Digital IHRM, People Analytics and Artificial Intelligence

Introduction

International human resource management (IHRM) is increasingly conducted through digital systems that connect employees, managers, HR specialists and external partners across countries. Global human resource information systems (HRIS), cloud platforms, collaboration tools, people analytics and artificial intelligence (AI) now influence recruitment, mobility, performance management, learning, compensation, workforce planning and employee experience.

Digitalisation does not simply make existing HR processes faster. It changes who can access information, how decisions are made, which work is visible, and how organisations exercise control. A cloud HR platform may standardise data globally while allowing local legal adaptations. An algorithm may identify patterns in turnover but may also reproduce historical discrimination. A dashboard may support evidence-based decisions, but poorly defined indicators can create false confidence.

The central managerial challenge is therefore sociotechnical: HR professionals must understand technology, data, organisational context, law, ethics and human behaviour together. Effective digital IHRM combines global integration with local responsiveness, analytical discipline with professional judgement, and innovation with accountability.

Learning Objectives

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

1. Explain the role and architecture of global HRIS and cloud HR platforms.
2. Evaluate digital employee experience across international workforces.
3. Define people analytics and distinguish descriptive, diagnostic, predictive and prescriptive analysis.
4. Calculate and interpret selected HR metrics and dashboard indicators.
5. Assess AI applications in recruitment, performance, learning and workforce planning.
6. Identify risks involving privacy, cybersecurity, bias, transparency and automated decision-making.
7. Recommend governance practices for responsible digital IHRM.

1. The Digital Transformation of IHRM

Traditional HR administration was often country-based, paper-heavy and operationally fragmented. Digital IHRM integrates information and workflows across geographical boundaries. Typical capabilities include:

• a common global employee data model;
• self-service for employees and managers;
• digital recruitment and onboarding;
• online learning and skills inventories;
• payroll and benefits interfaces;
• workforce planning and scenario analysis;
• employee listening and collaboration platforms; and
• analytics and decision-support tools.

Digital IHRM has four strategic purposes. First, it improves administrative efficiency by reducing duplicate data entry and automating routine transactions. Second, it supports global visibility, allowing authorised users to compare workforce information across business units. Third, it enables employee and manager empowerment through self-service. Fourth, it can improve strategic decision-making by linking people data with business outcomes.

However, global digitalisation involves tensions. Standardisation can promote consistency, yet excessive standardisation may ignore national labour law, language, culture or employment practices. Centralised data can improve planning, yet may increase surveillance and cybersecurity exposure. Automation can reduce delays, yet may remove opportunities for explanation and human discretion.

2. Global HRIS and Cloud Platforms

2.1 HRIS architecture

A global HRIS generally contains several layers:

1. Data layer: employee identity, job, organisation, compensation, absence, skills and employment records.
2. Process layer: recruitment, onboarding, performance, learning, payroll interfaces and offboarding.
3. Integration layer: application programming interfaces (APIs), identity management and connections to finance, timekeeping and local payroll systems.
4. Analytics layer: reports, dashboards, models and data visualisation.
5. Governance layer: access control, retention, audit logs, data quality and legal compliance.

A single global system does not necessarily mean a single global process. A useful design distinguishes between:

• global core processes, such as employee identification, organisational hierarchy and minimum data definitions;
• regional or country extensions, such as statutory leave, tax, benefits and consultation requirements; and
• local configuration, such as language, approval paths and culturally appropriate communication.

This is sometimes called a “global template with local extensions.” It is more realistic than either complete uniformity or completely separate country systems.

2.2 Cloud computing in IHRM

In a cloud model, HR software and infrastructure are hosted by a service provider and accessed through a network. Common service arrangements include:

• Software as a Service (SaaS): the provider operates the application and normally manages updates.
• Platform as a Service (PaaS): the organisation uses a provider’s platform to develop or integrate applications.
• Infrastructure as a Service (IaaS): the provider supplies computing, storage and networking resources.

Cloud platforms can offer scalability, frequent updates, common data structures and lower internal infrastructure requirements. They also create dependence on vendors, connectivity and provider security controls. International organisations must examine where data are stored, who can access them, subcontracting arrangements, incident notification, business continuity and exit arrangements.

2.3 Evaluation criteria

An HR professional evaluating a global platform should consider the following dimensions:

Implementation should begin with requirements and process mapping rather than software demonstrations alone. A realistic business case includes licence costs, implementation, data cleansing, integration, training, change management, support and eventual replacement costs.

3. Digital Employee Experience

Digital employee experience (DEX) is the totality of employees’ interactions with digital tools, information, processes and support. It includes the usability of HR portals, collaboration software, virtual meeting tools, mobile access, learning systems and digital work practices.

A positive DEX is not equivalent to having many applications. It depends on coherence, reliability and trust. Important dimensions include:

• accessibility: employees can use systems despite disability, location, device or connectivity constraints;
• simplicity: common tasks require few steps and understandable language;
• consistency: interfaces and processes are reasonably coherent across countries;
• responsiveness: systems and support operate when needed;
• inclusion: tools do not disadvantage language groups, remote workers or employees with limited digital skills;
• privacy and control: employees understand what is collected and how it is used; and
• human support: employees can reach a person when automated service is inadequate.

International DEX is especially affected by time zones, bandwidth, language, cultural expectations and national regulation. A system designed for office-based employees may exclude factory, field or frontline workers. A global collaboration platform may improve connection but also intensify after-hours work and monitoring.

DEX should be evaluated through usage data, task completion, support requests, accessibility testing and employee research. A high login rate does not prove a good experience; employees may log in because the system is mandatory. Combining quantitative indicators with interviews and usability testing produces better evidence.

4. People Analytics: From Data to Decision

People analytics is the systematic use of workforce data, analytical methods and contextual knowledge to improve decisions affecting people and organisational performance. It is broader than HR reporting because it investigates relationships, explanations and possible actions.

4.1 Four levels of analysis

1. Descriptive analytics: What happened? Examples include headcount, absence and turnover reports.
2. Diagnostic analytics: Why might it have happened? Examples include examining turnover by manager, tenure, location or occupation.
3. Predictive analytics: What is likely to happen? Examples include estimating the probability of voluntary turnover.
4. Prescriptive analytics: What actions might be considered? Examples include comparing retention interventions under defined constraints.

These levels should not be confused with certainty. A predictive model produces an estimate, not a fact. A correlation between training participation and performance does not establish that training caused the performance difference. Employees who volunteer for training may already be more motivated or supported.

4.2 A disciplined analytics process

A defensible people analytics project normally follows this sequence:

1. Define the decision and its organisational purpose.
2. Identify the population, outcome and relevant comparison groups.
3. Specify data definitions, sources, time periods and ownership.
4. Assess data quality, representativeness and legal basis.
5. Explore patterns without prematurely claiming causation.
6. Select an appropriate analytical method.
7. Validate results, including subgroup performance and error rates.
8. Translate findings into options and safeguards.
9. Implement, monitor outcomes and review unintended effects.

The decision question should come before the dashboard or algorithm. Otherwise, organisations risk collecting large amounts of data without improving decisions.

5. Metrics, Measures and Dashboards

A metric is useful only when its numerator, denominator, population, period and purpose are clear. Metrics should also be interpreted alongside business context and qualitative evidence.

5.1 Common HR metrics

Turnover rate is commonly calculated as:

Average headcount is often approximated as:

This approximation assumes that the beginning and ending values reasonably represent the period. If headcount changes substantially during the period, a monthly or daily average may be more accurate. The metric should specify whether it includes voluntary separations, involuntary separations, internal transfers or only selected employee groups.

Absence rate can be expressed as:

The organisation must define absence days and available scheduled workdays consistently. Public holidays, approved leave and partial days may be treated differently under different policies.

Selection rate is:

A high selection rate may reflect a small applicant pool rather than effective recruitment. It should not be interpreted without considering applicant quality, labour-market conditions and hiring-manager requirements.

Cost per hire is often calculated as:

Costs may include advertising, agency fees, assessment, recruiter time, travel and onboarding. The organisation must state which costs are included and whether internal labour costs are estimated.

5.2 Worked example: interpreting turnover

An international service centre begins a quarter with 480 employees and ends with 520. During the quarter, 36 employees leave.

Average headcount is:

Therefore:

The result is a quarterly turnover rate, not automatically an annual rate. An annualised estimate might multiply by four only as a simple approximation, yielding 28.8%, but that assumes comparable quarterly conditions and does not account for changing headcount or repeated employee exposure. It should therefore be labelled clearly as an approximation rather than an observed annual rate.

5.3 Dashboard design

A useful dashboard connects indicators to decisions. It should include:

• a clearly stated purpose and audience;
• definitions and calculation dates;
• filters that do not enable inappropriate identification of individuals;
• trend and comparison views;
• thresholds linked to action, not merely colour coding;
• data-quality warnings;
• subgroup analysis where legally and statistically appropriate; and
• an explanation of uncertainty and limitations.

Dashboards can create risks. Ranking managers by turnover without adjusting for job type, labour market or inherited workforce composition may encourage gaming. Small-country results may be unstable or identify individuals. A dashboard should not become a surveillance instrument simply because the data are technically available.

6. Artificial Intelligence in IHRM

AI refers broadly to computational systems that perform tasks associated with human cognitive activities, such as classification, prediction, language processing, recommendation or generation. In IHRM, applications include:

• candidate sourcing and résumé classification;
• chatbots for employee enquiries;
• personalised learning recommendations;
• skills inference from work records;
• workforce demand forecasting;
• suggested performance comments;
• sentiment or text analysis; and
• scheduling and workforce optimisation.

AI systems differ in their technical and organisational risks. A rules-based workflow is not the same as a machine-learning model. A generative AI system that drafts interview questions creates different risks from a model that ranks candidates. The risk depends on the decision’s consequences, the data used, the model’s behaviour, the degree of automation and the possibility of human review.

6.1 Model performance and classification

For a binary classification system, such as predicting whether an employee may leave, outcomes can be represented as:

• true positive: a positive prediction for a person who experiences the outcome;
• false positive: a positive prediction for a person who does not experience the outcome;
• true negative: a negative prediction for a person who does not experience the outcome; and
• false negative: a negative prediction for a person who experiences the outcome.

Important measures include:

These measures assume a defined positive class and a labelled evaluation dataset. Accuracy can be misleading when the outcome is rare. Precision and recall change when the decision threshold changes. Evaluation should be conducted on data not used to train the model and should examine relevant employee groups.

6.2 Bias and fairness

Bias can enter through historical decisions, incomplete data, proxy variables, measurement choices, labels, sampling, model design or human use of outputs. Removing a protected characteristic does not necessarily remove bias because other variables may act as proxies. For example, postcode, employment gaps or language patterns may correlate with protected characteristics.

Fairness is not one universal mathematical condition. Organisations must define the relevant harm and legal or ethical requirement. Possible checks include comparing selection rates, error rates, calibration or outcomes across groups. These measures may conflict, particularly when groups have different base rates. Statistical testing can identify differences, but a non-significant result does not prove that a system is fair; small samples may lack power and legally significant harm may still exist.

Responsible practice includes impact assessment before deployment, representative validation, human review, an appeal route, documentation, monitoring and the ability to suspend the system.

6.3 Human oversight and automated decisions

Human oversight must be meaningful rather than symbolic. A manager who can only click “approve” on an algorithmic recommendation is not exercising independent judgement. Meaningful oversight requires access to relevant information, understanding of the system’s limitations, authority to disagree and responsibility for the decision.

High-impact decisions—such as hiring rejection, dismissal, promotion, compensation or immigration-related assignment—require stronger safeguards than low-risk administrative recommendations. Employees and candidates should receive understandable information about relevant automated processing, subject to applicable law and legitimate security constraints.

7. Privacy, Cybersecurity and International Governance

7.1 Privacy principles

International HR data governance should address purpose limitation, data minimisation, accuracy, retention, access, transparency, security and lawful transfer. The organisation should ask:

• Why is this data needed?
• Is the purpose specific and legitimate?
• Is the collection proportionate to that purpose?
• Who can access it, and why?
• How long will it be retained?
• Can the individual correct or challenge it?
• Where is it processed and transferred?
• What happens if the vendor or organisation suffers a breach?

Consent is not always a reliable basis in employment because of the power imbalance between employer and employee. Organisations must follow the applicable legal framework in each jurisdiction and obtain specialist advice where necessary. Cross-border transfers require particular attention to local requirements, contractual protections and vendor practices.

7.2 Cybersecurity

HR systems are attractive targets because they contain identity, compensation, health, performance and banking information. Controls should include:

• multi-factor authentication;
• least-privilege access;
• segregation of duties;
• encryption in transit and at rest;
• secure configuration and patching;
• logging and monitoring;
• tested backups and recovery;
• phishing and social-engineering training;
• vendor risk assessment; and
• an incident response plan.

Cybersecurity is not solely an information-technology responsibility. HR determines what data are collected, who needs access and how long records are retained. Managers and employees also influence risk through everyday behaviour.

7.3 Governance model

A practical AI and digital HR governance structure may include:

1. Inventory: record systems, data sources, owners, purposes and affected groups.
2. Risk classification: assess impact, sensitivity, scale and degree of automation.
3. Pre-deployment assessment: review legality, necessity, bias, security, accessibility and employee impact.
4. Documentation: maintain data sheets, model descriptions, versions, assumptions and decision rules.
5. Validation: test accuracy, robustness, subgroup performance and security.
6. Human accountability: assign a named owner with authority to intervene.
7. Transparency and contestability: provide explanations and routes for correction or appeal.
8. Monitoring: review drift, complaints, outcomes and unintended consequences.
9. Retirement: remove systems that are unsafe, ineffective or no longer necessary.

8. Illustrative Case: Global Talent Platform at Meridian Foods

Background

Meridian Foods, a fictional multinational with 18,000 employees in 12 countries, introduced a cloud talent platform. The platform integrated employee profiles, learning records, skills, performance ratings and internal vacancies. Its stated objective was to improve international mobility and identify future skills.

Problem

After one year, headquarters reported increased internal applications and faster vacancy filling. However, employees in two countries complained that the system recommended fewer development opportunities to them. Employee representatives also questioned whether performance comments were being analysed by a generative AI tool without adequate notice. The cybersecurity team discovered that many local HR administrators had broader access than necessary.

Analysis

The project team examined data definitions, recommendation logic, access permissions and outcomes by country, gender, employment status and language group. They found three problems:

1. Historical promotion data favoured employees who had worked in headquarters roles, so the model treated headquarters experience as a signal of potential.
2. Learning records were incomplete for frontline employees, causing the system to infer lower skill levels from missing data.
3. The AI text-analysis feature had been enabled globally even though local consultation and privacy review were incomplete.

The team also found that the apparent improvement in vacancy filling was partly due to a change in the reporting definition. The new measure counted a vacancy as filled when an offer was accepted, whereas the previous measure counted the employee’s start date.

Approach and solution

Meridian paused automated opportunity recommendations and used the platform only to display verified vacancies. It established a global data dictionary, while allowing country-specific legal and process fields. Missing data were reported as missing rather than treated as evidence of low capability. The company introduced role-based access, multi-factor authentication and quarterly access reviews.

An interdisciplinary governance group—including HR, legal, cybersecurity, data specialists, country representatives and employee representatives—approved a staged pilot. The pilot used human review, documented reasons for recommendations and an employee correction process. The generative AI feature was disabled until a separate assessment addressed purpose, retention, vendor access and transparency.

Outcome

After two quarters, internal mobility increased modestly, but more importantly, recommendation disparities narrowed and data-correction requests improved record quality. Managers reported that the system was less convenient than the original automated design, but employees expressed greater confidence in the process. Meridian retained analytics for workforce planning while prohibiting solely automated decisions for promotion or termination.

Learning points

• A global platform requires global data governance, not only global software.
• Missing data must not silently become negative evidence.
• Report definitions can create misleading claims of improvement.
• Human oversight, appeal rights and employee representation improve legitimacy.
• Pausing a system can be a sign of mature governance rather than technological failure.

9. Implementation Guidance for HR Professionals

A responsible implementation can follow these stages:

Stage 1: Clarify value. Define the business or employee problem and the decision to be improved.

Stage 2: Map stakeholders. Include country HR teams, employees, managers, works councils or representatives, legal specialists, IT, security and vendors.

Stage 3: Establish data foundations. Create common definitions, ownership, quality controls, retention rules and access roles.

Stage 4: Design for local legitimacy. Adapt language, accessibility, legal requirements and consultation processes without losing essential global comparability.

Stage 5: Pilot safely. Use a limited population, define success criteria, document risks and retain a manual alternative.

Stage 6: Evaluate outcomes. Examine effectiveness, cost, user experience, subgroup effects, security incidents and unintended consequences.

Stage 7: Scale with accountability. Train users, publish appropriate information, monitor performance and provide appeal mechanisms.

Employability in digital IHRM requires more than software familiarity. Graduates should be able to define a metric, question a data source, communicate uncertainty, identify discriminatory risk, collaborate with technical specialists and explain a digital HR decision to employees and managers.

Challenges and Limitations

Digital IHRM can be limited by poor data quality, incompatible country systems, weak connectivity, vendor lock-in, insufficient analytical skills and unclear ownership. Analytics may encourage an illusion of objectivity because numerical outputs appear neutral. Models can also drift as labour markets, organisational structures and employee behaviour change.

There are further limits to prediction. Human behaviour is context-dependent, and intervention can change the very outcome being predicted. If employees learn that a turnover model labels them as “high risk,” they may alter behaviour or lose trust. A model trained on one country or occupational group may not generalise to another. Finally, efficiency gains do not automatically equal better work. Digital tools should be judged by fairness, employee experience, organisational outcomes and respect for rights, not only by speed or cost.

Contemporary Developments and Best Practices

Contemporary digital IHRM is moving toward skills-based workforce planning, employee-facing generative AI, integrated talent marketplaces, process mining and continuous listening. These developments increase the importance of data provenance, model literacy and governance. Generative AI is especially sensitive because outputs may be inaccurate, fabricated, culturally inappropriate or based on confidential input.

Best practices include:

• use the least intrusive data capable of answering the question;
• separate exploration from high-impact decision-making;
• document definitions, assumptions and model limitations;
• test systems across countries and relevant employee groups;
• combine metrics with qualitative evidence;
• provide human review and meaningful appeal;
• protect confidential HR information from unauthorised AI tools;
• review vendors and subcontractors continuously;
• train HR professionals in data, privacy, security and algorithmic risk; and
• measure whether digitalisation improves fairness and employee experience, not only efficiency.

Chapter Summary

Digital IHRM uses HRIS, cloud platforms, employee-facing technologies, analytics and AI to coordinate and improve HR across national boundaries. Global systems require a common data and process core, but also local adaptations for law, language, culture and work arrangements. Digital employee experience depends on accessibility, simplicity, reliability, inclusion, privacy and access to human support.

People analytics progresses from descriptive reporting to diagnostic, predictive and prescriptive analysis. Metrics such as turnover, absence, selection rate and cost per hire are useful only when their definitions, populations, periods and assumptions are explicit. Dashboards should support decisions rather than encourage superficial ranking or surveillance.

AI can assist recruitment, learning, mobility, service delivery and workforce planning, but algorithmic outputs are not automatically objective. Bias may arise from historical data, proxies, missing values, labels and implementation. Responsible digital IHRM therefore requires privacy protection, cybersecurity, data-quality management, impact assessment, human accountability, transparency, appeal rights, subgroup monitoring and vendor governance. The professional role of HR is to connect technical capability with evidence, organisational purpose and respect for employee rights.

Key Terms

Artificial intelligence: Computational systems performing tasks such as prediction, classification, language processing or recommendation.

Cloud computing: Delivery of computing services, applications or infrastructure through a network, usually by an external provider.

Data minimisation: Collecting and retaining no more personal data than is necessary for a defined purpose.

Dashboard: A visual display of selected measures designed to support monitoring and decisions.

Digital employee experience: Employees’ total experience of digital tools, processes, information and support.

Global HRIS: An integrated information system supporting HR data and processes across countries.

Human oversight: Meaningful human ability and responsibility to review, challenge or override an automated output.

Machine learning: Methods that infer patterns from data to make predictions or classifications.

People analytics: Systematic use of workforce data and analysis to improve people-related decisions.

Precision: The proportion of positive predictions that are true positives.

Predictive analytics: Analysis estimating the likelihood of future events or outcomes.

Privacy: Appropriate control, protection and transparency concerning personal information.

Recall: The proportion of actual positive cases correctly identified.

Selection rate: The number hired divided by the number of applicants, expressed as a percentage.

Turnover rate: Separations during a period divided by average headcount, usually expressed as a percentage.

Vendor governance: Processes for evaluating, controlling and monitoring external technology providers.

Data drift: Change over time in the data or relationships on which a model depends.

Algorithmic bias: Systematic disadvantage or distortion arising from data, model design or use of an algorithm.

Review Questions

Short Answer Questions

1. What is the difference between a global HRIS and a local HR system?
2. Define digital employee experience and identify two of its dimensions.
3. Distinguish descriptive analytics from predictive analytics.
4. Why can accuracy be misleading when an outcome is rare?
5. What is meant by meaningful human oversight?

Long Answer Questions

1. Evaluate the advantages and risks of implementing a cloud-based HR platform across several countries.
2. Explain how an organisation can design a dashboard that supports evidence-based HR decisions.
3. Discuss the principal sources of bias in AI-supported recruitment or promotion.
4. Explain the relationship between privacy, cybersecurity and data governance in IHRM.
5. Develop a governance framework for introducing generative AI into employee services.

Analytical / Application Questions

1. A business begins a year with 800 employees, ends with 900 and records 102 separations. Calculate the approximate annual turnover rate and state one limitation of the calculation.
2. An AI screening tool produces 45 true positives, 15 false positives, 30 false negatives and 210 true negatives. Calculate precision and recall, and explain why both measures matter.
3. A global talent model recommends headquarters employees for international assignments at twice the rate of employees in regional offices. Design an investigation and propose at least four corrective safeguards.

Multiple Choice Questions with Answers

Question 1

Which feature best describes a global HRIS designed with a global template and local extensions?

A. Every country uses completely separate software and definitions.

B. All countries must use identical processes regardless of law.

C. Core data and processes are standardised while legally necessary local variations are configured.

D. Only headquarters employees can access the system.

Correct Answer: C. Core data and processes are standardised while legally necessary local variations are configured.

Question 2

A service centre begins a quarter with 480 employees, ends with 520, and has 36 separations. Using average beginning and ending headcount, what is the quarterly turnover rate?

A. 6.9%

B. 7.2%

C. 7.5%

D. 8.0%

Correct Answer: B. 7.2%

Question 3

Which type of people analytics asks, “What is likely to happen?”

A. Descriptive analytics

B. Diagnostic analytics

C. Predictive analytics

D. Prescriptive analytics

Correct Answer: C. Predictive analytics

Question 4

Which practice is most consistent with data minimisation?

A. Collecting all potentially useful employee data indefinitely

B. Collecting only data necessary for a defined and legitimate purpose

C. Allowing every manager to access all employee records

D. Retaining data because storage costs are low

Correct Answer: B. Collecting only data necessary for a defined and legitimate purpose

Question 5

If a classification model has 80 true positives and 20 false positives, what is its precision?

A. 20%

B. fifty percent

C. 80%

D. 100%

Correct Answer: C. 80%

Question 6

Why may accuracy be a poor evaluation measure for a rare outcome?

A. Accuracy cannot be calculated for classifications.

B. A model can appear accurate by predicting the majority class almost every time.

C. Accuracy always ignores true negatives.

D. Accuracy is measured only in financial units.

Correct Answer: B. A model can appear accurate by predicting the majority class almost every time.

Question 7

Which is the strongest example of meaningful human oversight?

A. A manager automatically accepts every algorithmic recommendation.

B. An employee is never told that automation was used.

C. A trained decision-maker can review evidence, challenge the output and explain the final decision.

D. A vendor alone determines whether the output is fair.

Correct Answer: C. A trained decision-maker can review evidence, challenge the output and explain the final decision.

Question 8

Which situation is a data-quality risk identified in the chapter’s case?

A. Missing learning records were treated as evidence of low capability.

B. Employees were allowed to correct inaccurate records.

C. Access permissions were reviewed quarterly.

D. Vacancies were displayed to employees.

Correct Answer: A. Missing learning records were treated as evidence of low capability.

Question 9

Which control most directly reduces the risk of unauthorised access to HR systems?

A. Removing audit logs

B. Least-privilege access and multi-factor authentication

C. Publishing employee salaries publicly

D. Allowing shared administrator accounts

Correct Answer: B. Least-privilege access and multi-factor authentication

Question 10

Why should a predictive turnover model not automatically be interpreted as causal evidence?

A. Predictions never use employee data.

B. A statistical association may reflect other factors and does not by itself show that one factor caused turnover.

C. Causation is guaranteed when a dashboard is used.

D. Predictive models can only analyse financial data.

Correct Answer: B. A statistical association may reflect other factors and does not by itself show that one factor caused turnover.