Julie Cullen on the Unintended Consequences of Education Reform

Julie Cullen at UCSD visited Duke this week and gave a talk this morning on the unintended consequences of education reform. Her talk is a general summary of the lessons drawn from three papers on 1) the impact of fiscal incentive on student disability rates, 2) school gaming under a performance accountability system, 3) strategic high school choice under Texas’ top ten percent plan.

These papers ask different specific questions, but they all address broadly the mechanisms through which education reforms can change individual behavior and create unintended consequences. The intuition is that whenever eligibility to funding/benefits are conditional on some manipulable measures, those measures will be manipulated. This same logic is applied to examine how fiscal incentives, performance standards, and policies aimed at expanding college access have distorted incentives of schools.

Although I am not very interested in the economics of education, I can imagine the same approach being used to assess other types of government policies, e.g. welfare program participation.

Advertisements

The effects of taxes and transfers (2): incorporating human capital accumulation in life cycle labor supply models

Following our last class on simple models of life cycle labor supply, we discussed about incorporating human capital accumulation into the individual decision-making process. Now the wages are no longer exogenous. Instead, current wage depends on the amount of labor supplied in the last period, which is a proxy of human capital accumulation. Everything else stays the same. If this model is a more realistic description of the world (which is likely), the traditional estimation of Frisch elasticity (intertemporal substitution elasticity) will be biased downwards. To see this, note that over time wage rates are rising as a result of human capital accumulation, but the returns to “learning by doing” is decreasing (as the individual gets nearer to retirement). The total marginal return on labor supply estimated by Frisch elasticity combines these two effects and understates the true intertemporal substitution effect.

These models are nice, but here are two questions worth considering:

The first question is due to my classmate DS: How should we perceive the process of human capital accumulation?

The model presented above assumes that human capital accumulation happens through working more hours, which implies labor supply and human capital are complements. But they might well be substitutes: think about workers who have accumulated a body of knowledge and can work more efficiently with shorter hours. In this case, the Frisch elasticity might not be downward biased.

The second question is due to my classmate MZ:

Suppose you are at a cocktail party and two people start arguing over the effect of raising or lowering taxes on the incomes of wealthy people. Someone mentions that you are a budding young economist, so the two parties temporarily stop arguing to hear your words of wisdom. What do you say?

After our discussion, I think a fair answer would be that tax changes affect people’s behavior along multiple margins. The more margins we allow people to respond to in our analysis, the more distortionary effects we are likely to find. Obviously, economists rarely agree and even if they agree, it takes a long time for them to channel through policy making and have a real-world impact.

The effects of taxes and transfers (1): from static to life cycle labor supply models

The second half of this semester I am taking a PhD module on the effects of taxes and transfers by Professor Joe Hotz. I have enjoyed it so much that I decided to write a series of reflective posts on it. This post is the first one. References to the papers are available upon request; I won’t post them here because of time constraint.

Taxes and transfer programs change individual behavior through various channels, among which labor supply is one of the most important and policy relevant. Knowledge about the responsiveness of labor supply towards tax changes allows us to better evaluate the distortionary effects of taxes and the magnitude of dead weight loss of different tax schedules (e.g. progressive vs. regressive).

The earliest models of labor supply assume individual decisions are made in a static setting (Heckman, 1974). Given exogenous wage and non labor income, individuals choose to their labor supply levels to maximize utility (which depends on their tastes). There is no dynamic concerns: “lifetime” is treated as the one and only period where the individual needs to make a decision.Tax changes figure into individual’s budget constraint through changing pre-tax wage rates. In more sophisticated models, taxes on wages and nonlabor income are modeled jointly.

Since labor supply decisions are made throughout the life cycle, static labor supply models yield incomprehensible estimates for labor supply elasticity that cannot be used for policy evaluation. Life cycle labor supply models are developed as a consequence. The simplest life cycle model (MaCurdy, 1980) assumes no uncertainty, i.e. wages, prices, and interest rates are all exogenous and known to the individual. From there we can derive the Frisch elasticity (holding the marginal utility of wealth constant) which measures the intertemporal substitution of labor supply. Frisch elasticities allow us to infer individual dynamic behavior along a given profile, but a structural model of individual effects needs to be estimated in order to derive compensated and uncompensated elasticities for cross-sectional comparisons.

Later models incorporate uncertainty into the evolution of wages, prices, and interest rates. In such settings, individuals maximize expected utility over their remaining life and adjust their labor supply/consumption choices after period-specific information is revealed. Using appropriate IVs that are not correlated with present-period “shocks”, we can still adopt the first-difference method used in the perfect certainty models to estimate intertemporal elasticity.

A few other papers relaxed the unrealistic assumptions that individual utility is separable contemporaneously between leisure and consumption and across periods. Ziliak and Kniesner (1999. 2005) found that contemporaneous consumption and leisure are complements, and ignoring this relationship can bias the estimates for labor supply elasticity.

The important takeaway is tax changes will affect people’s behavior along multiple margins. The significance of the earliest models lies in their simplicity. The subsequent development in the literature enriches the life cycle model and allow people to respond to foreseeable as well as unpredicted changes in wage rates and prices, to endogenously determine their wages (through human capital accumulation), to have preference for joint consumption of leisure and other goods, etc. I will write a post on human capital accumulation after tomorrow’s class.

For this course, each student is required to lead one class, where he/she has to write summary notes of the papers, do a presentation, and draft a few discussion questions. I found it’s harder than I thought to summarize a few related papers (to extract the “juice” out of them) and to explicitly point out how they complement each other or where the dispute is. You need to really digest the papers, think through about the fundamental modelling assumptions and empirical methodologies, and what are the pros and cons of each approach. It took me one and a half week to fully digest and to organize the four papers such that my classmates would spend the minimum time getting the essence out of them. But I felt really glad to receive positive feedback on my summary notes and presentation. It was also rewarding to see my questions generated a fair bit of insightful discussion. Keep it up!

What I’ve learned in my first three weeks as a PhD student

The first three weeks of the fall semester flashed by, and I feel the urge of summarizing a few things I’ve learned so far before I lose track of them. Unlike most first-year PhD students, I only need to take two of the six core courses (micro I and macro II) and am therefore taking two field classes — Demand Estimation with Prof. Jimmy Roberts and Topics in Public Econoimcs with Prof. Juan Carlos Suarez Serrato. I’m also attending lunch groups and seminars to have a better idea of what frontier research is like in different fields.

In Demand Estimation we are studying models of firm behavior such as pricing strategies, collusion vs. perfect competition, introduction of new products, etc. Horizontal and vertical product differentiation are the two conceptual frameworks often used to model the product space. In vertical product differentiation, consumers agree on the desirability of products but have different willingness or ability to pay. With horizontal product differentiation, consumer tastes vary, and product characteristics affect demand. Although I’m not an IO expert yet, I’ve learned that industry knowledge is essential for you to succeed in this field, whether you learn it from experience or from reading. An entrepreneurial spirit is valuable when it comes to finding data.

I’ve also learnt a few things from practice job market talks, seminars, and student presentations.

First, establish your research question before you address it. Clearly framing the question, positioning it in the literature, and describing its contribution are essential before going into the details.

Second, always know what you are explicitly and implicitly assuming in your model. This is especially important for people using the reduced-form approach. Not having a structural model shouldn’t be the excuse for not thinking through the underlying mechanisms.

Third, make sense of your results, in numerical and economic sense. Comparison with existing well-known results can be useful. This is especially true for policy-relevant questions.

Last but not least, interpret, or at least speculate the mechanisms behind your results if they are unexpected.

Now that I’m formally in the PhD process, I’ve come to realize that “work-life balance” is such a vacuous claim. The best time management strategy for me (at least for now) is to have a clear timeline of the things that need to be done (i.e. with a strict deadline) but also bear in mind the long run goals. Don’t work so hard to get burned out early. Instead, organize your life so that you work efficiently and deliver what you need to deliver. As JG has wisely pointed out, no one will get an award for working all day without a rest.

Shelly Lundberg on Educational Inequality and Returns to Skills

test-prep
DuPRI‘s seminar series in the fall started with a fascinating talk by Shelly Lundberg (UCSB) on her recent research on differential returns to skills in different environments. Abstract is below:

I examine the effects of cognitive ability and personal traits on college graduation in a recent cohort of young Americans, and how the returns to these traits vary by family background, and find very substantial differences across family background groups in the personality traits that predict successful completion of college, particularly for men. The implications are two-fold. First, the returns to non-cognitive traits may be highly context-dependent. Second, policy discussion concerning educational inequality should include, not just the possibilities for re-mediating the skill levels of poor children, but also approaches to changing the environments that limit their opportunities.

Inequality in education in the US is widely documented and extensively researched. Policy interventions have often focused on helping children from disadvantaged backgrounds to develop skills which will supposedly help them succeed, but the effect of family background is largely viewed as a secondary source of educational inequality.

The skills of children are often measured through reports of behavior by teachers or parents. These often include internalizing vs. externalizing behaviors, antisocial behaviors, and tendency towards violence (e.g. fighting with classmates). While this is convenient, using behaviors as proxies for skills is problematic. Children’s behaviors, like adults’, are determined not only by their personal traits but also by the environment and constraints they are faced with.

Lundberg used the widely cited NLSY data in the US. NLSY data contain measurement of the “Big 5” personality inventory which are acknowledged metrics for personality in psychology. The five dimensions of personality traits are openness to experience, conscientiousness, extroversion, agreeableness, and neurotic-ism (emotional stability). Lundberg took the measures of personal traits for people from different backgrounds and estimated the correlation between these traits and educational attainment (defined as graduation from college, and broader educational achievement). She divided the sample into four groups according to two criteria: 1) mother’s education (some college and above vs. no college), and 2) whether the child was living with BOTH BIOLOGICAL parents at the first wave of the survey. These categorization breaks the sample down into four groups, with people having highly educated mothers and living with both biological parents being the most advantaged one.

A simple model about children’s educational outcome was presented. Children’s educational achievement is a function of focus and information. Focus is the personal efforts and information is the resources available. Her argument was that children in more advantaged households have more resourceful parents who will make their “focus” more rewarding. Parental resources are also assumed to be a perfect substitute for the information children can get at school. This assumption can be challenged if parents with higher social-economic status (SES) can improve the quality of the information.

Her findings suggest that people from more advantaged families receive a higher return (in terms of education) on conscientiousness, while people from disadvantaged backgrounds receive a higher return on openness. These patterns are the same regardless of gender, though the effects appear to be more distinct for men. There are racial differences as well — black men seems to receive a higher return on their openness.

It is not surprising that family background affects the returns to your skills. Being born in a rich and educated family gives a person access to good education, nice living environment, less obligation to shoulder family responsibilities (though this might not be a good thing) and more opportunities to explore personal interests. The resources of the parents have an instrumental impact on how children perceive which skills will be useful. Expectations of the parents also matter. If parents see the education of their children as an investment, they would invest more in promising children, thus making the returns to “focus” higher for more gifted children. Peer effects (spillover) matter as well. The level of openness should matter more for someone who grow up in a poor neighborhood where a role model is nonexistent — he might assume this is the way it is and never dare to push himself nearer to success.

Although the topic is tremendously interesting, there is a caveat. These results cannot be interpreted as causality because personality may well be endogenous. Although personality seems to be stable during an individual’s lifespan, there is no definitive evidence that life-cycle events cannot change one’s personality. Moreover, some personal traits (e.g. openness) might interact with educational attainment and evolve. For instance, being in college might make an introverted person more open to new ideas and more adaptable to new cultures, which in turn increases his earning potentials.

Caroline Hoxby on Expanding College Enrollment for Low-income High-achieving Students

Professor Caroline Hoxby from Stanford gave a talk about her project with Sarah Turner (Virginia) on expanding college access for low-income capable students. This study sheds light on the importance of information on making life choices and the often underestimated impacts of peers in personal development.

She motivated the topic by showing the audience the college application decisions of high-achieving students from rich and poor families. The differences are striking. Students from high-income families apply to a carefully constructed portfolio of schools, most of which closely match their abilities (in terms of test scores). They apply to a few “safe choices” just to back up. Students from low-income families, however, rarely have a portfolio. Many of them apply to nonselective colleges which are much worse than what they can possibly get. A question naturally arises: Since low-income high-achieving students know their ability levels by pre-test scores, why don’t they apply for good schools which provide them with better resources?

Several explanations are proposed. The most convenient explanation is students from poor families are worried about high costs of better schools. But a closer look at the college costs lowers the credibility of this claim. Colleges which are more selective are actually cheaper to attend — probably because they have better finances and offer more scholarships. Note that low-income students may not know this fact. And lack of such information is likely to deter them from applying for good schools. Another hypothesis is that the matching between universities and poor students do not happen as efficiently as we would like to. Low-income students are spread out and universities may not have publicized enough to this geographically dispersed population. The third explanation is that they simply don’t want to go to a selective college. The distant, uncertain benefits of going to a school with higher ranking may not be high enough to justify the pain from leaving home and friends. Sticking with what you are familiar with is always a safe choice.

With these thoughts in mind, Professor Hoxby designed a randomized controlled trial measuring how different interventions impact college application decisions among low-income high-achieving student. For one treatment group, she mailed application brochures containing guidance from informed high school counsellors about college choices. The brochures are considerately customized, including comparisons between schools outside their states and schools within states which they are more familiar with. There are two other treatment groups: fee waiver group and parent intervention group. The former provides coupons for application fees (which these students are eligible for in the first place) , while the latter provides customized college application guidance emphasizing parents’ concerns (e.g. costs) and written in simpler language.

Their econometric model is very simple. They didn’t include co-variates because none of those were significant. Results suggest that fee waiver coupons seem to serve as “earnest money”. Students are much more likely to look at the materials inside the envelop if they see the free money first. This may also release the guard of parents, who often open their kids’ envelops. Treatment group students are indeed more likely to apply schools which match their abilities.

The data have just come out and is only a one-year survey. Much remains to be learned about the academic performance, social networks, earnings, and other choices of these individuals in the future. With survey data, response rate is always a problem. In this case, a big proportion of the treatment group students didn’t even look at the materials. This restricts the power of the conclusion as we now have a limited and probably censored sample of treated observations. Making things clear is an inherent difficulty in this survey: they were only able to provide students with very limited information of this study or students will find out the aim and incentives will be distorted.

New evidence against healthcare exceptionalism

Chad Syverson from Chicago Booth gave a talk at Duke about his new paper “Healthcare exceptionalism? Productivity and Allocation in the U.S. Healthcare Sector” with Amitabh Chandra, Amy Finkelstein and Adam Sacarny.

This paper was motivated by the large variation in medicare costs across regions and hospitals as well as the little correlation between health outcomes and costs. The authors ask whether healthcare industry allocate resources less efficiently than other industries. They find evidence against the widely held argument that healthcare industry does not operate with standard market forces. Specifically, they find that more productive hospitals have a higher market share in the current time period and are more likely to expand.

Standard productivity models are used to estimate the TFP of hospitals and then estimate the relationship between TFP and market share. Data on heart attack treatment are used to reduce the patient sorting (because heart attack is emergent and therefore people do not have much time to shop around) and to make use of relatively well-measured inputs.

Several key issues need to be resolved for their estimation. First, estimations of production function are subject to endogeneity problems. In this case, hospitals’ choice of inputs are likely to be correlated with their productivity. This issue is addressed by identifying the coefficients on inputs only from variation within hospitals. Second, input costs across regions and hospital may be incomparable because prices are different. The authors adjust for this effect using Diagnosis Related Group (DRG) and standard estimated costs averaged across the country. Third, censoring may occur in measurement for outputs. Their main analysis has a one-month horizon. But they changed the time window (shorter and longer) to show robustness of results. They also used a Bayesian shrinkage estimation method to adjust for estimation error in hospital TFP.

Prof. Syverson sees this paper as the starting point of a series of investigations into the productivity and allocation of healthcare industry. Many other interesting topics remain to be researched on: Does hospital management or doctors have a bigger impact on the productivity? Does change in ownership affect productivity? There is much to be done to demystify this industry.