Leadership Decision-Making, Evidence and Analytics

Introduction

Leadership is expressed not only through vision, influence and relationship-building, but also through the quality of decisions made under pressure. Senior leaders routinely decide which strategy to pursue, how to allocate scarce resources, whom to promote, whether to enter a new market, how to respond to risk and when to change course. These decisions are rarely made with complete information. They involve uncertainty, competing interests, time constraints, ethical consequences and imperfect human judgement.

Leadership decision-making therefore requires more than choosing an option. It involves defining the problem accurately, identifying relevant evidence, evaluating alternatives, involving appropriate stakeholders, recognising bias, managing risk and accepting accountability for consequences. Analytics can improve this process, but data do not eliminate judgement. A dashboard may reveal a pattern without explaining its cause; a predictive model may estimate risk without determining what ought to be done; and a statistically significant relationship may be practically unimportant or ethically unacceptable.

This chapter integrates rational, behavioural and participative approaches to leadership decision-making. It also examines evidence quality, people analytics, dashboards, uncertainty, groupthink, cognitive bias, privacy and fairness. The central argument is that effective leaders develop a disciplined decision architecture: they match the decision process to the situation, use evidence critically, involve people appropriately and make assumptions visible.

Learning Objectives

By the end of this chapter, learners should be able to:

1. Explain rational, bounded-rationality, behavioural and participative decision models.
2. Select an appropriate decision process according to urgency, complexity, expertise and stakeholder impact.
3. Distinguish risk from uncertainty and apply structured methods for evaluating alternatives.
4. Diagnose groupthink, cognitive bias and escalation of commitment.
5. Interpret leadership and people analytics, dashboards and predictive evidence critically.
6. Evaluate validity, reliability, privacy, fairness and governance in data-informed decisions.
7. Design an evidence-based decision process that remains accountable and humane.

1. The Nature of Leadership Decisions

A decision is a commitment to a course of action among alternatives. Leadership decisions differ from routine administrative choices in four ways.

First, they are consequential. They affect employees, customers, communities, investors and other stakeholders. Second, they are often non-routine: previous procedures may not provide a clear answer. Third, they are interdependent. A hiring decision can affect culture, performance and future leadership capacity. Fourth, they are made under incomplete knowledge. Even accurate data about the past cannot guarantee the future.

A useful distinction is between a problem, a symptom and a decision opportunity. Falling sales may be a symptom; the underlying problem may be poor customer retention, inappropriate pricing or a change in market demand. A decision opportunity arises when a leader can intentionally choose among alternative responses. If the problem is framed incorrectly, even a technically excellent analysis may produce a poor decision.

Decision quality should be distinguished from decision outcome. A sound decision can produce a disappointing outcome because of external events, while a careless decision can occasionally succeed by chance. Leaders should therefore evaluate both the quality of the process and the result. A useful post-decision review asks:

• What information was available at the time?
• Which assumptions shaped the choice?
• What alternatives were considered and rejected?
• What risks and uncertainties were identified?
• Which outcomes were controllable, and which were not?
• What evidence would justify changing course?

2. Rational and Analytical Decision Models

2.1 The classical rational model

The rational model assumes that decision-makers can define objectives, identify alternatives, obtain relevant information, compare consequences and select the option that best satisfies the objectives. Its usual sequence is:

1. Define the decision and desired outcomes.
2. Establish criteria and their relative importance.
3. Generate feasible alternatives.
4. Gather and assess evidence.
5. Estimate consequences, risks and resource requirements.
6. Compare alternatives.
7. Select, implement and monitor the preferred option.

The model is valuable as a discipline even when its assumptions are unrealistic. It encourages explicit criteria, comparison and accountability. A simple weighted-scoring model can support structured comparison:

Here, S_j is the total score of alternative j, w_i is the weight assigned to criterion i, and r_ij is the rating of alternative j on criterion i. Usually, weights are non-negative and sum to 1, while ratings use a defined scale such as 1–5. The method assumes that criteria can be meaningfully rated and combined, and that the weights represent genuine priorities. It should not be treated as an objective truth: changing the criteria or weights can change the result.

Illustrative example. A leadership team evaluates three learning technologies using cost, learning impact and implementation feasibility. If the weights are 0.30, 0.45 and 0.25, and Technology A receives ratings of 4, 5 and 3, its score is:

S_A = (0.30 × 4) + (0.45 × 5) + (0.25 × 3) = 1.20 + 2.25 + 0.75 = 4.20

The result is meaningful only if the ratings and weights are transparent and comparable. Sensitivity analysis should test whether a small change in a weight or rating reverses the ranking.

2.2 Expected value and risk

When alternatives have different possible outcomes, expected monetary value or expected utility can support comparison. Expected value is:

EV(a) is the expected value of action a, p_k is the probability of outcome k, and v_k is the value of that outcome. The probabilities must be non-negative and sum to 1. Expected value is most appropriate when probabilities are reasonably estimable, outcomes can be represented on a common value scale and repeated decisions or long-term consequences justify averaging. It can be misleading when probabilities are speculative, outcomes are irreversible or ethical effects cannot be reduced to money.

For example, a new service may produce a gain of £200,000 with probability 0.6 and a loss of £100,000 with probability 0.4:

EV = (0.6 × £200,000) + (0.4 × −£100,000) = £120,000 − £40,000 = £80,000

An expected value of £80,000 does not mean that the organisation will receive £80,000. It is a probability-weighted average, not a forecast of a single outcome. Leaders must also consider cash-flow constraints, downside exposure, strategic value, legal obligations and stakeholder harm.

2.3 Decision trees and scenario analysis

A decision tree represents choices, chance events and consequences in sequence. It is useful where an initial decision is followed by information, a market response or a later choice. Scenario analysis complements decision trees by examining coherent futures rather than assigning false precision to one forecast. Robust decisions often perform acceptably across several plausible scenarios rather than optimally in only one.

A practical scenario process is to identify the major external drivers, construct two or three plausible futures, test each strategic option and specify signposts that indicate which future is emerging. This supports adaptive leadership: leaders make an initial commitment while preserving options and defining review points.

3. Behavioural Decision-Making

3.1 Bounded rationality

Herbert Simon’s concept of bounded rationality recognises that leaders have limited time, information, attention and computational capacity. Rather than maximising, they often satisfice: they select an option that is good enough relative to a threshold. Satisficing is not necessarily irrational. In urgent or low-stakes decisions, an adequate solution may be preferable to an exhaustive search. However, satisficing becomes dangerous when the threshold is unexamined or when powerful stakeholders define “good enough” for others.

Leaders can improve bounded rationality by using decision rules, checklists, staged commitments, independent review and pre-mortems. These mechanisms reduce cognitive load without pretending that complete rationality is possible.

3.2 Cognitive biases

Biases are systematic tendencies that can distort judgement. Important examples include:

Bias is not eliminated by telling people to “be objective.” It is managed through process design. For example, a hiring panel can use structured questions, common scoring criteria, independent ratings before discussion and evidence-based documentation. These practices reduce opportunities for irrelevant impressions to dominate.

3.3 Dual-process judgement

Leaders use rapid, intuitive judgement as well as slower, analytical reasoning. Intuition can be valuable where a leader has deep, relevant experience and receives rapid feedback. It is less reliable in novel environments, politically charged situations or systems with delayed and ambiguous feedback. The appropriate response is not to reject intuition but to ask when it has been earned and when it should be tested.

A useful discipline is to separate signal detection from explanation. A leader may intuit that morale has declined; that intuition can motivate inquiry, but it is not proof of the cause. Interviews, survey data, turnover patterns and observation may then test competing explanations.

4. Participative and Collaborative Decision-Making

Participation ranges from consultation to delegated authority. A leader may decide alone, seek individual input, consult a group, facilitate a joint decision or delegate the decision within agreed boundaries. The appropriate level depends on several factors:

• Decision quality: Does the issue require specialised knowledge distributed across people?
• Acceptance: Will implementation depend on commitment from those affected?
• Urgency: Is there time for consultation?
• Conflict: Are interests aligned or deeply contested?
• Capability: Do participants understand the problem and decision criteria?
• Accountability: Who has formal responsibility for the outcome?

Participative leadership can improve information quality, creativity and commitment. It can also produce delay, conformity or an unclear accountability structure. Participation should therefore be designed, not assumed to be beneficial in every situation.

4.1 The Vroom–Yetton–Jago perspective

The Vroom–Yetton–Jago model proposes that leaders select a decision method according to situational characteristics. Methods range from autocratic decision-making to consultation and group-based decision-making. The model directs attention to decision quality, the leader’s information, problem structure, the need for acceptance and the likelihood of conflict.

A practical adaptation is a decision-rights statement:

Participation must be psychologically safe. Employees should be able to challenge assumptions without retaliation. Leaders should also explain how input influenced the final decision, including why some recommendations were not adopted.

5. Groupthink and Constructive Dissent

Groupthink occurs when the desire for cohesion or agreement suppresses realistic appraisal of alternatives. Warning signs include self-censorship, pressure on dissenters, illusions of unanimity, stereotyping of outsiders and the belief that the group is morally correct. Groupthink is more likely when a group is highly cohesive, insulated from outside views, led by a directive chair and operating under stress.

Countermeasures include:

1. The leader withholds an initial opinion.
2. Members develop independent views before discussion.
3. A devil’s advocate or red team challenges the preferred option.
4. External experts or affected stakeholders are invited.
5. The group conducts a pre-mortem: “Assume the decision failed. What caused it?”
6. Anonymous input is used where status differences inhibit candour.
7. A final review checks evidence, assumptions, alternatives and dissent.

Constructive dissent is not disloyalty. It is a governance resource. The leader’s task is to distinguish evidence-based challenge from personal obstruction, while ensuring that disagreement is evaluated on substance rather than rank.

6. Evidence for Leadership Decisions

Evidence varies in quality and relevance. A useful hierarchy is not a rigid ranking but a set of questions:

• Is the evidence relevant to the decision?
• Is the measurement valid—that is, does it represent the intended construct?
• Is it reliable and consistent?
• Is the sample appropriate and sufficiently complete?
• Could selection, survivorship or reporting bias distort it?
• Does the evidence show correlation or support a causal claim?
• Are results practically meaningful, not merely statistically significant?
• Can the finding be transferred to this context?

6.1 Correlation, causation and experimentation

A correlation indicates that variables move together; it does not establish that one causes the other. A relationship between remote work and performance, for example, may be affected by job type, employee autonomy, manager quality or workload. Leaders should ask what alternative explanations exist.

Causal evidence is stronger when there is temporal order, plausible mechanism, control of confounding factors and, where feasible, randomisation. In organisational settings, controlled experiments, phased pilots and quasi-experimental comparisons can be useful. However, experimentation must respect consent, fairness and the possibility that a control group may receive less beneficial treatment.

A practical evidence statement should specify: the claim, the population, the measure, the comparison, the time period, the uncertainty and the decision implication. This prevents a colourful chart from becoming an unsupported conclusion.

6.2 Qualitative and experiential evidence

Interviews, observations, focus groups, narratives and professional judgement are legitimate evidence when collected systematically and interpreted transparently. Qualitative evidence is particularly valuable for understanding meaning, context, unintended effects and mechanisms. It should not be dismissed because it is not numerical. Conversely, a small number of vivid stories should not be treated as representative without considering sampling and saturation.

Strong leadership evidence often combines methods. For example, employee survey trends may be interpreted alongside interviews, absenteeism data and team-level observations. Convergence strengthens confidence; divergence signals the need for further inquiry.

7. People Analytics and Leadership Dashboards

People analytics applies workforce data to questions such as retention, capability, inclusion, wellbeing and workforce planning. Common measures include voluntary turnover, internal mobility, absenteeism, time to fill, engagement, learning participation and representation. Each metric requires a precise definition.

For example, a basic turnover rate may be calculated as:

The period and denominator must be specified. If a company has 48 separations during a year and an average headcount of 1,200, the turnover rate is (48 ÷ 1,200) × 100 = 4%. A different denominator, such as end-of-period headcount, may produce another figure. Comparisons are valid only when definitions are consistent.

Dashboards should support decisions rather than merely display data. An effective dashboard identifies the strategic question, presents a small number of relevant indicators, shows trends and comparisons, makes definitions accessible and signals data quality. It should distinguish:

• Descriptive analytics: What happened?
• Diagnostic analytics: Why might it have happened?
• Predictive analytics: What may happen?
• Prescriptive analytics: What actions could be considered?

A prediction is not a command. If an algorithm predicts that an employee may leave, the ethical response is not automatic surveillance or denial of opportunity. Leaders should investigate organisational conditions, offer fair support and avoid treating a probabilistic classification as a personal verdict.

7.1 Metric validity and unintended consequences

Measures shape behaviour. If managers are evaluated only on speed of recruitment, they may sacrifice quality or inclusion. If sales leaders are rewarded solely on revenue, they may encourage aggressive practices that damage trust. Good measurement systems combine outcome, process and balancing measures.

A metric can be mathematically accurate yet conceptually invalid. Counting training hours does not establish learning; employee presence does not establish contribution; engagement scores do not fully measure wellbeing. Leaders should ask whether the metric is a proxy, what it omits and how people may game it.

8. Uncertainty, Risk and Adaptive Decisions

Risk refers to situations in which possible outcomes and their probabilities can be estimated with some credibility. Uncertainty refers to situations in which probabilities, outcomes or causal relationships are poorly known. The distinction is practical rather than absolute.

Leaders can manage uncertainty by:

• identifying assumptions and confidence levels;
• separating reversible from irreversible commitments;
• using pilots and staged investment;
• establishing trigger points and stop-loss rules;
• seeking diverse information sources;
• updating beliefs when new evidence appears;
• communicating uncertainty honestly.

Bayesian reasoning provides a conceptual model for updating beliefs. If P(H) is an initial belief in hypothesis H, new evidence E should change the belief according to:

Here, P(H|E) is the posterior probability, P(E|H) is the likelihood of observing the evidence if the hypothesis is true, P(H) is the prior probability and P(E) is the overall probability of the evidence. The formula applies when these probabilities are defined coherently. In managerial practice, the value lies less in exact calculation than in asking how strong the new evidence is and how plausible the initial belief was.

9. Ethics, Privacy and Fairness

Data-informed leadership creates ethical obligations. Employees may not know what data are collected, how they are combined or how decisions affect them. Ethical governance requires purpose limitation, data minimisation, access controls, retention limits, security and transparent communication. Applicable data-protection law must be followed, including requirements concerning lawful processing, individual rights and automated decision-making where relevant.

Fairness requires more than removing sensitive attributes from a dataset. Other variables may act as proxies for protected characteristics, and historical data may encode past discrimination. Leaders should examine:

• representation and missingness across groups;
• differences in selection, error and benefit rates;
• whether the target variable is itself biased;
• whether the system creates disparate impact;
• who can challenge or appeal a decision;
• whether human oversight is meaningful rather than symbolic.

Fairness metrics can conflict. For example, equal selection rates, equal false-positive rates and equal predictive accuracy may not all be achievable simultaneously when groups have different base rates. Therefore, fairness is a normative and governance question, not only a technical one. The organisation must state which conception of fairness is relevant and why.

Leaders should apply a proportionality test: Is the data use necessary for a legitimate purpose? Is there a less intrusive alternative? Are benefits sufficient to justify risks? Can affected people understand and contest the decision? High-impact decisions concerning employment, promotion, discipline or access to essential services require especially strong safeguards.

10. Artificial Intelligence in Leadership Decisions

Artificial intelligence can summarise evidence, detect patterns, forecast demand and support scenario analysis. It can also reproduce bias, generate inaccurate content, obscure reasoning and encourage automation bias—the tendency to accept machine output because it appears technical.

Responsible use requires a human decision owner, documented purpose, validated data, performance testing across relevant groups, monitoring for drift, security controls and an appeal process. Leaders should not place confidential employee information into unapproved systems. AI-generated recommendations should be treated as inputs requiring verification, not as authoritative decisions.

A sound governance question is not simply “Is the model accurate?” It is “Accurate for whom, for what purpose, under which conditions, and with what consequences?” A model with high average accuracy may still be unacceptable if its errors are concentrated among vulnerable groups or if the decision cannot be challenged.

11. An Integrated Leadership Decision Protocol

The following protocol combines analytical discipline with participation and ethical governance.

Frame. Define the decision, affected stakeholders, scope, time horizon and decision owner. Separate symptoms from causes.

Clarify criteria. Identify desired outcomes, constraints, non-negotiable ethical requirements and trade-offs. Make weights or priorities explicit.

Generate alternatives. Include the status quo, a conservative option, an innovative option and, where relevant, a staged or reversible option. Avoid evaluating only the first idea.

Gather evidence. Combine quantitative and qualitative sources. Assess validity, reliability, causal strength, timeliness and limitations.

Engage the right people. Decide who provides information, who is consulted, who decides and who must implement. Protect dissent.

Test the decision. Use scenarios, sensitivity analysis, a pre-mortem, red-team challenge and fairness review.

Decide and communicate. State the choice, rationale, assumptions, risks, responsibilities and review date. Communicate uncertainty rather than disguising it.

Implement and monitor. Define leading indicators, outcome measures, trigger points and escalation routes.

Learn and update. Compare expectations with results, record lessons and revise the decision when evidence changes.

12. Illustrative Case: A Retention Dashboard and the Risk of Misinterpretation

Background

A fictional organisation, Northbridge Services, reports rising voluntary turnover among early-career analysts. The executive team proposes an algorithm that identifies employees likely to resign. Managers would then receive alerts and retention recommendations.

Problem

The chief people officer wants rapid action, but employee representatives question whether the system would monitor private behaviour or label individuals unfairly. The existing dashboard shows that turnover is highest in two departments, but it does not distinguish resignation, dismissal, internal transfer or retirement. Survey participation is also lower among night-shift workers.

Analysis

The leadership team first improves the metric definition. It separates voluntary turnover from other separations and calculates rates by department, tenure and work pattern. It discovers that one department has high turnover because new analysts are assigned unstable schedules and receive limited feedback. The dashboard initially suggested a general “engagement problem,” but interviews reveal a more specific workload and management issue.

The team then examines the proposed algorithm. Historical data reflect earlier managerial decisions, including fewer development opportunities for part-time workers. A model trained on those data might reproduce the pattern by treating part-time status as a risk indicator. The team concludes that prediction alone would not address the underlying cause.

Approach and solution

Northbridge adopts a staged intervention:

1. Managers receive training on scheduling, feedback and career conversations.
2. Employees are offered a confidential channel to report workload concerns.
3. A voluntary development programme is piloted in the affected department.
4. Turnover is monitored with clearly defined measures and subgroup checks.
5. Any future predictive tool must pass privacy, validity and fairness review, and may recommend organisational interventions but may not automatically deny promotion or impose surveillance.

Outcome

After two review periods, voluntary turnover declines in the pilot department and employee reports indicate improved schedule predictability. The organisation does not claim that the intervention alone caused the improvement; seasonal demand and hiring changes may also have contributed. It therefore continues monitoring and compares results with a similar department not yet receiving the full intervention.

Learning points

The case illustrates that analytics is most valuable when it improves problem definition and organisational learning. A prediction of who may leave is less useful than evidence about why people leave and which conditions leaders can change. It also demonstrates that privacy and fairness are integral to decision quality, not obstacles added after technical development.

Challenges and Limitations

Decision frameworks can create false confidence. A weighted score may hide disputed values; a forecast may disguise fragile assumptions; participation may become symbolic; and a dashboard may reward what is measurable rather than what matters. Data may be incomplete, delayed, strategically manipulated or shaped by previous decisions. Leaders must therefore retain professional judgement and moral accountability.

There is also a tension between speed and deliberation. Excessive analysis can delay action, while premature action can create avoidable harm. The appropriate balance depends on reversibility, stakes, uncertainty and the cost of delay. In high-stakes decisions, leaders should prefer transparent assumptions, independent challenge and documented review over apparent certainty.

Best-Practice Principles

1. Define the decision before collecting data.
2. Use multiple forms of evidence and explain their limitations.
3. Distinguish correlation, prediction and causation.
4. Make criteria, assumptions and decision rights visible.
5. Match participation to expertise, acceptance and urgency.
6. Invite dissent before commitment, not merely after failure.
7. Use pilots, review points and exit criteria for uncertain decisions.
8. Treat metrics as imperfect proxies and monitor unintended effects.
9. Protect privacy and test systems for disparate impact.
10. Keep a human decision owner accountable for consequential outcomes.

Chapter Summary

Leadership decision-making combines rational analysis, behavioural awareness and participative judgement. The rational model provides structure through objectives, alternatives, criteria and consequences, while bounded rationality recognises limits of time, information and attention. Expected value, weighted scoring, scenarios and decision trees can clarify choices, but they depend on credible assumptions and should not replace ethical judgement.

Behavioural approaches show how anchoring, confirmation bias, overconfidence, status quo bias and escalation of commitment can distort decisions. Structured dissent, pre-mortems, independent estimates and review criteria help manage these risks. Participative models clarify when leaders should decide, consult, facilitate or delegate, while preserving accountability.

Evidence must be assessed for relevance, validity, reliability, causal strength and practical significance. People analytics and dashboards are useful only when measures are well defined and connected to decisions. Predictive analytics and AI require human oversight, privacy protection, fairness assessment, security and avenues for challenge. Effective leaders make uncertainty visible, protect dissent, test assumptions, monitor consequences and update decisions as evidence changes.

Key Terms

Bounded rationality: Decision-making constrained by limited information, time and cognitive capacity.

Cognitive bias: A systematic tendency that can distort judgement.

Decision tree: A representation of sequential choices, chance events and outcomes.

Descriptive analytics: Analysis of what has happened.

Expected value: A probability-weighted average of possible outcomes.

Fairness: The ethically and socially justified treatment of people in a decision system.

Groupthink: Suppression of critical evaluation caused by excessive pressure for group agreement.

People analytics: Use of workforce data to understand and improve organisational decisions.

Participative decision-making: Involving affected or knowledgeable people in the decision process.

Pre-mortem: An exercise that assumes failure and identifies possible causes in advance.

Predictive analytics: Analysis used to estimate likely future outcomes.

Proxy variable: A variable that indirectly represents another characteristic, sometimes creating unfair effects.

Risk: A situation in which possible outcomes and probabilities can be estimated to some degree.

Satisficing: Selecting an option that meets an acceptable threshold rather than maximising an objective.

Scenario analysis: Examination of decisions across several plausible future conditions.

Sensitivity analysis: Testing how results change when assumptions or inputs change.

Uncertainty: A condition in which outcomes, probabilities or causal relationships are poorly known.

Validity: The extent to which a measure or method represents what it claims to represent.

Review Questions

Short Answer Questions

1. What distinguishes decision quality from decision outcome?
2. Define bounded rationality and satisficing.
3. State two methods for reducing groupthink.
4. What is the difference between correlation and causation?
5. Why should a leadership dashboard include metric definitions and data-quality information?

Long Answer Questions

1. Compare the rational, behavioural and participative models of leadership decision-making. Explain when each is most useful.
2. Discuss how cognitive biases can affect promotion or hiring decisions and propose a process to reduce their influence.
3. Explain the difference between risk and uncertainty and evaluate how staged commitments can help leaders manage uncertainty.
4. Critically assess the benefits and limitations of people analytics in employee retention decisions.
5. Explain why privacy and fairness are leadership responsibilities when AI or predictive analytics is used in employment decisions.

Analytical / Application Questions

1. A team gives a proposed project a high weighted score, but the result changes when the cost criterion increases slightly. What does this reveal, and what should the leader do next?
2. A dashboard shows that one demographic group has lower promotion rates. Design an evidence-gathering process that distinguishes possible performance differences, opportunity differences and measurement bias.
3. You chair a committee deciding whether to close a poorly performing service. Apply a decision protocol that includes stakeholder participation, a pre-mortem, uncertainty management, ethical review and post-decision monitoring.

Multiple Choice Questions with Answers

Question 1

Which feature most directly characterises bounded rationality?

A. Decision-makers always maximise expected utility

B. Decision-makers operate with limited information, time and cognitive capacity

C. Decisions are made entirely by group consensus

D. All relevant outcomes have known probabilities

Correct Answer: B. Decision-makers operate with limited information, time and cognitive capacity

Question 2

In a weighted-scoring model, what does the weight assigned to a criterion represent?

A. The number of alternatives available

B. The reliability of the data source only

C. The relative importance of that criterion

D. The probability that the alternative will succeed

Correct Answer: C. The relative importance of that criterion

Question 3

An expected value of £80,000 means that a decision will definitely produce £80,000. Is this statement correct?

A. Yes, because expected value is the guaranteed outcome

B. Yes, if the largest outcome is positive

C. No, because expected value is a probability-weighted average, not a guaranteed result

D. No, because expected value can only be used for non-financial outcomes

Correct Answer: C. No, because expected value is a probability-weighted average, not a guaranteed result

Question 4

Which practice is most effective for reducing groupthink?

A. Requiring immediate unanimity

B. Allowing the leader to state a preferred answer first

C. Using independent assessments and structured dissent

D. Excluding external experts from the discussion

Correct Answer: C. Using independent assessments and structured dissent

Question 5

A relationship between employee engagement and performance does not by itself establish that engagement causes performance because:

A. Correlation cannot identify all alternative explanations

B. Engagement can never be measured

C. Performance data are always qualitative

D. Causation requires a larger dashboard

Correct Answer: A. Correlation cannot identify all alternative explanations

Question 6

If an organisation records 30 separations during a period and has an average headcount of 750, what is the turnover rate using the formula in the chapter?

A. 2%

B. 4%

C. 25%

D. 780%

Correct Answer: B. 4%

Question 7

Which is an example of diagnostic analytics?

A. Reporting last year’s turnover rate

B. Estimating which employees may resign next year

C. Investigating why turnover increased in one department

D. Automatically selecting employees for promotion

Correct Answer: C. Investigating why turnover increased in one department

Question 8

Why can removing a protected attribute from an algorithm fail to eliminate unfairness?

A. Algorithms cannot process numerical data

B. Other variables may act as proxies for the protected attribute

C. Protected attributes are never relevant to fairness analysis

D. Removing variables always increases accuracy

Correct Answer: B. Other variables may act as proxies for the protected attribute

Question 9

Which decision is most suitable for a staged pilot?

A. A reversible intervention with uncertain effects

B. A decision where no learning is possible

C. A fully routine transaction with a fixed procedure

D. A decision that must remain secret from all implementers

Correct Answer: A. A reversible intervention with uncertain effects

Question 10

What is the primary leadership implication of an AI-generated recommendation in a high-impact employment decision?

A. The recommendation should be accepted automatically because it is data-based

B. The model replaces the need for accountability

C. The recommendation should be treated as an input subject to validation, oversight and challenge

D. The recommendation is fair if its average accuracy is high

Correct Answer: C. The recommendation should be treated as an input subject to validation, oversight and challenge