Two-level measurement-based dynamic model of learning with α, β, and γ transition processes¶

Theory · All theories
Evidence · 6 claims (6 for) · 1 study (1 theoretical), q2 · 1 of 1 report an effect size · 6 claims rest on one study

Description¶

The article models a student's measured knowledge as a two-level system in which the measured-correct level holds the score and the measured-wrong level holds its complement. Learning incidents cause transitions between levels, with three processes: α-type changes driven by instruction acting on wrong knowledge, β-type changes acting on correct knowledge, and γ-type associative changes from internal interactions between correct and incorrect knowledge. The dynamics are expressed as differential equations, e.g. the overall form ds_k/dt = α_k s_k(1−s_k) + β_k s_k + γ_k s_k(1−s_k). The article states it is a measurement-based probabilistic model that 'doesn't model any cognitive processes of how the learning occurs'.

Design Implications¶

Context¶

Requirements¶

• Scores scaled to 0–1 with a fixed measurement window defining the instrument's coverage of the knowledge domain

Constraints¶

• The model treats guessing and general mistakes as random errors not included in the transition processes
• The β-process is assumed to have a very small success rate in current education settings and is not analyzed in detail
• The model does not describe actual cognitive processes of learning, only probabilistic relations between measured states

Target Learners¶

• physics students in pre-post assessed courses

Target Learning Objectives¶

• conceptual knowledge gains measured by pre-post testing

Claims¶

• FCI change-score versus pretest-score data from three classes show linear relations at high pretest scores and curving-down at low scores, diagnosable as α- versus γ-process dominance [+M]
• When the γ-process (associative interaction of correct and incorrect knowledge) is considered, the normalized gain correlates positively with pretest score [+M]
• Hake's survey found interactive-engagement courses achieved average normalized gains about two standard deviations greater than traditional courses, with overall gain–pretest correlation of +0.02 [+M]
• High-ability students' learning behaves as a dominant α-process even at low pretest scores, while average-ability students show more γ-process behavior [+W]
• Random measurement noise in pretest scores produces a negative contribution to the correlation between normalized gain and pretest score [+W]
• Under a dominant α-process, the normalized gain contains no pretest-score term and is uncorrelated with pretest score if α is uncorrelated with pretest score [+M]

Related Theories¶

• Generalized non-constant α model framing learning as an error-reduction process

Examples¶

• Fit change-score versus pretest-score data with α and γ process models to diagnose learning processes and distinguish student ability from instruction impact

Key Sources¶

• Lei Bao. (2007). Dynamic Models of Learning and Education Measurement. https://arxiv.org/abs/0710.1375