Run better research, build conclusions that hold up
Practical, reviewable guidance on research design, surveys, sampling, data management, statistics, and reporting.
What Should Ad Creative Testing Measure? The Favorite Version May Not Have Completed the Communication Task
A layered evaluation framework—spanning exposure opportunity, message comprehension, brand attribution, joint valid reception, and action intent—explains why the most-liked ad version can still lose. A 1,000-person demo-vs-story × forced-view-vs-feed experiment illustrates why liking alone cannot decide the winner.
How to Conduct Informed Consent and Protect Privacy in AI Interviews? Start with Data Flows, Not Checkboxes
Implement layered disclosure covering AI roles, recording media, purposes, access, retention, sharing, withdrawal, and incidental disclosure. Support commitments through object-level data inventories, minimization, and re-identification review.
How to Write an AI Interview Guide: Probing for Evidence, Not Reinforcing Assumptions
Translate research questions into topic objectives, primary questions on concrete experiences, neutral evidence probes, transitions, and stopping conditions. Pre-test each version using coverage, completeness, leading, and safety metrics.
AI Interview: Text, Voice, or Video? Decide by Evidence Yield, Not Richness
Compare the three modalities on evidence capability, completion burden, mode effects, transcription errors, privacy, and accessibility, then select a single or mixed mode based on the usable evidence per invitee.
Are Employees Actually Using Generative AI? Measure Access, Behavior, Trust, and Avoidance Separately
Baseline adoption, adoption drivers, and governance readiness are three different questions. Measure availability and authorization, concrete usage behavior, trust split by object, and boundary concerns as four separate evidence layers so that an access problem is not misread as an employee-attitude problem.
How to Code, Validate, and Turn Open-Ended Responses into Reliable Conclusions?
From analytical paradigms, sampling for codebook development, and dual coding to multi-label proportions, AI assistance, and rechecking original text, a complete worked example using 600 responses illustrates how open-ended responses can lead to auditable conclusions.