Evidence for Advanced Concept Maps #

Every substantive claim on the Advanced Concept Maps page is checked against current research. Here is each claim, how well today’s evidence supports it, and the sources. The full, de-duplicated source list lives on the references page.

Supported · weak evidence — Systems (causal-loop) concept maps let you hold a whole dynamic system on one page – balancing loops that seek a goal, reinforcing loops that amplify, and the delays that make systems oscillate – so you see the whole and its feedback, not just the parts.

The bucket example, the balancing/reinforcing distinction, and the ‘see the whole, not the parts’ framing are taken directly from Senge’s The Fifth Discipline, which the page credits. This is a faithful summary of a foundational, widely adopted conceptual framework, not an empirical effect; the value is in the modelling lens, and the page’s own caveat (a tidy loop is a hypothesis, not a proof) keeps it honest.

Sources: Senge, P. M. (1990/2006), The Fifth Discipline: The Art and Practice of the Learning Organization — https://www.penguinrandomhouse.com/books/163962/the-fifth-discipline-by-peter-m-senge/ · full reference ›

Supported · moderate evidence — Concept and knowledge maps help learners model complex, interrelated material more effectively than studying text alone.

Nesbit & Adesope’s 2006 meta-analysis (55 studies) found that learning with concept and knowledge maps produced better retention and transfer than reading text, attending lectures, or class discussion, with effects generally moderate. The benefit is real and replicated, though it varies with how maps are used (constructing vs studying a given map) and with learner and task – so ‘moderate’ rather than ‘strong’.

Sources: Nesbit, J. C., & Adesope, O. O. (2006), Learning With Concept and Knowledge Maps: A Meta-Analysis. Review of Educational Research — https://doi.org/10.3102/00346543076003413 · full reference ›

Supported · strong evidence — Learning is itself a feedback loop – you attempt something, get a result, compare it against the goal, and adjust, which is a balancing loop seeking a target – and retrieval practice and Refresh Reviews are deliberately built feedback that closes the gap between what you think you know and what you can actually recall.

The empirical core – that testing yourself (retrieval practice) outperforms restudy – is strongly supported by Adesope, Trevisan & Sundararajan’s 2017 meta-analysis. The ’learning is a balancing loop’ systems framing and the self-regulated plan-act-review cycle behind it are interpretive but faithful (cf. Zimmerman’s SRL model). The closing claim that short feedback delays beat long ones is broadly consistent with feedback-timing research.

Sources: Adesope, O. O., Trevisan, D. A., & Sundararajan, N. (2017), Rethinking the Use of Tests: A Meta-Analysis of Practice Testing — https://journals.sagepub.com/doi/10.3102/0034654316689306 · Zimmerman, B. J. (2002), Becoming a Self-Regulated Learner: An Overview — https://doi.org/10.1207/s15430421tip4102_2 · full reference ›

Supported · moderate evidence — A decision tree lays out a choice and its uncertain outcomes, and where you know the payoffs and probabilities you can fold the tree backwards to give each option an expected monetary value – a probability-weighted average – choosing the highest-EMV path, while at decision nodes you simply take the better branch rather than averaging.

Decision trees, the decision-node/chance-node distinction, backward induction (‘roll-back’), and expected-value maximisation are standard, textbook-settled results of decision analysis (a field founded by Raiffa, a co-author of the cited Smart Choices and of the classic Decision Analysis, 1968). The method itself is mathematically uncontested. ‘Moderate’ reflects that EMV is a normative prescription whose real-world appropriateness depends on the caveats the page correctly flags (averages mislead on one-shot bet-the-company calls; garbage-in-garbage-out on the inputs).

Sources: Hammond, J. S., Keeney, R. L., & Raiffa, H. (1999), Smart Choices: A Practical Guide to Making Better Decisions — https://www.hbs.edu/faculty/Pages/item.aspx?num=131 · full reference ›

Supported · moderate evidence — A decision tree counts only what you feed it – expected monetary value is an average that can mislead on a single bet-the-company decision where the spread of outcomes matters as much as the mean, and a tree that prices only money misses effects like damage to corporate image.

These are well-recognised limitations of naive EMV in decision analysis: risk attitude matters when outcomes are non-repeatable (hence utility theory and risk tolerance, central in Raiffa’s work and in Smart Choices), and unmodelled non-monetary objectives bias single-attribute trees (hence multi-attribute methods). The caveats are accurate and standard; rated moderate because they are accepted theory rather than a measured effect.

Sources: Hammond, J. S., Keeney, R. L., & Raiffa, H. (1999), Smart Choices: A Practical Guide to Making Better Decisions — https://www.hbs.edu/faculty/Pages/item.aspx?num=131 · full reference ›

Supported · strong evidence — Deductive reasoning moves from premises to a conclusion that must be true if the premises are (it guarantees the link, not the inputs), whereas inductive reasoning generalises from examples to a conclusion that is a reasonable bet but may be false – so an induced conclusion should be treated as a hypothesis to keep testing.

The deductive/inductive distinction, the truth-preserving (validity) guarantee of a valid deductive argument, and the defeasible, probabilistic nature of induction are foundational, uncontested results of logic, presented in any standard text (e.g. Hurley’s A Concise Introduction to Logic, the best-selling logic textbook, ch. 1). ‘Strong’ here means definitionally settled within its domain, not an empirical effect size. The page’s penguins-vs-‘birds fly’ and rainy-Seattle illustrations are textbook-accurate.

Sources: Hurley, P. J. (2014), A Concise Introduction to Logic (12th ed.), Cengage Learning — https://www.cengage.com/c/a-concise-introduction-to-logic-12e-hurley/ · full reference ›

Supported · weak evidence — Software-based concept maps (force-directed graph views, hyperbolic/fisheye trees, treemaps, 3D models) earn their place only when a map is too large, too connected, or too dynamic to hold on a page; for everything else a pen still wins.

This is a practical, design-level recommendation rather than a tested claim, and no single source establishes it. The general principle that concept/knowledge maps aid learning is supported by Nesbit & Adesope (cited loosely here for the underlying map-benefit), but the specific guidance about when software outperforms paper is the author’s reasoned judgement, well-aligned with information-visualisation practice yet not directly evidenced. Rated weak and flagged as ‘related’ rather than directly supported.

Sources: Nesbit, J. C., & Adesope, O. O. (2006), Learning With Concept and Knowledge Maps: A Meta-Analysis — https://doi.org/10.3102/00346543076003413 · full reference ›