)	−0.0173 (0.0508)	0.0379 (0.0325)
	Two years	0.0999*** (0.0344)	0.0639* (0.0336)	0.116*** (0.0419)	−0.0167 (0.0167)	0.110*** (0.0398)	−0.0432* (0.0254)
Observations		1,782	1,312	1,782	1,312	1,782	1,312
R-squared		0.323	0.320	0.274	0.241	0.249	0.207

Notes: Cluster-robust standard errors in parentheses. *** indicates p<0.01, ** p<0.05, * p<0.1.

Source: University College London, UCL Institute of Education, Centre for Longitudinal Studies (2022) Next Steps: Linked Education Administrative Datasets (National Pupil Database), England, 2005-2009: Secure Access. [data collection]. 6th Edition. UK Data Service. SN: 7104, DOI: 10.5255/UKDA-SN-7104-6.

One potential explanation for these findings is that students with higher prior attainment are often better academically prepared and may have higher educational aspirations. For these students, EMA might serve as an additional resource that supports their already established post-compulsory education plans. For students with lower prior attainment, the opportunity cost of continuing education might be higher. They might be more inclined to enter the workforce immediately rather than investing in further education. Thus, EMA might not be sufficient to change their future plans for these students.

Finally, I also explore the impact of EMA by parental education. High or low parental education here is determined by whether the highest qualification held by either main or second parent is at least A level or equivalent.¹⁸ As shown in Table 11, for students with high parental education, a two-year receipt of EMA is associate with increases in the likelihood of higher education participation by 11.3 percentage points, degree completion by 8.2 percentage points and NVQ achievement by 7.0 percentage points. For students with low parental education, the impact is positive but smaller and less consistent, with a significant increase in HE participation (5.9 percentage points) and a non-significant increase in degree and NVQ achievement.

Table 11. Impact of EMA on educational attainments, by parental education (with entropy balancing).

		HE		Degree		NVQ
		Male	Female	Male	Female	Male	Female
EMA (base=Never received)	One year	0.0290 (0.0371)	−0.0314 (0.0331)	−0.0312 (0.0408)	−0.0144 (0.0270)	−0.0219 (0.0428)	0.00985 (0.0334)
	Two years	0.113*** (0.0348)	0.0589* (0.0331)	0.0823* (0.0443)	0.0272 (0.0292)	0.0700** (0.0329)	0.0198 (0.0322)
Observations		1,367	1,827	1,367	1,827	1,367	1,827
R-squared		0.522	0.403	0.413	0.269	0.408	0.247

Notes: Cluster-robust standard errors in parentheses. *** indicates p<0.01, ** p<0.05, * p<0.1.

Source: University College London, UCL Institute of Education, Centre for Longitudinal Studies (2022) Next Steps: Linked Education Administrative Datasets (National Pupil Database), England, 2005-2009: Secure Access. [data collection]. 6th Edition. UK Data Service. SN: 7104, DOI: 10.5255/UKDA-SN-7104-6.

Compared to students whose parents have lower educational qualifications, those have parents with higher educational qualifications might have more academic resources at home, such as books or information for higher education, which can complement the financial support from EMA. Besides, parents with more education tend to value education more and have higher expectations in academic achievement for their children, thereby increasing the likelihood that EMA will be used effectively to further educational aspirations.

5.6. Cost-benefit discussion

In this section, I undertake a ‘back-of-the-envelope’ calculation to evaluate the economic returns of increased HE attendance attributable to the EMA, drawing upon the degree returns reported by Belfield et al. (2018). They exclude those with fewer than five A*-C GCSE grades, as this level of attainment is a near-universal prerequisite for entry to university. Thus, here, I will also focus on those with at least five A*-C GCSE grades, that is the high prior attainment group in our sample.

According to Bolton (2011), the average cost of EMA per pupil is £996 and £911 in academic years 2007/08 and 2008/09, respectively. The discount rate is set at 3%, and the aggregate cost of EMA over two years is calculated to be £1,880. Using 2004 to 2017 data from the Labour Force Survey (LFS), Belfield et al. (2018) estimate median earnings across the lifecycle for HE and non-HE graduates. For both males and females, they find the annual average earnings difference between HE and non-HE graduates between ages 30 to 60 is larger than £10,000. For the purposes of this analysis, a conservative estimate of £10,000 per annum is employed to simplify the calculation.

To estimate the incremental earnings attributable to the EMA, the enhanced likelihood of degree attainment (11.6 percentage points) for recipients of the two-year EMA is multiplied by the projected earnings differential over the 30-year period (£140,279), resulting in an additional lifetime earnings of £16,272 per pupil. This figure takes into account the deadweight cost associated with individuals who would have pursued HE without the EMA. Even after taking into account of the deadweight cost of those who received EMA for one year, the lifetime returns to EMA per pupil exceeds £13,000.

These findings align with the research conducted by Britton and Dearden (2015), who examined the consequences of replacing the EMA with the 16 to 19 Bursary Fund. Their study revealed that the policy shift led to a decline in participation and achievement in Years 12 and 13, particularly among students from the lowest-income families who would have qualified for the EMA. Furthermore, their cost–benefit analysis suggests that short-run savings from the reform are overall outweighed by the long-run losses.

6. Conclusions and recommendations

In the UK, the Education Maintenance Allowance has been an effective way to encourage young people from low-income families to stay in education after the compulsory school-leaving age. The results in this study show that EMA not only influences the education-related decisions and behaviours at the time when young people were receiving the payments but also has lasting effects on education outcomes, such as degree participation and completion. The results also imply that the allowance has a more significant impact on male students, those with higher prior academic attainment, and students whose parents have higher educational qualifications.

This study uses Next Steps, which Henderson, Shure, and Adamecz-Völgyi (2020) have shown to produce higher education outcomes that are comparable to external datasets and representative of the target population. To estimate the medium-term impact of EMA on educational attainments regression analysis with entropy balancing is conducted using rich longitudinal data. It is worth noting that the finding in this paper could provide a ‘upper bound’ estimate of the impact of EMA as those who applied and obtained EMA could have been more motivated than those who never applied. The estimates suggest that receiving EMA for two years has a statistically significant impact on higher education participation and, even after controlling for demographic factors, prior attainment and behaviours and attitudes. After balancing the treatment and control groups, two-year EMA recipients are 7.1 percentage points more likely to attend higher education. However, there is no statistically significant result for one-year EMA recipients. One possible explanation for this result is that long-term financial incentives are more effective than short-term ones. Young people need to receive incentives for a certain amount of time before they can change their decisions and behaviours. Moreover, this study finds EMA has no positive impact on attending of high-status institutions or obtaining a first or upper second-degree. The implication is that while EMA helps disadvantaged young people to stay in education, it cannot do much about their performance after secondary education.

Furthermore, the estimated impacts of EMA have a gender heterogeneous effect. Receiving EMA for two years is associated with a larger impact on higher education participation and degree completion for male students. This suggests that the benefits of the allowance have a more prolonged effect on young men compared to their female counterparts. The study also indicates that EMA is particularly beneficial for students with higher prior academic achievements and those whose parents have higher educational levels. In general, the results confirm that financial difficulties and credit constraints do play important roles in education decisions among young people from low-income backgrounds. Policies targeting these disadvantaged young people, like EMA, will be beneficial in the long-run as the returns to higher education are substantial (Blundell et al. 2000; Moretti 2004; I. Walker and Zhu 2011). Furthermore, the cost–benefit analysis presented in this paper also suggests that the short-term costs of the EMA are far outweighed by the long-term returns to the programme.

In 2011, the EMA was scrapped because of its high ‘deadweight’ cost—only 12% of the EMA recipients said their participation decisions were affected by the receipt. The scheme was replaced by the 16–19 Bursary Fund, which provides financial support to a much smaller group of students, with a rise in the compulsory age at which young people must be in some form of education or work-based training from 16 to 17 in 2013 and to 18 in 2015. The introduction of the new scheme reduced the annual expenses by more than two-thirds, from £560 million to £180 million. Although the combination of compulsory post-16 education and discretionary support funds is more economically favourable in the short run, it cannot keep those from low-income families on the academic track. This can pose risks to social mobility as the scrapping of EMA combined with a massive rise in higher education tuition from £3,000 to £9,000 in 2012 could discourage those from lower SES backgrounds from participating in higher education which leads to higher lifetime earnings in a longer time frame. To address the barriers to higher education, policymakers need to offer specific financial aid to those in need in order to encourage them to stay on the academic track but should also provide them with enough information to weigh the immediate costs of education and the benefits of later returns.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by Yuyan Jiang’s UBEL DTP PhD scholarship from the Economic and Social Research Council [Grant Reference ES/P000592/1].

Notes

1 General Certificate of Secondary Education (GCSE) is a qualification earned by students in England, Wales, Northern Ireland, and other British territories, typically taken by students aged 14–16 after two years of study.

2 National Vocational Qualification (NVQ) is a work-based qualification in England, Wales and Northern Ireland. It was withdrawn and replaced by the Regulated Qualifications Framework (RQF) in 2015. NVQ Level 3 qualifications includes A level, access to higher education diploma and advanced apprenticeship. More information can be found at www.gov.uk/what-different-qualification-levels-mean/list-of-qualification-levels.

3 In the UK, a bachelor's degree can be classified as either an honours degree (bachelor's with honours) or an ordinary degree (bachelor's without honours). Honours degrees are classified into First Class Honours, Second Class Honours (upper and lower divisions), and Third Class Honours. This classification is typically based on a weighted average of marks gained in exams and other assessments. Students who achieve a first or an upper second-class degree generally have a weighted average of at least 60%.

4 In England, EMA was replaced by the 16 to 19 Bursary Fund in 2011. More information about the new bursary can be found at https://www.gov.uk/1619-bursary-fund.

5 The pound had an average inflation rate of 3.41% per year between 2007 and 2023, meaning the buying power of £30 in 2007 is equivalent to about £51 in 2023.

6 EMA is still available in Scotland, Wales and Northern Ireland.

7 This includes: 1) stay in full-time education; 2) start an apprenticeship or traineeship; 3) spend 20 h or more a week working or volunteering, while in part-time education or training.

8 The Russell Group is an association of 24 leading UK public research universities, including Oxford and Cambridge.

9 The answers to ‘why young person's EMA application was unsuccessful?’ include unclear responses such as ‘was turned down’, ‘did not take up’, ‘was accepted’, ‘other’, ‘no answer’, and ‘do not know’.

10 The National Statistics Socio-economic classification (NS-SEC) is a measure the employment relations and conditions of occupations widely used in the UK.

11 In the England, the national curriculum is organised in to blocks of years called ‘Key Stages’ (KS) and the performance of pupils will be formally assessed at the end of each KS. For example, KS2 results refer to the Standard Assessment Tests (SATs) results at age 11 and KS4 results are the General Certificate of Secondary Education (GCSE) results at age 16.

12 A logit or probit model will not converge here because of the presence of schools with single sampling unit.

13 That is the mean, variance and skewness. For binary covariates, only their first moment will be considered.

14 For RGU and degree classification, the treatment and control groups are balanced with respect to the first and second moments of demographic factors only, because of the small sample size.

15 IPWRA estimators use weighted regression coefficients to compute averages of treatment-level predicted outcomes, where the weights are the estimated inverse probabilities of treatment.

16 However, there are one-off payments which are based on both attendance and performance against set learning goals.

17 ‘Equivalent’ here refers to other qualifications that are considered to have a similar level of academic rigour, such as the Scottish National 5 or the International General Certificate of Secondary Education (IGCSE) for international students.

18 ‘A level’ stands for Advanced Level, which is a subject-based qualification that can lead to university, further study, training, or work. It is typically taken in the final two years of high school (ages 16–18) in the UK and is a standard entry qualification for universities in the United Kingdom. "Equivalent" qualifications may include the International Baccalaureate (IB) diploma, Scottish Highers, or vocational qualifications like BTEC Nationals, which are all recognised for university entry and considered to be of a similar level to A levels.