Educational Measurement: Issues and Practice Summer 2014, Vol. 33, No. 2, pp. 5–13 Evaluating the Predictive Value of Growth Prediction Models Daniel L. Murphy and Matthew N. Gaertner, Pearson, Austin This study evaluates four growth prediction models—projection, student growth percentile, trajectory, and transition table—commonly used to forecast (and give schools credit for) middle school students’ future proficiency. Analyses focused on vertically scaled summative mathematics assessments, and two performance standards conditions (high rigor and low rigor) were examined. Results suggest that, when “status plus growth” is the accountability metric a state uses to reward or sanction schools, growth prediction models offer value above and beyond status-only accountability systems in most, but not all, circumstances. Predictive growth models offer little value beyond status-only systems if the future target proficiency cut score is rigorous. Conversely, certain models (e.g., projection) provide substantial additional value when the future target cut score is relatively low. In general, growth prediction models’ predictive value is limited by a lack of power to detect students who are truly on-track. Limitations and policy implications are discussed, including the utility of growth projection models in assessment and accountability systems organized around ambitious college-readiness goals. Keywords: growth prediction, on-track indicators O ne of the chief critiques of the original No Child Left Behind Act (NCLB, 2001) concerned its focus on sin- gle snapshot measurements of student proficiency to deter- mine whether schools were making adequate yearly progress (AYP). Accountability models based on proficiency status (i.e., status models) do not account for students beginning each school year at different achievement levels with different achievement histories, nor do they provide information about how far students progress over the course of a school year. Under status-based accountability systems, it is possible that schools responsible for great gains in achievement throughout the year would go unrecognized if their students begin with low achievement levels (Linn, 2004; Novak & Fuller, 2003). Critics have argued, therefore, that accountability systems based on student progress (i.e., growth) are more valid than those based on status alone. In response to the criticism, Education Secretary Mar- garet Spellings introduced the Growth Model Pilot Program (GMPP; U.S. Department of Education, 2005) to provide a method for giving credit to schools for students who demon- strate adequate growth throughout the school year. Defining criteria for what constitutes an adequate amount of growth throughout a school year, however, is difficult (Betebenner, 2009). It is possible, for example, for nonproficient students to remain nonproficient despite substantial growth. Short- term growth without a long-term proficiency goal does not align with the primary aim of NCLB that all students must be proficient by 2013–2014. Therefore, to align the definition of growth with the goals of NCLB, the GMPP defines growth Daniel L. Murphy and Matthew N. Gaertner, Pearson, 400 Center Ridge Dr., Austin, TX 78753; dan.murphy@pearson.com; matthew.gaertner@pearson.com in terms of progress toward future proficiency. The family of models developed to measure growth toward proficiency has been termed growth prediction models (Castellano & Ho, 2012). There are two assumptions underlying accountability sys- tems that incorporate growth prediction models. The first assumption is that students exist who are truly progressing along a trajectory from nonproficiency to proficiency. If stu- dent proficiency is constant or decreases over time, then accu- rate growth prediction models would give no credit to schools for students’ growth under the terms of the GMPP. Inaccu- rate models, on the other hand, would “overrate” schools (i.e., give underserved credit) by overestimating future achieve- ment outcomes for some students. The second assumption is that, if some nonproficient students truly are progress- ing toward proficiency, growth models can accurately dis- criminate between students who are on-track and students who are not. Models that fail to identify students who are truly on-track will underrate schools in an accountability system. The validity of an accountability model anchored to future proficiency goals is therefore affected by its growth model’s power (its ability to identify students who are truly on-track) and Type I error (the proportion of students incorrectly iden- tified as on-track). The purpose of this article is to examine how growth model power and Type I error, which relate to prediction accuracy at the student level, affect school-level accountability ratings. The ideal growth model combines high power with low Type I error rates—two features critical to establishing va- lidity evidence for an accountability system based on growth to proficiency. The extent to which commonly used growth models provide such evidence has not been thoroughly inves- tigated. C  2014 by the National Council on Measurement in Education 5