Interpretation, Decision-Making and Managerial Recommendations

Introduction and context

A dissertation does not end when data analysis is complete. Statistical tables, interview themes, or evaluation results become academically and managerially valuable only when they are interpreted in relation to the research question, theory, prior evidence, organisational context, and practical decisions. Interpretation explains what the findings mean; decision-making determines what should be done; managerial recommendations translate justified conclusions into feasible action.

In human resource management (HRM), this final stage is especially important because HR decisions affect employees, managers, organisational performance, equity, privacy, and legal compliance. A statistically significant relationship between training and performance does not automatically justify a large training budget. A non-significant result does not prove that an intervention has no value. Sound dissertation work therefore distinguishes evidence from inference, association from causation, statistical significance from practical significance, and desirable action from merely plausible action.

This chapter presents a structured approach for moving from results to defensible HR recommendations. It integrates theory and previous research, evaluates limitations and alternative explanations, and develops recommendations with implementation, governance, and evaluation plans.

Learning objectives

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

1. Interpret quantitative and qualitative findings in relation to research questions, theory, and prior evidence.
2. Distinguish statistical, practical, and managerial significance.
3. Evaluate limitations, bias, validity, reliability, and alternative explanations.
4. Make transparent decisions under uncertainty.
5. Develop ethical, feasible, measurable, and prioritised HR recommendations.
6. Construct implementation and evaluation plans using appropriate indicators.
7. Calculate and interpret HR investment returns without confusing costs, benefits, and net benefits.

From results to interpretation

Description, interpretation, and implication

These three activities should be kept separate:

A result should first be reported accurately, then interpreted cautiously, and only then connected to a recommendation. Overclaiming occurs when researchers move directly from an observed association to a causal managerial prescription.

A disciplined interpretation sequence

A useful sequence is:

1. Return to the research question. Identify exactly which question the result addresses.
2. State the principal finding. Report direction, size, uncertainty, and relevant conditions.
3. Connect with theory. Explain whether the finding supports, extends, qualifies, or challenges the proposed mechanism.
4. Compare with prior evidence. Identify convergence, divergence, and possible reasons for differences.
5. Assess credibility. Examine design, measurement, sampling, missing data, and competing explanations.
6. Assess importance. Ask whether the effect is large enough to matter to employees or the organisation.
7. Derive a proportionate implication. Recommend action that matches the strength of evidence.

Integrating theory and prior evidence

Theory provides an explanatory lens rather than a guarantee. For example, social exchange theory suggests that employees may reciprocate perceived organisational support through commitment and extra effort. If a recognition programme is associated with higher commitment, the result may be consistent with this theory. However, the finding may also reflect selection: employees who already feel committed may participate more actively in recognition activities.

A strong discussion compares the present study with prior evidence across four dimensions:

• Direction: Do the findings point in the same direction?
• Magnitude: Is the observed association larger or smaller?
• Context: Does the setting differ by industry, occupation, country, or workforce composition?
• Method: Were different measures, samples, time periods, or research designs used?

Contradictory findings are not necessarily failures. They may reveal boundary conditions. For example, flexible work may improve wellbeing for employees with autonomy and suitable technology but increase strain where workloads and communication expectations remain uncontrolled.

Understanding significance

Statistical significance

Statistical significance concerns compatibility between the observed data and a specified null hypothesis, given the assumptions of the statistical model. A p-value is not the probability that the research hypothesis is true, nor is it a measure of effect size or managerial importance.

Confidence intervals provide more information than a significance label because they show a range of values reasonably compatible with the data. For a difference in group means, a confidence interval that is narrow and close to zero suggests a precise, small difference; a wide interval indicates considerable uncertainty.

Statistical conclusions depend on sample size, measurement quality, model specification, and assumptions. A very large sample may make a trivial effect statistically significant, while a small sample may fail to detect a practically important effect.

Practical significance

Practical significance asks whether the size of a finding matters in real conditions. Useful measures include:

• percentage-point differences;
• standardised mean differences;
• correlation or regression coefficients;
• predicted probabilities;
• changes in absence or turnover rates;
• confidence intervals;
• minimum practically important differences.

For a two-group mean comparison, a commonly used standardised difference is Cohen’s d:

Here, M₁ and M₂ are group means and sₚ is the pooled standard deviation. This measure assumes that standard deviations are reasonably comparable and that the outcome is measured on an appropriate scale. It is a descriptive effect-size index, not proof of causality.

For groups with sample sizes n₁ and n₂ and standard deviations s₁ and s₂, the pooled standard deviation is:

Managerial significance

Managerial significance concerns the decision consequences of a finding. A modest effect may matter greatly when the affected population is large, the intervention is inexpensive, or the issue concerns safety, discrimination, or legal compliance. Conversely, a large effect may not justify action if implementation is prohibitively costly or evidence quality is weak.

A useful decision frame considers:

This is a decision concept rather than a universal accounting identity. Benefits and costs should be defined consistently, and non-financial consequences should not be disguised as monetary precision.

Worked example: interpreting an effect

An organisation compares mean engagement scores between employees receiving structured manager feedback and those receiving usual supervision. The intervention mean is 4.10, the comparison mean is 3.80, and the pooled standard deviation is 0.75 on a five-point scale.

The difference is 0.30 scale points and the standardised difference is 0.40. This may be practically meaningful, but interpretation still requires the confidence interval, sample design, baseline differences, response rates, and whether engagement improvement leads to outcomes such as retention or performance. The result supports further testing; it does not by itself establish that feedback caused the improvement.

Evaluating limitations and alternative explanations

Limitations are not an apology section. They define how far the conclusions can travel. A credible dissertation identifies limitations before presenting recommendations.

Common validity threats in HR research

Causal claims are strongest when temporal order, covariation, and credible exclusion of alternative explanations are present. Random assignment supports causal inference under compliance and measurement assumptions, but even experiments may have spillovers, attrition, or limited generalisability. Cross-sectional studies generally establish association and perceived experience, not temporal causation.

Alternative explanations

Before recommending action, ask:

• Could a policy change, leadership transition, seasonal cycle, or economic event explain the result?
• Did the intervention group differ in workload, pay, location, or employee composition?
• Could employees with favourable attitudes have been more likely to participate?
• Could survey wording or manager enthusiasm have changed responses?
• Is reverse causality plausible—for example, high-performing teams receiving more development resources?

Sensitivity analysis can test whether conclusions change under alternative specifications, exclusion of influential cases, or different definitions of the outcome. Qualitative evidence can help explain mechanisms but should not be used selectively to confirm a preferred numerical result.

Decision-making from evidence

Evidence-informed HR decision-making

Evidence-informed practice combines four sources:

1. the best available research evidence;
2. internal organisational data;
3. professional expertise and operational knowledge; and
4. the values, needs, and rights of affected stakeholders.

A decision matrix can make trade-offs explicit. Each option may be scored against criteria such as expected impact, evidence confidence, cost, feasibility, equity, legal risk, and reversibility. Scores should not create false objectivity; the assumptions and weights must be documented.

Choosing among action types

Evidence does not always imply immediate full implementation. Appropriate options include:

• Adopt: evidence and feasibility are strong.
• Pilot: potential value is promising but uncertainty remains.
• Adapt: the intervention is modified for local context.
• Monitor: evidence is insufficient for intervention but the issue matters.
• Do not proceed: expected harm, cost, or legal risk outweighs likely benefit.

A pilot should have a defined population, duration, comparison strategy, success criteria, stopping rules, and ethical safeguards. “Pilot” must not become a way to avoid accountability.

Developing high-quality managerial recommendations

A recommendation should be specific enough to act upon and cautious enough to reflect evidence. A useful structure is:

Recommendation = action + owner + target population + timing + resources + success measure + safeguard.

For example: “The HR director should pilot structured monthly career conversations in two departments for six months, train participating managers, protect confidentiality, and assess improvement in career clarity, internal mobility, and employee experience against two comparison departments.”

Criteria for recommendations

Recommendations should be:

• Evidence-based: linked to a finding and its confidence level.
• Aligned: connected to organisational strategy and workforce needs.
• Feasible: realistic given budget, capability, technology, and time.
• Ethical: respectful of dignity, autonomy, privacy, and fairness.
• Inclusive: assessed for differential effects across groups.
• Measurable: associated with defined indicators and baselines.
• Prioritised: sequenced rather than presented as an unmanageable list.
• Reversible where possible: tested before irreversible commitments.

Financial appraisal and correct ROI terminology

Financial appraisal should distinguish investment cost, total benefit, net benefit, and return on investment. If net benefit already means total benefits minus investment cost, the correct ROI formula is:

where:

Equivalently, if the numerator is written directly:

Investment cost and total benefit must use the same currency and time basis. ROI is not the same as a benefit–cost ratio, which is commonly defined as total benefit divided by cost. Intangible benefits, opportunity costs, discounting, and uncertainty should be reported rather than hidden.

Worked example. A leadership programme costs £40,000 and produces estimated measurable benefits of £58,000 during the evaluation period.

The result means that net benefit equals 45% of the investment cost under the stated assumptions. It does not mean the organisation receives £45 for every £100 of total benefit; that interpretation would confuse ROI with a benefit–cost ratio. Estimates should be accompanied by assumptions and, where possible, sensitivity scenarios.

Implementation and evaluation planning

A recommendation becomes operational through an implementation plan.

Logic models

A logic model links inputs to outcomes:

Inputs → Activities → Outputs → Short-term outcomes → Longer-term outcomes

For a mentoring intervention, mentors and training are inputs; mentoring meetings are activities; completed meetings are outputs; improved career clarity is a short-term outcome; and reduced regretted turnover may be a longer-term outcome. The model prevents the common error of treating activity completion as proof of organisational impact.

Evaluation design

Evaluation should include baseline measurement and a specified follow-up point. Possible designs include:

• Before-and-after: simple but vulnerable to history and maturation.
• Comparison group: stronger when groups are reasonably comparable.
• Difference-in-differences: compares change over time in intervention and comparison groups. Its key assumption is that, without intervention, both groups would have followed approximately parallel trends.
• Randomised trial: strong internal validity when ethically and operationally feasible.
• Process evaluation: examines reach, fidelity, participation, and implementation quality.
• Outcome evaluation: examines employee and organisational effects.

For difference-in-differences, the estimated intervention effect is:

Y₁ refers to the intervention group and Y₀ to the comparison group. The formula is appropriate when the outcome is measured comparably across groups and the parallel-trends assumption is plausible. It does not automatically remove all confounding.

Evaluation indicators should include leading and lagging measures. Participation and completion are leading indicators; retention, absence, productivity, and employee wellbeing may be lagging indicators. Disaggregating results by relevant demographic or employment groups can reveal unequal benefits or harms, while protecting confidentiality when subgroup numbers are small.

Ethics, governance, and technology

Recommendations must comply with applicable employment, equality, health and safety, data-protection, and consultation requirements. Ethical HR decision-making requires informed communication, proportional data collection, secure access, retention limits, and a clear explanation of how data will be used.

AI-enabled analytics may identify patterns in turnover, recruitment, or performance data, but prediction is not explanation and correlation is not causation. Historical HR data can reproduce discriminatory patterns. Before deployment, organisations should document data provenance, test for differential error, establish human oversight, provide review or appeal routes, and avoid using sensitive attributes or their proxies irresponsibly. Employees should not be subjected to opaque automated decisions without accountability.

A recommendation that improves average performance but disadvantages a protected group may be unacceptable. Equity analysis should therefore be part of the decision, not an optional later audit.

Illustrative case: the retention analytics decision

Background

This fictional case concerns Northbridge Services, a 1,200-employee customer-support organisation. Annual turnover among employees with less than two years’ service is 24%. HR introduces a structured six-month career-conversation programme in three branches. Three similar branches continue usual practice.

Problem and evidence

After six months, turnover is 18% in the programme branches and 22% in comparison branches. The programme branches also had slightly higher starting pay and a new regional manager. Survey response was 61% in the intervention branches and 48% in comparison branches. Interviews suggest that employees valued clearer development information, but some managers reported inadequate time.

The raw difference is four percentage points. It is promising, but the pay difference, leadership change, response imbalance, and short follow-up weaken a causal conclusion. The intervention may have contributed, but the evidence does not isolate its effect.

Decision and approach

The HR committee chooses an expanded but controlled pilot rather than immediate organisation-wide adoption. It will:

1. standardise the conversation guide;
2. train managers and provide protected time;
3. include matched comparison branches;
4. record baseline and quarterly outcomes;
5. monitor participation and employee experience;
6. analyse retention by relevant workforce groups;
7. publish a plain-language evaluation summary to stakeholders.

Success criteria are a sustained reduction in regretted early turnover, improved career clarity, no material widening of subgroup disparities, and acceptable manager workload. If participation is low or harms emerge, the programme will be adapted or paused.

Outcome and learning points

After a further six months, the programme shows improved career-clarity scores, but turnover reduction is uncertain because a major competitor opened nearby. The organisation continues the programme with revised workload arrangements while collecting longer-term evidence.

The case demonstrates that a responsible recommendation can be neither unconditional adoption nor rejection. It can be a monitored, ethical, theory-informed pilot that recognises uncertainty and creates better evidence.

Challenges and best practices

Common errors include treating p < 0.05 as proof of importance, presenting recommendations unrelated to findings, ignoring null results, using ROI with unverified benefit estimates, confusing outputs with outcomes, and concealing limitations. Best practice is to maintain an evidence-to-action table linking each recommendation to the result, confidence, theory, owner, cost, risk, indicator, and review date.

Researchers should also distinguish dissertation conclusions from organisational policy. A study may justify further investigation without justifying a permanent policy. Recommendations should identify what is known, what is uncertain, and what evidence will be collected next.

Chapter Summary

Interpretation is the disciplined movement from findings to meaning, not a repetition of results. Strong interpretation reconnects evidence with research questions, theory, prior studies, organisational context, and competing explanations. Statistical significance must be distinguished from effect size, practical importance, and managerial significance. Limitations concerning sampling, measurement, design, missing data, confounding, and generalisability determine how confidently findings can support action.

Evidence-informed HR decisions combine research, internal data, professional expertise, and stakeholder values. Recommendations should be specific, feasible, ethical, inclusive, measurable, prioritised, and proportionate to evidence strength. Implementation plans identify activities, ownership, resources, risks, safeguards, and timelines. Evaluation plans use appropriate baselines, comparison strategies, process and outcome indicators, and explicit assumptions.

Financial claims require precise terminology: net benefit equals total benefit minus investment cost, and ROI equals net benefit divided by investment cost, multiplied by 100. Finally, HR recommendations must protect privacy, fairness, dignity, and accountability, particularly when AI or employee analytics are used.

Key Terms

Interpretation: Explanation of what research findings mean in context.

Statistical significance: Evidence judged relative to a specified statistical null hypothesis and model assumptions.

Practical significance: The real-world importance or size of a finding.

Managerial significance: The relevance of a finding to organisational decisions and consequences.

Effect size: A quantitative description of the magnitude of a difference or relationship.

Confidence interval: A range expressing uncertainty around an estimated quantity.

Confounding: A situation in which another variable influences both the predictor and outcome.

Selection bias: Systematic differences between participants and non-participants or comparison groups.

Common-method bias: Bias arising when variables are measured using the same source or method.

Causal inference: Reasoning about whether an exposure or intervention produced an outcome.

Alternative explanation: A plausible account of a finding other than the proposed explanation.

Evidence-informed HRM: HR practice combining research evidence, organisational data, expertise, and stakeholder values.

Pilot: A limited implementation used to test feasibility and effects before wider adoption.

Logic model: A framework linking inputs, activities, outputs, and outcomes.

Leading indicator: An early measure expected to precede a later result.

Lagging indicator: A result measure observed after activities or interventions.

Net benefit: Total benefit minus investment cost.

Return on investment (ROI): Net benefit divided by investment cost, expressed as a percentage.

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

Algorithmic governance: Processes for transparency, oversight, fairness, accountability, and control of automated systems.

Review Questions

Short Answer Questions

1. Distinguish description, interpretation, and implication in dissertation writing.
2. Why is statistical significance not sufficient for an HR recommendation?
3. State two alternative explanations for an apparent intervention effect.
4. Define net benefit and ROI.
5. What is the purpose of a logic model?

Long Answer Questions

1. Explain how a researcher can integrate findings with theory and prior evidence without overstating causality.
2. Discuss the differences between statistical, practical, and managerial significance, using HR examples.
3. Evaluate the main validity threats that can affect an HR intervention study.
4. Explain how ethical and governance considerations should shape people-analytics recommendations.
5. Describe the components of a feasible implementation and evaluation plan.

Analytical / Application Questions

1. An intervention reduces absence from 8.0 to 7.2 days per employee, but the intervention group had a new manager and lower workloads. How would you assess the evidence before recommending adoption?
2. A programme costs $25,000 and produces estimated total benefits of $34,000. Calculate net benefit and ROI, and identify two assumptions that should be tested.
3. Design a six-month evaluation for a flexible-working policy, including a comparison strategy, three indicators, one ethical safeguard, and a possible alternative explanation.

Multiple Choice Questions with Answers

Question 1

Which statement best distinguishes practical significance from statistical significance?

A. Practical significance concerns whether the effect matters in real conditions; statistical significance concerns compatibility with a null model.

B. Practical significance is determined only by the p-value.

C. Statistical significance proves that an intervention caused the outcome.

D. The two terms have identical meanings.

Correct Answer: A. Practical significance concerns whether the effect matters in real conditions; statistical significance concerns compatibility with a null model.

Question 2

An intervention costs £40,000 and produces total benefits of £58,000. What is the ROI?

A. 31.0%

B. 45.0%

C. 68.9%

D. 145.0%

Correct Answer: B. 45.0%

Question 3

Which is an alternative explanation for lower turnover after an HR intervention?

A. The intervention was described in the report.

B. The outcome was measured in percentages.

C. A new manager or pay increase affected the intervention group.

D. The study had a research question.

Correct Answer: C. A new manager or pay increase affected the intervention group.

Question 4

What does a logic model primarily connect?

A. Inputs, activities, outputs, and outcomes

B. P-values, confidence intervals, and sample sizes only

C. Job titles and salary grades only

D. Legal rules and disciplinary penalties only

Correct Answer: A. Inputs, activities, outputs, and outcomes

Question 5

Which condition is important for a difference-in-differences interpretation?

A. The intervention group must have no baseline measurement.

B. The groups must have identical sample sizes.

C. The parallel-trends assumption should be plausible.

D. The outcome must be qualitative.

Correct Answer: C. The parallel-trends assumption should be plausible.

Question 6

If net benefit is defined as total benefit minus investment cost, the correct ROI formula is:

A. Investment cost ÷ net benefit × 100

B. Total benefit ÷ net benefit × 100

C. Net benefit ÷ investment cost × 100

D. Investment cost − total benefit × 100

Correct Answer: C. Net benefit ÷ investment cost × 100

Question 7

Which recommendation is most evidence-informed?

A. Implement every suggested intervention immediately.

B. Adopt a monitored pilot with an owner, safeguards, indicators, and review date.

C. Ignore non-significant findings completely.

D. Use an AI prediction without human oversight.

Correct Answer: B. Adopt a monitored pilot with an owner, safeguards, indicators, and review date.

Question 8

Which measure is generally a lagging indicator?

A. Manager training completion

B. Number of mentoring meetings

C. Employee participation rate

D. Annual regretted turnover

Correct Answer: D. Annual regretted turnover

Question 9

Why can a very large sample produce a statistically significant but unimportant result?

A. Large samples always contain measurement error.

B. Statistical tests can detect very small effects when estimates are precise.

C. Large samples prevent calculation of effect sizes.

D. Statistical significance measures financial return.

Correct Answer: B. Statistical tests can detect very small effects when estimates are precise.

Question 10

Which governance practice is appropriate for AI-enabled HR analytics?

A. Remove all human review to improve efficiency.

B. Use historical data without checking for bias.

C. Document data use, test differential effects, and provide accountability.

D. Hide the model from employees and managers in every circumstance.

Correct Answer: C. Document data use, test differential effects, and provide accountability.