«Financing lifelong learning Hessel Oosterbeeka, Harry Anthony Patrinosb,∗ a Universiteit van Amsterdam School of Economics b World Bank Abstract ...»
The authors ﬁnd that service take-up rates were much higher for students oﬀered both services and scholarships than for those oﬀered services alone. They also ﬁnd that females used services more than males. No program had an eﬀect on grades for males. First-term grades were signiﬁcantly higher for females in the two scholarship treatment groups. These eﬀects faded somewhat by year’s end, but remain signiﬁcant for females who planned to take enough courses to qualify for a scholarship. There also appears to have been an eﬀect on retention for females oﬀered both scholarships and services. This eﬀect is large enough to generate an overall increase in retention. The authors conclude that the results suggest that a combination of services and incentives is more promising than either alone, especially for females.
34 H. Oosterbeek and H. A. Patrinos A randomized experiment where ﬁrst-year university students could earn ﬁnancial rewards for passing all ﬁrst year requirements within one year was conducted (Leuven et al., 2005). They ﬁnd small and non-signiﬁcant average eﬀects of ﬁnancial incentives on the pass rate and the numbers of collected credit points. There is, however, evidence that high ability students collect signiﬁcantly more credit points when assigned to (higher) reward groups. Low ability students collect less credit points when assigned to higher reward groups. After three years these eﬀects have increased, suggesting dynamic spillovers. The small average eﬀect in the population is, therefore, the sum of a positive eﬀect for high ability students and an (partly) oﬀsetting negative eﬀect for low ability students. A negative eﬀect of ﬁnancial incentives for less able individuals is in line with research from psychology and recent economic laboratory experiments, which shows that external rewards may be detrimental for intrinsic motivation.
Partly motivated by the negative incentive eﬀect on low-ability students, Leuven et al. (2007) conducted another randomized ﬁeld experiment in which ﬁrst year students attending an introductory microeconomics course were invited to participate in a rank-order tournament. Three leagues were created, one with a high prize (5,000), one with a medium prize (3,000) and one with a low prize (1,000). The best performing student in each group wins the prize. Upon application, students were asked to choose for which of the three prizes they wanted to play. The motivation for this is that this could (self-)select students into more homogeneous ability groups such that low-ability students compete with each other. Within each group students were randomly assigned to treatment groups (these could actually win a prize) and control groups (they could not win a prize). The results are somewhat disappointing. Students in treatment groups have the same eﬀort (measured as attendance and preparation time) and achievement (measured as exam score) as students in the control groups. The only exception is that treated students tend to attend the very ﬁrst meeting after announcement of the results of the randomization more often than control students. This suggests that ﬁnancial incentives for students (of the size used in this experiment) only have short run eﬀects but no long run eﬀects.
3.2 Subsidization Mechanisms
Another category of ﬁnancial instruments consists of subsidization mechanisms. In this subsection we discuss the following: (1) scholarships and grants, (2) vouchers, entitlements and learning accounts, and (3) tax instruments.
Scholarships/grants. In many countries, tuition fees are heavily subsidized and cover only a small part of the actual costs of education. Motives to do so can be related to the alleged relevance of externalities and may be due to equity considerations. To the extent that equity considerations matter, the subsidies should somehow be related to recipients’ social background.
The impact of tuition on enrollment choices has received considerable attention. Kane (2006) provides a brief summary of such studies pertaining to the United Financing lifelong learning 35 States. Using diﬀerent datasets and diﬀerent identiﬁcation methods, diﬀerent studies arrive basically at the same conclusion: an increase in tuition of $1,000 translates in a decrease in college enrollment of 4 to 6 percentage points. The lower estimate comes from a study by Dynarski (2003) who exploits the sudden discontinuation of the Social Security Student Beneﬁt program to measure the impact on college attendance. The higher estimate comes from studies that exploit variation in tuition policies across states. Only studies that include state ﬁxed eﬀects tend to report smaller estimates of the eﬀect of a change in tuition on enrollment (1 to 2 percentage points per $1,000).
Studies looking at the impact of college costs on enrollment in countries outside the United States tend to ﬁnd smaller eﬀects. An example is the study by Kodde and Ritzen (1984) for the Netherlands. In a survey they asked prospective students about the earnings expectations with and without further education and about parents’ income. They use the estimates on the impacts of these variables to simulate the eﬀect of a change in tuition and ﬁnd that enrollment is insensitive to such changes. Similar conclusions are drawn from studies that exploit variation in tuition and enrollment over time (Pissarides, 1981, 1982; Canton and De Jong, 2005).
Vouchers. The most elaborated and consistent voucher plan applicable to postcompulsory education was proposed by Levin (1983). As in all voucher models, participants receive the entitlements and the funding follows their choices. Implementation will aﬀect both the supply- and demand-side of the schooling market. On the supply-side, new courses will be oﬀered which are aimed at persons who currently do not enroll in post-compulsory schooling. That such new supply will indeed be oﬀered is shown by experience in the United States with the so-called GI Bill. In a sense, this law constitutes a pilot study with vouchers (O’Neill, 1977). Under the GI Bill, veterans of war are entitled to attend up to 45 months of education during a 10-year period after their active duty. They are entitled to receive an allowance if they attend an accredited schooling or training program. The allowance may be used either to meet the direct schooling costs or to cover costs of living. Many newly established courses have been approved in relation to the GI Bill. Many of these courses were geared towards low-ability veterans, and these courses are believed to have a positive eﬀect on earnings. This contradicts the belief that the voucher mechanism leads to the supply of inferior quality.
Bound and Turner (2002) and Turner and Bound (2003) have looked at the effects of the GI Bill on educational attainment of veterans. Using variation in service during World War II between cohorts, Bound and Turner (2002) estimate signiﬁcantly positive eﬀects on years of college completed and on the probability of college completion. Turner and Bound (2003) show that this has been accompanied by a widening of the gap in educational outcomes between African-Americans and others.
Messer and Wolter (2009) report on a randomized experiment analyzing the use of vouchers for adult training. The vouchers had a signiﬁcant impact on participation in training, especially by people who otherwise would not have enrolled in such 36 H. Oosterbeek and H. A. Patrinos activities. The rate of participation in adult education in the experimental group increased by 20 percent.
Individual Learning Accounts. Individual learning accounts (ILAs) encourage savings for education while providing vouchers to people interested in pursuing training. An ILA is a base amount of resources set aside for an individual to use for his or her learning. ILAs can be used to develop knowledge, skills and abilities that increase their human capital.
An ILA initiative has been running in the Netherlands since 2001. It involves eight pilot projects, each serving up to 150 people. The project includes contributions from learners, employers, and the state. State contributions are budgeted at about $400 per learner; employers contribute about $130-$400 per learner. The pilots have been conﬁned to particular training ﬁelds. Renkema (2006) conducted an in-depth study of the eﬀect of ILAs on recipients’ educational intentions. To this end, he focused on two sectors: elderly care and technical installation services. In the ﬁrst sector he fails to ﬁnd any eﬀect at all, for the second sector he reports modest positive impacts on intentions; the experimental condition of respondents accounted for only 5 percent of the variation of educational intention, compared to 27 percent for age and 10 percent for prior participation.
Tax instruments. Some countries subsidize training participation through tax instruments. This can be done either by allowing ﬁrms to deduct training expenditures from the tax bill, or to allow individuals (workers) to deduct their training expenditures from their income tax. As ﬁrms’ training expenditures are part of their normal operation costs, ﬁrms will normally be allowed to deduct such costs from their tax bill. This is possible in many countries including for instance the Netherlands, Japan, Chile and Canada.
To study the extent to which ﬁrms’ tax deduction for training expenditures affects training participation, Leuven and Oosterbeek (2004) exploited the feature that the Dutch tax scheme allowed ﬁrms to deduct an extra amount in case the training expenditures pertained to the training of workers older than 40 years. This policy created a discontinuity in ﬁrms’ training costs at the age of 40. For a worker (just) over 40 years old, training is 14 percent cheaper than for a worker (just) under 40 years old. While the policy was implemented with the aim to stimulate training participation among older workers, the empirical results suggest that this did not happen. Training participation among workers just above 40 is substantially above training participation among workers just below 40. This diﬀerence is, however, not the result of increased training rates among older workers but results from decreased training rates among younger workers. Apparently, training participation by workers just below 40 was postponed.
Another possibility is to allow individuals to deduct direct training expenditures from their taxable income. Such tax deduction of training expenditures is possible in various countries including Germany, Italy and the Netherlands (in Italy against the lowest marginal tax rate), but not in other countries such as France, Sweden, Norway and the United Kingdom (where it was recently been replaced by the now Financing lifelong learning 37 abandoned individual learning accounts). In some other countries, including the United States, Canada and Australia, training expenditures can be deducted as long as they are made to maintain existing skills. The diﬀerences across countries show that tax (non-)deductibility of training expenditures is a policy variable, which is used by some countries, but not by others, as an instrument to stimulate training participation.
The deductibility of direct training expenditures from taxable income was evaluated using two diﬀerent approaches (Leuven and Oosterbeek, 2007). The main challenge is to isolate the eﬀect of tax deductibility of direct training expenditures from the (implicit) tax deductibility of opportunity costs of training investment and from the taxation of returns to training investments. The ﬁrst method exploits differences in deductibility rates around kinks in the tax schedule. By choosing the intervals around the kinks such that average net wage rates are equal, they get rid of the tax deductibility of opportunity costs. They also show that future marginal tax rates for individuals who are above and below kinks in a given year are very similar.
This eliminates diﬀerences in taxation of returns to training. Results based on this approach indicate that a 10 percentage point increase in the tax deductibility rate of direct training expenditures increase training participation by 0.33 percentage points (10 percent increase in training rate).
Their second method takes advantage of the 2001 tax reform, which implied a substantial change in marginal tax rates. Investment costs in 2000 were subject to the old tax code, while investment costs in 2001 were subject to the new tax code.
Because returns to training materialize with some delay, returns to investments made in 2000 and 2001 were both subject to the new tax code. Accordingly, this method isolates changes in taxation of costs from changes in taxation of returns. It does not, however, isolate tax deductibility of direct training expenditures from tax deductibility of opportunity costs. This method identiﬁes the joint eﬀect of these two deductibility rates, and since these operate in the same direction, it will overestimate the eﬀect of interest. Results based on this approach indicate that a 10 percentage point increase in the tax deductibility rate of training costs increase training participation by 0.8 percentage points (a 25 percent increase in training rate). The authors show that the ratio of the results from the two methods are informative about the ratio of the opportunity costs of training investments and the direct expenditures of training investments, implying that opportunity costs are 1.5 times as large as direct expenditures.
There is reason to believe the true eﬀect of tax deductibility of direct training expenditures is somewhere in between the estimates from the two methods. To the extent that the ﬁrst approach does not fully neutralize diﬀerences in the taxation of returns, the estimates based on this method underestimate the true eﬀect. Moreover, this method assumes that individuals are fully aware of the marginal tax rate applicable to their training expenditures. If this assumption does not hold for some individuals with incomes close to a kink, these individuals will not act on their tax treatment and their responsiveness will thus be zero. This also biases the estimate 38 H. Oosterbeek and H. A. Patrinos from the local identiﬁcation method downwards.