Questions and contributions: distinguish competing explanations

Audience and output

For readers with subject knowledge preparing a dissertation or proposal. Produce a two-page question memo and a claim–evidence map. The example and exercises below are platform-designed, fictional teaching material, not findings or assignments from the cited scholars.

Uri Alon's problem-choice essay and Simon Peyton Jones's writing talk offer researchers' perspectives. Use them to examine decisions, rather than treating scholarly reputation as evidence that a strategy succeeds.

Build competing explanations

Suppose you study AI-assisted paper reading. Improved immediate answers might reflect conceptual understanding, more fluent summaries or extra practice time. Delayed transfer, withdrawal of the tool and an equal-time comparison distinguish these explanations differently.

Refine the candidate question to: does structured AI questioning improve reasoning about an unfamiliar paper under equal reading time? Specify population, material, intervention, comparator and time window. Novelty remains unestablished until a scoped literature search.

Establish the gap and a possible refutation

1. Read directly relevant studies and a recent review. Follow citations backward to method sources and forward to subsequent tests.
2. Extract claims, measurements, comparators, units of analysis, limitations and available materials, with full-text locations.
3. Explain disagreements through differences in populations, measurement or identification conditions.
4. State an observation that would weaken your preferred explanation: improved fluency without transfer would challenge the understanding claim.

An unsuccessful search establishes only what was not found within its databases, date and query scope. Explain why the remaining uncertainty matters.

Compare feasible designs

A self-selected-user survey is inexpensive but leaves selection concerns. Randomized equal-time tasks address an intervention question but still require valid measurement and attrition handling. Interviews can examine practices and meanings, without estimating an average intervention effect. Match the design to the question.

For each candidate, record resources, likely failure, the smallest informative check and a stopping condition. If relevant data are unavailable, narrow the claim or change the question.

Review criteria

• Map each contribution to its comparator, required evidence, available evidence, strongest objection and scope.
• Distinguish planned tests from observed findings. New methods need fair baselines; measurements need validity arguments; theory needs assumptions and derivations.
• Ask a subject reader for the strongest alternative explanation and document the resulting design revision.
• Explain what the project would teach if its preferred hypothesis failed.

Open the scholar resources, then continue to causal design or evidence synthesis.

Advanced practice and review evidence

Separate delayed retention, transfer and answer fluency in the claim that AI improves learning. For each of three questions, identify two explanations and an observation that would change your preferred explanation. Without recruitment, compare how published studies measure transfer, acknowledging that this does not identify your own intervention effect.

Review operational definitions, traceable gap evidence and a genuinely disconfirming observation. Transfer: distinguish instrument drift from sample differences in a measurement problem; redesign the evidence rather than reusing learning outcomes.

Open practice, AI review and work export.