Research Note - Vol.2501 - Using Student Scores on FAST PM1 Assessment to Estimate Student Performance on PM3 Assessment

Q: What is the main goal of using FAST PM1 scores to estimate FAST PM3 performance?

To estimate FAST PM3 achievement outcomes using PM1 assessment results, improving predictive power and giving teachers and schools actionable information to support student performance.

Q: Why was ordinal logistic regression selected instead of linear regression?

Linear regression produced scale score estimates that were harder to interpret and were affected by regression toward the mean. Ordinal logistic regression better suited achievement-level prediction probabilities.

Q: How were the prediction outcomes validated?

Data were randomly split into two halves: one half to fit the model and the other half to validate it. The validation compared predicted versus observed percentages in achievement levels 3-5 and compared expected versus observed mean achievement levels.

Oct 15, 2026

Upload by วิน

Brief research note explaining how FAST PM1 and PM2 assessment results can be used to estimate student performance on FAST PM3. It describes an ordinal logistic regression approach that models PM3 achievement-level probabilities from PM1 scale scores, run separately by PM1 grade and subject area. The document validates prediction accuracy via half-sample fitting and comparison of expected versus observed achievement levels 3-5 and mean achievement levels, reporting good overall adequacy.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages3

Reading Time8 min

AccessFree to download

Methodology

Validating the Model

Results

Fitting of the Model

Validating the Model

Discussion and Limitations

Research Note Vol.2501 Nastasia Schreiner, Ph.D. November 2025 Aleksandr Shneyderman Ed.D. Using Student Scores on FAST PM1 Assessment to Estimate Student Performance on PM3 Assessment This brief explains how the results of the FAST PM1 and PM2 assessments can be used to provide FAST PM3 student performance estimates. Using scores from the same assessment to make such estimates may improve the predictive power and provide teachers and schools with actionable information to improve student performance. Methodology It is possible to use a linear regression to estimate the prediction equation that would provide scale scoreson PM3 from those on the PM1 assessment. Such a regression fits the data well with the correlation coefficients varying between 0.73 and 0.81 across all grade levels. On the other hand, estimated scale scores are more difficult to interpret and may provide less actionable information than the achievement levels. In addition, the results of the ordinary linear regression are subject to the regression toward the mean phenomenon, where the expected scores on PM3 for students who score either very low or very high on the PM1 assessment would gravitate toward the mean scale score for a given grade level. Because of these considerations, ordinal logistic regression was selected as an analytical tool to carry out the prediction. Specifically, the 2024-2025 FAST PM3 achievement levels were used as the outcome while the scale scoreson the PM1 were used as the sole predictor in the model. Because many students participate in the FAST assessment above their enrolled grade, especially in mathematics, the model was run separately for each of the PM1 test grades and for each of the tested areas of ELA and Mathematics. This model predicts the probability of scoring in each of the five achievement levels on PM3 using the scale score from PM1 . In order to validate the outcomes ofthe model, all the data were randomly split in half. One-half was used to fit the model, and the other half was used to validate the outcomes. Such an approach was taken because when the same data are used for both fitting the model and examining the prediction success, the results generally appear more positive than they are if the model parameters are applied to the new data and then the actual and the predicted scores are compared. 1 Research Services Office of Assessment, Research, and Data Analysis 1450 NE Second Avenue, Suite 208, Miami, Florida 33132 (305) 995 7512 Fax (305) 995-7522 Results Fitting of the Model The ordinal regression model fit the 2024-2025 FAST ELA and Mathematics data well. The goodness-offit outcomes were statistically significant (p < .001) for both ELA and Mathematics and for all test grades. The Nagelkerke pseudo R-squared statistic varied from 0.533 to 0.667 across test grades in ELA and from 0.437 to 0.639 in Mathematics. These values indicate sufficient adequacy of the model. It is important to note that pseudo R-squared does not have the same interpretation as the R-squared in the ordinary linear regression. That is, one cannot say that when the pseudo R-squared is 0.65, 65% of the variations in the outcome are accounted for by the predictor in the model. Validating the Model Once the model was fit on one-half of the data, its parameters were used to calculate the outcomes using the other half of the data. These calculated outcomes included the probabilities of scoring within each of the five achievement levels as well as the expected values of the achievement levels, and cumulative probabilities of scoring within achievement levels 3-5. Then, two validation comparisons were performed for each tested subject and grade: 􀁸 The observed percentage of students scoring within achievement levels 3-5 was compared to the expected percentage of such students. 􀁸 The mean observed achievement level on PM3 was compared to the mean of the expected achievement level. The table below shows t

FAQs

What is the main goal of using FAST PM1 scores to estimate FAST PM3 performance?

To estimate FAST PM3 achievement outcomes using PM1 assessment results, improving predictive power and giving teachers and schools actionable information to support student performance.

Why was ordinal logistic regression selected instead of linear regression?

Linear regression produced scale score estimates that were harder to interpret and were affected by regression toward the mean. Ordinal logistic regression better suited achievement-level prediction probabilities.

How were the prediction outcomes validated?

Data were randomly split into two halves: one half to fit the model and the other half to validate it. The validation compared predicted versus observed percentages in achievement levels 3-5 and compared expected versus observed mean achievement levels.

Oct 15, 2026

Upload by วิน