Industrial Relations Analytics and Evidence-Based Management

Introduction and Context

Industrial relations (IR) analytics is the systematic use of quantitative and qualitative evidence to understand relationships among employees, managers, trade unions, and employers, and to improve decisions concerning work, conflict, representation, and employment conditions. It extends conventional HR reporting. A report may state that absenteeism increased; analytics asks why it increased, whether the increase is concentrated in particular groups, what consequences it has, and which intervention is likely to reduce it without creating new risks.

Evidence-based management (EBM) complements analytics. It requires managers to integrate four forms of evidence: the best available research, organisational data, professional expertise, and the values and experiences of affected stakeholders. In industrial relations, the last category is especially important because formal indicators may conceal fear of retaliation, unequal voice, informal conflict, or deteriorating trust.

IR analytics must therefore be technically competent and institutionally sensitive. Employees are not merely observations in a database; they are rights-bearing participants in a relationship governed by law, collective agreements, organisational policies, and norms of fairness.

Learning Objectives

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

1. Explain the role and scope of analytics in industrial relations.
2. Construct and interpret indicators of grievances, absence, turnover, disputes, bargaining outcomes, engagement, and productivity.
3. Diagnose IR problems using triangulated quantitative and qualitative evidence.
4. Apply forecasting, evaluation, and causal reasoning appropriately.
5. Recognise validity, privacy, bias, and governance risks in IR data.
6. Recommend evidence-based interventions and communicate findings responsibly.

1. From HR Reporting to Industrial Relations Intelligence

A useful analytical cycle has six stages:

1. Frame the decision. Specify the managerial or employment-relations question, such as whether a new shift schedule is increasing grievances.
2. Construct the evidence base. Combine administrative records, collective-bargaining documents, surveys, interviews, focus groups, observation, and external benchmarks.
3. Assess data quality. Examine definitions, missingness, reliability, comparability, and possible selection effects.
4. Analyse patterns and mechanisms. Identify distributions, trends, differences, associations, and plausible causal pathways.
5. Evaluate action. Estimate whether an intervention changed outcomes relative to an appropriate comparison.
6. Govern and learn. Protect confidentiality, involve legitimate representatives, document assumptions, and revise the analysis as new evidence appears.

Analytics can be descriptive, diagnostic, predictive, or prescriptive. Descriptive analytics reports what happened. Diagnostic analytics investigates why. Predictive analytics estimates what may happen. Prescriptive analytics compares possible actions. The latter two should never be presented as certainty: forecasts are conditional on assumptions and available information.

The unit of analysis must be chosen carefully. A grievance may be recorded at employee, issue, incident, work-group, or case level. Counting cases rather than affected employees can exaggerate repeated complaints; counting employees can conceal the seriousness of a complex dispute. Analysts should define the numerator, denominator, time period, and population before calculating an indicator.

2. Core Indicators and Their Interpretation

2.1 Grievances and complaints

A grievance rate can be expressed as:

The result is grievances per 100 employees for the specified period. The denominator should match the population covered by the grievance system. A rate does not measure the underlying level of dissatisfaction directly. A rise may indicate worsening conditions, improved reporting confidence, a change in policy, or better access to representation. Useful supplementary measures include median resolution time, proportion resolved within the contractual deadline, recurrence rate, issue category, escalation rate, and employee perceptions of procedural fairness.

2.2 Absenteeism

A standard lost-time absence rate is:

“Available” should be defined consistently and should normally exclude approved leave categories that the organisation does not intend to treat as absence. Analysts should distinguish frequency from duration. Two units may have the same lost days, while one experiences many short absences and the other a few long-term cases. The Bradford factor, sometimes used to highlight repeated short absences, is:

where S is the number of absence spells and D is total days absent in the selected period. It is a screening device, not a measure of employee culpability or health. Its use can disadvantage employees with disabilities or recurring medical conditions and should never replace reasonable accommodation, medical confidentiality, or contextual review.

2.3 Turnover and retention

A conventional turnover rate is:

Voluntary, involuntary, retirement, contract-end, and regrettable turnover should be reported separately. Survival analysis is often more informative than a single annual rate because it examines the probability that employment continues over time. A simple retention rate is:

Cohort definitions are essential. Comparing a newly recruited cohort with a long-established workforce can produce misleading conclusions.

2.4 Disputes and industrial action

Useful indicators include disputes per 1,000 employees, working days lost, duration, settlement route, issues in dispute, and recurrence. Working days lost may be calculated as the number of participating employees multiplied by working days not worked, adjusted for partial participation. Comparisons require attention to workforce size, industry seasonality, legal restrictions on action, bargaining cycles, and whether unofficial disputes are captured.

A low strike rate is not automatically evidence of healthy relations. Conflict may be suppressed, channelled into litigation, expressed through presenteeism, or absent because workers lack effective voice. Qualitative evidence is therefore indispensable.

2.5 Bargaining outcomes

Collective bargaining analytics should examine both substantive and process outcomes. Substantive outcomes include wage growth, benefit changes, working-time provisions, job-security clauses, and flexibility. Process outcomes include bargaining duration, number of negotiation rounds, mediation use, ratification margin, grievances arising from ambiguous clauses, and implementation delays.

A nominal wage increase should not be confused with a real wage increase. If wages rise by w percent and consumer prices rise by p percent, the exact real wage change is:

Here w and p are expressed as decimals. The approximation w − p is reasonable only for small percentages. For example, a 6% wage increase and 4% inflation produce (1.06 ÷ 1.04 − 1) × 100 ≈ 1.92%, not exactly 2%.

2.6 Engagement, trust, and voice

Engagement surveys may measure commitment, discretionary effort, supervisor support, perceived fairness, psychological safety, union-management trust, and confidence in speaking up. A mean score can conceal polarisation, non-response bias, or small but vulnerable subgroups. Report response rates, item wording, scale properties, and confidence intervals where appropriate. Open-ended comments and representative interviews help explain why scores change.

Engagement is not equivalent to compliance. Employees can be highly engaged in their work while distrusting management, or satisfied with pay while feeling excluded from decision-making. IR analysis should separate constructs rather than collapse them into a single index without validation.

2.7 Productivity and quality

A basic labour productivity measure is:

Output may be units, cases, or value added; labour input may be hours worked or full-time-equivalent employment. Units must be defined and quality must be incorporated. Increasing output per hour through excessive work intensity, unpaid overtime, safety compromises, or rework is not sustainable productivity. A balanced dashboard should include quality, safety, customer outcomes, absence, working time, and employee voice.

3. Diagnosing Problems Through Triangulation

Triangulation means comparing different data sources, methods, investigators, or theoretical perspectives. It does not mean that the largest dataset automatically determines the truth. For example, a rise in absence may coincide with a new attendance policy. Administrative data shows timing; interviews may reveal that employees fear disciplinary action for reporting stress; workload data may show understaffing; and union records may document unresolved scheduling grievances.

A practical diagnostic matrix is useful:

Analysts should distinguish correlation from causation. A department with high grievances and low productivity may have poor supervision, but the association could also reflect difficult work, a recent restructuring, or better grievance access. Causal claims require a credible counterfactual: what would have happened without the intervention?

Confounding occurs when a third factor affects both the presumed cause and outcome. For example, a new technology may be introduced first in the most overloaded departments; their subsequent absence cannot be attributed solely to the technology. Stratification, regression, matching, interrupted time-series analysis, and controlled pilots can improve inference, but none eliminates the need for institutional knowledge.

4. Forecasting and Risk Assessment

Forecasting can support early intervention, but risk scores should identify conditions requiring review, not determine employment decisions automatically. A basic time-series model might represent an outcome as:

where y_t is the outcome at time t, α is the intercept, β is the estimated average trend per period, and ε_t is unexplained variation. This linear trend assumes that the relationship is approximately stable over the forecast horizon and that residuals do not contain serious autocorrelation or structural breaks. Those assumptions often fail around bargaining rounds, restructures, pandemics, or legal changes.

More advanced models may estimate the probability of an event, such as voluntary exit, using logistic regression:

Here Y = 1 denotes the event, X₁ … Xₖ are predictors, β₁ … βₖ are estimated coefficients, and e is the base of natural logarithms. The model estimates association conditional on included variables; it does not prove that changing a predictor will cause the outcome to change. Calibration, discrimination, subgroup performance, and practical consequences must be assessed before use.

Fairness requires checking whether data reflect unequal monitoring or historical discrimination. A model trained on past disciplinary actions may reproduce biased managerial practices. Sensitive attributes may be excluded from a model yet reconstructed through proxies such as location, shift, tenure, or job title. Governance should prohibit automated dismissal, discipline, or anti-union targeting based solely on a risk score.

5. Evaluating Interventions

An intervention might involve supervisor training, a revised grievance procedure, joint union-management problem-solving, workload redesign, or a new scheduling arrangement. Evaluation should begin with a theory of change:

Intervention → mechanism → intermediate outcome → final outcome.

For example, joint scheduling committees may improve perceived voice, which reduces schedule-related grievances, which may improve attendance and retention.

A before-and-after comparison is informative but weak because other events may explain the change. A difference-in-differences design compares changes in an intervention group with changes in a comparison group:

where Y₁ is the intervention group outcome and Y₀ is the comparison group outcome. The key assumption is parallel trends: absent the intervention, both groups would have followed similar trends. Analysts should inspect pre-intervention trends and report uncertainty, missing data, spillovers, and implementation fidelity.

Statistical significance is not the same as practical importance. A small change can be statistically significant in a very large workforce, while a meaningful change may not reach significance in a small sample. Report effect size, confidence intervals, operational implications, and distributional effects across groups.

6. Privacy, Validity, Bias, and Governance

IR data often include health information, union membership or activity, complaints, disciplinary records, and inferred attitudes. Organisations should apply purpose limitation, data minimisation, access control, retention limits, secure storage, and lawful processing consistent with applicable data-protection and labour law. Collective agreements and consultation duties may impose additional requirements.

Anonymity means identity cannot reasonably be established; confidentiality means identity is protected by controls but may be known to authorised personnel. Small-cell reporting can re-identify individuals, especially when demographic, shift, and job information are combined. Suppression or aggregation is necessary where disclosure risk is material.

Validity has several dimensions:

• Construct validity: Does the measure represent the intended concept? A “culture score” may not validly measure trust.
• Content validity: Does it cover relevant dimensions, including voice and fairness?
• Criterion validity: Does it relate appropriately to an external outcome?
• Reliability: Would the measure produce consistent results under similar conditions?
• Internal validity: Is the estimated intervention effect credible?
• External validity: Can findings transfer to another site, occupation, or bargaining regime?

Common biases include survivorship bias, non-response bias, social desirability bias, measurement error, omitted-variable bias, algorithmic bias, and Simpson’s paradox, in which an aggregate pattern reverses within relevant subgroups. Analysts should predefine key measures where feasible, maintain a data dictionary, record transformations, conduct sensitivity analyses, and invite union and employee representatives to challenge interpretations.

7. Technology and AI in IR Analytics

Modern platforms can integrate case-management systems, workforce schedules, survey text, payroll data, and operational measures. Natural-language processing can classify themes in grievance narratives, while dashboards can display trends. These tools are useful only when human review remains meaningful.

Text classification can misinterpret dialect, sarcasm, multilingual expression, or legally significant nuance. Generative AI may produce plausible but unsupported summaries and can expose confidential information if data are sent to an unauthorised service. Approved systems should use access controls, audit logs, documented model cards, human verification, bias testing, and clear escalation procedures. Employees should know when automated tools influence decisions and have a route to challenge inaccurate data or conclusions.

8. Illustrative Case: The Northbridge Components Agreement

Background. Northbridge Components, a fictional manufacturer with 1,200 employees, introduced rotating weekend shifts six months before collective bargaining. Grievances rose from 24 to 48 per quarter, voluntary turnover increased from 8% to 12% annually, and lost-time absence rose from 3.2% to 4.1%.

Problem. Senior management proposed attendance sanctions and argued that the figures demonstrated declining commitment. The union argued that unpredictable schedules and inadequate rest were the primary causes.

Analysis. The HR analytics team first verified definitions and found that the grievance count referred to cases, while one employee could open several cases. It segmented the data by plant, shift, tenure, and supervisor. The increase was concentrated among rotating-weekend employees and was greatest where schedules were published less than seven days in advance. A survey showed lower schedule predictability and trust in those units. Interviews identified childcare difficulties, fatigue, and perceptions that grievance meetings were slow. Productivity per labour hour had risen slightly, but quality rework and overtime had also increased.

The team avoided claiming that scheduling caused every absence. It compared pre-change trends with a similar plant that retained fixed shifts and found a larger post-change increase at Northbridge. Because the plants differed in workload and workforce composition, the comparison was treated as supportive rather than definitive causal evidence.

Approach. Management and the union agreed to a three-month pilot: schedules would normally be issued 14 days ahead; employees could exchange shifts through a jointly governed process; supervisors received training; and a rapid grievance panel reviewed scheduling cases within five working days. Measures included schedule notice, grievances per 100 employees, absence rate, overtime, quality rework, turnover intentions, and perceived fairness. Data were reported in aggregated form.

Outcome. Schedule-related grievances fell by 35% in the pilot units, the absence rate declined to 3.6%, and survey confidence in voice improved. Overtime declined modestly, while quality remained stable. The evaluation could not establish that all improvements resulted from the pilot because the bargaining climate also improved. Nevertheless, the convergence of administrative, survey, interview, and operational evidence supported continuation with further monitoring.

Learning points. The case illustrates the importance of valid denominators, subgroup analysis, qualitative explanation, joint governance, balanced productivity measures, and cautious causal language. Sanctions would have treated a symptom while potentially worsening trust.

Managerial and Professional Implications

IR analysts should act as translators between data and institutional decision-making. Their recommendations should state the decision, evidence, assumptions, uncertainty, affected groups, legal constraints, and proposed review date. Managers should not ask only whether an indicator improved; they should ask who benefited, who bore costs, whether voice was strengthened, and whether the intervention remains legitimate under the collective agreement.

Best practice includes:

• define indicators jointly with relevant stakeholders;
• maintain consistent denominators and a documented data dictionary;
• combine quantitative and qualitative evidence;
• disaggregate where it improves diagnosis, while protecting privacy;
• distinguish monitoring from causal evaluation;
• test models for calibration, validity, and disparate impact;
• avoid automated high-stakes decisions;
• communicate uncertainty plainly;
• preserve employee and union voice throughout the analytical cycle; and
• review interventions for unintended consequences.

Chapter Summary

Industrial relations analytics applies disciplined quantitative and qualitative methods to employment relationships, conflict, representation, bargaining, and work outcomes. Indicators of grievances, absence, turnover, disputes, bargaining, engagement, and productivity are useful only when their definitions, denominators, time periods, and limitations are explicit. Diagnosis requires triangulation because administrative patterns rarely explain themselves. Forecasting can support prevention but is conditional, susceptible to bias, and inappropriate as an automatic basis for discipline. Intervention evaluation is stronger when it uses a credible comparison and states assumptions such as parallel trends. Evidence-based management combines research, organisational data, professional judgement, and stakeholder values. Privacy, validity, fairness, transparency, and worker voice are not peripheral concerns; they are conditions of legitimate IR analytics.

Key Terms

Industrial relations analytics: Systematic analysis of evidence about employment relationships and workplace conflict.

Evidence-based management: Decision-making that integrates research, organisational data, expertise, and stakeholder values.

Grievance rate: Grievances relative to a defined employee population over a stated period.

Absence rate: Scheduled workdays lost as a proportion of scheduled workdays available.

Turnover rate: Separations relative to average headcount over a stated period.

Retention rate: The proportion of an initial cohort remaining employed at a specified date.

Working days lost: Labour time not worked because of industrial dispute participation.

Engagement: A multidimensional construct involving connection, energy, commitment, and involvement in work.

Labour productivity: Output produced per unit of labour input.

Triangulation: Comparing multiple evidence sources or methods to improve interpretation.

Construct validity: The extent to which a measure represents the intended concept.

Reliability: Consistency of measurement under comparable conditions.

Confounding: Distortion caused by a factor related to both an apparent cause and outcome.

Counterfactual: The outcome that would have occurred without an intervention.

Difference-in-differences: An evaluation method comparing changes over time between intervention and comparison groups.

Calibration: Agreement between predicted probabilities and observed event frequencies.

Data minimisation: Collecting and retaining only data necessary for a legitimate purpose.

Algorithmic bias: Systematic disadvantage produced or reproduced by an analytical system.

Procedural fairness: Perceived and actual fairness of processes used to make decisions.

Theory of change: A reasoned sequence linking intervention, mechanism, intermediate outcomes, and final outcomes.

Review Questions

Short Answer Questions

1. Distinguish descriptive, diagnostic, predictive, and prescriptive analytics.
2. Why should grievance counts be accompanied by a clearly defined denominator?
3. Explain the difference between anonymity and confidentiality.
4. What is triangulation in industrial relations analytics?
5. Why is productivity per labour hour alone an incomplete performance measure?

Long Answer Questions

1. Explain how an organisation could use quantitative and qualitative evidence to diagnose rising absenteeism.
2. Critically assess the use of predictive turnover models in employment decision-making.
3. Discuss the validity and ethical risks of using employee engagement surveys.
4. Explain the assumptions and practical value of difference-in-differences evaluation.
5. Develop a governance framework for analytics involving union and employee data.

Analytical / Application Questions

1. An organisation has 800 employees, 36 grievances in a quarter, and an average headcount of 750 during that quarter. Calculate the grievance rate per 100 employees and identify two reasons why the rate alone may be misleading.
2. A pilot unit’s absence rate falls from 5.0% to 4.0%, while a comparison unit’s rate falls from 4.0% to 3.5%. Estimate the difference-in-differences effect and explain the parallel-trends assumption.
3. Design an evidence-based evaluation of a new joint union-management scheduling committee. Specify outcomes, comparison strategy, qualitative evidence, privacy safeguards, and possible unintended consequences.

Multiple Choice Questions with Answers

Question 1

Which statement best describes evidence-based management in industrial relations?

A. Using only the largest available dataset

B. Combining research, organisational data, professional expertise, and stakeholder values

C. Allowing algorithms to make employment decisions

D. Replacing collective bargaining with statistical forecasting

Correct Answer: B. Combining research, organisational data, professional expertise, and stakeholder values

Question 2

A grievance rate is calculated by dividing grievances by which denominator?

A. Total payroll cost

B. Average or otherwise appropriately defined employee population

C. Number of supervisors only

D. Total working days lost

Correct Answer: B. Average or otherwise appropriately defined employee population

Question 3

What is a major limitation of interpreting a rise in grievances as proof that conditions worsened?

A. Grievances can also rise because reporting access or confidence improved

B. Grievances are never measurable

C. Grievances always represent identical issues

D. The denominator is irrelevant

Correct Answer: A. Grievances can also rise because reporting access or confidence improved

Question 4

If wages increase by 6% and prices increase by 4%, the exact real wage change is approximately:

A. 0.00%

B. 1.92%

C. 2.50%

D. 10.00%

Correct Answer: B. 1.92%

Question 5

The key identifying assumption in a basic difference-in-differences design is:

A. The intervention group must be larger

B. Both groups must have identical outcomes at baseline

C. Parallel trends absent the intervention

D. There can be no qualitative evidence

Correct Answer: C. Parallel trends absent the intervention

Question 6

Which measure is most directly concerned with the consistency of a measurement instrument?

A. Reliability

B. External validity

C. Calibration

D. Counterfactual validity

Correct Answer: A. Reliability

Question 7

Why should a turnover prediction model not automatically trigger discipline?

A. Predictions are always random

B. Models estimate conditional risk and may reproduce historical bias

C. Turnover has no organisational consequences

D. All employee data are public

Correct Answer: B. Models estimate conditional risk and may reproduce historical bias

Question 8

Which is an example of triangulation?

A. Using only a manager’s opinion

B. Combining absence records, survey findings, interviews, and scheduling data

C. Deleting contradictory evidence

D. Reporting only an annual average

Correct Answer: B. Combining absence records, survey findings, interviews, and scheduling data

Question 9

Which practice best protects privacy in a small work group?

A. Publishing individual-level case histories

B. Removing all access controls

C. Aggregating or suppressing small cells where re-identification is possible

D. Sharing raw union records widely

Correct Answer: C. Aggregating or suppressing small cells where re-identification is possible

Question 10

Why should productivity analytics include quality and safety measures?

A. Output per hour always measures employee wellbeing

B. Higher output can be accompanied by rework, unsafe intensity, or unsustainable workload

C. Quality and safety are unrelated to productivity

D. Productivity cannot be measured quantitatively

Correct Answer: B. Higher output can be accompanied by rework, unsafe intensity, or unsustainable workload