Evaluation of the Growth Model Pilot Project

Ω

Final Report

GMPP was initiated to allow states to experiment with adjustments to the No Child Left Behind Act (NCLB) status accountability system, in order to improve the validity of AYP determinations by giving schools credit for students who are making significant growth. The pilot allowed states, districts, and schools to count students who were "on track" to being proficient but not yet there. Under NCLB, such students were not counted as proficient for the purpose of AYP determinations.

The pilot was initiated in November 2005 with the goal of approving up to ten states to incorporate growth models in school AYP determinations. The project was written into regulation in late 2008; now any state may apply to use a growth model meeting certain core principles.

Currently, 15 states are implementing growth models under this authority: Alaska, Arizona, Arkansas, Colorado, Delaware, Florida, Iowa, Michigan, Minnesota, Missouri, North Carolina, Ohio, Pennsylvania, Tennessee, and Texas.

Key findings include:
• Growth models enabled additional schools to make AYP compared to status and safe-harbor rules alone, but the percentages of schools that made AYP because of the growth models were generally not large.
• The impact of growth models varied widely across states.
• Most (but not all) schools that made AYP by status would also have met their reading and math AMOs under a hypothetical "growth-only" model (i.e., one using neither status nor safe harbor but only growth).
• Controlled simulations comparing the impacts of different types of growth models on student and school growth results show that the "projection model" functions in stark contrast with "transition" and "trajectory" models.
• Simulations comparing the results of different growth models using the same data show that projection models have the highest correct classification rates for future proficiency: over 80 percent. These rates are 5 to 20 percentage points higher than trajectory and transition models, depending on the grade level and proximity to the growth model time limit. While the projection model is more accurate, it is theoretically more difficult to implement and to explain to practitioners and parents than the other models.
• Although not an option under the Growth Model Pilot guidelines, growth models not tied directly to proficiency standards could identify a broader contingent of students as making adequate growth than current models. One alternative to the GMPP-permissible growth-to-proficiency models that could be used with vertical test score scales is the difference between proficiency cut scores in successive grade levels.