A Midpoint Model: Match the Feedback Loop to the Learning Job

For digitally delivered instructional feedback, the useful question is no longer simply whether a message arrives immediately or later. This synthesis connects the feedback loop, the heterogeneous meta-analytic evidence, and the timing contrast into one practical judgment: choose a feedback arrangement by the learning job it must support, the learner who must use it, and the later performance that will show whether it worked. The model is designed to make a timing decision more precise without pretending that the evidence supplies a universal recipe.

The unit of judgment is the whole feedback situation

The mechanism and the evidence operate at different levels, and keeping those levels separate prevents two common mistakes. The mechanism explains how a response could become learning: a learner attempts a task, an external source compares the response with a standard, a message is delivered, and the learner interprets and acts on it. The meta-analysis asks what happens when that loop is implemented across many different situations. Its moderate average effect shows that digitally delivered feedback can improve measured learning performance. Its heterogeneity shows that the loop is not instantiated in one stable form.

Those statements support each other. The mechanism tells us why timing might matter: a correction can be easier to compare with a response that is still accessible. It also tells us why timing is insufficient: the learner must notice the message, understand what it addresses, and decide how to use it. The meta-analysis supplies the population-level warning that these conditions vary. Immediate feedback, delayed feedback, the task being assessed, the message’s form and focus, and the way learners control the interaction are not interchangeable settings around a single underlying intervention.

This distinction changes how the broad numbers should be used. The overall Hedges’ g of 0.41 is evidence that feedback is worth taking seriously across the included comparisons. The I² of 61.02 is evidence that an average alone cannot identify the best arrangement for an individual course. The immediate and delayed timing estimates add another layer: both categories were positive, but the delayed category rested on five interventions rather than the 79 supporting the immediate category. The results describe a field of possibilities, not a winner’s podium.

A design judgment therefore needs more than a preferred clock. It needs an explicit account of the learner’s starting point, the work they are doing, the information the message supplies, the opportunity to process it, and the outcome that matters. The five questions below are not a generic teaching checklist. They are five ways of specifying the same feedback system so that a timing choice can be interpreted rather than admired.

Five questions before choosing a clock

What is the learning job and the desired outcome? A feedback system cannot be evaluated until “learning” has been made concrete. The desired result might be accurate recall, durable retention, a corrected procedure, better self-monitoring, or transfer to a new task. These outcomes overlap, but they are not identical. A learner can improve an answer on the next item because they have just seen the correction without yet understanding the process well enough to use it elsewhere. Conversely, a learner may spend more time working through a difficult comparison and show a smaller immediate gain while developing a more transferable understanding.