Behavioral Decision-Making Concept Card Template (Biases, Nudges)

Behavioral Decision-Making Concept Card: [Your name, course or decision, and date]

A concept card for judgment and decision-making as it is taught in the leadership, negotiation and decision courses of a leading MBA: the heuristics people use to judge probabilities and values, the systematic biases those heuristics create, prospect theory and loss aversion, noise in professional judgment, the techniques that reduce bias and noise in organisations (decision hygiene), and choice architecture, the design of the setting in which other people choose. It is for MBA students in organisational behaviour, leadership, negotiation or decision analysis courses and for managers who make or review high-stakes decisions. Worked example A takes a fictional product launch through base rates, a reference class forecast, a regression correction, a loss-aversion test, a noise audit and a sunk-cost decision; Worked example B is half solved; Exercise C asks you to design a premortem and a choice architecture. Source basis: Tversky and Kahneman (1974, 1981, 1992), Kahneman and Tversky (1979), Thaler (1980), Samuelson and Zeckhauser (1988), Kahneman and Lovallo (1993), Johnson and Goldstein (2003), Benartzi and Thaler (2004), Milkman, Chugh and Bazerman (2009) and the other primary works listed under Sources, written up in original words with fictional numbers. Replace every [bracketed] value and every row marked Example. This is a learning aid, not psychological, medical, legal or financial advice.

Document control

Contents

• When to use it
• The idea in plain language
• Formulas and notation
• Bias catalogue
• Assumptions and limits
• Debiasing and decision hygiene
• Choice architecture
• Worked example A (fully solved): Corvane Robotics launches Model W
• Worked example B (half solved): a second forecaster
• Exercise C (unsolved): premortem and a default
• Connections to other cards
• Common mistakes and misconceptions
• Self-test questions
• My notes and next steps
• Sources and further reading
• Answer key

When to use it

Use this card when:

• a decision is large, hard to reverse and made rarely, so the people making it get little feedback on their judgment;
• a forecast (sales, cost, schedule, probability of success) comes from the team that wants the project approved;
• several professionals judge similar cases (credit limits, hiring, performance ratings, underwriting) and their judgments should be consistent;
• you are designing a form, a default, a price display, an enrolment process or a policy that other people will respond to;
• you prepare a case discussion in leadership, negotiation, marketing or finance where the protagonist's judgment is in question.

Do not use it to label colleagues ("that is just anchoring") in place of evidence, and do not assume that knowing about a bias protects you from it: the research suggests it rarely does without a process change.

The idea in plain language

Heuristics and biases. Tversky and Kahneman (1974) showed that people judge uncertain quantities with a few mental shortcuts. Representativeness: judging probability by how similar a case is to a stereotype, which leads people to neglect base rates and sample size. Availability: judging frequency by how easily examples come to mind, which overweights vivid and recent events. Anchoring and adjustment: starting from an initial value, even an irrelevant one, and adjusting too little. These shortcuts are usually useful, but they produce errors that lean in predictable directions: biases.

Prospect theory. Kahneman and Tversky (1979) and the cumulative version of Tversky and Kahneman (1992) describe how people choose between risky options. Outcomes are judged as gains and losses from a reference point, not as final wealth. Losses loom larger than gains of the same size (loss aversion). Sensitivity diminishes with distance from the reference point, so people tend to be risk averse for gains and risk seeking for losses. Probabilities are weighted: small probabilities are overweighted and moderate to large ones underweighted. The same choice described as a gain or as a loss can therefore produce different decisions (framing; Tversky and Kahneman 1981).

Mental accounting, sunk costs and the status quo. Thaler (1980) showed that people treat money differently depending on the mental account it sits in and that they let past, unrecoverable costs affect current choices. Arkes and Blumer (1985) documented the sunk-cost effect in experiments; Staw (1976) showed that decision makers who were responsible for an earlier failing choice invested more in it (escalation of commitment). Samuelson and Zeckhauser (1988) documented status quo bias: people stick with the current or default option more than its merits justify.

Overconfidence and the planning fallacy. People are too sure of their estimates and plans. Kahneman and Lovallo (1993) argued that managers combine "bold forecasts", optimistic estimates from an inside view of the project, with "timid choices", excessive caution when each risky decision is considered alone. The cure they proposed is the outside view: forecast from the record of a reference class of similar projects. Flyvbjerg (2006) turned this into reference class forecasting for large projects.

Noise. Bias is the average error; noise is the unwanted variability of judgments that should be identical. Two underwriters pricing the same risk or two managers rating the same employee often disagree far more than their organisations expect. Measured as mean squared error, overall error is bias squared plus noise squared, so the two enter in the same way and reducing either one lowers the error. Kahneman, Sibony and Sunstein develop this argument in their book Noise.

From individuals to organisations. Organisations can reduce bias and noise with process: independent estimates before discussion, the outside view, checklists, premortems, structured criteria and decision records (Milkman, Chugh and Bazerman 2009; Kahneman, Lovallo and Sibony 2011). And because people respond predictably to defaults, framing and effort, the design of choices, choice architecture, changes behaviour without forbidding options (Thaler and Sunstein, Nudge).

Formulas and notation

Bayes' rule for the probability of a hypothesis $H$ (the product succeeds) after a signal $S$ (a positive pilot):

Prospect theory value of a gain $x \ge 0$ or a loss $x < 0$, with the parameters estimated by Tversky and Kahneman (1992): $\alpha = \beta = 0.88$, $\lambda = 2.25$:

Probability weighting with $\gamma = 0.61$ for gains and $\delta = 0.69$ for losses (same source), and the value of a two-outcome mixed gamble:

Regression correction of an intuitive forecast toward the base rate, where $\rho$ is the correlation between the evidence used and the outcome:

Reference class forecast: multiply the inside-view estimate by a percentile $q$ of the overrun ratios $R$ in the reference class:

Error equation for $J$ judgments $y_j$ of a case whose true value is $T$:

Brier score for $n$ probability forecasts $p_i$ of events with outcomes $o_i \in \{0, 1\}$ (lower is better):

Excel: posterior =Prior*Hit/(Prior*Hit+(1-Prior)*FalsePos); value of a gain =x^0.88, of a loss =-2.25*(-x)^0.88; reference class P80 =PERCENTILE.INC(Ratios,0.8)*Budget; noise =STDEV.P(Judgments); Brier score =SUMXMY2(Forecasts,Outcomes)/COUNT(Forecasts).

Bias catalogue

Assumptions and limits

• Parameters are estimates. The prospect theory parameters are median estimates from one set of experiments with students and modest stakes; individuals, cultures and stakes differ.
• Effects vary in size and some replicate weakly. Several famous effects are robust (loss aversion in many settings, defaults, anchoring); others, especially some priming and ego-depletion results, have replicated poorly. Check the current evidence before relying on a specific effect.
• Heuristics are often adaptive. In familiar settings with fast, clear feedback, intuition can be accurate. The card targets settings where feedback is slow or absent.
• Bayes needs credible inputs. The posterior is only as good as the base rate and the hit and false positive rates; estimate them from data where you can.
• The regression correction needs a correlation. If you have no evidence about how predictive your intuition is, be conservative and stay close to the base rate.
• Choice architecture has ethical limits. A nudge should be transparent, easy to avoid and in the interest of the person nudged; designs that exploit biases against people's interests (sometimes called dark patterns) may also breach consumer protection law. Check the current rules in your jurisdiction with a qualified lawyer.

Debiasing and decision hygiene

The order of a decision hygiene process for a large decision:

Choice architecture

Thaler and Sunstein define choice architecture as the context in which people make decisions, and a nudge as a change in that context that alters behaviour predictably without forbidding options or changing economic incentives much. Evidence that it matters: Madrian and Shea (2001) found that automatic enrolment raised participation in a US employer's retirement savings plan sharply, and that many automatically enrolled employees kept the default contribution rate and fund. Johnson and Goldstein (2003) found much higher rates of consent to organ donation in countries with an opt-out default than with an opt-in default. Benartzi and Thaler (2004) showed that employees who committed in advance to raise their saving rate with future pay rises (Save More Tomorrow) saved much more over time.

Worked example A (fully solved): Corvane Robotics launches Model W

Situation (fictional). Corvane Robotics, a growth technology company, is deciding whether to launch Model W, a warehouse robot. The product team, led by Ana Reyes, presents a plan: a 75% chance of commercial success, a development budget of USD 4.0 million, first-year sales growth of 40% for the robot line, and a request to approve today. A pilot with one customer was positive. Chief executive Priya Raman asks you to review the proposal as the independent reviewer.

Step 1: base rate and Bayes (BI-01)

Corvane's record of comparable launches, and industry data, give a base rate of 20% for commercial success. Past pilots were positive in 80% of products that later succeeded and in 30% of those that failed.

The team's 75% treats the pilot as if it were decisive. With a 30% false positive rate, one pilot doubles the base rate to 40%; a second independent pilot would raise it to 64%, still below the team's number.

Step 2: reference class forecast for the budget (BI-04, DB-01)

Twelve comparable development projects at Corvane and its peers had cost overrun ratios (actual over budget) of 1.05, 1.10, 1.12, 1.18, 1.20, 1.25, 1.30, 1.35, 1.42, 1.55, 1.70 and 2.05 (Example).

A budget of 4.0 has almost no chance of being met on this record: every project in the reference class overran. Approve 5.10 as the expected cost and hold contingency up to 6.10.

Step 3: regression correction of the sales forecast

The team expects 40% first-year growth. Similar product lines grew 15% on average, and Corvane's past launch forecasts correlated about 0.3 with outcomes (Example).

Use 22.5% as the planning case and keep 40% as an upside scenario.

Step 4: loss aversion and narrow framing (BI-05, DB-06)

A plant manager is offered a process change with a 50% chance of saving USD 150,000 and a 50% chance of losing USD 100,000 (expected value +25,000). He refuses. In thousands:

Under prospect theory the refusal is predictable: with loss aversion of 2.25 the gain must be about two and a half times the loss before the bet looks attractive. For Corvane, which runs dozens of such process changes a year, the right frame is the portfolio: twenty independent bets like this one have an expected gain of 500,000 and about a 13% chance of an overall loss (fewer than 8 wins out of 20). Evaluate them together and reward managers on the portfolio, not on each bet.

Step 5: noise audit of credit limits (BI-12, DB-08)

Corvane sells robots on credit terms. Five credit managers set limits for the same three customer files without conferring (USD thousand, Example).

Average pairwise difference relative to the pair's mean: 22.2%, 26.4% and 16.7%. Management had expected about 10%. For customer 1, the credit limit that later proved right was 100 (Example):

Bias and noise both contribute; removing the bias alone would leave 43% of the error (=366/850). Remedy: a scored credit template with defined criteria, averaging two independent assessments for large limits, and a noise audit each year.

Step 6: sunk cost (BI-07, DB-09)

A year later, Corvane has spent 6.0 on Model W. Finishing needs another 3.0, and the present value of the cash flows from finishing is now 2.5. The net present value of continuing is -0.5 (=2.5-3). The 6.0 already spent is the same whether Corvane continues or stops, so it does not enter the decision: stop, or find a cheaper path to finish. Because escalation is strongest in the person responsible for the original choice (Staw 1976), ask someone who did not approve the original budget to make this call.

Worked example B (half solved): a second forecaster

A second Corvane team proposes Model X. The base rate for this kind of product is 10%; its test has a hit rate of 90% and a false positive rate of 20%.

The team lead's last ten probability forecasts and outcomes (1 = happened) were: 0.9 and 1, 0.8 and 1, 0.8 and 0, 0.7 and 1, 0.9 and 1, 0.6 and 0, 0.9 and 0, 0.8 and 1, 0.7 and 0, 0.9 and 1.

Questions to answer: is the team lead calibrated or overconfident? What does the comparison with always forecasting 0.5 tell you? [ ]

Exercise C (unsolved): premortem and a default

C1. Run a 20-minute premortem on Corvane's Model W launch: assume it is two years later and the launch has failed. Each member writes reasons alone for five minutes; then list and group them, and pick the three most likely. Write one mitigation and one early warning indicator for each.

C2. Corvane's safety training for warehouse technicians has 55% completion. Design a choice architecture with at least three tools from CA-01 to CA-06, state the default, and pass each tool through its ethical check.

C3. Restate the following two options in both a gain frame and a loss frame, then say which you would recommend and why: a restructuring that keeps 200 of 600 jobs for certain, or one with a one-third chance of keeping all 600 jobs and a two-thirds chance of keeping none.

Connections to other cards

Common mistakes and misconceptions

1. Using only the hit rate. A test that catches 80% of successes is weak if it also passes 30% of failures.
2. Believing that knowing a bias removes it. Process changes work better than awareness.
3. Treating every error as a bias. Scatter around the right answer is noise, and it needs different remedies.
4. Overcorrecting. Pulling every forecast to the base rate throws away real information; use the correlation.
5. Quoting prospect theory parameters as constants. They are median estimates from specific experiments.
6. Counting sunk costs in a continue or stop decision, or, the opposite error, ignoring the lessons they teach about the forecast.
7. Evaluating risky choices one by one when the company makes many of them.
8. Nudging against people's interests or hiding the opt-out.
9. Running a premortem after the decision, when it can only defend the choice.
10. Assuming a classic effect replicates at full size. Check later replications before you build a policy on it.

Self-test questions

1. Name the three heuristics of Tversky and Kahneman (1974) and one bias each produces.
2. Compute the posterior probability of success with a 20% base rate, a hit rate of 80% and a false positive rate of 30%. Give the Excel formula.
3. State the four features of prospect theory.
4. With $\lambda = 2.25$ and $\alpha = \beta = 0.88$, what gain makes a 50/50 bet against a loss of 100 just acceptable, ignoring probability weighting?
5. What is the difference between bias and noise, and how does the error equation combine them?
6. Why did every project in Step 2's reference class overrun, and what should Corvane budget?
7. Correct an intuitive forecast of 30% growth when the base rate is 10% and the correlation is 0.5.
8. What is a premortem and when must it be run?
9. Why should sunk costs not enter a continue or stop decision?
10. Give two pieces of evidence that defaults change behaviour.
11. What Brier score does a forecaster get by always saying 0.5 on events that happen half the time?
12. Name two ethical tests a nudge should pass.

My notes and next steps

Sources and further reading

Primary works:

• Tversky, A. and Kahneman, D. (1974). Judgment under uncertainty: heuristics and biases. Science, 185(4157), 1124-1131.
• Kahneman, D. and Tversky, A. (1979). Prospect theory: an analysis of decision under risk. Econometrica, 47(2), 263-291.
• Tversky, A. and Kahneman, D. (1981). The framing of decisions and the psychology of choice. Science, 211(4481), 453-458.
• Tversky, A. and Kahneman, D. (1992). Advances in prospect theory: cumulative representation of uncertainty. Journal of Risk and Uncertainty, 5(4), 297-323.
• Thaler, R. (1980). Toward a positive theory of consumer choice. Journal of Economic Behavior and Organization, 1(1), 39-60.
• Staw, B. M. (1976). Knee-deep in the big muddy: a study of escalating commitment to a chosen course of action. Organizational Behavior and Human Performance, 16(1), 27-44.
• Fischhoff, B. (1975). Hindsight is not equal to foresight: the effect of outcome knowledge on judgment under uncertainty. Journal of Experimental Psychology: Human Perception and Performance, 1(3), 288-299.
• Lord, C. G., Ross, L. and Lepper, M. R. (1979). Biased assimilation and attitude polarization: the effects of prior theories on subsequently considered evidence. Journal of Personality and Social Psychology, 37(11), 2098-2109.
• Arkes, H. R. and Blumer, C. (1985). The psychology of sunk cost. Organizational Behavior and Human Decision Processes, 35(1), 124-140.
• Samuelson, W. and Zeckhauser, R. (1988). Status quo bias in decision making. Journal of Risk and Uncertainty, 1(1), 7-59.
• Kahneman, D. and Lovallo, D. (1993). Timid choices and bold forecasts: a cognitive perspective on risk taking. Management Science, 39(1), 17-31.
• Madrian, B. C. and Shea, D. F. (2001). The power of suggestion: inertia in 401(k) participation and savings behavior. The Quarterly Journal of Economics, 116(4), 1149-1187.
• Johnson, E. J. and Goldstein, D. (2003). Do defaults save lives? Science, 302(5649), 1338-1339.
• Benartzi, S. and Thaler, R. H. (2004). Save More Tomorrow: using behavioral economics to increase employee saving. Journal of Political Economy, 112(S1), S164-S187.
• Flyvbjerg, B. (2006). From Nobel Prize to project management: getting risks right. Project Management Journal, 37(3), 5-15.
• Milkman, K. L., Chugh, D. and Bazerman, M. H. (2009). How can decision making be improved? Perspectives on Psychological Science, 4(4), 379-383.
• Klein, G. (2007). Performing a project premortem. Harvard Business Review, September 2007.
• Kahneman, D., Lovallo, D. and Sibony, O. (2011). Before you make that big decision. Harvard Business Review, June 2011.

Books (further reading, by name only): Bazerman and Moore, Judgment in Managerial Decision Making; Kahneman, Thinking, Fast and Slow; Thaler and Sunstein, Nudge; Kahneman, Sibony and Sunstein, Noise: A Flaw in Human Judgment.

Open resource: OpenStax, Organizational Behavior (CC BY 4.0), https://openstax.org.

Answer key

Worked example B: posterior 33.3% (=0.09/(0.09+0.18)), far below the team's 80%. Mean forecast 0.80 against a hit rate of 0.60: the team lead is overconfident by about 20 points. Brier score 0.25, no better than always saying 0.5, so the stated probabilities carry no more information than a coin-flip forecast. Say: "Your forecasts have been 20 points too high on average; start from the base rate and show how the test moves it."

Exercise C:

• C1: typical reasons include a slower sales cycle than planned, integration problems with customers' warehouse systems, a competitor's price cut, and a key engineer leaving. Each needs an owner, a mitigation and an early indicator (for example, pilot-to-order conversion below a set level after six months).
• C2: a strong answer makes enrolment in the next session the default with an easy one-click deferral (CA-01), shortens the booking steps (CA-02), sends a reminder a week before (CA-06) and shows each site's completion against similar sites (CA-05); every tool passes the checks because training is in the technicians' safety interest and they can still reschedule.
• C3: gain frame: "keep 200 jobs for certain" against "one-third chance to keep all 600"; loss frame: "lose 400 jobs for certain" against "two-thirds chance to lose all 600". Expected jobs kept are 200 in both options. People tend to pick the certain option in the gain frame and the risky one in the loss frame (Tversky and Kahneman 1981). A recommendation should rest on criteria stated in advance (variance tolerance, effects on communities, the company's survival), not on the frame.

Self-test:

1. Representativeness (base-rate neglect), availability (overweighting vivid events), anchoring and adjustment (insufficient adjustment from a starting value).
2. 40%: =0.2*0.8/(0.2*0.8+0.8*0.3).
3. Reference dependence, loss aversion, diminishing sensitivity, probability weighting.
4. 251.31 (=100*2.25^(1/0.88)).
5. Bias is the average error, noise the scatter of judgments; mean squared error equals bias squared plus noise squared.
6. Inside-view budgets ignore the problems that typically arise; budget 5.10 as the expected cost with contingency up to 6.10 at P80.
7. 20% (=10%+0.5*(30%-10%)).
8. Imagining that the decision has already failed and listing the reasons; it must be run before the decision is final.
9. They are the same whichever option is chosen; only future costs and benefits differ between the options.
10. Automatic enrolment raised retirement plan participation (Madrian and Shea 2001); opt-out countries had much higher organ donation consent (Johnson and Goldstein 2003).
11. 0.25.
12. It should be in the interest of the person nudged and easy to avoid; it should also be transparent.

Revision history

Standard and licence note

Based on the primary research works listed above on judgment, decision-making and choice architecture, summarised in original words. The text, the examples, the numbers and the exercises are original; no textbook, article or published case is copied or paraphrased, and Corvane Robotics, its people and every number in this card are fictional. The template is independent and not affiliated with or endorsed by any author, publisher or business school. Check the original studies and later replications before you rely on a specific effect. Template licence: MIT.