Chapter 3: Choosing A Value-Added Model

Sanders and Rivers argue that value-added assessment emerges from the principle that each student, regardless of their current academic ability, should make at least a year’s growth in a year’s time. The authors note that value-added provides three benefits: (1) the data are useful for instructional and administrative decisions (such as where to assign students and teachers), (2) it is a fairer way to measure the performance of schools because it is about growth, not overall performance (which is closely linked to economic status of the students), and (3) value-added is the best measure to make the basis of an accountability system designed to reward highly effective administrators and teachers and remediate ineffective administrators and teachers.

Value-added is a measure of growth for a student that compares a student’s actual growth with their expected growth. Variance from expectation, positive or negative, is considered a measure of school or teacher impact. Past student scores are used to estimate future scores by mapping the student trajectory. The student’s projected score is based on their prior scores. The more prior data available, the more reliable the projected score will be. Value-added is not just the score a student receives on n assessment, but the difference between the score and the projected score.

Since all value-added models are not created equal, it is important to do a short review of the most common methods and their advantages and limitations.

Class average gain: Each students previous year score is subtracted from the current year assessment score. An average growth for the class is calculated and teachers are compared.

• Simple to calculate.


• Only students with current and previous year scores.


• Might be unstable.


• Tests must be on same scale.


• Requires random distribution of students.

ANCOVA, one previous score: A student’s score for the current year is predicted using the previous year’s score by regression.

• Simple model to fit.


• Tests do not have to be on the same scale. • Bias because only one year used to predict.


• Only students with previous year re included in the calculation.

ANCOVA, multiple scores: At least three previous scores are used to calculate the predicted score.

• Dampens error measurement.


• Can be easy model to fit if the software is available.


• Tests scores do not need to be on same scale.


• Many students will not have sufficient testing history to be included.

Univariate Response Model (URM): This model is similar to the previous model in that it uses multiple previous tests, but it is different in that the student need only have three prior scores and not scores from all previous test periods are not required.

• Test scores not required to be on the same scale.


• Minimizes concern about student selection bias because fewer students are excluded.


• Difficult to calculate using available software and computing power required is extensive.

Multivariate Response Model (MRM): Uses all student test scores from all grades and test periods available. These model tend to be conservative.

• All data from students are used regardless of how complete.


• Reduces measurement error.


• Student selection bias is reduced.


• Since the student score is linked to all past teachers there is opportunity to reduce potential interference.


• Software are not currently available to do these calculations.

This chapter ties in well with another new book out on value-added: “Value-Added Measures in Education: What Every Educator Needs to Know” (Douglas Harris 2011). Harris, a professor of education policy, has written a book specifically for educators explaining value-added measurement. Harris sets out to dispel eight common misconceptions about value-added:

1. We cannot evaluate educators based on value-added because student variability, because value-added model have flaws, because the assessments are inadequate, because teaching I complicated, and because the individual needs of student are too diverse.


2. Value-added is fair for teachers, but not for students.


3. Value-added are not useful because they are summative, not formative.


4. Because value-added involves comparing teachers to one another, and there are no absolute value-added standards value-added measures are not useful for accountability.


5. Because we know o little about the effects of value-added, we cannot risk our kids’ futures by experimenting with it.


6. Value-added is too complicated for educators to enderstand.


7. Value-added imply represents another step in the process of industrializing education, making it more traditional and less progressive.


8. Value-added is a magic bullet that by itself will transform education.

Harris explores each of these misconceptions and concludes that value-added can be used and that policy makers must consider a series of tradeoffs when implementing the use of value-added. Harris’s book is a useful compliment to OPE for the study and implementation phase of any effort to use value-added for evaluation and compensation.