Learn With Jesse
Menu

Method and corrections

How the evidence is labeled and checked

What the labels mean, how citations were verified, what I could not verify, what was corrected along the way, and how AI was used.

Evidence labels

Each technique carries one of three labels. The labels describe the research behind the technique as a whole, in settings like college courses. They are my judgment, made from the sources listed on each page, and each page explains why it got its label.

Strong evidence

Many studies, often including meta-analyses or large classroom studies, point the same way in college or closely comparable settings. Details such as the best timing can still vary.

Moderate evidence

Consistent results from several studies, but a narrower base, results concentrated in a few subjects, or effects that depend a lot on how the technique is carried out.

Emerging evidence

Promising early results, often from one project or setting, or a design whose parts are supported but whose combination has not been tested.

Strong: Retrieval practice, Spaced practice, Worked examples, Cooperative groups.

Moderate: Peer instruction, Feedback with a next step, Instructional video design.

Emerging: Explain It Yourself, Transparent assignments, Process-based assessment, AI tutor for practice.

A technique can have a strong label and still be a poor fit for your course. That is what the best-when and weak-when sections on each page are for.

How citations were checked

  • Every reference was checked at the level of authors, title, journal, year, volume, pages, and DOI against DOI registration records (Crossref and OpenAlex).
  • Numbers and findings in the text, such as sample sizes, effect sizes, and percentages, were checked against the abstract or an open full text. Where I could read only a summary, the page states the finding in general terms.
  • Where I could not confirm something, it is listed under "What I could not confirm" in that page's About section, so you can see the gap.
  • Where a widely quoted number does not hold up, the page says so. For example, the often-cited .70 effect size for formative assessment is not supported by a careful K-12 meta-analysis, which found about .20 (see feedback with a next step).
  • Findings are attributed to what the study measured. A laboratory result is described as one, and a result from one course or one school is described that way.

Limits of this method

  • This is an index of well-supported practices, not a systematic review. I searched for each technique's main evidence, not the whole literature. Where a page says "I did not find," it describes the searches done.
  • Some publisher sites and PubMed were not reachable during checking, and one DOI service rate-limited some lookups, so a few records were checked through OpenAlex instead. In places I could read only abstracts.
  • Labels are judgments. Reasonable people could place a technique one level higher or lower.
  • The research changes. Pages carry an updated date, and the AI-related pages in particular are likely to need revision.

Corrections made while building

Checking changed the pages. These are the changes worth knowing about.

  • Two DOIs I first wrote from memory, for the peer instruction implementation review and for the video-in-higher-education review, were wrong. The lookups caught them and I replaced them with the verified records.
  • Two other sources I had planned to cite could not be verified, so I dropped them.
  • The first draft of Explain It Yourself had three placeholder citations, which were replaced with checked details. It also cited the 1998 Phi Delta Kappan summary by Black and Wiliam and a 1998 cooperative-learning book. I replaced them with the full 1998 Black and Wiliam review in Assessment in Education and the 2009 Johnson and Johnson review, whose details I could verify.
  • The same draft said that students who generate explanations remember better than students who read them, and that preparing to teach produces more learning. The cited studies tested generated words and effort toward learning with a computer agent. The page now says that applying them to explanations is an extension, not a tested result.
  • The Bastani and colleagues high school study in PNAS has a published correction. I checked it: it fixes one author's affiliation and does not change the findings.

AI disclosure

Model used for research, drafting, editing, and verification: claude-sonnet-5-5 (Claude, by Anthropic). The author is responsible for every claim and citation.

The first draft of the Explain It Yourself page was written with claude-haiku-5-5, then rebuilt and re-checked with claude-sonnet-5-5. The site's pages and code were also produced with AI assistance.

The disclosure is on the site because the pages teach about using AI well, and the same standard applies to them.

License and reuse

Text on this site is licensed Creative Commons Attribution 4.0 (CC BY 4.0). You can copy, adapt, and use it in lesson plans, syllabi, and course-design forms, including commercially, if you credit the source and say if you changed it. Every technique page has an AMA-style "Cite this page" line. The references themselves belong to their authors and publishers.

This is an independent project by Jesse Rogers. It is not an official resource of Palm Beach State College.

Accessibility

The site targets WCAG 2.1 level AA. It has a skip link, a keyboard-operable menu and finder, visible focus, text that scales, labels that do not rely on color, and respect for reduced-motion settings. The launch animation is decorative and has a pause button, and visitors who prefer reduced motion or data saving see a still frame. It has not had a formal third-party audit.

Report a problem

If you find a wrong number, a broken link, or an accessibility barrier, message me on LinkedIn. I will check it against the source and update the page with a new date.