People Analytics and Evidence-Based OB Decisions

Introduction

Organizational behavior (OB) traditionally relied on managerial experience, employee interviews, engagement surveys, and established theories of motivation, leadership, groups, and culture. These sources remain valuable, but contemporary organizations also generate extensive digital evidence: collaboration records, workforce information systems, customer outcomes, performance data, learning histories, recruitment data, and employee sentiment. People analytics is the disciplined use of such evidence to understand, predict, and improve organizational decisions.

People analytics is not simply the production of dashboards or the application of machine learning to employee data. Its central challenge is epistemic and managerial: converting an organizational problem into a defensible question, a measurable construct, an appropriate research design, and an action whose consequences can be evaluated. Evidence-based OB combines the best available research evidence, organizational data, professional expertise, and the interests and values of affected stakeholders.

For example, a manager may ask, “Why is morale low?” This question is too vague for rigorous analysis. A stronger sequence is:

1. Which employees are experiencing lower work engagement?
2. How will engagement be defined and measured?
3. Is the concern associated with workload, supervisor behavior, role ambiguity, or career opportunity?
4. Which relationships are merely correlational, and which interventions might plausibly cause improvement?
5. How will risks to privacy, fairness, and employee autonomy be controlled?

Analytics can improve decisions, but it can also amplify biased measures, encourage surveillance, confuse prediction with explanation, and turn employees into objects of automated classification. Advanced OB practice therefore requires technical competence together with theoretical reasoning, ethical judgment, and institutional governance.

Learning Objectives

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

• translate managerial questions into clearly defined constructs and measurable variables;
• distinguish constructs, indicators, outcomes, predictors, mediators, moderators, and confounders;
• select and interpret surveys, experiments, quasi-experiments, network analysis, workforce metrics, and predictive models;
• distinguish association, prediction, explanation, and causation;
• evaluate reliability, validity, sampling, missing data, and measurement bias;
• interpret dashboards without mistaking descriptive metrics for causal evidence;
• assess privacy, fairness, transparency, accountability, and responsible-AI risks; and
• design an evidence-based people decision with appropriate safeguards and evaluation.

1. From Organizational Questions to Measurable Constructs

Constructs and operationalization

A construct is an abstract concept that cannot usually be observed directly. Examples include psychological safety, burnout, trust in leadership, inclusion, organizational commitment, role conflict, and perceived justice. A variable is an observable representation of a construct, such as a survey score, absence count, turnover status, or response time.

Operationalization is the process of specifying how a construct will be observed. A useful operational definition identifies:

• the conceptual meaning of the construct;
• the population and unit of analysis;
• the measurement instrument or data source;
• the time period;
• the scale and coding rules; and
• limitations and plausible sources of bias.

A single indicator is often insufficient. Psychological safety, for instance, may be assessed through several survey items concerning interpersonal risk, error reporting, speaking up, and response to dissent. Administrative data may supplement, but not replace, these perceptions: incident-reporting rates can reflect safety, but they can also reflect workload, reporting norms, or fear of consequences.

A construct-to-decision map

A disciplined analysis distinguishes the following elements:

A causal diagram can sharpen reasoning. Suppose schedule unpredictability is associated with turnover, partly because it increases work–family conflict. Tenure and local labor-market conditions may affect both scheduling and turnover. Without accounting for them, the observed relationship may be misleading.

Measurement levels and summary statistics

Nominal variables classify categories without numerical distance, such as department. Ordinal variables rank categories, such as low, medium, and high satisfaction. Interval-like scales assume approximately equal distances, although this assumption may be contestable for rating scales. Ratio variables have a meaningful zero, such as hours worked or days absent.

For observations x₁, x₂, …, xₙ, the arithmetic mean is:

Here, x̄ is the sample mean, xᵢ is observation i, and n is the number of observations. The mean is most informative when the scale is reasonably quantitative and extreme values do not dominate. The median may better represent skewed variables such as overtime hours or compensation.

The sample standard deviation is:

The denominator n − 1 provides the usual unbiased estimate of population variance under random sampling assumptions. Standard deviation is expressed in the same units as x. These summaries describe data; they do not establish why differences exist.

2. Evidence Designs in Organizational Behavior

Surveys and psychometric measurement

Surveys are efficient for measuring perceptions, attitudes, climate, and self-reported behavior. Good survey design requires clear items, appropriate response options, attention to wording and order effects, protection from coercion, and a sampling strategy aligned with the decision.

Reliability concerns consistency. Internal consistency is sometimes summarized by coefficient alpha, but alpha is not a universal measure of validity and can be inflated by many redundant items. A scale may be reliable yet measure the wrong construct. Validity requires evidence that scores support the intended interpretation. Important forms include:

• content validity: coverage of relevant aspects of the construct;
• construct validity: convergence with related measures and distinction from different constructs;
• criterion-related validity: association with a meaningful outcome; and
• consequential validity: consideration of effects produced by using the measure.

Common threats include nonresponse bias, social desirability, acquiescence, common-method bias, translation problems, and fear that responses are identifiable. Reporting a response rate alone is inadequate: respondents and nonrespondents may differ systematically.

Experiments and causal inference

A randomized experiment assigns units to treatment or control conditions using a random mechanism. If implementation is adequate, randomization tends to balance observed and unobserved characteristics in expectation. The average treatment effect can be represented as:

Y(1) is the potential outcome if a unit receives treatment, Y(0) is the potential outcome without treatment, and E denotes the average across the target population. The fundamental problem is that each unit is observed under only one condition. Randomization permits estimation of the average contrast, subject to assumptions such as consistency, appropriate treatment assignment, limited interference, and adequate implementation.

For example, a company might randomly assign eligible teams to a structured manager-feedback program or business-as-usual practice. If treated teams subsequently show higher engagement, the difference is more credible as an effect of the program than a simple comparison of volunteers with non-volunteers.

Organizational experiments face practical difficulties: contamination between groups, attrition, small samples, managerial noncompliance, ethical concerns, and outcomes that take time to emerge. Cluster randomization may be preferable when employees within the same team influence one another, but it requires attention to the number of clusters and intra-cluster dependence.

When randomization is impossible, quasi-experimental designs can improve inference. Difference-in-differences compares changes over time in a treated group with changes in a comparison group:

T and C denote treatment and comparison groups; pre and post denote periods before and after the intervention. A central assumption is parallel trends: absent treatment, the groups would have followed similar average trends. This assumption must be examined using pre-intervention data and substantive reasoning; it cannot be guaranteed by the formula.

Regression discontinuity, interrupted time series, matched comparisons, and natural experiments can also provide useful evidence, but each depends on design-specific assumptions. Statistical adjustment for many variables does not automatically create causal validity.

Correlation, prediction, explanation, and causation

The Pearson correlation between X and Y is:

Cov(X,Y) is covariance and σₓ and σᵧ are standard deviations. The coefficient describes linear association and ranges from −1 to +1 under conventional conditions. It does not show that X causes Y. A third variable may cause both, reverse causality may operate, or the relationship may be a statistical artifact.

A predictive model is evaluated by how accurately it estimates outcomes for new cases. A causal model asks what would happen under an intervention. A variable can be useful for prediction without being an appropriate intervention target. For example, recent sick leave may predict resignation, but punishing sick leave is unlikely to be a valid retention intervention and may create legal and ethical harm.

Workforce metrics

Common metrics include turnover, retention, absence, time to fill, internal mobility, promotion rates, span of control, overtime, compensation equity, and training completion. Every metric needs a denominator, time window, inclusion rule, and interpretation.

A crude turnover rate for a period may be calculated as:

The result is a proportion for the specified period. Annualizing a short-period rate assumes a stable process and may be misleading when seasonal patterns are strong. Voluntary, involuntary, regrettable, and avoidable turnover should not be treated as interchangeable categories.

The validity of a metric depends on its relationship to the decision. A high training-completion rate may indicate compliance, not learning or behavior change. A low absence rate may reflect healthy attendance or presenteeism. Metrics can become targets, and when people optimize the target rather than the underlying purpose, Goodhart’s law becomes relevant: a measure used as a target can cease to be a good measure.

Social network analysis

Organizational network analysis studies relationships such as advice, information exchange, collaboration, or friendship. A network consists of nodes and ties. Degree centrality counts direct connections; betweenness centrality reflects the extent to which a person lies on shortest paths between others; eigenvector-like measures reflect connection to well-connected actors.

Network measures are descriptive and context-dependent. A person with high betweenness may be a valuable broker, overloaded coordinator, or an information bottleneck. Network data are especially sensitive because employees may not expect relational disclosures to be used for evaluation. Consent, purpose limitation, de-identification, and restrictions on individual-level managerial use are essential.

3. Predictive Models and Interpretable Decision Support

A basic linear model can be written as:

Here, yᵢ is the outcome for case i, β₀ is the intercept, βⱼ are coefficients, xᵢⱼ are predictors, p is the number of predictors, and εᵢ is unexplained error. The model is useful for estimating conditional associations under assumptions concerning functional form, independence, error behavior, and measurement. A coefficient is not automatically a causal effect.

For binary outcomes, logistic regression models the log odds:

pᵢ is the predicted probability of the outcome for case i. The exponential of a coefficient, e^βⱼ, is an odds ratio for a one-unit increase in xⱼ, holding the other included predictors constant. Odds ratios are not risk ratios, especially when outcomes are common.

More complex models, including random forests, gradient boosting, and neural networks, can improve predictive performance but may reduce interpretability. Model selection should consider discrimination, calibration, stability, subgroup performance, operational usefulness, and the consequences of errors. A model with high overall accuracy can still perform poorly for a small subgroup.

A sound validation process separates training, validation, and test information where feasible. Temporal validation is often appropriate for workforce data because future decisions should be tested on later cases rather than randomly mixed historical cases. Data leakage occurs when information unavailable at the time of decision enters the model, producing unrealistically optimistic performance.

A risk score should support human judgment, not silently determine employment outcomes. Managers should know the model’s purpose, evidence base, limitations, appeal process, and appropriate action. Predictive scores should generally trigger supportive inquiry—for example, a conversation about workload or development—not automatic punishment or exclusion.

4. Dashboards and Evidence-Based Decisions

A dashboard is an interface for monitoring selected information. It is not an analysis by itself. Effective dashboards connect a decision to a small set of meaningful measures, show trends and uncertainty, provide definitions, and distinguish descriptive, predictive, and causal evidence.

A useful evidence chain is:

Question → construct → measure → design → analysis → decision → evaluation

For example, the question “Has the new hybrid-work policy improved collaboration?” requires a definition of collaboration, such as advice ties, project coordination, or perceived coordination. It requires a baseline, comparison strategy, attention to job type, and a consideration of unintended effects such as meeting overload or exclusion of remote employees.

Dashboards should display denominators, sample sizes, missingness, comparison periods, confidence intervals where appropriate, and suppression rules for small groups. Color coding can exaggerate small differences. Ranking managers by unstable engagement scores can create competition and discourage honest reporting. A dashboard should therefore provide context rather than invite simplistic league tables.

Evidence-based management is not evidence-only management. Managerial experience helps interpret local context; employee voice identifies consequences and feasible solutions; research evidence provides tested concepts and methods; organizational data reveal patterns and constraints. A defensible decision makes these sources explicit and records uncertainty.

5. Data Quality, Validity, and Reproducibility

People data are often generated for administrative purposes rather than research. Payroll data may be accurate for payment while being unsuitable for measuring effort. Collaboration metadata may record digital activity but not the quality of collaboration. Performance ratings may combine results, visibility, favoritism, and role differences.

A data-quality review should assess:

• accuracy and consistency;
• completeness and missingness mechanisms;
• timeliness and version history;
• duplicate records and identifier errors;
• representativeness of included employees;
• changes in definitions over time; and
• whether the data-generating process itself changed.

Missingness is not automatically random. Employees who distrust management may skip a survey, and employees under severe workload may have less time to respond. Complete-case analysis can therefore distort results. Analysts should describe missingness, conduct sensitivity analyses where appropriate, and avoid imputing sensitive outcomes without defensible assumptions.

Reproducibility requires documented data definitions, code or analytic procedures, transformation logs, model versions, decision thresholds, and records of who approved changes. Privacy protection may require that analysts cannot access direct identifiers, while authorized personnel retain a controlled linkage key. Reproducibility and confidentiality are compatible when access is governed carefully.

6. Ethics, Privacy, Fairness, and Responsible AI

People analytics involves power asymmetry. Employees may technically consent to a policy while reasonably believing refusal could harm their employment. Ethical governance should therefore rely on necessity, proportionality, transparency, purpose limitation, data minimization, security, retention limits, and meaningful accountability.

Important questions include:

• Is the data collection necessary for a legitimate and clearly stated purpose?
• Would employees reasonably expect this use?
• Are less intrusive alternatives available?
• Who can access individual-level information?
• How long will data be retained?
• Can employees challenge inaccurate data or consequential decisions?
• Are vulnerable groups exposed to disproportionate risk?

Privacy is not solved merely by removing names. Re-identification may be possible through combinations of job title, location, timing, and network position. Aggregation and suppression can reduce risk, but small teams may still be identifiable. Differential privacy and secure computation may be relevant in advanced settings, although they introduce technical trade-offs and do not repair invalid constructs.

Fairness is multidimensional. Depending on the decision, organizations may examine demographic parity, equalized error rates, calibration, subgroup validity, or individual consistency. These criteria can conflict, particularly when base rates differ. No single fairness metric resolves the normative question of what equal treatment means. Fairness assessment must include the complete decision process: data collection, labels, feature selection, model outputs, human review, and consequences.

Bias can enter through historical decisions, underrepresentation, selective measurement, proxy variables, label error, unequal access to opportunities, or differential implementation. Removing a protected attribute does not remove bias if correlated proxies remain. Conversely, protected-group information may be needed for auditing fairness, subject to lawful and carefully governed use.

Responsible AI requires documented purpose, impact assessment, human oversight, explainability appropriate to the decision, monitoring for drift, incident response, auditability, and a route for appeal. High-impact employment decisions deserve stronger safeguards than low-stakes workforce planning. Compliance with applicable employment, privacy, anti-discrimination, and data-protection law is a minimum, not a substitute for ethical judgment.

7. An Integrated Decision Process

A robust people-analytics project can follow these stages:

1. Frame the decision. Specify who must decide, what action is possible, the time horizon, and what would count as improvement.
2. Develop a theory of change. Explain how the proposed action is expected to influence behavior and outcomes.
3. Define constructs and measures. Separate perceptions, behaviors, outcomes, and contextual variables; document operational definitions.
4. Choose the design. Prefer randomization when feasible; otherwise consider credible quasi-experimental or longitudinal designs.
5. Audit data and risks. Examine quality, representativeness, privacy, fairness, legal constraints, and unintended consequences.
6. Analyze uncertainty. Report effect sizes, intervals, sensitivity checks, subgroup results, and practical significance rather than relying only on statistical significance.
7. Make the decision. Combine evidence with expertise, employee voice, resources, and organizational values.
8. Evaluate implementation. Monitor whether the intervention was delivered, whether outcomes changed, and whether harms emerged.
9. Learn and revise. Treat the intervention as a learning cycle rather than a one-time analytics product.

Worked example: interpreting an intervention effect

Suppose 200 employees are randomly assigned equally to a manager-coaching program or control condition. After three months, average engagement is 4.10 in the coaching group and 3.70 in the control group on a five-point scale. The estimated difference is:

The units are engagement-scale points. This difference is an estimate of the average treatment effect under the experiment’s assumptions. It does not show that every employee improved, that the effect will persist, or that coaching caused improved business performance. The analyst should also examine uncertainty, attrition, implementation fidelity, subgroup effects, cost, and possible adverse effects. A statistically detectable 0.40-point improvement may be operationally unimportant if the program is expensive or difficult to sustain; conversely, a modest average effect may be valuable if it is inexpensive and concentrated among employees at high risk of disengagement.

8. Illustrative Case: Northstar Services

Background

Northstar Services, a fictional regional customer-support organization, reports rising voluntary turnover among experienced agents. Senior leaders propose an automated “flight-risk” model based on attendance, schedule changes, performance records, survey responses, and collaboration data.

Problem

The initial proposal treats predicted turnover as an individual employee problem. It does not specify whether the goal is to reduce regrettable exits, improve working conditions, or reduce recruitment costs. Employees are concerned that being labeled a flight risk will affect promotion and scheduling.

Analysis

The people-analytics team reframes the project. The primary outcome becomes voluntary resignation within six months, with a secondary outcome of employee-reported workload and supervisor support. The team discovers that schedule instability is more common in one site because of staffing shortages. Historical performance ratings are also less favorable for part-time employees, partly because they receive fewer high-visibility assignments.

A model trained on historical resignations predicts well overall but has higher false-positive rates for part-time workers. Attendance is predictive, but the team concludes that using it as a punitive signal could penalize employees with health or caregiving responsibilities. Network data are excluded from individual risk scoring because employees were not clearly informed of that use.

Approach and solution

Northstar adopts a three-part design:

• conduct a confidential survey measuring workload, schedule predictability, psychological safety, and supervisor support;
• pilot manager coaching and schedule-stability changes in randomly selected sites, with comparison sites continuing usual practice; and
• use the predictive model only in aggregate to identify teams requiring review, not to rank or discipline individuals.

Employees receive a plain-language explanation of the purpose, data sources, retention period, and appeal process. Results are reported with subgroup checks and confidence intervals. Managers are instructed to respond with supportive conversations and resource adjustments rather than assumptions about employees’ intentions.

Outcome

After six months, the pilot sites show a reduction in voluntary turnover and improved schedule predictability. The largest improvement occurs in teams with initially high workload. However, the evaluation finds no clear change in customer satisfaction, and one site reports increased administrative burden. Northstar therefore continues the intervention with revised implementation guidance rather than claiming universal success.

Learning points

The case demonstrates that analytics quality depends on problem framing, construct validity, causal design, fairness analysis, employee participation, and responsible use. A model can be statistically accurate yet ethically inappropriate. Aggregate prediction can support resource allocation without converting uncertain probabilities into individual surveillance. Most importantly, the intervention targeted working conditions rather than treating employees as the sole source of turnover risk.

Challenges and Contemporary Developments

People analytics is becoming more real-time through digital exhaust, natural-language analysis, generative AI, and integrated workforce platforms. These developments can identify patterns quickly, but speed increases the risk of premature action. Generative systems may produce persuasive explanations unsupported by valid evidence, expose confidential information, or reproduce stereotypes from training data. Synthetic data can assist development but may preserve structural patterns and should not be assumed private automatically.

The future of OB analytics is likely to emphasize causal machine learning, continuous experimentation, employee-centered data governance, privacy-enhancing technologies, and measurement of algorithmic impact over time. Advanced practice will also move beyond prediction toward organizational learning: identifying which conditions produce sustainable performance, inclusion, health, and meaningful work.

Chapter Summary

People analytics applies measurement, research design, statistics, and responsible technology to organizational decisions. Its foundation is the translation of broad OB questions into constructs, indicators, outcomes, and a theory of change. Surveys provide evidence about perceptions but require reliability, validity, representative participation, and protection from coercion. Experiments and credible quasi-experiments provide stronger evidence about causal effects than cross-sectional correlations, although each depends on assumptions and implementation quality.

Workforce metrics describe organizational patterns but require clear denominators, definitions, time periods, and attention to unintended incentives. Network analysis reveals relational structure but creates heightened privacy concerns. Predictive models estimate future outcomes; they do not automatically explain causes or identify appropriate interventions. Dashboards are decision-support interfaces, not substitutes for research design or judgment.

High-quality analytics requires data-quality audits, transparent assumptions, appropriate validation, uncertainty reporting, subgroup evaluation, and reproducible procedures. Ethical practice requires necessity, proportionality, privacy, fairness, transparency, human oversight, accountability, and appeal. Evidence-based OB integrates research evidence, organizational data, professional expertise, and employee voice. The ultimate standard is not whether an organization can analyze more data, but whether it can make better, fairer, and more accountable decisions.

Key Terms

Construct: An abstract concept, such as trust or psychological safety, that must be represented through indicators.

Operationalization: The specification of how a construct will be measured.

Indicator: An observable variable used to represent a construct.

Reliability: The consistency of a measurement procedure or score.

Validity: Evidence that a measure supports the interpretation and use intended for it.

Confounder: A variable related to both a predictor and an outcome that can distort their observed relationship.

Mediator: A mechanism through which one variable influences another.

Moderator: A variable that changes the strength or direction of a relationship.

Correlation: Statistical association between variables, not necessarily causal influence.

Causal effect: The difference between potential outcomes under alternative conditions.

Average treatment effect: The average causal effect of treatment relative to a comparison condition.

Quasi-experiment: A design that estimates effects without full random assignment, using structured comparisons.

Turnover rate: Separations during a period divided by average headcount for that period.

Social network analysis: The study of nodes and relationships among organizational actors.

Predictive model: A model designed to estimate outcomes for new or future cases.

Data leakage: Inclusion of information in model development that would not be available at decision time.

Calibration: Agreement between predicted probabilities and observed outcome frequencies.

Fairness: Normative and statistical assessment of whether decisions produce unjustified disparities or unequal errors.

Purpose limitation: Restricting data use to a specified, legitimate purpose.

Responsible AI: Governance ensuring that AI systems are valid, transparent, fair, secure, accountable, and subject to appropriate human oversight.

Review Questions

5 Short Answer Questions

1. What is the difference between a construct and an indicator?
2. Why does correlation not establish causation?
3. What does the denominator contribute to a workforce metric?
4. State two threats to survey validity.
5. Why can removing a protected attribute fail to eliminate algorithmic bias?

5 Long Answer Questions

1. Explain how an organizational question can be translated into a construct-to-decision map. Illustrate your answer with a retention problem.
2. Compare randomized experiments and difference-in-differences designs, including their principal assumptions and practical limitations.
3. Discuss how reliability, validity, and consequential validity should influence the use of employee surveys.
4. Explain why a predictive turnover model should not automatically be used to discipline employees.
5. Develop a governance framework for a people-analytics dashboard used by senior leaders.

3 Analytical / Application Questions

1. An organization observes that teams with more overtime have higher turnover. Identify at least three possible confounders or alternative explanations, and propose a design that would better assess causality.
2. A model predicts resignation with high overall accuracy but has a substantially higher false-positive rate for part-time employees. Analyze the fairness and managerial risks, and recommend corrective actions.
3. Design an evidence-based evaluation of a hybrid-work policy. Specify the main construct, outcome measures, comparison strategy, data-quality checks, privacy safeguards, and criteria for deciding whether to continue the policy.

Multiple Choice Questions with Answers

Question 1

Which statement best distinguishes a construct from an indicator?

A. A construct is an abstract concept; an indicator is an observable measure representing it.

B. A construct is always numerical; an indicator is always qualitative.

C. A construct is a causal effect; an indicator is a confounder.

D. A construct is a dashboard; an indicator is a research design.

Correct Answer: A. A construct is an abstract concept; an indicator is an observable measure representing it.

Question 2

What is the principal identifying assumption in a basic difference-in-differences design?

A. The treatment group must be twice as large as the comparison group.

B. The groups would have followed parallel trends without treatment.

C. All employees must receive the intervention simultaneously.

D. The outcome must be measured only after the intervention.

Correct Answer: B. The groups would have followed parallel trends without treatment.

Question 3

A turnover rate is most appropriately calculated as:

A. Average headcount divided by separations during the period.

B. Voluntary separations divided by total employees ever hired.

C. Separations during the period divided by average headcount during the period.

D. Total employees at the end of the period divided by new hires.

Correct Answer: C. Separations during the period divided by average headcount during the period.

Question 4

Which statement about predictive models is most accurate?

A. A predictive variable is necessarily an appropriate intervention target.

B. High prediction accuracy proves that the model’s predictors cause the outcome.

C. Removing protected attributes guarantees fair predictions.

D. A model can predict an outcome without identifying a causal mechanism.

Correct Answer: D. A model can predict an outcome without identifying a causal mechanism.

Question 5

In the formula ATE = E[Y(1) − Y(0), what do Y(1) and Y(0) represent?

A. The average and standard deviation of the outcome.

B. Potential outcomes under treatment and no treatment.

C. The treated and untreated sample sizes.

D. Two independent survey indicators.

Correct Answer: B. Potential outcomes under treatment and no treatment.

Question 6

Which practice best addresses data leakage in workforce prediction?

A. Including all variables available anywhere in the organization.

B. Randomly shuffling future and historical records together.

C. Using only information that would have been available at the time of the decision.

D. Deleting variables with missing values regardless of their timing.

Correct Answer: C. Using only information that would have been available at the time of the decision.

Question 7

Why may a high training-completion rate be an inadequate measure of learning?

A. Completion measures participation or compliance, not necessarily knowledge or behavior change.

B. Completion rates cannot be represented numerically.

C. Training completion is always a causal outcome.

D. Learning cannot be measured using any form of evidence.

Correct Answer: A. Completion measures participation or compliance, not necessarily knowledge or behavior change.

Question 8

Which is an appropriate privacy safeguard for organizational network analysis?

A. Publish individual network positions to encourage competition.

B. Use relational data for any purpose once collected.

C. Avoid informing employees because network data are objective.

D. Limit purpose, access, retention, and individual-level managerial use.

Correct Answer: D. Limit purpose, access, retention, and individual-level managerial use.

Question 9

If a coaching group has an average engagement score of 4.10 and a control group has 3.70, what is the unadjusted difference?

A. 0.04 points

B. 0.30 points

C. 0.40 points

D. 7.80 points

Correct Answer: C. 0.40 points

Question 10

Which feature most clearly distinguishes evidence-based management from evidence-only management?

A. It rejects managerial experience and employee voice.

B. It integrates research, organizational data, expertise, and stakeholder values.

C. It uses only statistically significant results.

D. It replaces judgment with automated rankings.

Correct Answer: B. It integrates research, organizational data, expertise, and stakeholder values.