psychology

Choice Architecture

How the way choices are presented affect how they are chosen.

Key: There is no such thing as a neutral presentation of choices.

Structuring a Choice Task: What to Present?

How many Choices

• more choices means better preference match but greater cognitive burden
  • Insurance: 100 options overwhelms
  • Medicare Drug Plans: Too many choices
• too few choices can cause the presence of one choice to influence the presence of another choice
• generally, 4 to 5 non-dominating choices seem like the sweet spot

Technology and Decision Aids

Decision aids are things that help with making decisions like google search and recommendation algorithms.

• secretly bias you to stick with your own world of stuff
• helps with filtering choices nonetheless
• can be designed to subtly steer consumers towards specific themes of choices (a handful of filter options that subtly make you eat a salad)

Defaults

A choice you can choose without having to actively choose it. Whatever happens when you do nothing.

THE DOING NOTHING OPTION IS OFTEN CHOSEN MORE.

Types of Defaults

1. Simple Defaults one default for everyone
2. Random Defaults different people get randomly different pre-selected options
3. Forced Choice - no default; must actively choose (heads or tails?)
4. Sensory Defaults change based on what can be inferred (phone detects you’re driving - default option switches to do not disturb)
5. Persistent Defaults remembers past choices (pizza order, and next order defaults to past order)
6. Reverting Defaults forget last changes (the next order defaults to the original default everytime)
7. Predictive Defaults intelligently alter based on user observation (thermostat detects time, default temp changes)

Choice Over Time

Lots of choices lead to outcomes that unfold overtime. This could lead to some biases:

• Myopia bias for caring about the present more than later
• Uncertainty about the future focus too much on the uncertainties of life, failing to consider satisfactory choices
• Unrealistic Optimism too optimistic of the future

Ways to address these biases

Tool	How it Works	Example
Order of consideration	Drawing attention to delayed options refocuses the decision	Show long term option first
Focus on Satisficing	Encourage the consideration of the “good enough” option	leads to higher satisfaction
Limited time windows	Giving deadlines makes the bias kinda go away	expiration dates
Translate to immediate outcomes	make future consequences feel immediate	calculating present value?

Task Structure

How the way the choices are structured affect the way people explore and search through the options to reach a final decision.

Single Choice vs Configuration Context

Single Choice Context is where you have to choose one things from a list Configuration Context is where you can make multiple sequential decisions. This uses a different way of choice thinking.

Consumers are more likely to choose defaults when bombarded with choices for the first choice.

Scenario A: Start with MANY options (interior color with 8 choices)

1. First choice: Interior color (8 options) - requires mental effort, depletes cognitive resources
2. Second choice: Engine type (3 options) - more likely to just pick the default
3. Third choice: Sound system (3 options) - more likely to just pick the default

Scenario B: Start with FEW options (engine type with 3 choices)

1. First choice: Engine type (3 options) - easy, doesn’t deplete much
2. Second choice: Sound system (3 options) - still have mental energy
3. Third choice: Interior color (8 options) - less likely to pick default, more likely to actively choose

Called Decision Fatigue

Two Stage Search: Screening then Comparing

Screening eliminates alternatives based on some subset of attributes Comparing make alternative-based comparisons among remaining options

If you screen out options based on a specific attribute, then choice makers will factor in that attribute in the available choices more.

• You focus the choice on just one thing like the ugliness of a car, then people will weigh choices more heavily on looks

Search Costs and When to Stop Searching

Choices might be cut short because people might not know when to stop searching. Perceiving the cost of search to be too high.

Describing Choice Options (HOW to Present)

Partitioning Options and Attributes

How you divide options dramatically affects choice behaviour.

Why? partitioning choices into categories affects how we allocate our choices (assuming its a choice that can be allocated, like money or time)

Naive Allocation, 1/n Heuristic

Assuming a choice where allocation is needed (like time, money, etc.), people tend to split their allocation evenly amongst the presented categories of the choices.

Domain	Pattern	Research
Investment	Allocate 1/n of savings to each of n options in a 401(k) plan	Benartzi & Thaler 2001; Thaler & Sunstein 2008
Consumption	Seek variety when choosing multiple goods for future consumption	Read & Loewenstein 1995; Simonson 1990
Time	Favor spreading consumption over different time periods	Loewenstein & Prelec 1993
Justice	Favor equal allocation of benefits and burdens among individuals (unless compelling alternative criterion)	Messick 1993
Probability	Assign equal probabilities to each event that could occur	Fox & Clemen 2005; Fox & Rottenstreich 2003
Importance weights	Assign equal importance to each explicitly identified attribute	Weber et al. 1988

Designing Attributes

People choose between alternatives by weighting pros and con of different attributes. ​ We can influence behaviour by making particular attributes more or less noticeable.

1. PARSIMONY - Keep It Simple​

   • Too many attributes overwhelm → people focus on only one, ignore important info, make worse decisions​
   • Solution: Use smaller attribute sets, highlight most important​
   • Research: Decision-makers understand more and weigh information better with choices requiring less cognitive effort (moreso for people bad at math)​

2. LINEARITY - Fix Non-Linear Relationships​

   • Problem: Attributes often have non-linear relationship to objectives​

3. COMPARABILITY - Use Same Scale​

   • Annual subscriptions vs. monthly bills vs. per-use downloads​
   • Put on common scale for accurate comparison​

4. EVALUABILITY - Add Context to Numbers​

   • Break into categories (grades) or label endpoints as good/bad​
   • Example: EPA rates cars 1-10 for CO₂ (no one knows “good” levels)​
   • Why it works: Labels allow quick affective reactions​

5. TRANSLATION & EXPANSION - Make Impact Salient​

   • Translation: Map attribute to consequences (gas consumption → cost AND CO₂)​
   • Expansion: Change scale to enlarge differences (100 miles vs. 10,000 miles)​
   • Expanded attributes receive more weight in choice

Implementation Issues

3.1 Individual Differences

THE CRITICAL PRINCIPLE: A nudge can have multiple effects that depend on characteristics of the decision-maker

Why one-size-fits-all fails:

Reason 1: Energy Use Feedback

Intervention: Inform households about their relative energy use compared to neighbors

Average effect: 2% decrease in energy usage

BUT when broken down by political affiliation:

• Liberal households: REDUCED consumption
• Republican households: INCREASED consumption

Why the difference: Presumably due to differences in environmental concerns (Costa & Kahn 2010)

Lesson: Same intervention can produce opposite effects in different populations

Reason 2: The “Curse of Knowledge” Problem

What it is: Choice architects may:

• Anchor first on what THEY themselves know or want
• Insufficiently adjust for OTHER people’s knowledge levels or preferences
• Nickerson 1999, 2001

Example:

• Expert nutritionists design food labels assuming people understand concepts like “saturated fat”
• But many consumers don’t know what saturated fat is or why it matters
• The label design reflects the designer’s knowledge, not the user’s

The Solution: Test in Diverse Populations

Implication: The intuitions of choice architects will not always be enough

Best practice:

• Test choice architectures in diverse populations of interest
• Don’t rely solely on intuition
• Run experiments before full implementation

Reason 3: What We Already Know About Individual Differences

We can bring existing knowledge about individual differences to bear on choice architecture:

Content domain differences:

• Cultural cognitions important in environmental domains
• Political ideology affects response to messaging
• Religious beliefs may affect health decisions

Decision problem type differences:

• Numeracy critical in decisions with unfamiliar numeric information
• Need for cognition affects how people process detailed information
• Risk tolerance affects financial decisions

Reason 4: Numeracy and Choice Architecture

What is numeracy: Ability to understand and work with numbers

Why it matters: Series of studies showing numeracy interacts with how information is presented

Finding 1: Cognitive Effort and Numeracy (Peters et al. 2006)

General principle: Requiring less cognitive effort helps decision-makers understand more information and weigh important information better

Numeracy interaction: This is PARTICULARLY TRUE for less numerate consumers

What this means for design:

• Simplifying numeric information helps everyone
• Helps less numerate people MUCH more
• Don’t assume everyone can handle complex calculations

Finding 2: Affective Meaning and Numeracy (Peters et al. 2007, 2009)

Intervention: Attach affective (emotional) meaning to numeric information

• Add evaluative labels (good/bad)
• Add color coding
• Add visual indicators

Effects:

• For everyone: Allows integration of MORE information into decisions
• For less numerate people: REDUCES reliance on less relevant emotional sources like mood states

Mechanism: Labels facilitate information processing by allowing affective reactions to be accessed more quickly

Example:

Without labels:          With labels:
CO2: 250 g/mile         CO2: 250 g/mile (7/10 - Good)
                        ↑ Emotional reaction accessed quickly

Finding 3: Organizing Frameworks (Sagara 2009)

Intervention: Use organizing framework (like a table or chart) to help consumers comprehend information

For less numerate consumers:

• ✅ Better comprehension of information SUMMARIZED in the framework
• ❌ Worse comprehension of information NOT in the framework

Interpretation: The framework focuses attention on what’s included, but draws attention AWAY from what’s excluded

Finding 4: Number Comparisons

For highly numerate decision-makers:

• May OVERUSE number comparisons when such information is provided
• Focus too much on quantitative differences
• Neglect qualitative factors

Conclusion on Individual Differences:

“A one-size-fits-all approach to choice architecture will not always work, particularly in diverse and sometimes highly politicized environments.”

Best practices:

1. Understand which individual differences matter for your domain
2. Test interventions in diverse populations
3. Consider providing customized information when possible
4. Be particularly careful in politicized contexts

3.2 Evaluating Outcomes: Prediction vs. Experience

The fundamental question: How can we tell if a choice architecture intervention has actually helped?

The obvious answer: Look at whether people are happier with outcomes The problem: People often cannot accurately predict how they’ll feel about outcomes

The Prediction-Experience Gap

What most choice theories assume:

• Utility estimated EX ANTE (before experience) = Utility experienced EX POST (after experience)
• If you think you’ll like something, you will
• If you think something will make you happy, it will

Reality from research: Numerous ways people fail to accurately predict how they’ll feel about choice outcomes (Hsee & Hastie 2006; Loewenstein & Schkade 1999)

Common Prediction Errors:

Error 1: Impact Bias People OVERESTIMATE the impact of various factors on well-being

Example - Income:

• People overestimate how much happier they’ll be with more money (Kahneman et al. 2006)
• Actual effect of income on happiness is smaller than predicted

Example - Academic tenure:

• Professors greatly overestimate how devastated they’ll be if denied tenure
• Actual emotional impact is much smaller and shorter-lived than predicted (Gilbert et al. 1998)

Example - Romantic breakup:

• People greatly overestimate the duration of emotional response to breakup
• Adaptation happens faster than predicted (Gilbert et al. 1998)

Error 2: Projection Bias People UNDERESTIMATE how their current state affects their predictions

Example - Grocery shopping:

• Shopping while hungry leads to buying much more food
• People underestimate how empty stomach affects decisions (Nisbett & Kanouse 1968)
• Once home and full, regret the purchases

Error 3: Focalism and Duration Neglect People focus too much on one aspect and UNDERPREDICT adaptation

The adaptation problem: People greatly underestimate how quickly they adapt to changes, both positive and negative (Schkade & Kahneman 1998)

Example - Major life changes:

• Moving to California (people think they’ll be much happier; adaptation makes effect small)
• Getting a promotion (people think they’ll be thrilled forever; adapt quickly)
• Becoming disabled (people think life will be terrible; adapt more than expected)

Error 4: Distinction Bias Things that seem very different when evaluating side-by-side feel more similar when experienced separately

Example:

• Comparing two houses side by side, small differences seem huge
• Living in either house, the differences barely matter day-to-day

Who Predicts Better: The Role of Experience

Key finding: People WHO HAVE EXPERIENCE in a situation make MORE ACCURATE predictions about adaptation (Schkade & Kahneman 1998)

Implications:

• Novices need more help from choice architecture
• Experts can rely more on their own predictions
• Experienced advisors can help bridge the gap

Existing Policy Responses (Already Recognize This Problem)

Many consumer protection policies reflect tacit knowledge of prediction-experience gap:

“Cooling Off” Periods:

• Allow consumers to cancel choices without penalty within certain timeframe
• Recognize that people may make different decisions once emotions cool
• Examples: Door-to-door sales (3-day right of cancellation), time-share purchases (rescission periods)

Why they exist:

• Acknowledge that decisions made in the heat of the moment may not match experienced utility
• Give people chance to reconsider once initial emotional state passes

Role of Experienced Agents and Advisors:

One function advisors serve: Encourage decision-makers to consider not just:

• Features salient AT TIME OF CHOICE
• But also features important WHEN OUTCOMES ARE EXPERIENCED

Example - Home buying: Realtor might point out:

• “The open floor plan looks great now, but think about the noise when you have kids”
• “The long commute seems fine on Saturday, but consider doing it twice a day, every day”

Implications for Choice Architecture

The toolbox should include interventions that bear on decision-maker’s knowledge about future outcomes:

Problem	Intervention	How It Helps
Focalism	Direct attention to day-to-day experience, not just headline features	”Think about a typical Tuesday evening in this house”
Impact bias	Provide information about adaptation	”Most people adjust to this change within 6 months”
Projection bias	Create opportunity to experience in relevant state	Test drive car in rain, not just sunshine; shop for groceries after eating
Lack of experience	Provide data on experienced utility from others	Customer reviews focusing on long-term use, not initial impressions

Example Application: Major Purchase Decisions

Car buying architecture that accounts for prediction-experience gap:

❌ Traditional approach:

• Test drive on sunny Saturday
• Focus on exciting features (acceleration, sound system)
• Compare specifications side-by-side

✅ Better approach:

• Extended test drive including rush hour commute
• Focus on daily use features (visibility, ease of parking, cargo space)
• Information on owner satisfaction after 1 year and 3 years
• “Cooling off” period to return vehicle

The Bottom Line:

“These and other interventions that bear on the decision-maker’s knowledge about their future outcomes should be considered part of the decision architect’s toolkit.”

Key principle: Don’t just help people choose what they THINK they want; help them understand what they’ll ACTUALLY experience

Part 4: Real-World Applications

4.1 Environmental Decisions

Domain scope includes:

• Energy consumption: Appliances, transportation, heating/cooling
• Water use: Showers, gardening, swimming pools, rice farming
• Land use: Deforestation, agriculture types, urban planning

Why This Domain Needs Choice Architecture

The climate change challenge:

• Perhaps the greatest sustainability challenge
• Requires drastic reductions in greenhouse gas (GHG) emissions
• Needs reduced energy consumption + better efficiency + conservation

Why it seems like a win-win:

• Produces financial gains for consumers (lower energy bills)
• Produces societal gains for environment (reduced emissions)
• Seemingly should be easy to promote

Why traditional economic approaches haven’t worked: Psychological biases are barriers to adoption that economic incentives alone can’t overcome (Weber 2012, in press)

Traditional Economic Solutions (Insufficient)

Approach 1: Regulate behavior

• Building codes
• CAFE (Corporate Average Fuel Economy) standards
• Result: Often without substantial effects

Approach 2: Raise price of energy

• Carbon tax (in some countries, not USA)
• Higher energy prices
• Result: Often without substantial effects

Why they fail:

• Present bias: Future savings heavily discounted
• Psychological distance: Climate change feels abstract and far away
• Status quo bias: Easier to keep current behavior
• Lack of evaluability: Hard to understand impact of individual actions

Behavioral Interventions That Work Better

Example 1: Social Comparison

• Show households their energy use compared to neighbors
• Effect: Average 2% reduction (though varies by political affiliation)

Example 2: Defaults

• Opt-out for renewable energy vs. opt-in
• Dramatic differences in adoption rates

Example 3: Translation

• Convert energy use to dollars per year
• Show CO2 emissions as “equivalent to X trees”
• Makes abstract concrete

Example 4: Limited time windows

• Tax credits for energy-efficient appliances with expiration dates
• Creates urgency that overcomes procrastination

4.2 Financial Decisions

Why financial decisions are prime for choice architecture:

• Complex products (retirement plans, mortgages, credit cards)
• Intertemporal tradeoffs (present sacrifice for future benefit)
• Present bias is strong
• Low financial literacy
• High stakes (mistakes are costly)

4.3 Eating Decisions

Why eating behavior is special:

• People make 200-300 food decisions per day (Wansink & Sobal 2007)
• Most occur without conscious thought
• People rely on heuristics and decision rules (Wansink 2010)
• Habitual behaviors become rigid and unresponsive to changes in health/nutrition understanding

Evidence of Non-Economic Behavior

Example: Pizza Buffet Study (Just & Wansink 2011)

Setup: All-you-can-eat pizza buffet with different prices

What economics predicts:

• People should eat the same amount regardless of price (it’s all-you-can-eat!)
• Sunk cost doesn’t matter

What actually happened:

• Doubling the price led people to eat MORE pizza
• Even though diminishing returns to taste and perceived quality quickly dropped
• Led them to eat more AND enjoy it less

Interpretation: “I paid more, so I should eat more” → irrational but real

Traditional Interventions That FAIL in Eating

Approach 1: Alter prices

• Taxing unhealthy food
• Subsidizing healthy food
• Result: Generally ineffective (Mytton et al. 2007)

Approach 2: Provide information

• Nutrition labels
• Calorie counts on menus
• Result: Generally ineffective in altering consumption

Why they fail:

• Decisions are mindless/habitual
• Made hundreds of times per day
• Cognitive capacity overwhelmed
• Present enjoyment > future health

Behavioral Interventions That Work

Intervention 1: Partitioning

• Separate shopping cart sections for fruits/vegetables
• Segregate healthy menu items into separate categories
• Result: Purchases/choices match partition structure (Wansink et al. 2012)

Intervention 2: Defaults

• Make healthy option the default side dish
• Require active choice to substitute unhealthy option
• Result: Significant increase in healthy choices

Intervention 3: Portion size/Packaging

• Smaller plates in cafeterias
• 100-calorie snack packs
• Result: People eat less without feeling deprived

Intervention 4: Convenience

• Place healthy foods at eye level
• Make unhealthy foods less convenient
• Result: Significant shifts in consumption

Intervention 5: “Trigger foods”

• Control what foods are available together
• Presence of bananas and green beans → decreased ice cream sales
• Presence of sugary sides → increased cake and chips sales
• (Hanks et al. 2012)

Key Insight: Modified Utility Theory

Standard economic model: Assumes rational time and risk preferences

Reality: Need to modify standard utility theory to allow for:

• Psychological processes
• Effects in combination with economic incentives
• (Just & Wansink 2009)

Synthesis: Why Behavioral Economics Works Where Economics Fails

The Limits of Traditional Economic Solutions

Traditional economic policy levers:

1. Alter prices (taxes, subsidies)
2. Provide information (labels, disclosure requirements)
3. Restrict behavior (bans, mandates)

Why These Often Fail:

Financial decisions:

• Purely economic incentives not enough
• Even company matching on 401(k) contributions doesn’t achieve 100% participation
• Present bias too strong

Food consumption:

• Behavior can’t be reconciled with standard economic models
• Altering prices generally ineffective
• Information generally ineffective

Environmental choices:

• Regulating behavior (building codes, CAFE standards) often without substantial effects
• Raising energy prices (carbon tax) often without substantial effects

The Issues Traditional Economics Doesn’t Address:

1. Excessive discounting - Future benefits heavily discounted
2. Status quo effects - Strong preference for current state
3. Information processing limitations - Can’t handle complexity
4. Present bias - Strong preference for immediate outcomes
5. Framing effects - How options are described matters enormously
6. Choice overload - Too many options reduces decision quality
7. Naive allocation - Tendency to allocate evenly across categories
8. Prediction-experience gap - What we think we’ll like ≠ what we actually like

Why Behavioral Interventions Work:

Key principle: Behavioral economics offers means to encourage more optimal behavior WITHOUT inducing resistance and reactance often associated with restrictive policies (Just & Wansink 2009)

Acceptance by the public:

• Such behavioral interventions are not necessarily objectionable to decision-makers
• Even when unaware of impacts on their own behavior (Johnson & Goldstein 2003; Wansink 2012)
• Individuals believe they’re better off
• Intervention encourages good behavior while NOT PROHIBITING bad behavior

Additional Situational Factors

Beyond the main tools, many subtle situational changes can have desirable effects:

Factor	Effect	Research
Website wallpaper	Changes what people buy	Mandel & Johnson 2002
Social setting	Affects choices	Milch et al. 2009
Information mode	Text vs. visual affects decisions	Weber & Lindemann 2007
Framing of outcomes	Gains vs. losses dramatically affects choice	Tversky & Kahneman 1981
Attribute labels	Word choice changes preference	Hardisty et al. 2010

Ethical Considerations and Transparency

The Manipulation Concern

Objection: “Choice architects influence behavior without decision-makers’ awareness—isn’t this manipulation?”

Response 1: Full Disclosure

Proposal: Choice formats could be accompanied by description of potential influences

What this means:

• Explain how the choice is structured
• Describe known effects of that structure
• Example: “This form uses defaults. Research shows defaults increase selection rates by 40%.”

Current state:

• Such full disclosure rarely done today
• Effects deserve further study

Goal: Making open disclosure a routine responsibility for choice architects

Response 2: No Neutral Alternative

Key point: There IS NO neutral presentation

• Every choice architecture influences outcomes
• The question isn’t “Should we influence?” but “How should we structure influence?”
• Even “random” presentation is a choice that has effects

Response 3: Preserves Freedom

Libertarian paternalism:

• Nudges people toward better choices
• While preserving freedom to choose otherwise
• Contrasts with bans or mandates

Example - Organ donation:

• Opt-out default increases donation rates
• But anyone can still opt out
• Freedom preserved, outcomes improved

Response 4: Benefits Decision-Makers

Public acceptance:

• People don’t object to nudges when they benefit profit-seeking sellers
• Should be even MORE willing when nudges benefit themselves or society
• Examples: Health, environment, financial security

The Complete Toolbox: Summary Reference

STRUCTURING THE TASK (What to Present)

Tool	When to Use	Key Principle
Reduce alternatives	When people face choice overload	4-5 options as starting point
Technology aids	When options are complex	Sort, filter, recommend
Defaults	Almost always applicable	Set to what most would actively choose
Focus on satisficing	When perfectionism causes paralysis	”Good enough” reduces deferral
Limited time windows	When procrastination is likely	Creates urgency
Decision staging	When process is complex	Break into manageable steps

---

DESCRIBING OPTIONS (How to Present)

Tool	Problem Addressed	Key Principle
Partitioning	Naive allocation across categories	Segregate favored, integrate disfavored
Parsimony	Attribute overload	Fewer, more important attributes
Linearity	Misunderstanding non-linear relationships	Transform to linear or provide translation
Comparability	Can’t compare across contexts	Put on same scale
Evaluability	Can’t judge if numbers are good/bad	Add labels, categories, endpoints
Translation	Don’t recognize attribute consequences	Map to relevant objectives explicitly
Expansion	Underweight important attributes	Express on larger scale

---

IMPLEMENTATION CONSIDERATIONS

Issue	Solution	Key Principle
Individual differences	Test in diverse populations; customize when possible	One size doesn’t fit all
Prediction-experience gap	Focus on experienced utility; cooling off periods	What people think they’ll like ≠ what they’ll actually like
Political sensitivity	Extra careful testing; consider full disclosure	Same nudge can have opposite effects