Why Assignment Gains Do Not Automatically Travel to Exams

With E-homework, the critical comparison is not simply screen versus paper. It is assisted assignment performance versus more independent exam performance. This chapter compares those two outcome channels using both their average estimates and their variability, then shows what a learner must carry beyond the platform if an assignment improvement is to become an exam improvement.

The distinction is easy to miss because both outcomes are reported as scores. But a score describes performance under particular conditions. An assignment completed with immediate correction and repeated attempts measures a different situation from an exam in which the student must recognize, retrieve, and apply knowledge without the same support. The evidence becomes clearer when those conditions remain visible.

The observed contrast is about size, stability, and uncertainty

The focal K–12 meta-analysis separates academic achievement into assignment performance and exam performance. Its estimate for assignment performance is Hedges’ g = 0.368, with a 95% confidence interval from 0.257 to 0.479. The estimate for exam performance is Hedges’ g = 0.253, with a 95% confidence interval from 0.116 to 0.390. Both estimates are positive and statistically significant within their own analyses.

The first observable contrast is therefore magnitude: the assignment estimate is larger. But the more important contrast for the title is stability. The assignment subgroup shows almost no heterogeneity across its included studies. The exam subgroup shows very high heterogeneity, meaning that its estimates vary sharply from one study to another beyond what sampling error alone would explain.

This combination supports a restrained interpretation. E-homework appears to have a stronger and more consistent relationship with assignment performance in the available evidence, while its exam relationship is positive on average but much less predictable. “More reliable” refers primarily to the pattern of variation: studies of assignment outcomes point in a more uniform direction, whereas exam outcomes depend more visibly on circumstances that differ across studies.

The word “appears” matters because the formal test comparing the two subgroups did not find a statistically significant difference between their pooled effects. A reader cannot take the fact that 0.368 is larger than 0.253 as proof that E-homework has a definitively greater effect on assignments. The data support the descriptive ordering and the sharp difference in heterogeneity, but they do not settle the direct comparison at the conventional level used by the review.

This is a common place for evidence to be overstated. Researchers may ask two separate questions—whether the assignment estimate differs from zero and whether the exam estimate differs from zero—and find a significant answer to both. That does not automatically answer a third question: whether the two estimates differ from each other. The third question requires its own comparison, and in this review that comparison was not conclusive.

flowchart LR A["Practice task"] --> B["Immediate feedback and revision"] B --> C["Submitted assignment score"] B --> D["Knowledge and strategy must travel"] D --> E["Independent retrieval"] E --> F["Changed or unfamiliar task demands"] F --> G["Exam performance"] D -. "transfer may weaken" .-> G

The diagram separates two ideas that are often combined.