How do adult returns to schooling affect children’s enrollment?

What type of assessment can be made?

The issue at hand is whether wage returns to schooling that an individual expects to enjoy as an adult determine school enrollment and the amount of schooling pursued when the individual is young. As it is impossible to see into the future, empirical assessment of the wage return-enrollment relationship must be at least somewhat indirect, and a wide variety of study types exist on the topic.

One group of studies examines the relationship between school enrollment and the wage returns to schooling in the labor market by analyzing whether the wage returns enjoyed by the adults who reside near a group of children affect the likelihood that those children are enrolled in school. Data from South Africa near the end of apartheid show that the likelihood of children’s enrollment in school was positively associated with the wage returns to schooling that adults earned in the children’s region of residence. Because apartheid limited interregional mobility, these adult returns were likely informative of how the children could reasonably expect to fare when they grew up. A couple of additional interesting insights are found in the data on these children. One is that enrollment rates for boys—particularly for boys older than 11—were more responsive to higher wage returns than enrollment rates for girls or younger boys. It is likely that in South Africa, at that time, older boys had more and higher paying alternative uses for their time, so that the rate of return was much more important to their decisions about how to spend their time than it was for either girls or younger boys. For the latter two groups, schooling was likely one of the few activities they had available, and may have just been something to do, as opposed to being a conscious investment choice among many options. Perhaps for those reasons, enrollment of girls and younger boys was less dependent on the economic calculation about current time usage versus future payoff and was rather more related to immediately available alternatives [1].

Somewhat similar results have been found in other studies that directly examine the association between “local” adult returns to education and children’s schooling. A study in Malaysia estimated rates of returns for mothers and fathers at the province level; it found a significant and positive relationship between higher wage returns for mothers and provincial rates of school enrollment. There was no significant relationship with fathers’ returns, and the relationship for mothers’ returns became insignificant when boys and girls were considered separately [2].

In a study of India, rates of return to schooling were estimated at the village level. Differences in the rates of return between villages were found to be more important for explaining village-level variation in school enrollment when there was greater variation in schooling levels within a village’s adult population. In villages with wide variation, the variation in schooling among adults and the associated (large) differences in their earnings may have enabled families to more clearly understand that a variation in payoff to different accumulations of years enrolled in school existed. Although payoff to schooling also existed in villages with narrower variation, the narrowness may have kept families from noticing the payoff [3].

An issue with this body of evidence is the presumption that the only expected future return considered in enrollment is the earnings- or wage-measured rate of return to education. But, better-educated workers tend to have other advantages in the labor market, in addition to higher wages. In developing countries in particular, this includes access to formal sector jobs, which are both more secure and more likely to convey access to social protection schemes in the event of leaving a job, including for reasons of disability or retirement. The studies discussed above do not assess the relative importance of these channels compared to the wage channel. Another issue with this body of evidence is that the rates of return enjoyed by the next generation of adults in a local area may rise or fall for subsequent generations as the general economic fortunes of the locality rise and fall. Finally, children may not end up working as adults in the same place where they went to school.

With respect to the last criticism, patterns of rural-to-urban migration have been used to infer that higher wage returns to education tend to incentivize higher school enrollment. One study hypothesizes that schooling decisions of rural Indian households are affected not only by returns to schooling in their region of residence, but also in places that their adult children may migrate to. It then shows that rural children receive more schooling if the differential in wages between better-educated and less-educated workers in a nearby urban area is higher [4]. In a similar vein, patterns of rural-to-urban migration in Latin America and Kenya have been interpreted as linking school enrollment to higher wages for better-educated workers [5].

Other evidence that school enrollment responds to the rate of return is found in studies that link increased schooling to the relevance of education in the labor market. Particularly interesting are studies of India that analyze where, how, and why the green revolution had an effect on schooling outcomes. The technological advancements from the green revolution increased school enrollment in rural areas that adapted the technology most extensively and rapidly, but only the enrollment of children from land-owning households increased. The authors argue that since only land-owning households possessed the means to implement the new technology, the technology was relevant only to them, and thereby provided returns only to their investments in physical and human capital. As only their children enjoyed new returns to schooling, only those children subsequently pursued more schooling [6], [7]. The policy takeaway from this setting is that education needs to be relevant to the work that children will eventually perform as adults. For many policymakers, the challenge will therefore be to find a way to apply this insight to a more inclusive group of children.

Household finances impact enrollment

In a number of countries, including Brazil, Cambodia, Colombia, Honduras, Mexico, Malawi, and Nepal, evaluation evidence from carefully constructed experiments suggests that cash payments given to families that are conditioned on school enrollment or attendance increase enrollment [8], [9]. By explicitly subsidizing schooling, these cash conditional transfer (CCT) programs reduce the costs incurred by families, and therefore may implicitly raise the returns their children enjoy from schooling (because the rate of return is a measure of the extent to which benefits exceed costs). Therefore, an interpretation of cash-for-schooling programs is that they raise the returns to schooling and, through that channel, raise enrollment.

Although experiments around CCTs do a good job of establishing cause and effect between the cash payments and higher school enrollment, they may not provide evidence of the connection between higher expected returns to education and higher school enrollment. They may instead be addressing a basic issue of affordability faced by poor families. Indeed, many of these experiments show greater responsiveness of enrollment to the CCT among poorer families, where sacrifices made to send children to school are most acute. This suggests that the CCT’s effect could come purely from having more income, and is thus not a response to the explicit enrollment requirement to receive the transfer.

There are a number of theoretical models suggesting that (the expectation of) permanently low income may lead families to make sub-optimal investments in education for their children [1], [5]. These models posit that if families could borrow against the future returns to education, they could have the resources to cover the out-of-pocket and opportunity costs of sending their children to school today and therefore fully exploit the opportunity to invest in education. In this conception, there is a liquidity problem where the investment being made in the child is effectively not liquid for many years; there is also an “agency problem,” in that the person who eventually collects the investment returns (the educated child, many years later) is typically not the same person who takes on the borrowing obligation (the parent). As a result, the poor parent places significant importance on today’s pressing household needs when considering how he/she wants to provide for the child’s future. In higher-income families, this particular calculus is not likely to affect the number of years enrolled in school because these families do not suffer from the day-to-day lack of resources that creates the need to borrow in order to invest in their children’s education. These latter families can pursue education investment opportunities without constraint.

For the poorer parent who wants to borrow, but cannot, an increase in the rate of return to schooling can have an interesting effect if the increase is related to any additional increase in time spent in school. That parent may then reason that a higher return per unit of time spent in school may mean that less total time needs to be spent to provide as well for their child’s future well-being (compared to the longer time they may have planned to enroll their child when the return was lower). It is thus theoretically possible that higher rates of return lead to lower rates of enrollment among poor children. As this calculus is made only by parents who are constrained from taking full advantage of the better investment opportunity (i.e. poor parents), this suggests that for a group of children who face the same rates of return to schooling, the poorer children are less likely to enroll in school than the wealthier ones; moreover, raising poorer children’s household income may increase their likelihood of enrollment, by easing the constraint that had kept poor parents from taking fuller advantage of the opportunity to invest in their children’s education.

Studies that attempt to better isolate the pure income effect on enrollment include the following: regression studies based on observational data that incorporate some measure of family or household income; studies that attempt to assess an income effect by examining how households react to unanticipated shocks to their income; and, studies that examine how anticipated increases in income matter, which include non-conditional cash transfer programs. The results based on regression analyses have been least helpful. One reason is that they typically find little significant effect of income on enrollment, which is difficult to believe. The other reason is that methodological flaws endemic to the approach make interpretation of the results—significant or not—suspect. This is because household income is correlated with so many other measures that might affect income and family decision making about school enrollment, including the rate of return to education, and it is difficult to ensure that regression analysis can appropriately net out the effect of income from these other effects [10].

Studies based on unanticipated shocks to income do not suffer from the same methodological problems because truly unanticipated shocks cannot, by definition, be correlated with the income that families anticipate when they make enrollment decisions. The interesting big-picture take away from these studies is that unanticipated decreases in family income interrupt children’s schooling, but do not tend to decrease eventual total years of enrollment. This suggests little systematic effect on enrollment rates per se, but does point to a concern about the efficacy of schooling when there are gaps in its progression [5], [10]. However, studies based on unanticipated interruption of income still suffer from a methodological flaw: they cannot measure a key variable, namely the effect of permanently lower incomes. This is either because shocks are temporary, or because the data in which the shocks are observed do not include a long enough period to determine whether the income effects are more than temporary.

Returning to studies based on anticipated increases in income: some cash transfer programs can be relatively long lived and do not condition transfer receipt on expectations of certain behavior. State-supported old-age-pension guarantees (which could matter to schooling decisions in households with grandparents and grandchildren) and needs-tested poverty supplements (so long as the supplement does not feasibly affect actions to escape poverty) are just two examples. From a measurement standpoint, unconditional cash transfer (UCT) from such programs may usefully identify a pure income effect on school enrollment.

Compared to analyses of CCT programs, there have been far fewer attempts to assess the impact of UCT programs. However, a recent systematic assessment of that small body of work suggests that UCTs can significantly increase the odds of school enrollment [11]. In terms of the household decision making problems discussed above (e.g. day-to-day liquidity issues), this may imply that lower income families receiving the transfer tend to “satisfice” when determining schooling for their children (i.e. they make decisions aimed at reaching the minimum satisfactory goals for their children’s future). That is, when returns to schooling are high, they decide that a little less of it can be pursued to still achieve an adequate payoff for their children when they grow up. The transfers seem to move families closer to the optimal duration of school enrollment, from an investment perspective.

Of particular interest is the fact that the same assessment also allows for a comparison of the effects on school enrollment from CCTs relative to UCTs, and suggests that the odds of a CCT increasing enrollment is about 15% higher than a similarly sized UCT [11]. Unfortunately, this estimate is not statistically significant, most likely because of the small number of UCT studies available and other measurement issues. However, the difference is significant to the discussion in this paper in the sense that it does provide some evidence for the argument that a CCT combines two effects: a pure income effect and an effect of implicitly raising returns to schooling (keeping in mind that returns to schooling in this case refers to the difference between the overall costs and benefits of attending school).

However, just because a CCT tends to increase school enrollment more than a UCT does not mean that a CCT is necessarily a better tool for promoting the optimal amount of schooling for children. If the UCT addresses the income or liquidity constraints that would otherwise have kept poor children from attending school, and this is the only barrier to optimal investment, then the absence of conditions in the UCT may allow poor families to optimally invest in schooling and also to invest more in items that are also critical to children’s future well-being, such as health care, nutrition, and housing. A CCT of similar generosity forestalls use of the transfer for investing in other things because families must satisfy the condition to receive the transfer (e.g. school enrollment). As a result, the CCT may lead to over-investment (too much time spent) in schooling. This suggests the importance of thinking carefully about whether and why conditionality is necessary or desirable.

Beyond the scope of this paper is a literature that would suggest assessing whether or not to pursue conditionality based on who makes decisions about spending in the household (the agency problem). That literature suggests the importance of asking whether the key decision maker(s) demonstrate that their objective is truly the well-being of everyone in the household, or whether they clearly prioritize themselves or some members first, at the expense of the children. In the latter cases, conditionality may serve to more effectively ensure that transfers improve the children’s well-being.

Can targeting school inputs increase returns to schooling?

Having established that there is reasonable evidence for expecting that higher wage returns to schooling lead to more schooling for children, a natural next question is whether something can be done about schooling itself to ensure higher future returns. From an economic perspective, this amounts to asking whether schooling outcomes are better if more of something, e.g. more teachers, is devoted to generating them; or if a higher quality of something, e.g. better trained teachers, is needed; or both. These types of quantity and quality considerations are henceforth discussed together using the word “inputs” to capture both dimensions of the resources devoted to schooling. Indeed, there is a vast literature on schooling inputs, motivated by the notion that more, better, or better use of certain inputs should improve schooling outcomes.

There is no direct empirical assessment of the relationship between school inputs and the rate of return to education, but there are many studies on school inputs and other schooling outcomes. A snapshot of that evidence is sufficient to illustrate the complexity and low likelihood of a generalizable relationship between school inputs and the rate of return to schooling.

Consider Figure 1, which is based on a recent extensive literature review covering the relationship between school inputs and student test scores [12]. Higher test scores may indicate students who are likely to be more productive workers in the future, and therefore to be better paid. This is interesting because if generalizable results about inputs that raise test scores are available then it may be possible to infer a relationship between inputs and wage returns to schooling as well.

The authors of this meta study started with a body of about 9,000 papers from which they systematically identified 79 of sufficient relevance and quality. These 79 studies make up the “sample” the authors assessed in detail, and they collectively cover countries from around the world. Within the 79 studies, the authors further separated out studies that they label “high quality” based on these studies’ use of certain statistical techniques that are more likely to reveal causality (rather than simple statistical correlation); and from among these, they also separately assessed random control trial (RCT) studies, which were designed according to the scientific method in an attempt to isolate and identify specific causal relationships (making these “highest” quality). For each input variable (row) and quality type (column), the numerical entries in Figure 1 give the number of statistical specifications in the authors’ sample in which the input was found to have a statistically significant effect (numerator) and the total number of specifications in which the input was studied (denominator).

A first point to note from Figure 1 (first column) is the long list of inputs to schooling that have been hypothesized as important. This reflects, to some degree, a lack of consensus among researchers as to what inputs should matter for schooling outcomes. The final three columns appear to justify this lack of consensus. From the column for “All 79 studies” the top five variables in terms of the number of statistical specifications in which they appear are pupil–teacher ratio (101 specifications), teacher education level (70), teacher experience (63), (use of) textbooks/workbooks (60), and female teachers (39). Remarkably, of these variables, only pupil–teacher ratio crosses a threshold for being statistically significant (i.e., having an effect sufficiently likely to be different from no effect) in 50% or more of the specifications in which it was included. At the same time, there is some lack of clarity on whether a higher or lower pupil–teacher ratio produces superior results. While in most studies, a lower ratio produces better test scores, in about 15% of the specifications, a higher pupil–teacher ratio is associated with higher test scores. In short, Figure 1 shows that the literature on the relationship between school inputs and test scores has produced little in the way of widely applicable generalizations that can be made with comfort; this is due to the low number of studies that deal with most of the listed inputs, and because little statistically significant or uniform evidence exists in favor of the few inputs that have been studied more extensively.