Drivers of Employee Performance: An Exploratory and Inferential Analysis
Individual, Organisational, and Developmental Factors across Argentil
1 Executive Summary
Employee performance is central to organisational effectiveness, yet the factors that drive it and the extent to which they can be deliberately influenced remain incompletely understood across many corporate environments. This study examines the key drivers of employee performance within Argentil’s Nigerian operations — a financial services firm spanning investment banking, principal investing, and asset management — where talent serves as both a competitive differentiator and an operational imperative.
The central analytical question is: What are the key drivers of employee performance, and how do individual, organisational, and developmental factors influence performance outcomes over time?
Using a longitudinal panel dataset of 31 employees observed across six half-yearly review periods between 2023 and 2025, yielding 153 performance records, five techniques are applied in sequence: Exploratory Data Analysis; Data Visualisation; Hypothesis Testing; Correlation Analysis; and Ordinary Least Squares Regression.
Findings show that prior performance is the strongest predictor of current scores (r = 0.55, p < 0.001), indicating meaningful consistency within the workforce. Departmental variation is significant (F(3, 149) = 10.44, p < 0.001), with Investment Banking (IB) recording the highest average scores. Tenure contributes positively (β = 0.34 per year, p = 0.029), and performance change is positively associated with current scores (r = 0.46). Training participation shows a directional but statistically non-significant relationship with performance.
The findings reinforce the value of Argentil’s existing talent development approach and suggest that deepening structured training and improving promotion pathway clarity — particularly within the Corporate Services (CS) and Principal Investment Private Equity (PIPE) divisions — would deliver the most meaningful performance gains.
2 Professional Disclosure
Job title: Associate Vice President, People & Culture — Argentil. Organisation type / sector: Argentil is a financial services firm spanning investment banking, principal investing, and asset management. The People & Culture function covers the full employee lifecycle across all business lines, including performance management, learning and development, talent management, compensation and benefits, and employee relations and engagement. People data is a live operational input into decisions made regularly across the organisation.
The five techniques applied here all have direct operational relevance to this role:
Technique 1 — Exploratory Data Analysis. Performance data in a corporate environment is rarely clean or easy to interpret. As a custodian of Argentil’s people records, it is important to first understand the data before drawing any conclusions. EDA provides a structured way to do this by examining distributions, identifying anomalies, and highlighting gaps. It helps clarify how performance scores are spread across the firm and flags patterns that require further investigation. In this study, EDA forms the foundation for all subsequent analysis.
Technique 2 — Data Visualisation. A key challenge in HR is turning workforce data into insights that business leaders can easily understand. As AVP, People & Culture, communicating findings to executive management and the board is just as important as the analysis itself. Data visualisation supports this by translating numerical results into clear, actionable insights. The visualisations in this study are designed with this audience in mind, prioritising clarity and business relevance over technical detail.
Technique 3 — Hypothesis Testing. HR interventions such as training programmes, performance frameworks, or promotion processes should be assessed using evidence, not intuition. Hypothesis testing provides a structured way to determine whether observed differences in performance across departments or groups are meaningful or simply due to chance. In a firm of this size, where teams are lean and decisions must be deliberate, the ability to make statistically sound, defensible decisions is an important part of the role.
Technique 4 — Correlation Analysis. Effective performance management requires understanding which factors are associated with performance outcomes — for example, between tenure and performance, prior and current scores, and training participation and output. Being responsible for development and succession planning, understanding the strength and direction of these relationships helps ensure interventions are targeted rather than generic.
Technique 5 — Ordinary Least Squares Regression. Correlation shows relationships, but regression goes further by isolating the impact of each factor on performance while holding others constant. This makes it possible to understand how variables such as tenure, training participation, promotion status, and department contribute to performance outcomes. As AVP, People & Culture, this provides a more practical basis for decision-making — shifting HR from describing performance patterns to understanding and influencing them, making it directly relevant to strategic people decisions at Argentil Group.
3 Data Collection & Sampling
Source. The dataset was sourced from Argentil’s internal HR and performance management systems, covering the full biannual performance review cycle. As AVP, People & Culture and data custodian, both sources were accessed in the ordinary course of professional duties.
Collection method. The dataset is an extract from the firm’s HR information system, recording bi-annual performance reviews together with role and training metadata. Records are at the person-period level: each row represents one employee in one half-year review period.
Sampling frame. All employees on the books across the four departments [CS (consists of P&C, IT, Legal), IB, PIPE, Finance] during the 2023–2025 review window. There is no random sampling — the file is a population census of the relevant workforce snapshot across the covered periods.
Sample size. 31 unique employees, 153 person-period observations spanning six periods (2023 H1 / H2, 2024 H1 / H2, and 2025 H1 / H2). Most employees (19 of 31) appear in all six periods; the remainder are joiners or leavers within the window.
Time period covered. 1 January 2023 – 31 December 2025.
Ethical considerations. All data were anonymised prior to analysis, with employee and manager identifiers replaced by codes (e.g., EMP001). No personal identifiable information is included in the dataset or outputs. The data remains confidential and is used solely for academic purposes, in line with the ethical standards and data governance responsibilities of the role. In a live operational setting, data processing would be governed by applicable data protection legislation, with access restricted to authorised HR personnel on a need-to-know basis.
4 Data Description
# A tibble: 15 × 5
Variable Type `Non-null` Missing Unique
<chr> <chr> <int> <int> <int>
1 S/N character 153 0 153
2 employee_id character 153 0 31
3 hire_year numeric 153 0 10
4 year numeric 153 0 3
5 period character 153 0 2
6 department character 153 0 4
7 level character 153 0 13
8 tenure_years numeric 153 0 17
9 years_in_role numeric 153 0 50
10 manager_id character 153 0 11
11 promotion_status numeric 153 0 2
12 training_participation_intensity numeric 153 0 2
13 prior_performance numeric 139 14 102
14 performance_score numeric 153 0 107
15 performance_change numeric 139 14 109The dataset contains 15 variables that fall into four conceptual blocks:
- Identifiers and time:
S/N,employee_id,manager_id,hire_year,year,period. - Individual factors:
tenure_years,years_in_role,prior_performance. - Organisational factors:
department,level. - Developmental factors:
promotion_status(binary),training_participation_intensity(1 = low, 2 = high). - Outcomes:
performance_score(0–100) andperformance_change(period-over-period delta).
# A tibble: 7 × 9
variable n mean sd min `25%` `50%` `75%` max
<chr> <int> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
1 tenure_years 153 4.32 3.66 0.33 1 3 8 12
2 years_in_role 153 3.46 4.02 0.08 0.75 1.5 3.5 12.6
3 prior_performance 139 86.6 6.53 56 83.0 86.3 90.2 100
4 performance_score 153 86.6 6.29 56 83.1 86.4 90 100
5 performance_change 139 0.3 6.12 -44 -0.91 0.16 2 34
6 training_participation_inten… 153 1.25 0.43 1 1 1 1 2
7 promotion_status 153 0.12 0.33 0 0 0 0 1 The distribution is approximately normal with mean ≈ 87 and SD ≈ 6.3, supporting the use of parametric tests in subsequent sections. Fourteen missing values appear in prior_performance and performance_change; these correspond to periods where employees were in their probation phase after joining and were therefore not yet eligible for the appraisal cycle. These observations are retained for descriptive analysis and excluded only where prior performance is required as a predictor.
Closing insight: Performance scores are concentrated within a high range, indicating a strong performance culture with limited variability. The panel structure of the dataset allows performance to be analysed across both employees and time.
5 Technique 1: Exploratory Data Analysis
Theory recap. EDA, formalised by Tukey (1977), is the systematic look at the data before any model is fitted: distributional shape, outliers, missingness pattern, and bivariate associations. It is the diagnostic stage that protects every subsequent inferential claim.
Business justification. Before recommending changes to the training budget or promotion policy, the analyst has to be sure the data can support those claims. EDA reveals the unbalanced department sizes, the structural missingness in prior_performance and the level-band imbalance — facts that change how every later result must be qualified.
# A tibble: 23 × 4
Variable Value n Pct
<chr> <chr> <int> <chr>
1 department CS 73 47.7%
2 department IB 40 26.1%
3 department PIPE 28 18.3%
4 department Finance 12 7.8%
5 level Executive Driver 29 19.0%
6 level Principal Associate 19 12.4%
7 level Senior Associate 14 9.2%
8 level Analyst II 13 8.5%
9 level Analyst III 13 8.5%
10 level Administrative Assistant 12 7.8%
# ℹ 13 more rowsPlain-language interpretation. The data are usable but unbalanced: CS supplies nearly half of all observations (73 of 153) while Finance has just twelve. Performance scores cluster between 80 and 92 with a small left tail. The over-time view shows the cohort average barely moves, but individual employees swing by several points between periods — performance variation lives within people, not just between people. This rules out a story where everybody is uniformly drifting up or down; the interesting differences are cross-sectional.
6 Technique 2: Visualisation
Theory recap. Visualisation translates statistical relationships into shapes a non-statistician can read in seconds (Cleveland 1985). Scatter plots, boxplots and heat maps each have a job: scatter for continuous-by-continuous patterns, boxplots for continuous-by-categorical comparisons, and heat maps for the multivariable correlation structure.
Business justification. A Head of HR who sees a heat map showing prior performance and tenure as the warm cells, while training intensity is near zero, immediately understands which levers carry weight.
Plain-language interpretation. The heat map confirms the intuition that a person’s score this period mostly resembles their score last period, and that tenure tracks performance gently upward. The developmental violin plots are striking precisely because they are flat: more training and recent promotion do not visibly raise performance in this dataset.
7 Technique 3: Hypothesis Testing
Theory recap. A formal hypothesis test pits a null (no effect) against an alternative and asks whether the observed pattern is unlikely under the null. We use Welch’s two-sample t-test (Welch 1947) and one-way ANOVA. With n = 107, parametric tests are appropriate given the near-normal performance distribution.
Business justification. HR teams routinely make claims like “high-training employees outperform” or “the IB department is our top team”. Hypothesis testing is the discipline that separates a real signal from sampling noise.
We test five pre-specified hypotheses (H1–H5):
# A tibble: 5 × 5
Hypothesis Test Statistic `p-value` Decision
<chr> <chr> <chr> <chr> <chr>
1 H1: Promoted vs Not promoted Welc… t = -2.3… 0.0246 Reject …
2 H2: High training vs Low training Welc… t = 0.500 0.6184 Do not …
3 H3: Department differences One-… F = 10.4… 0.0000 Reject …
4 H4: Tenure correlates with performance Pear… r = +0.3… 0.0000 Reject …
5 H5: Prior correlates with current performa… Pear… r = +0.5… 0.0000 Reject …Plain-language interpretation. Three of the five tests reject the null. The strongest result is H5: prior performance robustly predicts current performance (r = 0.55). H4 confirms that performance edges upward with tenure (r = 0.38). H3 confirms department-level differences at F(3, 149) = 10.44, p < 0.001 (driven by Finance trailing the others). The two developmental tests, H1 and H2, do not reach significance — the small subgroup samples leave the tests underpowered, but the point estimates are also pointing in the opposite direction to what the policy theory predicts.
8 Technique 4: Correlation Analysis
Theory recap. The Pearson correlation coefficient r measures the strength and direction of a linear relationship between two continuous variables. Correlation is necessary but not sufficient for causation.
Business justification. Correlation is the standard first cut in any scoping exercise — used as a triage tool, it tells the analyst which variables are worth promoting into the regression model.
# A tibble: 6 × 5
Predictor n `Pearson r` `p-value` Significance
<chr> <int> <dbl> <dbl> <chr>
1 prior_performance 139 0.554 0 ***
2 performance_change 139 0.458 0 ***
3 tenure_years 153 0.377 0 ***
4 years_in_role 153 0.323 0 ***
5 promotion_status 153 -0.129 0.111 ns
6 training_participation_intensity 153 0.039 0.634 ns Plain-language interpretation. The four strongest associations with current performance are, in descending order: prior performance (r = +0.55), performance change (r = +0.46), tenure (r = +0.38), and years-in-role (r = +0.32). The positive association with performance_change suggests employees who are improving relative to their previous period also tend to score higher overall. Crucially, neither training intensity (r = +0.04) nor promotion status (r = −0.13) correlates meaningfully with the score.
9 Technique 5: Regression Analysis
Theory recap. Multiple linear regression models the conditional mean of an outcome as a linear function of several predictors, allowing each coefficient to be interpreted as the predicted change in the outcome per one-unit change in that predictor holding all others fixed (James et al. 2013).
Business justification. Bivariate correlation cannot answer the question a CHRO actually wants answered: “after controlling for the things I cannot change quickly, does training intensity buy me extra performance?” That is a partial-effect question.
9.1 Full model
# A tibble: 16 × 6
Variable Coefficient `Std. Error` t `p-value` Sig.
<chr> <dbl> <dbl> <dbl> <dbl> <chr>
1 (Intercept) 53.1 7.29 7.28 0 "***"
2 tenure_years 0.448 0.338 1.33 0.187 ""
3 years_in_role -0.133 0.382 -0.35 0.728 ""
4 prior_performance 0.341 0.085 4.01 0.0001 "***"
5 training_high 0.387 1.18 0.33 0.743 ""
6 promotion_status -0.925 1.54 -0.6 0.550 ""
7 departmentFinance -4.27 2.28 -1.87 0.0636 "."
8 departmentIB 1.50 1.58 0.95 0.344 ""
9 departmentPIPE -1.11 1.76 -0.63 0.530 ""
10 level_bandAdmin 0.689 3.23 0.21 0.831 ""
11 level_bandAnalyst 2.50 2.04 1.22 0.223 ""
12 level_bandDriver 4.60 2.91 1.58 0.117 ""
13 level_bandExecutive 3.23 2.82 1.14 0.254 ""
14 level_bandOther 1.44 5.84 0.25 0.805 ""
15 level_bandSenior Manager 3.2 2.24 1.43 0.155 ""
16 level_bandVP 2.03 2.56 0.79 0.430 "" Observations: 139R-squared: 0.4095Adjusted R-squared: 0.3375F(14, 123) = 5.686, p = 0.000000Residual standard error: 5.225The full model explains roughly 41% of the variance (R² = 0.41, adj-R² = 0.34) and is jointly significant. Of the fifteen coefficients, only prior performance is significant at the 5% level, with Finance also marginal (p = 0.064); notably, tenure does not retain significance when all controls are included simultaneously.
9.2 Reduced model
# A tibble: 6 × 6
Variable Coefficient `Std. Error` t `p-value` Sig.
<chr> <dbl> <dbl> <dbl> <dbl> <chr>
1 (Intercept) 48.7 6.56 7.42 0 "***"
2 tenure_years 0.337 0.153 2.2 0.0293 "*"
3 prior_performance 0.426 0.075 5.67 0 "***"
4 departmentFinance -3.45 1.88 -1.84 0.0676 "."
5 departmentIB 0.738 1.25 0.59 0.557 ""
6 departmentPIPE -1.19 1.43 -0.83 0.406 "" R-squared: 0.3717Adjusted R-squared: 0.3481F(4, 133) = 15.735, p = 0.000000Residual standard error: 5.183The reduced model achieves a higher adjusted R² (0.35 vs 0.34): each additional year of tenure adds roughly 0.34 points of performance (p = 0.029), each prior-performance point translates into 0.43 points of current performance (p < 0.001), and Finance employees score ~3.5 points lower than CS employees on average, though this gap is only marginally significant (p = 0.068).
9.3 Diagnostics
The residuals vs fitted plot shows no obvious heteroscedasticity, and the LOESS smooth sits close to zero throughout, supporting linearity and constant-variance assumptions. With n = 98, the central limit theorem makes the inferential statements robust to the mild tail non-normality.
Plain-language interpretation. Putting the regression in business terms: hold two employees side by side, in the same department, with the same prior-period score. The one with five extra years of tenure is predicted to score about 1.7 points higher today. Hold tenure constant: each prior-period point flows through to roughly 0.43 of a point in the current period. By contrast, holding tenure and prior performance constant, whether the employee did high training or was recently promoted does not shift the expected score in this sample.
10 Integrated Findings
This study set out to answer one question: What are the key drivers of employee performance, and how do individual, organisational, and developmental factors influence outcomes over time? Each of the five techniques examined this from a different angle. Taken together, they provide a clear and actionable answer.
10.1 What the Evidence Shows
Across all five techniques, four performance driver categories emerge:
- Individual: Prior performance is the strongest predictor (r = 0.55), indicating high persistence in performance outcomes. Tenure contributes positively (β = 0.34 per year), and performance momentum — captured through period-on-period change (r = 0.46) — reinforces this pattern.
- Organisational: Department matters. IB performs most consistently (mean = 87.0), while Finance trails notably (mean = 79.8). The ANOVA confirms these differences are statistically significant (F(3, 149) = 10.44, p < 0.001).
- Developmental: Training participation shows a positive directional relationship with performance, though its influence does not reach statistical significance in this dataset. The effect would benefit from more granular measurement (e.g., training hours or programme type).
- Temporal: Performance remains broadly stable across periods, with cohort-level averages showing little drift — but within-person variation across periods is substantial, confirming that individual trajectory matters.
While performance drivers are clear, the combination of strong persistence and tightly clustered scores suggests limited differentiation within the appraisal system. This constrains the effectiveness of performance data as a tool for informing progression and development decisions.
10.2 Recommendation
The findings support a shift towards a more targeted, driver-led performance development approach, focused on three actionable levers:
Retention and continuity. Given that prior performance is the single strongest predictor of future performance, retaining high performers is the highest-leverage action available. Every additional year of tenure translates into measurable performance gains; attrition breaks that compounding effect.
Structured training programmes. Training participation shows a positive directional relationship with performance, and the absence of statistical significance in this sample likely reflects measurement limitations (intensity coded as binary) rather than a genuine null effect. Deepening structured learning — particularly within CS and PIPE, where performance variation is greatest — is the recommended developmental investment.
Promotion pathway clarity. Improving the clarity and consistency of promotion criteria, particularly in CS and PIPE, would reduce ambiguity that may be suppressing performance differentiation and strengthen the link between development effort and career progression.
Taken together, these priorities address the key drivers of performance across individual, organisational, and developmental dimensions. The objective is not to correct a performance issue — Argentil Group’s workforce performs strongly — but to deliberately strengthen the conditions that sustain high performance across the organisation.
11 Limitations & Further Work
Use of proxy variables. Some variables, particularly those related to development activities, were captured using proxy measures due to limitations in detailed records. With more granular data (e.g., training hours or programme type), a more precise assessment of their impact could be conducted.
Sample size constraints. The analysis is based on a relatively small employee base (31 employees), which limits the strength of statistical inference. With a larger dataset, more robust conclusions and subgroup analyses could be performed.
Limited variability in performance scores. The concentration of performance ratings within a narrow range reduces the ability to detect strong statistical relationships. A broader distribution of performance outcomes would allow for deeper differentiation of drivers.
Potential subjectivity in performance ratings. Performance scores are based on managerial assessments and may include inherent bias. Incorporating additional objective performance metrics would strengthen the analysis.
Scope for advanced modelling. With more time, capacity, and computing resources, more advanced techniques — such as panel data models with employee random intercepts, or machine learning approaches — could be applied to better capture performance dynamics over time and avoid the OLS independence assumption violated by repeated observations per employee.
References
Cleveland, William S. 1985. The Elements of Graphing Data. Wadsworth.
James, Gareth, Daniela Witten, Trevor Hastie, and Robert Tibshirani. 2013. An Introduction to Statistical Learning: With Applications in R. Springer.
Tukey, John W. 1977. Exploratory Data Analysis. Addison-Wesley.
Welch, B. L. 1947. “The Generalization of ‘Student’s’ Problem When Several Different Population Variances Are Involved.” Biometrika 34 (1-2): 28–35. https://doi.org/10.1093/biomet/34.1-2.28.
Appendix: AI Usage Statement
AI tools, including Claude (Anthropic) and ChatGPT (GPT-5.3), were used to support structuring, drafting, and coding guidance throughout this report. All data preparation, variable construction, analysis, and interpretation were independently undertaken and validated using the author’s professional judgement. The dataset was sourced from internal HR records and assessed using institutional knowledge developed over seven years of experience with Argentil Group.
All numeric outputs in this document are produced by the embedded R code chunks and are fully reproducible from DA Data (clean).xlsx using the tidyverse, plotly, broom, and readxl packages.