TTR ValAdd NBER
TTR ValAdd NBER
TTR ValAdd NBER
Reviewed By
Dale Ballou Vanderbilt University February 2012
Summary of Review
A new report titled The Long-Term Impacts of Teachers concludes that teachers whose students tend to show high gains on their test scores (called high value-added teachers) also contribute to later student success in young adulthood, as indicated by outcomes such as college attendance and future earnings. To support this claim, it is not sufficient for researchers to show an observed association between teacher value-added and later outcomes in young adulthood. It is also necessary to rule out plausible alternative explanationsfor example, that parents who did the most to promote their offsprings long-term success also endeavored to secure high value-added teachers for their children. This review explains that, for the most part, the evidence needed to rule out these alternatives is missing from the report. Thus, policy-makers should tread cautiously in their reaction: the case has not been proved.
http://nepc.colorado.edu/thinktank/review-long-term-impacts
1 of 11
Kevin Welner
Editor
William Mathis
Managing Director
Erik Gunn
Managing Editor
This is one of a series of Think Twice think tank reviews made possible in part by funding from the Great Lakes Center for Education Research and Practice. It is also available at http://greatlakescenter.org.
This material is provided free of cost to NEPC's readers, who may make non-commercial use of the material as long as NEPC and its author(s) are credited as the source. For inquiries about commercial use, please contact NEPC at nepc@colorado.edu.
http://nepc.colorado.edu/thinktank/review-long-term-impacts
2 of 11
I. Introduction
Value-added assessment of teachers is an attempt to measure the contribution of teachers to student learning, as gauged by progress on standardized tests. As a procedure for evaluating teachers, it remains controversial and widely unpopular with teachers. Part of this controversy involves concerns that a policy of high-stakes teacher evaluation linked to students test scores would intensify practices such as the narrowing of curriculum and teaching to the test. A second part involves questions of whether value-added models are reliable and valid. This review focuses on the latter, modeling and measurement questions, setting aside for the moment those other elements of the policy debate. Within the measurement discussion, there are concerns that the statistical methods do not sufficiently control for other factors that influence test-score gains. But this is not all. The focus on testing is deemed unsatisfactory, in light of the following: (1) Tests are often poorly aligned with the curriculum and fail to reflect what is taught; (2) Test performance is noisy, measuring student ability with considerable measurement error; (3) Other important parts of a teachers job (e.g., building character, teaching citizenship) are not captured by performance on tests. Thus, evaluating teachers on the basis of students performance on a single test (an examination that frequently holds no consequences for students) can yield a reading on teacher performance that is incomplete at best, and possibly seriously misleading. A new report, The Long-Term Impacts of Teachers: Teacher Value-Added and Student Outcomes in Adulthood, by Raj Chetty, John N. Friedman, and Jonah E. Rockoff, adds important new evidence that helps address these concerns. Using data from a large urban school system, the report asks whether teachers who are effective at raising students test scores also have positive impacts on students subsequent transition from school to young adulthood. 1 Among these indicators are the probability of going to college, earnings once students enter the workforce, the probability that female students give birth as teenagers, the quality of the neighborhood in which students live at age 25, and the establishment of a retirement savings account. The study finds a broad pattern of positive effects associated with teachers who are effective at raising student test scores. Given these observed positive effects, the report
http://nepc.colorado.edu/thinktank/review-long-term-impacts
1 of 11
goes on to consider how public policy might enhance students opportunities to learn from such instructors.
also more likely to go on to college, obtain higher-paying jobs, etc. Such associations might arise for a variety of reasons other than the quality of the teacher. (For example, parents who care enough to get their children into the classrooms of the best teachers might also do a better job of preparing them for lifes later challenges). The report claims that these positive outcomes are the result of having had higher value-added teachers, above and beyond any association that might arise for other reasons. The methods used to support these causal claims are described in Section V below. Finally, it should be borne in mind that the foregoing findings are estimates and subject to error. However, even allowing for estimation error, the evidence is strong enough to reject with a high degree of confidence the hypothesis that the true impact of high -value-added teachers on these long-term indicators is zero. The report also investigates whether the long-term impact of high value-added teachers varies over student subgroups: female and male, low and high income, and minority and non-minority. For each of these subgroups, only one long-term outcome is examined: the
The report is highly persuasive on a key point: teacher value-added has been measured free of significant bias. However, measuring value-added is only a step along the way to the reports larger findings: that high value-added teachers improve students long-term outcomes.
quality of the college attended at age 20. The impact of teacher value-added is greater for females, high-income students, and non-minorities, suggesting the presence of complementarities between teacher value-added and family inputs. Students more likely to experience better outcomes anyway appear to be more receptive to the impact of these high value-added teachers. However, the impact is positive in all of these groups. The estimated effect of a high-value-added teacher is also greater when that teacher is encountered in middle school rather than in elementary school, though the impact, again, is positive at both levels. Given the broad pattern of positive long-term effects, the report goes on to consider implications for public policy: what can be done to enhance students opportunities to learn from such instructors. Bonuses might be paid to effective teachers, for instance, to increase the probability that they remain in the profession. However, the authors conclude that such an approach is not cost-effective, since most of these bonuses would be paid to teachers who would remain in the classroom anyway. The net gain in students future incomes would barely exceed the cost of financing the bonuses. 2 A second option described in the report explores whether replacing teachers who have the lowest value-added scores with teachers of average effectiveness has much larger net gains. Applied to teachers in the bottom 5% of the value-added distribution, such a policy generates substantial economic benefits. The expected gain increases the longer one waits to make the replacement, because more years of data are then available to determine
http://nepc.colorado.edu/thinktank/review-long-term-impacts
3 of 11
whether such a teacher is truly in the bottom 5%. However, waiting imposes a still greater cost, as these teachersmost of whom would likely remain low value-added teachers remain in the classroom longer. Assuming a class size of 28 students, the report contends that a policy of replacing a teacher who turns up in the bottom 5% of the distribution in her first year is expected to raise the lifetime earnings of that class by $135,000 (in present value). Given the number of years such a teacher might otherwise remain on the job, the total gains are potentially substantial.
http://nepc.colorado.edu/thinktank/review-long-term-impacts
4 of 11
factors have biased estimates of teacher value-added. The report concludes on the basis of these tests that there are no significant biases. The second stage of the analysis investigates whether the measures of teacher value-added obtained in the first stage are related to students long-term outcomes. Information on these outcomes is obtained from tax records, which are merged with school district student records. College attended is identified from 1098-T forms filed by postsecondary institutions on tuition payments received for every student. Neighborhood quality is derived from zip codes on 1040 forms. Teenage births are detected based on whether a young woman claims a new dependent on her tax returns. Information on students parents is obtained from the earliest 1040 form on which the student was claimed as a dependent. The use of tax records means some outcomes are not measured with perfect accuracy (for example, teenage births to women not filing tax returns). However, the overall match rate is above 80% and the amount of information obtained in this manner is impressive. To ascertain whether teacher value-added has affected these outcomes, it is again necessary to estimate a model. The mere fact of an association between teacher value added and a subsequent outcome, such as earnings, does not in itself establish a causal relation. It is also necessary to rule out the possibility that other factors, related to both value-added and earnings, is responsible for the observed associationa point to which we will return in Section V.
http://nepc.colorado.edu/thinktank/review-long-term-impacts
5 of 11
outcomes to the impact of teachers. Once again, the principal difficulty arises with regard to the way students are assigned to teachers. To the extent that these assignments can be affected by parents, it may be that parents with a high concern for their offspring s longterm success take steps to place them in the classrooms of the most effective (i.e., high value-added) teachers. Or they may be the passive beneficiaries of school policies such as ability grouping that make such assignments more likely. That the offspring of these same parents then enjoy greater success in pursuing higher education, in obtaining jobs, etc., may say more about their parents than the quality of their instructors. An association between high value-added teachers and subsequent life success would be observed, but it would not be a casual relation, but the result of good parenting. Thus, the second task is to establish that no such selection mechanism has been at work: that high value-added teachers are not more likely to have been assigned students who were for other reasons (e.g., parents) destined for greater long-term success. This section reviews the methods employed in the report to establish these two claims. To anticipate, the report contains compelling evidence supporting the first: that measured value-added is free of bias. However, the report contains much less evidence supporting the second: that students on a path to greater life success were not more likely to have been placed in classes taught by high value-added teachers. The same methods used so persuasively to establish the first point are not consistently applied to establish the second, though there is no apparent reason why they could not have been.
http://nepc.colorado.edu/thinktank/review-long-term-impacts
6 of 11
second kind of test, based on quasi-experimental variation in the assignment of teachers to students: specifically, changes in average value-added caused by teachers moving in and out of a particular grade level within a school. (These measures of value-added are based on teachers performance in other years, not the year of the move.) For the most part, parents do not have time to respond to such moves by changing schools, nor is it likely most would do so based on the movement of a small number of teachers. Of cour se, parents can still attempt to pick and choose among the teachers at their childs grade level. For this reason, the quasi-experimental test relies not on the value-added of a given students teacher. Instead, it relies on mean value-added of the teachers providing instruction in that students grade. Sorting of students into particular classrooms within the grade does not affect this mean. Thus, if there is an association between mean teacher value-added and subsequent test scores for students in that grade, it cannot be due to the way students were assigned to teachers within that grade, but only to the movement of teachers in and out of that grade. In other words, the association would be insulated from the factors that assign certain students to certain teachers in such a way as to bias valueadded estimates. Two questions are now asked. First, do test scores increase in the year of the move when teachers coming into a grade have higher value-added than the teachers who left? Second, can we rule out the occurrence of other suspicious changes at the same time? For example, are there simultaneous changes in parental characteristics that would predict improved scores? Are there changes in test results in subjects where mean teacher value -added has not changed, or in adjacent grades that are unaffected by teacher moves? The answer to the first question is yes. An increase in mean teacher value-added at the grade level is highly predictive of an increase in average year-end test results. This cannot be attributed to the favorable matching of students to teachers, because it applies to the entire grade: it is the mean change. The answer to the second question is no. There is no detectable change in the make-up of the students attending the school that would have predicted this increase. The placebo tests (looking for higher scores in subjects and grades not affected by teacher moves) also show no change. Thus, it is unlikely that the change in scores is attributable to anything other than the improvement in teacher quality resulting from teacher movement, as measured by teacher value-added. It is worth noting that these results could not have been taken for granted. The movement of teachers among schools is not the same as a controlled experiment. Depending on district policies, the configuration of neighborhoods, etc., there may be a tendency for better teachers to seek and obtain jobs in better schools. Or teachers may be moving along with students as attendance zones change and as schools open and close. Thus it is important to show that the movement of a high-value-added teacher into a particular school was not accompanied by other changes that would have produced higher test scores. The fact that no such changes were detected strengthens the conclusion that t he subsequent improvement in test scores had no other cause than the arrival of high valueadded teachers.
http://nepc.colorado.edu/thinktank/review-long-term-impacts
7 of 11
It must be established that the observed association between value-added and later earnings, college attendance, etc., is not the result of some third factor, such as good parenting.
now is to test whether family background factors that predict long-term success in employment, college attendance, etc., are distributed in such a way as to favor high-valueadded teachers. It should not be assumed that the same mix of factors that predict s high test scores necessarily predicts the others. While one would expect some overlap, it is likely that success in finding a job or in avoiding a teen pregnancy also depends on the values that parents impart and on their ability to shape character, factors that may not be of equal importance in predicting cognitive function on a standardized achievement test. Thus, the tests for bias need to be run again, allowing for the mix of family background factors to change depending on which long-term outcome is under examination. Indeed, assuming that all we need do is test whether high-value-added teachers have benefitted by being assigned students whose family backgrounds predict high test scores is to beg the question that is the focus of this report: do the same factors that predict high scores also predict these other outcomes? Second, a set of quasi-experimental tests could be conducted, relying once again on variation in teacher value-added that arises when teachers move between grades or schools. The same questions should be asked as before. When high-value-added teachers move into a grade, do we see improved long-term outcomes? Do we see improved longterm outcomes where they are not expected (e.g., for students in an adjacent grade that did not experience a change in the mix of teachers)? The reason for conducting these quasiexperimental tests is the same as the reason given in the preceding paragraph. The fact that similar tests have already validated the estimates of teacher value-added does not imply that they are not needed to validate inferences about the impact of teachers on long term outcomes. Indeed, it is probably even more important that these tests be conducted with respect to outcomes like earnings and the avoidance of teen pregnancy. Data ava ilable
http://nepc.colorado.edu/thinktank/review-long-term-impacts
8 of 11
from tax returns are not likely to distinguish well between families that nurture the development of character and families that are much less successful in this task. Unobserved differences between families are likely to be very important. The way to test whether high-value-added teachers have been systematically assigned more students whose families are richer with respect to these unobservable factors is to conduct the quasi-experimental tests described above. Absent that, we will not know whether such differences were present and were the underlying reason for the observed association between teacher value-added and students long-term success. The report contains the results of few such tests. There are no tests of the first kind, using additional data on parents from tax records to see whether the inclusion of this information in models predicting long-term outcomes takes away from the explanatory power of teacher value-added. As for quasi-experimental tests, only two are reported. An increase in mean teacher value-added caused by teacher movement in and out of a grade raises the probability of enrolling years later in college and improves college quality. However, no tests of this kind are reported for the other long-term outcomes investigated: earnings, residential neighborhood quality, the probability of a teenage birth, and contribution to a retirement savings account.
http://nepc.colorado.edu/thinktank/review-long-term-impacts
9 of 11
earnings are not sufficiently validated; more evidence on that point needs to be presented before policy and practice are shaped in response to this report. The report raises a question that is both important and timely. If the reports conclusions regarding long-term effects can be substantiated, it would strongly suggest that high value added teachers do more than simply raise test scores. To at least a modest extent, these teachers would be shown to be transformative, changing students for the better in ways that do not show up for years to come. Given the number of students with whom a teacher comes into contact over the course of a career, these modest impacts could have a large cumulative effect.
http://nepc.colorado.edu/thinktank/review-long-term-impacts
10 of 11
1 Chetty, R., Friedman, J. N. & Rockoff, J. E. (2011). The Long-Term Impacts of Teachers: Teacher Value-Added and Student Outcomes in Adulthood. NBER Working Paper No. 17699. Retrieved January 29, 2012, from http://obs.rc.fas.harvard.edu/chetty/value_added.pdf. 2 The report does not estimate the impact of bonuses on the probability of a teachers remaining in the classroom, but it uses an estimate from a study of North Carolina teachers, finding that a $1,800 bonus reduced teacher attrition by 17%. See: Clotfelter, C., Glennie, E. Ladd, H., & Vigdor, J. (2008). Would higher salaries keep teache rs in high-poverty schools? Evidence from a policy intervention in North Carolina. Journal of Public Economics 92, 1352-70. Whether this estimate would hold good in the urban system considered in this study is open to question. The bonus program examined in the NC study was also short-lived, raising doubts about the long-term versus the short-term response. It is possible that the long-term response to such a bonus differs from the short-term response: teachers might defer a decision to quit in order to obtain a bonus, but ultimately they would leave anyway. If 17% overstates the decline in attrition in response to a bonus of $1,800, such a policy would be still less cost-effective. 3 The description in this review omits many technical details. Omission of a particular issue should not be taken as an indication that the report likewise fails to mention it. 4 The term teacher of average ability must be qualified. It is the teacher of average ability for students with of the same characteristics. If there are differences in average teacher quality based on student, class, and school characteristics, these show up in the estimated effect of those variables on achievement, and are not part of the residual variation in student achievement that is the basis of teacher value-added.
http://nepc.colorado.edu/thinktank/review-long-term-impacts
11 of 11
DOCUMENT REVIEWED:
The Long-Term Impacts of Teachers: Teacher Value-Added and Student Outcomes in Adulthood Raj Chetty, John N. Friedman, and Jonah E. Rockoff National Bureau of Economic Research December 2011 February 16, 2012 Dale Ballou, Vanderbilt University dale.ballou@vanderbilt.edu (615) 322-8039
AUTHORS:
PUBLISHER: DOCUMENT RELEASE DATE: REVIEW DATE: REVIEWER: E-MAIL ADDRESS: PHONE NUMBER: SUGGESTED CITATION:
Ballou, D. (2012). Review of The Long-Term Impacts of Teachers: Teacher Value-Added and Student Outcomes in Adulthood. Boulder, CO: National Education Policy Center. Retrieved [date] from http://nepc.colorado.edu/thinktank/review-long-term-impacts.