Are parents' investments in education distorted by inaccurate beliefs? Guest Post by Tianqi Gan

 

This is the 14^th in this year’s series of posts by PhD students on the job market.

How much to invest in children's education is one of the most important decisions that households make. It is also a tough decision for two reasons. First, past and current performance may not be enough to guide parents on alternative outcomes (e.g., college admission) for different levels of investment. Second, for college admissions, there is information friction at the scale level. What matters is a child’s relative position among their cohort of college applicants nationwide or within a province, yet a parent can only observe their child’s relative rank among a local peer group (students in the same high school). The two information frictions may distort parents' beliefs about their children's ability and result in suboptimal educational investments. Studies have shown that there are negative consequences to misbelief in efforts and limited access to performance records, exacerbated for poorer households.

In my job market paper, I examine the impact of parents' belief in children's school performance on educational investment and, subsequently, children's actual academic performance using a randomized experiment with parents of high-school students in China. To start, I find two types of information frictions that cause belief biases among well-educated parents with full access to students' academic records. 

First, although parents have access to performance records and understand the information literally, most parents are optimistic about the improvements their children can achieve (in terms of improving in-school rank). Parents’ overprediction (overconfidence) of future improvements in rank can make them underestimate the additional help their children may need and result in suboptimal investment decisions. I test parents' prediction accuracy by comparing their predictions with children's actual future in-school rank, defined as parents’ prediction errors (left side of Figure 1) and find that most parents (almost 80%) have arguably large prediction errors about future improvements in their child’s in-school rank.

Second, the information available to parents on in-school rank (on average a cohort of 1,000 students) does not correlate well with in-province ranking (400,000 students). The latter is what determines college admissions. Therefore, parents cannot match in-school ranking to corresponding tier of colleges properly. For example, the baseline survey shows that on average parents believe the best 42 students in this school can get into the top tier of colleges, whereas only the top 35 students can be admitted to the top tier of colleges according to previous cohorts. Parents’ lack of accurate information on their children’s competitiveness in college admission can result in distorted investments.

Experiment

I tested two easy, low cost, and scalable interventions to eliminate these frictions. I sent parents an individualized online report in which I provided parents with two novel orthogonal information shocks. 

The first information is a prediction of their children's future in-school rank generated by a machine learning algorithm. This intervention targets parents' misprediction of children's future in-school rank. The algorithm is trained by LASSO using historical data on alumni who graduated in the same school and has proven to be effective in predicting students' rank. The prediction accuracy in the test sample is over 95%. The right side of Figure 1 is a summary of the algorithm's prediction errors. It's much more accurate than parents' predictions.

The second information aims at removing the scale-level information friction – parents only know children's in-school ranking – that is, among local peers -- whereas college admission depends on their ranking in a much broader peer group (in-province). I provide parents a report that lists the tier of colleges corresponding to children's current performance.

I conduct follow-up surveys after two, four, and six months to measure the intervention impacts.

How do parents update their beliefs?

Both interventions lead to dramatic reductions in belief biases shortly after the intervention, and the effects persist over time. Specifically, compared to the control group, the belief inaccuracies (absolute gap between believed ranking and the actual ranking at each period) for parents receiving machine-learning prediction have been reduced by 49% in the follow-up surveys. The second intervention, the information on colleges corresponding to children’s current in-school ranking also made parents' belief about the difficulty of getting into the ideal tier of colleges becomes 5.6% higher. Moreover, around one-third of parents changed the goals for their children, toward universities with lower reputations but closer to likelihood of acceptance for the student.

How do parents adjust investments?

Both interventions – which address the two types of information frictions-- significantly increase parents' educational monetary investments. Parents’ total monetary investment in children’s education in the past two months significantly increased by 4.8% with the future performance prediction intervention and 3.1% with the feasible colleges information intervention.  In contrast, I observe no impacts on parents' time investments. Finding no effects for educational time investment is reasonable in this setting because students have a busy schedule, so that parental educational time investment is very inelastic.

The changes in monetary investments are heterogeneous by parents’ prior beliefs about their children. I find significant non-monotonic impacts of ability belief and aspirations on parents’ educational investment around their aspirations. If parents think children haven’t reached their aspiration (parents’ belief about children’s current performance is below their belief about the performance needed for their ideal tier of colleges), they invest more when their belief about their children's ability decreases or their aspirations become harder to reach (the performance needed for their ideal tier of colleges is proven to be higher) . When the gap between what parent’s believe their child can achieve and the threshold to their college aspirations grows due to the interventions, parents recognize that more is needed for their child to reach the ideal tier of colleges. Whereas if parents think children have reached their aspirations (in parents’ belief, their children’s current performance is better than the performance needed for the ideal colleges), their optimal investment level is at the marginal cost equal to marginal benefit point, and they invest more when their belief about their children’s ability increases. The underlying logic is that the higher the ability and the higher the marginal benefit of educational investment, the higher the optimal educational investment level.

Does academic performance improve?

Students' academic performance is proxied using both in-school rankings and in-province rankings. Compared to the control group, the performance of children in both treatment groups significantly improved two months after the intervention. The effects on in-school ranking (the lower the ranking, the better the performance) decreased by 2.9% and 2% in the future performance prediction and feasible college information groups, respectively. The in-province ranking decreased by 6.5% and 4.8%, respectively.

The change in academic performance can be partly attributed to the increase in education investments. I use the treatment assignments as instrumental variables to identify the effects of educational investment on students' performance. The results suggest that the additional parental educational monetary investments initiated by the treatments are effective and can significantly enhance both the in-school and in-province rankings. When parental monetary investment in education increases by 1%, students' in-school rank and in-province rank decrease (better performance) by 0.43% and 1.37%, respectively.

I explore the dynamics of the effects by tracking parents' adjustments in beliefs and behaviors over time. As shown in Figure 2, parents update their beliefs shortly after receiving intervention information. The changes persist over time. However, educational monetary investment adjustments are lagged by one survey round (two months). This finding shows that parental educational investment is hard to adjust in the short term. Like the investments, the differential changes in children's performance become significant from the second round and stays elevated to the third follow-up.

Policy implications

The findings of this paper show that what parents believe about their child’s future academic performance, as well as their competitiveness to get into colleges, impacts investments parents make in their child’s education. Unfortunately, lacking sufficient information, these beliefs can be very biased. And this bias can, in turn, result in suboptimal investments in education.  Combing big-data techniques with readily available information from education administrative data sources, policymakers can provide parents with information to address such bias in a scalable and low-cost manner. 

This paper finds removing the information frictions result in a general increase in educational investment and improvement in relative ranking. One concern is about the benefit at the general equilibrium. First, the interventions are good tools to improve parents’ investment levels in societies struggling with under-investment issues (e.g. a lot of developing countries). Second, the general increase in investment observed in this paper is driven by the fact that most parents are too optimistic and have unreasonable aspirations on their children. The findings of the non-monotonic correlations around parents’ aspirations suggests policymakers should take measures to help parents set more reasonable aspirations. When parents become satisfied with their children’s performance, they will behave quite differently, and the intervention information will initiate more optimized, not necessarily higher, educational investments.

 

Tianqi Gan is a PhD candidate at the University of Maryland.