UX Psychology Principles¶

Design from mechanism: why the mind does this. A principle you can explain predicts where it stops working; a trick you can only name does not. Distilled from ux-psychology-research, which carries the primary sources.

Evidence grades¶

Quote the grade, never the statistic. Popular numbers ("2000%", "70–90%", "2x") are almost always the most extreme figure from one condition, stripped of its denominator.

replicated: survives replication or meta-analysis. Build on it.
contested: real, but size or generality disputed. Treat as a hypothesis.
folklore: repeated without solid primary support. A convention at most.

Many "laws" are maxims: Fitts and Hick–Hyman are fitted models; Jakob's and Tesler's are essays.

Perception¶

The visual system groups and ranks a screen before conscious attention arrives.

Principle	Grade	Mechanism → design
Gestalt grouping (proximity, similarity, common region, continuity, figure–ground)	replicated	Near / alike / enclosed things read as one group. Common region overrides proximity (Palmer 1992), so containers are a strong, use-sparingly cue.
Visual hierarchy / pop-out	replicated	A feature unique on one dimension is found pre-attentively; varied distractors kill it. Saturated colour everywhere = nothing pops.
Von Restorff (isolation)	replicated	Distinct items are remembered better, via memory, not only attention (Hunt 1995). One isolate per view; banner blindness is the limit.

Attention and memory¶

Principle	Grade	Mechanism → design
Working-memory limit	replicated (Cowan ~4); folklore ("7±2 menu items")	A visible menu is recognition, not recall. Don't make users carry information between screens; chunk long strings in groups of ~4.
Cognitive load	replicated effects, criticised accounting	Design can only cut extraneous load: split attention (put help next to its target), redundancy (drop labels that repeat).
Serial position	replicated	First and last items recalled best.
Zeigarnik	contested: memory claim fails meta-analysis (Ghibellini & Meier 2025)	What holds is Ovsiankina: the tendency to resume. Justify checklists by resumption, not memory.
Peak–end	replicated (r ≈ .58 meta-analysis)	Experiences are judged by peak and end; duration is ignored. Invest in error states and confirmations.

Interaction cost¶

Principle	Grade	Mechanism → design
Hick–Hyman	replicated for practised, ordered choices	Unfamiliar lists are scanned linearly. Group and order; not the same thing as choice overload.
Fitts	replicated	Bigger and closer is faster. Sticky bottom action bars; thumb reach.
Response-time bands	replicated (0.1 s / 1 s / 10 s); "400 ms Doherty" contested	Acknowledge within ~100 ms, show progress past ~1 s.
Tesler's law	folklore (useful maxim)	Complexity is conserved — move it to the system (defaults, lookups), don't hide it.
Progressive disclosure	contested	Hide by frequency of use, not business preference.
Jakob's law	replicated mechanism (skill transfer)	Follow conventions; spend novelty only on actual value.

Decision-making¶

Principle	Grade	Mechanism → design
Defaults	replicated, large variance (d = 0.68, 58 studies)	Endorsement, endowment, effort. A default is advice — own it.
Choice overload	contested (meta-analytic mean ≈ 0)	Only with complex sets, uncertain preferences, or a wish to decide fast.
Anchoring / relative thinking	replicated for explicit anchors	The first price seen is the ruler.
Decoy effect	contested	Holds mainly in numeric comparison tables.
Framing	replicated	Wording is design.
Loss aversion	contested (λ ≈ 2 from risky gambles; Gal & Rucker dispute generality)	State real losses accurately; it doesn't reliably transfer to upgrade dialogs.
Status-quo bias	replicated	Respect familiar settings; new defaults for new users.

Motivation and habit¶

Principle	Grade	Mechanism → design
Fogg B = MAP	contested as model, useful diagnostic	Missing motivation, ability or prompt? Friction is cheapest to fix.
Goal gradient / endowed progress	replicated / contested	Credit real completed steps; never fabricate progress.
Variable reward	replicated in animals; app application contested	Pair novelty with stopping cues.
IKEA effect	replicated	Value rises only when the build is completed.
Endowment effect	contested (Plott & Zeiler)	Trials create ownership; be clear about the end.
Commitment & consistency	replicated, modest	Start with one small useful step.

Principle	Grade	Mechanism → design
Reciprocity	replicated as a norm	Give unconditional value first.
Social proof	replicated, small	Real ratings at the decision point.
Aesthetic–usability	replicated for perception	Polish hides problems in testing — measure task success.
Credibility signals	replicated in direction	Trust is judged in ~50 ms from consistency, low complexity, typicality.

Corrections worth remembering¶

Jam study: Iyengar & Lepper at Stanford (Draeger's, Menlo Park), not "Columbia". The 24-jam table drew more shoppers; "3% vs 30%" is coupon redemption among those who stopped.
Car wash: Nunes (USC) & Drèze, ~34% vs ~19% (≈1.8x, not "double"). The Columbia study is the coffee card (Kivetz, Urminsky & Zheng) — goal gradient, not endowed progress.
Cialdini never ranked his principles; "+2000% from samples" is an industry best case from a magazine.
"Losses weigh 2x" is Tversky & Kahneman 1992's λ ≈ 2.25 from gambles; a 2024 meta-analysis gives ≈ 1.96 with wide spread.

dark-patterns — every persuasion principle above, turned against the user, and the test that catches it.
ui-craft-rules — the visual rules these mechanisms justify, and the accessibility overlap.