How biased self-assessments distort schooling decisions

Research from numerous corners of psychology suggests that self-assessments of skill and character are often flawed in substantive and systematic ways. For instance, it is often argued that people tend to hold rather favorable views of their abilities – both in absolute and relative terms. In spite of a recent and growing literature on the extent to which poor information can negatively affect educational choices (e.g. Hasting and Weinstein, 2008;Jensen, 2010; Dinkelman and Martinez, 2014), there is little systematic evidence establishing how inaccurate self-assessments distort schooling decisions.

In joint work with my colleague Veronica Frisancho at the IDB, we design and implement a field experiment to study whether and how youths’ inaccurate perceptions about their own academic skills generate misallocations through inadequate school choices. In particular, we consider the choice between attending a high school from the academic track (i.e., general education modality) and a high school from technical or vocational tracks. In most countries, curricular tracking within upper secondary education is crucial in shaping subsequent access to college and, consequently, labor market earnings. To the extent that students’ preferences over tracks partly depend on expected performance therein, biased perceptions about own academic potential will distort perceived individual and track-specific returns potentially leading to sub-optimal choices.

Context and Experimental Design

We nest our study within the school assignment mechanism currently in place in the metropolitan area of Mexico City, which allocates students to public high schools according to their relative placement in a standardized achievement test. Due to some of its specific features, the system offers an ideal setting for our purposes. First, most applicants apply to high schools without a clear idea of their own academic skills. Students are required to submit their final preference sheets early during the application process, not only before the admission test, but even before the majority of applicants start preparing for the exam. Second, administrative records capture individual revealed preference rankings over a wide range of available high school alternatives, which allows us to accurately measure student-specific ranked orderings over high school tracks. Third, the matching mechanism in place and the large portfolio size that can be submitted imply that the assignment mechanism is strategy-proof (Pathak, 2011). Since admission probabilities to each school do not play a role when choosing an application portfolio, any effect on choices due to our intervention is enabled through changes in net expected benefits from each schooling option.

We provide a sample of ninth graders with a mock version of the admission test prior to the submission of schooling preferences. We elicit subjective probabilistic distributions about test performance as a proxy for expected academic achievement both before and after the mock-exam and communicate individual scores to a randomly chosen sub-sample. This framework allows us to assess whether providing students with an informative signal about their academic skills enables them to update their beliefs about future achievement and, if so, how this change affects their revealed preferences over high school tracks.
Large ex-ante discrepancies are found between students’ expected and realized test performance. Relative to the mock exam score, students’ mean priors are on average off by almost 15 points out of a 128 point scale, with 80% of the students overestimating their performance in the test. These patterns are largely unchanged when we alternatively use the score in the actual achievement test used for the allocation in the system, which shows that the result of upwardly biased beliefs is quite robust in our sample. We also find that students’ self-assessments are strong predictors of school track choices and outcomes in the control group. Conditional on test performance, a one standard deviation increase in expected academic skills is associated with an average increase in the share of academic options requested in the system of about 4.5% – which on average leads to an increase of 7.5% in the probability of being assigned to a school from the academic track.

Results

Consistent with Bayesian updating, providing feedback about individual performance in the mock exam substantially reduces the gap between expected and actual performance for both over and under-confident students. This effect is stronger among the former due to greater initial average gaps. On average, preferences for the academic-oriented track become (more) responsive to students’ individual achievement, as measured by their performance in the admission test. We also find that these changes in students’ revealed preferences over school tracks pass through into final assignment outcomes, thereby suggesting the potential scope for longer-term impacts of the intervention on subsequent academic trajectories.
However, only under-confident students adjust their school choices as a result of the treatment; relative to their counterparts in the control group, these students record an average increase of 18% in the share of academic programs requested within the assignment system. Instead, the estimated average treatment effect on school choices among overconfident students is very small and statistically insignificant.

Interpretation

Economists have traditionally sought to explain over-confidence by either postulating some form of biased processing of information or by assuming that decision makers form their own beliefs without adequate information. How can then one reconcile the presence of overconfidence in our baseline data with the observed Bayesian beliefs’ updating patterns induced by the treatment? More so, why are those who receive positive performance feedback the only ones altering their high school track choices accordingly?
We argue that this evidence is consistent with the presence of “self-image” motives in an otherwise standard rational model of choice under uncertainty (Koszegi, 2006). Students’ incentives to learn about their academic skills depend on whether the expected gain from getting more accurate information outweighs the potential damage that new signals may cause to the consumption value of their beliefs. In the context of this model, our intervention provides an additional (and unsolicited) signal about students’ academic skills. Those who receive a signal that is above their mean prior beliefs are better off due to the resulting positive views about themselves and, accordingly, they are prone to increase their demand for the academic-oriented track. Instead, students provided with a signal that is below their average prior beliefs are likely to keep searching for more signals that can at least partially undo the negative effect of the treatment on their egos. Consequently, these students are less likely to align their school choices with their actual academic performance in response to the treatment.
Further evidence exploiting random variations in the time elapsed between the provision of information about exam performance and the submission of school preferences indirectly confirms this hypothesis. Among over-confident students who only had one extra day between treatment exposure and the submission of their preferences, we find that the treatment effectively reduces by 34% the share of academic options vis-a-vis the mean in the control group. This effect gets quickly diluted among those with more time to gather additional signals after the treatment delivery. No such heterogeneous responses are found among under confident students.

Conclusions

Our analysis can contribute to a recent body of evidence documenting the role of informational barriers in school choice decisions (Avery and Hoxby, 2012; Hoxby and Turner, 2014). We show that access to salient and individual-specific information about academic performance only translates into changes in school choices among a particular group of agents in our setting. Further research is needed to evaluate the effectiveness of alternative informational nudges in closing the gap between perceptions and reality about students’ academic skills.

This post first appeared on The World Bank Blog

Author: Matteo Bobba is a guest blogger for The World Bank Blog.

Image: Joy Cheng, a foreign exchange student from Taipei, Taiwan, listens to her junior high science teacher at Grant-Deuel School in Revillo, South Dakota February 13, 2012. REUTERS/Jim Young.