Does higher education affect pro-environmental behavior? Evidence from household waste recycling in Greece

We examine the effect of higher education attainment on pro-environmental behavior focusing on household waste recycling. To address the endogeneity of higher education attainment, we exploit a set of reforms that increased opportunities for university studies in Greece, affecting cohorts graduating from high school in year 2000. We leverage the exogenous variation introduced by the school enrollment age cutoff and estimate the local average treatment effect of higher education employing a fuzzy regression discontinuity design, comparing educational attainment and recycling behavior between households that were just, and just not exposed to the reforms. We find little evidence that higher education increases the probability of recycling, and no evidence that the share of household waste recycled responds to higher educational attainment. Our results suggest that human capital accumulation alone may not deliver green behavior returns.


Introduction
Realizing the transition towards sustainable consumption is crucial for addressing global social and environmental challenges (Lucas et al 2008, European Commission 2008, 2020, DEFRA 2011. To encourage the shift to sustainability, national governments enact incentive-targeting policies (Convery et al 2007, Douenne andFabre 2020) and behavioral interventions (Jacobsen et al 2012, Allcott and Rogers 2014, Byerly et al 2018, Ling and Xu 2021. At the same time, environmental behavioral norms, perceptions, and attitudes are conditioned on individual-level sociodemographic characteristics (Poortinga et al 2019), that respond to policies external to environmental or sustainability objectives. Fine-tuning sustainability policy requires understanding the influence of these individual-level determinants, and accounting for the impact of non-environmental policy on pro-environmental behaviors and attitudes.
Here, we examine the effect of higher education on pro-environmental behavior, focusing on household waste recycling. Assessing the relationship between pro-environmental behavior and higher education is complicated by unobserved confounders. Family background, individual willingness to have a positive societal impact, and time preferences among others, can simultaneously determine both the decision to attain higher education and proenvironmental behavior. To address the endogeneity of higher education attainment, we employ a quasi-experimental approach, exploiting the natural experiment created by an expansion in the supply of higher education in Greece in the late 1990s that raised participation and the number of graduates. We use the variation introduced in the probability of attaining higher education by the school enrollment age cutoff and apply a fuzzy regression discontinuity design comparing higher education attainment and recycling behavior between households containing individuals born after the cutoff, who were just exposed to the expansion in higher education opportunities, against households composed of individuals born earlier, who were not.
Human capital accumulation is a prominent lever for socioeconomic change with significant potential to influence attitudes and behaviors through its impact on earnings and preferences (Oreopoulos and Salvanes 2011). To this end, governments around the world are expanding higher education supply to take advantage of education's multidimensional nonpecuniary individual and social benefits (Poterba 1994, Oreopoulos andSalvanes 2011), while promoting innovation and advancing rates of economic growth (Valero and Van Reenen 2019). Education can affect pro-environmental action by positively impacting on labor market outcomes and earnings (Card 1999, Heckman et al 2006, Oreopoulos and Salvanes 2011, Cole et al 2014, improving the affordability of pro-environmental behavior and increasing individual willingness to pay for environmental quality (Hökby andSöderqvist 2003, Jacobsen andHanley 2009). On the other hand, by increasing earnings, education also raises the opportunity cost of time, plausibly disincentivizing time-consuming proenvironmental action (Akar et al 2019). By affecting time preferences and increasing patience (Becker and Mulligan 1997, Oreopoulos and Salvanes 2011, Perez-Arce 2017, Jung et al 2019, education can raise the value of distant benefits of pro-environmental behavior, while by impacting on generalized trust (Yang 2019), it contributes to improving the public's confidence in scientific evidence and the effectiveness of environmental policy (Volland 2017Tam and Chan 2018, Fairbrother et al 2019, Hao et al 2020. Furthermore, education promotes pro-social civic behavior (Dee 2004, Milligan et al 2004, Witschge et al 2019, Harring et al 2020, plausibly encouraging adherence to pro-environmental social norms. Tertiary studies have large potential to influence pro-environmental behavior. Financial returns from higher education can be larger, and enjoyed later in life demanding and promoting more patience (Becker and Mulligan 1997), while the content of university studies systematically improves student understanding of natural and social systems, and their interaction. Attending university can facilitate social identity development through interaction with peers and offer opportunities for informal learning about environmental issues through participation in student societies, volunteering, and sustainability awareness-raising events (Cotton and Alcock 2013).
The paper contributes to the literature on the relationship between education, and pro-environmental behavior in general and recycling in particular. A large literature explores the drivers of pro-environmental behavior including recycling (Jenkins et al 2003, Hage et al 2009, Evans et al 2013, Concari et al 2022, while increasing the share of waste that is recovered and recycled is an important policy goal to facilitate the transition to a circular economy (European Commission 2018, Tallentire andSteubing 2020). Existing evidence suggests a positive association between higher education and pro-environmental behavior (Videras et al 2012, Diederich andGoeschl 2017, Amore et , Eiswerth et al 2011, Franzen and Vogl 2013, perceptions of climate change (Deryugina 2013, Poortinga et al 2019, Douenne and Fabre 2020, Levi 2021, support for environmental policy (Harring et al 2020) and stated willingness to pay for ecosystem goods and services (MacKerron et al 2009). Few studies examine the causal effect of education on pro-environmental behavior, while their focus is on primary and secondary education, exploiting reforms that expanded the duration of, and access to compulsory schooling. Meyer (2015) uses compulsory schooling reforms in 14 European countries to compare attitudes between treated and untreated year-of-birth cohorts, finding that an additional year of schooling has positive and large effect on proenvironmental attitudes. Similarly, Wang et al (2022) use compulsory schooling reforms in China to find evidence of a positive relationship between education and pro-environmental behaviors. On the other hand, Powdthavee (2021) exploiting the variation from a one-year extension of compulsory schooling in England and Wales in the early 1970s, finds little evidence of an effect on environmental attitudes. Chankrajang and Muttarak (2017) instrument educational attainment with the number of primary school teaching staff across regions and over time in Thailand, finding mixed evidence of an impact on pro-environmental behavior. To our knowledge, the present paper is the first to assess the causal effect of higher education on pro-environmental behavior.

Recycling policy and practice
Waste management policy, planning and regulation in Greece are designed by the Ministry of Environment 3 , with household waste collection under the responsibility of local municipalities. Much of the infrastructure for recyclable waste collection and transport is provided by the Hellenic Recovery Recycling Corporation (HERRCO), a joint venture between municipalities and private firms with legal obligation to collect and recycle their product packaging. For many households, waste recycling refers to packaging recycling facilitated by the 'Blue Bin Project' network, operated by HERRCO. Recycling centers where households can dispose of their waste and networks for electronic waste collection also exist but their use is not widespread, while organic waste recycling infrastructure is very limited. The Blue Bin network consists of large curbside bins, shared across several buildings, placed next to conventional waste bins in every neighborhood. Households are 3 The European Waste Management directive was transposed in national law in 2012. At the time of the census waste management was regulated under the Joint Ministerial Decision 50910/2727/2003 'On measures and terms for solid waste management' (Bakas and Milios 2013). encouraged to separate their waste and dispose of packaging materials (glass, plastic, paper, cardboard, aluminum and metal) in their nearest blue bin. Collection frequency varies by municipality. In 2011, 241 out of 325 municipalities covering 79% of the population had access to recycling facilities provided by the blue bin network (HERRCO 2022). By 2018, the share rose to 96% (HERRCO 2022).

Educational system and expansion of higher education supply
Undergraduate higher education in Greece is supplied exclusively by the State free of charge (Psacharopoulos 2003). Schooling is organized in six years of primary education, three years of lower secondary education and three years of upper secondary education. Primary and lower secondary education are compulsory for students under 16. At the time examined here, higher education consisted of Universities and Higher Technological Educational Institutes (HTEIs) offering 4-6 and 3-3.5 year courses respectively. While private tertiary educational institutes exist, their qualifications are not treated as equivalent to those of public institutions by the State and the private sector (Tsakloglou and Antoninis 1999). Access to higher education is controlled by a competitive annual entry examination regulated by the Department for Education. Free provision and better employment opportunities for graduates resulted to demand for undergraduate higher education far exceeding available places (Psacharopoulos and Papakonstantinou 2005). The competitiveness of the admissions process led to the creation of an informal parallel educational system of private institutes focused on exam preparation, discounted the educational contribution of high school, and incentivized many to move abroad for undergraduate studies. Aiming to balance supply and demand for tertiary studies and to decrease the number of Greek undergraduate students abroad, the government expanded the supply of higher education in the late 1990s establishing new University and HTEI courses, and increasing places in already existing courses. Around 70 new undergraduate courses were created, distributed in institutions across the country. At the same time, the first cohort impacted by the 1997-98 educational reform 4 of the secondary educational system that modified university entry mechanism, graduated in year 2000. Between 1999 and 2000 the number of undergraduate students admitted in higher education institutions increased by around 20%, from 68 430 to 82 225. 5

Data
We use data from the 10% sample of the 2011 Greek Population and Housing Census collected and made publicly available by the National Statistical Office of Greece (Minnesota Population Center 2018), reporting individual, household and housing stock characteristics from over from 1 million individuals and 400 000 households residing in Greece in 2011. Importantly for the empirical approach applied in the present paper, the dataset includes information on each household member's month and year of birth, and sociodemographic characteristics including educational attainment.
The 2011 census collected information on recycling behavior, asking households: 'Do you recycle your waste?' . Available responses were '1. Yes' and '2. No' . Households answering 'Yes' were then asked in an open-ended format: 'What share of your waste do you recycle?' . The main outcome variable is binary, indicating households that answered 'Yes' to the recycling question 6 . It captures whether households engage with recycling or not. To assess whether higher education impacts on the intensity of household recycling activity, we also show results when the outcome is the share of household waste recycled, derived from the second (open-ended) recycling question. These were the only environmental-behavior related questions asked in the census. We drop households containing members born abroad who moved to Greece after the age of 5.5, as their exposure to the educational reforms is uncertain. We also exclude individuals who did not progress past primary education as they were not impacted by the expansion in higher education supply.
Recycling behavior is reported at the household level, while biographical information and educational attainment 7 are reported for each individual household member. Our analysis is at the household level. We show results from three samples: (i) Single person households, (ii) Multi-person households (containing 2+ members) and (iii) All households. For the first sample the main independent variable is binary indicating individuals with higher education qualifications. For the latter samples, the main independent variable is binary, indicating households containing 6 It is possible that some households overstate the level of recycling to conform to social norms and to portray themselves as responsible citizens (Larson 2019). In this case, the actual recycling rate would be lower than reported. However, we do not expect this to affect our results as there is no reason to expect misreporting to be systematically different on either side of the cutoff of our RDD approach. 7 The census does not include information on the subject studied. at least one member with higher education qualifications 8 .

Empirical approach
To address the endogeneity of higher education attainment, we focus on the expansion in the supply of publicly provided higher education that took effect in year 2000, and use the exogenous variation introduced in the probability of attending higher education by the school enrollment age cutoff. Children enrolled in primary school conditional on turning 5.5 years by 1 October of the enrollment year. Given the 5.5 years minimum enrollment age, the birth-date cutoff, and the 12-year duration of primary and secondary schooling, the first cohort exposed to the reforms were born on April 1982. 9 Assignment to the treated month-year cohorts does not guarantee that individuals attained higher education. Nevertheless, the probability of obtaining higher education increases discontinuously for the treated cohorts, allowing the use of a fuzzy regression discontinuity design to estimate the effect of higher education on household recycling behavior (Lee and Lemieux 2010, Feir et al 2016, Bertanha and Imbens 2020. As more than one household members may have higher education qualifications, multiple effects can be estimated. We focus on comparing the probability of pro-environmental behavior and the share of household waste recycled, between households containing at least one member with higher education, against households containing no members with higher education. Household treatment by the educational reforms depends on each individual member's month-year of birth relative to the cutoff. We define a household as treated, if it contains at least one member born after the cutoff. To compare between treated and untreated households we define an assignment score v ji = Age ji − c, for each member j ∈ {1, . . . , n} of household i, where c and Age ji are the cutoff and age in year-months respectively. We then create a single running variable at the household level, assigning each household to the maximum score of its members V i = max{v 1i , . . . , v ni } (Wong et al 2013). Our approach is a fuzzy regression discontinuity design with multiple running variables (Wong 8 We have also estimated models where household recycling behavior is assigned to each individual household member. In this case, we performed the analysis at the individual level, estimating the effect of higher education on the probability an individual lives in a recycling household. Results are the same as in the baseline analysis and are available upon request. 9 Those born on 31 March 1982 were 5.5 years on 31 September 1987. They started primary school in 1987, went through 12 years of schooling and sat the university entry examination in 1999. They were not affected by the expansion in higher education supply. Those born on 1 April 1982 were not 5.5 years on 31 September 1987 and missed the cutoff. They joined primary school in 1988 and sat the university entry examination in 2000. They were affected by the expansion in higher education supply. et al 2013). The local average treatment effect (LATE) of higher education on the probability of household waste recycling is: where R i is the household recycling indicator or the share of household waste recycled, and HE i indicates households containing a member with higher education. We estimate β 1 using two stage least squares, instrumenting household higher education attainment with household treatment status. The first stage estimates: where f(V i ) is a local polynomial of the running variable, T i = I{V i ⩾ 0} indicates households containing at least one member born after the cutoff, and ζ r is a set of municipality-of-residence controls capturing location specific differences in recycling behavior. Coefficient γ 1 captures the impact of the expansion of higher education opportunities on the probability a household contains a member with higher education qualifications. The second stage is: We also show estimates from the reduced form, capturing the impact of the expansion in higher education supply on household recycling behavior: For the baseline analysis we control for local linear polynomials of the running variable (f(V i )) allowing the slope to differ on either side of the cutoff (Gelman and Imbens 2019), and use a uniform kernel for the local linear regressions. We also show results controlling for quadratic local polynomials, using a triangular kernel, and excluding individuals born within three months on either side of the threshold to account for students retaking the university entry examination to improve their placement (Barreca et al 2011). Bandwidths (h) are selected using the approach of Calonico et al (2014). We diagrammatically illustrate share of households with higher education and the share of recycling households on either side of the cutoff, using a common 36 month bandwidth for all samples to harmonize the presentation. We cluster standard errors at the running variable level (Cameron and Miller 2015).
In robustness tests we test the results' stability to the choice of the bandwidth, and to expanding or contracting the set of controls. To further account for the possibility of lower recycling infrastructure coverage outside urban centres we also show estimates using households residing in municipalities with population over 20 000, and households in major metropolitan areas where recycling infrastructure was complete. Finally, we show estimates of the first stage and the reduced form from placebo reforms.
Identifying the LATE of higher education requires that the only systematic difference between individuals born just before and just after the 1982 school enrollment threshold is the opportunity to attend a higher education course, that is due to the reforms that took effect in year 2000 (Lee andLemieux 2010, Wong et al 2013) 10 . A challenge to identification comes from parents either delaying or accelerating births and registrations, to increase children's relative age in the classroom (Robertson 2011) or lower childcare costs (Dickert-Conlin and Elder 2010) respectively. As the desired direction of manipulation is unclear, it is unlikely that parents would systematically either expedite or delay births and registrations. At the same time, educational reforms in 2000 could not be anticipated by expecting parents in early 1982. Testing for manipulation of the discrete running variable using the Frandsen (2017) test does not reject the null of continuity (p = 0.611). Another threat to identification can arise from individuals deliberately delaying progression through school grades, aiming to benefit from the increase in higher education supply and variety. This is improbable due to the reputational cost from failing and repeating high school grades, while the prevalence of delayed graduation or failed progression is small (Kountouris 2020). Finally, identification is contingent on parents not deliberately delaying or expediting children's school enrollment. Early enrollment is not possible as school registrations require a birth certificate, and are refused for children below the birth date cutoff. Late enrollment is also unlikely as attendance is compulsory and parents may be prosecuted for not enrolling children 11 , while late registrations are few (Kountouris 2020). Table 1 shows averages and standard deviations for the variables used in the analysis. Columns 1-3 use information from the full samples of all, multi-and single-person households. Around 35% of households have at least one member with higher education qualifications and 58% of households recycle their waste. Among recycling households, the reported share of recycled waste is approximately 33%.

Data descriptive characteristics
Columns 4-6 show descriptive statistics from households included within the bandwidths used later in the analysis for each sample. Approximately 45% of those households contain a member with higher education while 57% recycle their household waste. Figure 1(a) shows the relationship between the expansion in higher education supply and higher education attainment, plotting the share of households containing at least one member with higher education qualifications against the running variable. Figures 1(b) and (c) focus on multi-and single-person households respectively. The prevalence of higher education jumps at the threshold for all samples indicating that the expansion of higher education supply discontinuously increased the probability of higher educational attainment. The share of households with higher education increases from 44% for those households whose youngest member was born on March 1982, to 51% for households whose youngest member was born on April 1982 (figure 1(a)). Figure 2 illustrates the reduced form relationship, showing the share of recycling households against the running variable, for the samples of all (figure 2(a)), multi-person (figure 2(b)) and singleperson (figure 2(c)) households. In all cases the probability of household waste recycling appears continuous around the threshold not suggesting an impact on the probability of household recycling from the expansion in higher education supply. Table 2 shows estimates of the LATE capturing the influence of higher education on the probability of household recycling behavior, and the first stage and reduced form equations, for each of the three samples (panels A-C). As anticipated by the diagrammatic presentation in figure 1, assignment to the justtreated group increases the probability a household contains a member with higher education by 9 percentage points relative to households assigned to the just-not-treated group (column 1). Treated households with 2 or more members are 8.7 percentage points more likely to include a member with higher education relative to untreated households (column 5). Exposure to the reforms increases the probability of higher education attainment by 10.5 percentage points for individuals in single-person households (column 9). Results are similar in terms of magnitude and statistical significance when controlling for quadratic local polynomials, omitting households with members born within three months on each side of the threshold, and employing a triangular kernel for the local linear regressions.

Higher education and the probability of recycling
Turning to the reduced form estimates, and consistent with figure 2, we do not find evidence that assignment to the just treated group impacts   the probability of recycling, for the samples of all households (panel A) and single person households (panel C). The estimated reduced form effect for these samples is generally statistically insignificant at conventional levels, and its magnitude close to zero. Results are similar when controlling for a second order polynomial of the running variable, excluding households near the threshold, or using a triangular instead of a uniform kernel. There is some evidence that the expansion in higher education supply increases the probability of recycling by 2 percentage points for households containing 2+ members (column 5), but the result is sensitive to the choice of polynomial order and kernel.
Estimates of the LATE of higher education are statistically insignificant for most specifications and samples, as anticipated by the reduced form effects. Neverhteless, the effect of higher education appears to be positive, large and statistically significant when focusing on multi-person households (column 5). However, as shown in later robustness tests, this result is sensitive to model specification and bandwidth choice. Figure 3 illustrates the reduced form relationship between the share of household waste recycled and the running variable, for recycling households 12 . There is no indication that the intensity of recycling responds to the expansion in higher education supply. Table 3 shows the corresponding estimates from the first stage, reduced form and structural equations, assessing the influence of higher education on the intensity of household waste recycling. The estimated influence of higher education is statistically insignificant for all samples and specifications (columns 1-12), not suggesting evidence of a higher education effect on the share of household waste recycled.

Spillover effects from offspring education
The baseline analysis relies on comparisons of proenvironmental behavior between households containing at least one, and no members with higher education, without accounting for their position in the household. Evidence suggests that increasing children's attainment can have positive spillover effects on parents health behavior (Berniell et al 2013) and longevity (Torssander 2012, Friedman  Note: The table presents estimates of the LATE, the first stage, and reduced form equations, assessing the influence of higher education on the probability a household recycles, for the samples of all (panel A), multi-(panel B) and single-person households (panel C). The outcome variable is binary, equal to 1 for recycling households. 'Higher Education' is binary indicating households containing at least one member with higher education qualifications. 'Treated' is binary indicating households with at least one member born on or after April 1982. All models control for municipality of residence. Columns 1, 5 and 9 use a linear local polynomial. Columns 2, 6 and 10 use a quadratic local polynomial. Columns 3, 7 and 11 omit subjects born within three months on either side of the threshold. Columns 4, 8 and 12 use a triangular kernel for the local regressions. Standard errors, clustered at the running variable level reported in parentheses. * * * p < 0.01 * * p < 0.05 * p < 0.1. and Mare 2014), while environmentally-oriented education targeted to children can influence parent's pro-environmental behavior (Damerell et al 2013) and energy use (Boudet et al 2016).We test whether the educational attainment of young adults who live with their parents influences household proenvironmental behavior. We restrict the sample to households with parent-child relationships, focusing on households where parents have no higher education qualifications. We then compare the probability of recycling between households containing young adults with higher education, against households containing young adults without higher education. As earlier, we instrument young adults' educational attainment with their month-year of birth relative to April 1982. When more than one young adult is present in the household, the running variable is determined by the youngest. Results are reported in table 4. There is some evidence that young adults' higher education impacts on household recycling behavior, but the result varies with model specification.

Robustness
Tables AT1 and AT2 in the appendix show the influence of higher education on the probability of recycling and the share of recycled waste respectively, when adding controls for sex, month of birth and region of birth in single-person household models, and for household size in multi-person household models. Similar to the main results there is some evidence that higher education increases the probability of recycling by 23.5 percentage points for just-treated households with two or more members. Tables AT3 and AT4 show estimates when excluding the region of residence controls, for the probability and share of household recycling respectively, with no change in results.
To test the results' sensitivity to the choice of the time window of the analysis, figure AF1 shows estimates from the first stage and the reduced form equations for each bandwidth between 12 and 60 months on either side of the threshold, for all samples, when the outcome is the binary recycling indicator. The impact of expanding higher education supply on the probability a household contains a member with higher education is positive and stable in magnitude and statistical significance regardless of the bandwidth choice for all samples (figures AF1(a)-(c)). On the other hand, the impact on the probability of household recycling is close to zero and statistically insignificant for the majority of bandwidths for all samples (figures AF1(d)-(f)). Figure AF2 repeats the analysis when the outcome of interest is the share of household waste recycled. As earlier, the impact of Note: The table presents estimates from the first stage, reduced form and structural equations, assessing the influence of higher education on the intensity of recycling, for recycling households, for the samples of all (panel A), multi-(panel B) and single-person households (panel C). The outcome variable is the share of household waste recycled ×100. 'Higher Education' is binary indicating households containing at least one member with higher education qualifications. 'Treated' is binary indicating households with at least one member born on or after April 1982. All models control for municipality of residence. Columns 1, 5 and 9 use a linear local polynomial. Columns 2, 6 and 10 use a quadratic local polynomial. Columns 3, 7 and 11 omit subjects born within three months on either side of the threshold. Columns 4, 8 and 12 use a triangular kernel for the local regressions. Standard errors, clustered at the running variable level reported in parentheses. * * * p < 0.01 * * p < 0.05 * p < 0.1. , the outcome variable is the share of household waste recycled ×100. 'Higher Education' is binary, indicating households that contain at least one child with higher education. 'Treated' is binary, indicating households containing at least one child born after the threshold. All models control for municipality of residence. Columns 1 and 5 a linear local polynomial. Columns 2 and 6 use a quadratic local polynomial. Columns 3 and 7 subjects born within three months on either side of the threshold. Columns 4 and 8 use a triangular kernel for the local regressions. Standard errors are clustered at the level of the running variable. * * * p < 0.01, * * p < 0.05, * p < 0.1. the reform on the probability of higher educational attainment is positive and statistically significant in all cases. In contrast the effect of the reforms on the share of household waste recycled is consistently close to zero and statistically insignificant, irrespective of the bandwidth choice, suggesting that the bandwidth choice does not drive earlier findings. Figures AF3 and AF4 show first stage and reduced form estimates for 300 placebo reforms, one for each year-month c − h − i, where c is the April 1982 cutoff, h is the bandwidth for each sample as shown in tables 2 and 3 in the text, and i = 1, . . . , 300, when the outcomes of interest are the recycling participation and the share of household waste recycled respectively. First-stage placebo estimates are concentrated around zero, while the true estimates lie on the far right of the distribution for all cases. In contrast, true reduced form estimates lie at the mass of the placebo estimates irrespective of sample or outcome variable. Table AT5 shows estimates estimates from a fuzzy RDD controlling for a local linear polynomial of the running variable for the sample of households in municipalities with population over 20 000, and for the sample of households in the municipalities of the main metropolitan areas where recycling infrastructure was complete. Results are similar to the baseline estimates.

Discussion and conclusion
Expanding the supply of education and promoting human capital accumulation is an attractive policy for countries aiming to bolster their economic growth rates and reap the pro-social returns of education. In this paper, we test whether the pro-social externalities of higher education extend to the environmental domain, focusing on recycling behavior. To overcome the challenges posed by unobserved confounders simultaneously affecting behavior and educational attainment, we apply a regression discontinuity design exploiting an expansion in the supply of publicly provided higher education in Greece.
We find no evidence that higher educational attainment impacts on the recycling behavior of single person households. There is some evidence that multi-person households containing at least one individual with higher education qualifications are about 20 percentage points more likely to recycle, relative to households containing no members with higher education. However this result is highly sensitive to the choice of bandwidth and specification. Our results agree with Powdthavee (2021) who finds limited evidence of an impact of compulsory and secondary schooling on pro-environmental attitudes. Previous research finds a positive association between higher education attainment and pro-environmental behaviors. Our findings suggest that this link may not be causal, at least for the population examined here, and that a relationship between the two variables may not exist. Nevertheless, findings may also be due to the bidirectional influence of education on behavior, driven by its impact on earnings and preferences: on the one side, education can promote civic behavior and encourage costly pro-environmental action; on the other, by increasing earnings education raises the opportunity cost of time plausibly discouraging time-consuming pro-environmental behavior (Bruvoll et al 2002, Hage et al 2009, Viscusi et al 2011, Reijonen et al 2021. Human capital accumulation can contribute to the transition towards sustainable consumption and production, by promoting innovation and technological change encouraging cleaner production (Bovenberg and Smulders 1996, Weitzman 1997, Capasso et al 2019, and through its influence on prosocial attitudes and preferences (Milligan et al 2004). Establishing a causal link between higher education attainment and pro-environmental behavior would imply that positive green externalities can be expected from the expansion of the provision of higher education. Our findings, however, suggest that investment in higher education alone may not deliver green behavior returns.
As with any observational study, the results cannot be readily extrapolated to other populations and circumstances. More evidence is needed to establish the role of higher education in fostering pro-environmental attitudes and action, drawing from interventions in different settings. It is worth mentioning that in the time elapsed since 2011, when the data we use here were collected, environmental challenges such as the climate crisis, have increased in visibility, highlighting the importance of transitioning towards sustainable production and consumption paradigms. An interesting area for future research would involve studying whether the relationship between higher education and environmental behavior has changed over time as global interest on environmental sustainability has increased.

Data availability statement
The data cannot be made publicly available upon publication because they are owned by a third party and the terms of use prevent public distribution. The data that support the findings of this study are available upon reasonable request from the authors. Note: The table presents estimates of the LATE, reduced form and structural equations, assessing the influence of higher education on the probability a household recycles, for the samples of all (panel A), multi-(panel B) and single-person households (panel C). The outcome variable is binary, indicating recycling households. 'Higher Education' is binary indicating households containing at least one member with higher education qualifications. 'Treated' is binary indicating households with at least one member born on or after April 1982. Models for single person households control for sex, month of birth and region of birth and region of residence. Models for multi-person households control for household size and municipality of residence. Columns 1, 5 and 9 use a linear local polynomial. Columns 2, 6 and 10 use a quadratic local polynomial. Columns 3, 7 and 11 omit subjects born within three months on either side of the threshold. Columns 4, 8 and 12 use a triangular kernel for the local regressions. Standard errors, clustered at the running variable level reported in parentheses. * * * p < 0.01 * * p < 0.05. single-person households (panel C). The outcome variable is the share of household waste recycled ×100. 'Higher Education' is binary indicating households containing at least one member with higher education qualifications. 'Treated' is binary indicating households with at least one member born on or after April 1982. Models for single person households control for sex, month of birth and region of birth and municipality of residence. Models for multi-person households control for household size and municipality of residence. Columns 1, 5 and 9 use a linear local polynomial. Columns 2, 6 and 10 use a quadratic local polynomial. Columns 3, 7 and 11 omit subjects born within three months on either side of the threshold. Columns 4, 8 and 12 use a triangular kernel for the local regressions. Standard errors, clustered at the running variable level reported in parentheses. * * * p < 0.01 * * p < 0.05. and single-person households (panel C). The outcome variable is the share of household waste recycled ×100, indicating recycling households. 'Higher Education' is binary indicating households containing at least one member with higher education qualifications. 'Treated' is binary indicating households with at least one member born on or after April 1982. Columns 1, 5 and 9 use a linear local polynomial. Columns 2, 6 and 10 use a quadratic local polynomial. Columns 3, 7 and 11 omit subjects born within three months on either side of the threshold. Columns 4, 8 and 12 use a triangular kernel for the local regressions. Standard errors, clustered at the running variable level reported in parentheses. * * * p < 0.01 * * p < 0.05.     Note: The table shows estimates of the structural form, the reduced form and the first stage for the samples of all, multi-and single-person households when restricting the sample to households residing in municipalities with population over 20 000 and in municipalities of the Athens and Thessaloniki metropolitan areas. Each column shows estimates from a different model. In panel A the outcome variable is binary, indicating households that recycle. In panel B the dependent variable is the share of household waste recycled ×100. 'Higher Education' is binary, indicating households with at least one member with higher education. 'Treated' is binary, indicating treated households. Reduced form and First stage estimates come from regressing the outcome and higher education respectively, on the treatment indicator, the running variable and their interaction. All models control for municipality of residence. Standard errors are clustered at the level of the running variable. * * * p < 0.01 * p < 0.1.