State-owned enterprises (SOEs) are created to perform economic activities as mandated by a government for the country’s benefit. In South Africa, they operate under various laws, such as the Companies Act (2008) and the Public Finance Management Amendment Act (1999). Over the past decade, employee retention in South African SOEs has become a major concern because of the high turnover of critical core skills (Khuluvhe et al., 2022). These enterprises have a diverse workforce with extensive expertise, particularly in technical areas, where retention remains problematic. In addition, South African SOEs face competition from private sector companies, which are seen as preferred employers and offer attractive incentives and benefits to retain their employees (Khuluvhe et al., 2022).

Researchers have identified several reasons why South African SOEs struggle to retain talented employees, including inadequate performance incentives, employee poaching by private sector companies, insufficient training and development and corruption (Khuluvhe et al., 2022; Othman & Lembang, 2017; Zondo Commission Report Part 1, 2022). According to Alnaqbi (2011), to improve employee retention, SOEs in South Africa need to focus on enhancing employees’ skills and involving them in decision-making processes while relying less on financial incentives compared to the private sector. This is supported by Halim et al. (2020), who stated that it is likely that retention strategies in SOEs will differ significantly from those in private institutions, and studies on private company retention strategies may yield different results.

While various studies have explored employee retention strategies in private sector organisations (Döckel, 2003; Hashish, 2017; Mitonga-Monga et al., 2020; Singh, 2019), there is a lack of research on the validity of these strategies for SOEs (Alnaqbi, 2011). A challenge is that private sector retention strategies may not be applicable to SOEs because of the differing contexts of the two sectors. In addition, employee retention factors vary across countries, economies and industry sectors (Halim et al., 2020). Various authors have also argued that employee retention is a complex, multifaceted issue that cannot be addressed with a one-size-fits-all approach (Bharath, 2021; Mzini, 2019).

Finding skilled employees is challenging, and attracting and retaining these talents in SOEs has proven to be difficult (Thusi & Chauke, 2023). Although retaining these employees is crucial, as failing to do so jeopardises the business’s performance and survival (Syarafina & Sushandoyo, 2022), the existing retention strategies and models used by South African SOEs have not been effective in preventing the mass resignation of skilled employees (Thusi & Chauke, 2023). Despite their significant contribution to South Africa’s economy, employee retention in South African SOEs is not well researched, and no empirical studies have validated the retention models they use (Thusi & Chauke, 2023). This study is thus unique in its focus on identifying the specific factors that should be included in an employee retention model tailored for South African SOEs to prevent the loss of skilled employees.

The main aim of the study was to develop and validate an employee retention model that could be applied by South African SOEs.

Employee retention has been described as an organisation’s ability to retain its employees for an extended period of time and is a challenge arising from the competition for talent, aimed at implementing measures to encourage employees to stay with the organisation for as long as possible (Zareen et al., 2013). Elaborating on this, Singh (2019) explained that retention involves various strategies employed by organisations to motivate their employees to remain with them longer. Hanafi and Baharin (2018) argued that organisations must operate efficiently through their workforce to stay economically viable and relevant, as strategic objectives cannot be achieved without people. Kaleem (2019) emphasised that the efficiency of organisations can only be accomplished if retention strategies are aligned with the organisation’s demand and supply needs. Importantly, Dhillon (2025) argued that employee retention is not merely the absence of turnover but a proactive, strategic process aimed at fostering long-term employee commitment and satisfaction.

A substantial body of literature underscores the complexity and multifaceted nature of employee retention. Coetzee et al. (2018), Mzini (2019) and Bharath (2021) contended that retention should not be conceptualised as a singular phenomenon; rather, the components of an effective retention model must be contextually adapted. Similarly, Al-sharafi et al. (2018) and Bharath (2021) argued that the factors influencing an employee’s decision to stay will differ from organisation to organisation. These views are supported by Dhillon (2025), who emphasised that employee retention encompasses various diverse and interrelated factors. These include organisational factors such as leadership style, organisational culture (OC) and human resource practices; job-related factors such as job satisfaction, role clarity, workload and autonomy and individual factors such as career aspirations, personal values and work–life balance (WLB) and relational factors such as supervisor support, coworker relationships and team cohesion. Moreover, Halim et al. (2020) highlighted that the determinants of employee retention are not only context specific but also vary across national, economic and sectoral boundaries.

Dhillon (2025) posited that employee retention is a strategic organisational priority, underpinned by multiple compelling factors. Retention practices have transitioned from reactive interventions to proactive, systematically becoming integrated components of organisational culture and policy. The rationale for this strategic emphasis can be delineated into four key dimensions: (1) Cost efficiency, where increased turnover can results in substantial expenses related to recruitment, onboarding and training; (2) knowledge continuity, as the retention of personnel safeguards institutional memory and specialised expertise; (3) team performance and employee morale, given that workforce stability is positively correlated with enhanced productivity and psychological well-being and (4) employer brand strength, where high retention rates contributes to a favourable organisational image that attractions and retains high-calibre talent.

Employee retention is conceptually distinct from related constructs such as employee turnover, turnover intention and employee engagement (EE), each of which contributes uniquely to understanding workforce dynamics:

An important consideration is that the Fourth Industrial Revolution (4IR) signifies a shift from a reliance on energy and physical resources to a future where human abilities are seamlessly integrated with emerging technologies such as artificial intelligence, robotics, nanotechnology and the Internet of Things (Revunit, 2022). Consequently, the global business landscape is rapidly evolving, with business models moving from selling products to embracing the sharing economy (Revunit, 2022). The Worldwide Economic Forum projected that by the end of 2022, the global economy would have seen a net loss of over 5 million jobs (Quaglietti & Wheeler, 2022). Although technology can eliminate many jobs, it also creates new ones that demand workers with new skills, as many current jobs did not exist a decade ago. Without opportunities for retraining and reskilling, organisations will struggle to meet their labour needs (Tamayo et al., 2023).

The World Economic Forum predicts that future employment will increasingly require both technical and interpersonal skills that support value creation within organisations (Quaglietti & Wheeler, 2022). According to Balalle and Balalle (2019), there is often a mismatch between where talent is needed and where it is available, posing a significant challenge for organisations in acquiring and retaining skilled employees. Although some jobs will need to be replaced because of the evolving work environment and the 4IR, it is crucial for organisations to implement effective retention strategies to maintain their current skilled workforce and provide the required reskilling (Balalle & Balalle, 2019; Bhatia, 2020; Tamayo et al., 2023). For South African SOEs, it is essential to retain and reskill current employees. Without this, the necessary skills for the new work environment and the 4IR will have to be sourced from the private sector or other countries. It is thus more practical to keep and reskill talented employees while promoting a culture of lifelong learning, rather than bearing the costs of recruiting external skills (Hewapathirana & Almasri, 2021).

A review of the literature on employee retention was conducted by the primary researcher and reported in Maphanga (2023). The review was conducted over a period of 14 months, which entailed a search in English journals, books, master’s dissertations and doctoral theses using Sabinet African Journals, Emerald Insight, Research Gate, JSTOR, ProQuest and Google Scholar. The phrase ‘constructs of employee retention’ was used, and the following inclusion criteria were applied: The source (1) must have been published between 1980 and the present; (2) must be a journal article, book, master’s dissertation or doctoral thesis and (3) must have provided empirical evidence of or theoretical justification for why the specified construct influenced employee retention. A total of 110 sources met all the criteria and were studied to identify constructs that influenced employee retention. Constructs that were identified by two or more sources as influencing employee retention were documented. This produced 12 constructs that are detailed in Table 1.

Six existing general organisational models of staff retention and three existing staff retention frameworks used by three major South African SOEs are shown in Table 2. None of these models and frameworks has been specifically developed to assess the 12 underlying staff retention constructs that have been the most globally researched and empirically shown to significantly influence staff retention as discussed earlier. A new conceptual model has therefore been proposed for this purpose.

Drawing from the employee retention constructs outlined in Table 1, a new theoretical model for employee retention in South African SOEs is proposed. This model includes 12 organisational employee retention factors (independent variables) and one dependent variable (IL). The proposed theoretical model is illustrated in Figure 1.

In the current study, IL is used as the measurement for employee retention (the dependent variable). This is supported by Cho et al. (2008), who stated that the construct of IL is frequently utilised as a practical measure of employee retention, serving as an alternative metric in empirical retention research.

The term turnover intention is considered interchangeable with the term IL (Cho et al., 2008). Specifically, IL refers to the subjective estimation of an individual regarding the probability of leaving an organisation in the near future (Cho et al., 2008; Dhanpat et al., 2018). This is in line with the suggestion by Nasir et al. (2019, p. 157) that ‘retention processes (staff retention) should be studied along with the quitting processes (IL)’. Staff retention can be better understood by studying factors leading to increased dissatisfaction among employees and, thus, their IL the organisation. Thus, measuring the factors contributing to an employee’s IL an organisation is the most valid measurement of the concept of employee retention (Yousuf & Siddqui, 2019).

According to Jaharuddin and Zainol (2019), turnover intention measures the likelihood of an employee leaving an organisation. Some employees’ general attitudes resulted from their idea of leaving and looking for other jobs (Joseph et al., 2007). It is considered a key determinant of the individual’s IL the current employer (Li et al., 2019). Turnover intention is also linked to voluntary and involuntary turnover. Employers are responsible for assessing these types of turnover to ensure that they retain employees who will assist them in reaching their organisational goals and objectives.

Although research indicates that an employee’s IL an organisation is a valid measurement of the concept of employee retention (Cho et al., 2008; Dhanpat et al., 2018; Nasir et al., 2019; Yousuf & Siddqui, 2019), the following limitations in using IL as a measurement of employee retention have been identified by Mohamed and Aboul-Ela (2023) and Contreras (2024): (1) Not all employees who express an IL actually resign, as intention does not always equal behaviour; (2) employee intentions can change over time, especially in response to organisational changes, personal circumstances or labour market conditions; (3) employees may underreport their IL because of fear of negative consequences or a desire to appear loyal, especially in non-anonymous surveys and (4) using IL as a standalone measure ignores other critical factors influencing employee retention.

A quantitative, cross-sectional survey design was employed, and primary data were gathered through a self-report questionnaire to collect statistical data for validating the proposed theoretical model of employee retention (Creswell & Creswell, 2022).

Following an extensive literature review, it was evident that no single questionnaire existed to measure all 12 identified retention factors in the newly proposed theoretical model of employee retention. Eight existing questionnaires, previously used in various studies to measure the 12 underlying constructs of employee retention and the construct of IL, were available, making it unnecessary to develop a new questionnaire. These eight questionnaires, with the studies in which they were used, are cited in Table 3, and they were combined into a consolidated Staff Retention Questionnaire (SRQ) for the empirical phase of this study. This combination approach was advocated by scholars such as Choudhary (2016) and Dhanpat et al. (2018), who argued that the advantages of using existing measurement instruments are established reliability and validity, and cost savings by avoiding the need to develop a new questionnaire. The final consolidated SRQ consisted of 162 items and comprised three sections: biographical information, items measuring the 12 underlying constructs of employee retention identified in Table 1 and items measuring IL, which was used as a proxy for the dependent variable employee retention. All items were adopted from the eight existing questionnaires and the names of the questionnaires, the number of items per construct as well as the reliability of the questionnaires are indicated in Table 4. The reliability of the eight questionnaires was evaluated based on Jain and Angural’s (2017) guidelines, where a Cronbach’s alpha (α) ≥ 0.9 is considered excellent, 0.9 > α ≥ 0.8 good, 0.8 > α ≥ 0.7 acceptable, 0.7 > α ≥ 0.6 questionable, 0.6 > α ≥ 0.5 poor and α < 0.5 unacceptable.

Based on Jain and Angural’s (2017) guidelines, Table 3 indicates that all the measurement instruments incorporated into the final SRQ recorded either an acceptable or good reliability score. The exception was the scale used to measure job characteristics, which recorded a reliability score of 0.41.

All six scales of Döckel’s (2003) Retention Factor Scale (RFS) were incorporated into the final SRQ. This decision was based on the fact that five of the six RFS scales recorded a good or acceptable internal consistency score based on the Jain and Angural (2017) guidelines. Only one scale, used to measure job characteristics, with a reliability score of 0.41, did not meet the acceptable reliability threshold of between 0.60 and 0.70. According to Jain and Angural (2017), this could be because of a construct being measured by a few items (the job characteristics scale only has four items). The low reliability score recorded for the job characteristics scale was noted as a limitation of the current study.

All dimensions, except for compensation and benefits, were measured on a 5-point Likert scale that ranged from 1 (strongly disagree) to 5 (strongly agree). The dimension of compensation and benefits was measured on a 5-point Likert scale using level of satisfaction, which ranged from 1 (strongly dissatisfied) to 5 (strongly satisfied).

The researcher, as an employee at the South African SOE under study, had assisted in accessing the organisation. To mitigate a possible conflict of interest, participants had to give their informed consent to participate in the study, responses were anonymous and not traceable to individual participants and a neutral third party was used to approach identified employees to invite them to participate in the study.

The researcher submitted a written request to top management to conduct the research within the SOE, which subsequently granted the necessary permission. Following this, ethical clearance to conduct the study was obtained from the Ethics Committee of the Industrial and Organisational Psychology (IOP) Department at the University of South Africa (UNISA) before data collection began. Once ethical clearance was obtained, a list of all SOE employees (N = 5395), along with their contact details, was provided to the researcher by the Human Resource Department. Stratified sampling categorised the population into six homogeneous job grades, followed by the selection of simple random samples from each grade (Creswell & Creswell, 2022). This produced a sample of 1000 employees whose email contact details were provided to a neutral third party. The third party invited all 1000 employees to participate in the study via email, which included the purpose of the study, that participation was voluntary and that anonymity would be maintained, as no information provided could be linked to any specific individual. Employees were also informed of their right to withdraw from the survey at any time before submission, after which it was not possible, as completed questionnaire could no longer be traced. The email contained a link to the online SRQ, and participants were asked to access the link if they agreed to participate. A total of 685 employees responded positively to the request and accessed the online questionnaire, which included a request for informed consent, without which the questionnaire could not be answered further. Completed surveys were submitted electronically through the online link to the organisation’s electronic communication system and then imported into an electronic spreadsheet for further statistical analysis.

The Statistical Package for Social Sciences software (SPSS) version 29 and its add-on package AMOS (Analysis of Moment Structures) were used to analyse the data and perform SEM (Arbuckle, 2022). The data from the SRQ were utilised to firstly generate descriptive statistics, including frequency distributions, means and standard deviations. Secondly, an internal consistency analysis was performed using Cronbach’s alpha to assess the reliability of the SRQ. Thirdly, a principal component analysis (PCA) was conducted to select the appropriate number of factors to include in the measurement model, and fourthly, SEM was used to specify and validate the measurement and structural models. The following six goodness-of-fit (GOF) indices, suggested by Hair et al. (2019), were utilised to validate the measurement and structural models:

Ethical clearance to conduct this study was obtained from the University of South Africa CEMS/OP Research Ethics Review Committee (No. 2019_CEMS/IOP_031). Informed consent was obtained via digital consent, without which the online questionnaire could not be answered further. If a participant withdrew midway through the survey, they were unable to submit the final questionnaire and their answers were not recorded, requiring no further action. Withdrawal after submission was not possible, as the researcher was unable to identify a participant’s completed questionnaire to cancel it as participation was anonymous.

The mean and standard deviation for each of the 12 employee retention dimensions and the IL dimension (dependent variable) measured by the SRQ were determined. Because of limited space, only the means are reported and interpreted. According to research by the Human Science Research Council (HSRC), a score of 3.2 on a 1–5 Likert scale serves as a useful benchmark for distinguishing between positive and potentially negative perceptions (Castro & Martins, 2010). This threshold is slightly above the neutral midpoint of 3.0, which typically reflects an undecided or neutral stance. By setting the cut-off at 3.2, the classification leans more confidently towards genuinely positive responses – those closer to ‘agree’ or ‘strongly agree’. This approach also helps minimise the risk of misinterpreting slightly above-neutral responses as positive. In essence, it creates a buffer between neutral and positive ratings. In addition, applying a consistent standard like 3.2 enhances the ability to compare results across different surveys and time periods.

The survey results showed that respondents generally had a negative perception of 7 of the 12 employee retention dimensions (ranging from 2.74 to 3.19) and a positive perception of 5 of the 12 dimensions (ranging from 3.32 to 4.04). However, some of these scores are close to the cut-off point of 3.2 and lack practical significance. The dependent variable, IL, had a mean score of 3.52, indicating that respondents tended towards the ‘Chances are good’ (Option 4) response rather than the ‘Undecided’ (Option 3) response when asked whether they intended to leave the SOE.

While PCA and exploratory factor analysis (EFA) often yield similar outcomes in many research contexts (Hair et al., 2019), PCA was selected for the current study. Exploratory factor analysis is primarily concerned with common variance and is typically used to explore underlying theoretical constructs. In contrast, PCA considers total variance and is particularly effective for reducing data by converting observed variables into a smaller number of uncorrelated components that account for most of the variance (Suhr, 2005). As the main objective of the current study was data reduction to identify the minimum number of components that captured the majority of the variance, PCA was deemed the most suitable method. Had the goal been to uncover latent factors explaining the correlation patterns among variables, EFA would have been more appropriate (Chumney, 2012).

With a large sample size of 685, the study was able to utilise a split-sample method as recommended by Fokkema and Greiff (2017). This entailed that PCA was performed on a random subset of 300 participants, while the remaining 385 participants were used for confirmatory factor analysis (CFA). The adequacy and suitability of the 300-participant sample for PCA were evaluated using the Kaiser-Meyer-Olkin (KMO) measure and Bartlett’s test of sphericity. The KMO value was 0.90, and Bartlett’s test of sphericity was statistically significant at the p < 0.001 level, indicating that the data were suitable for conducting a PCA (Fokkema & Greiff, 2017). Hair et al. (2019) recommended that more than one method should be used for identifying the optimal number of factors to retain in PCA. The current study used two methods: Kaiser’s (1974) eigenvalue-greater-than-one rule and parallel analysis (Hayton et al., 2004). Principal component analysis of the 162 original SRQ items identified 11 components with eigenvalues greater than 1, a reduction from the initial 13 constructs measured by the SRQ. These 11 components, detailed in Table 4, collectively accounted for 56.82% of the variance in the data. The decision was made to retain these 11 components, as they explained over 50.0% of the total variance (Hair et al., 2019).

Parallel analysis (PA) was used as the second method to determine how many components to retain in PCA. This method involves comparing the eigenvalues from the data matrix with those from a Monte-Carlo simulated matrix of random data of the same size. Components with eigenvalues that surpass those from the random data set are kept (Hayton et al., 2004). Parallel analysis was conducted following the procedure outlined by Hayton et al. (2004), which stipulated that either the mean or the 95th percentile of each eigenvalue from the random data (n = 300) could be used for comparison with the actual eigenvalues. The current study used the mean, and components from the actual data with eigenvalues exceeding the corresponding mean eigenvalues from the random data were retained. Table 5 shows that the first 11 actual eigenvalues are greater than the mean values from PA, leading to the retention of these 11 components. The factor structure derived from the eigenvalues and PA was interpreted, and a final solution was chosen (Hair et al., 2019). This structure was based on the unrotated factor matrix, which included the factor loadings for each variable on each factor.

Hair et al. (2019) stated that while an unrotated factor solution can reduce data, it does not adequately interpret the variables being studied. They recommended using rotation to simplify the factor structure, making it more straightforward and theoretically meaningful. The current study utilised the Varimax orthogonal rotation method because of its simplicity, which revealed a clear structure, with each of the 11 components showing several significant loadings after loadings below 0.4 were excluded, as recommended by Hair et al. (2019). The 11 components extracted through PCA were named based on the component pattern and loadings from the rotated solution, employee retention constructs identified in the literature review and the original employee retention construct names used in the SRQ. The components were named as follows:

The measurement scales for the 11 extracted components were created by averaging the items that loaded onto each scale, and these scales were then tested for internal reliability. Jain and Angural (2017) and Hair et al. (2019) suggest that a Cronbach’s alpha of 0.70 be used as the generally accepted minimum for internal reliability. Following this guideline, all 11 components showed acceptable internal reliability, with Cronbach’s alpha coefficients ranging from 0.72 to 0.96, as shown in Table 6.

This research provides important insights for retaining employees in South African SOEs. As compensation and benefits have the most significant negative impact on employees’ intentions to leave, managers should prioritise offering competitive and fair compensation packages. This could involve salary increases, bonuses and comprehensive benefits to retain employees. However, such solutions could be difficult to implement given that SOEs must operate within rigid budget frameworks, often prioritising operational survival over employee incentives (Bernstein, 2024). Cultivating a positive OC is essential as it greatly reduces the IL. Managers should strive to create a supportive, inclusive and engaging work environment by encouraging teamwork, recognising employee achievements and maintaining transparent communication. In addition, providing training and development opportunities is crucial for employee retention. Managers should offer continuous learning opportunities, career development programmes and skill enhancement workshops to keep employees motivated and engaged.

Even though factors such as ethical leadership, meaningful work, SS, internal and external relationships, OS, WLB and EE were not statistically significant in this study, they should not be overlooked. Managers should regularly assess these areas to avoid potential problems in the future. In addition, as various factors influence employees’ intentions to leave, managers should implement personalised retention strategies. Understanding the unique needs and preferences of individual employees can help in creating more effective retention plans. By addressing these areas, managers can develop a more stable and dedicated workforce, resulting in improved organisational performance and lower turnover rates.

It is acknowledged that implementing these recommended employee retention strategies could prove challenging, as South African SOEs operate within complex structural and contextual environments. Structural challenges include ambiguous mandates, governance challenges (Maphanga, 2023), bureaucratic rigidity, fragmented human resource systems and skills mismatch (Dlaminia et al., 2021). Contextual challenges include transformation pressures (Govender, 2009), budget constraints, brain drain, labour relations (Dlaminia et al., 2021) and public scrutiny (Maphanga, 2023). These realities emphasise the need for context-specific retention strategies.

The practical recommendations discussed earlier can also be linked to broader human resource management (HRM) theory:

This study had several limitations. Firstly, it concentrated on employee retention within SOEs, requiring caution when applying the results to other organisations. Secondly, the questionnaire used to assess employee retention was not specifically designed for South African SOEs. Thirdly, the study employed a cross-sectional survey design, which limits the ability to infer causal relationships between variables. Fourthly, the removal of 50% of the questionnaire items in the measurement model during CFA because of low factor loadings to improve model fit indices could have compromised the content validity of the constructs. Fifthly, the study relied on a self-completed survey questionnaire, which may have introduced common method bias and self-reporting bias, as all data were collected from the same respondents using a single method. Sixthly, a limitation was that the TLI value obtained for the structural model was 0.88, which falls slightly below the commonly accepted threshold of 0.90 for indicating good model fit. This could have limited the precision of the model’s explanatory power. Seventhly, a limitation was using IL as a proxy for measuring employee retention. Eighthly, the low reliability score of 0.41 recorded for the job characteristics scale could have affected the validity of findings related to job characteristics.

It is recommended that the employee retention model developed and validated in this study undergo further validation in other organisations. In addition, a new questionnaire tailored to the developed and validated employee retention model should be created and validated for use across all South African SOEs.