A middle ground on performance pay

See all posts
Incentive systems based on both individual or school-level results have their fans and detractors. Individual-level systems establish clear lines of responsibility but are viewed as encouraging competition and stifling collaboration. School-based systems remove the competition, but muddy responsibility:  the most effective teacher may think her efforts get lost in aggregation - that others may not be pulling their weight.

District administrators take note: the answer may lie somewhere in between. Incentives can be structured so that teachers can see how their efforts will impact the students for whom they are responsible without focusing solely on individual performance.

At least, that's the conclusion Michael Lovenheim and Scott Imbermann reached in a new NBER paper using seven years of data from the Houston Independent School District's ASPIRE program.

ASPIRE has been through a few iterations. The version studied in the paper awards bonuses to high-performing high school teachers in each of the core subject areas (English, mathematics, history, and science) based on value-added measures for each subject-grade combination district-wide.

The study uses variation in teacher assignments and course loads (in addition to extensive school and student controls) to estimate the impact "share" has on the effectiveness of ASPIRE. (Share is a measure of the proportion of students a teacher is responsible for in each subject-grade combination at a particular school.)

This approach finds a happy medium between most performance pay models by grouping teachers who teach the same subject and grade and comparing their value added scores to the same groups in other schools. The findings were significant for student achievement gains in mathematics and social studies, and to a lesser extent in English; there was no significant impact on science scores. 

In fact, the authors recommend a very specific design that is quite different than most of the incentive structures that have been studied (and found for the most part to have little or no effect): connect student data to groups of 3 to 5 teachers.

--Rob Rickenbrode