A quantitative, cross-sectional survey design was used. The design was appropriate for assessing doctors’ perceptions of rewards, perceived adequacy, reward expectations and retention intention at one point in time. Because the design was cross-sectional, the structural model tested hypothesised theoretical relationships and indirect effects; it did not establish temporal or definitive causal relationships.

The study was conducted among medical doctors working in public healthcare facilities in Ghana. Facilities represented three broad service levels: Primary-level facilities, including clinics and health centres where medical doctors provide primary care or supervisory clinical services; district and secondary hospitals and tertiary hospitals. These levels were selected because they reflect key points in the public health system where doctor retention affects continuity of care, referral quality and service delivery.

Eligible participants were fully qualified medical doctors who were actively involved in clinical service delivery and had worked long enough in the public health system to have experience of organisational reward practices. Doctors on internship, in temporary placement, on leave without regular clinical service or not directly involved in clinical care were excluded.

A stratified sampling approach was used to improve representation across healthcare facility levels. The strata were primary-level facilities, district and secondary hospitals and tertiary hospitals. Lists of eligible doctors were obtained through professional and facility-level networks, and recruitment links were circulated proportionately across the strata. Within each stratum, eligible doctors were invited electronically and were allowed to participate only once. Participation was voluntary, and no identifying information was collected.

The minimum sample size was assessed using statistical power and PLS-SEM requirements. For a medium effect size (f² = 0.15), alpha of 0.05, statistical power of 0.80 and a maximum of three predictors pointing to an endogenous construct, the minimum sample required was below the final achieved sample. In addition, PLS-SEM guidance recommends considering model complexity, indicator reliability and bootstrapping precision rather than relying only on the traditional ten-times rule.^15,16 The final sample of 400 valid responses was therefore considered adequate for estimating the measurement and structural models.

A total of 410 doctors were invited, of whom 400 submitted complete and usable questionnaires, giving a response rate of 97.6%. Records with substantial missing responses, duplicate entries or straight-line response patterns were excluded before analysis.

Data were collected using a structured, self-administered electronic questionnaire between August 2025 and November 2025. The instrument had two sections. Section A captured demographic and work-related characteristics, including age, sex, years of professional experience and facility level. Section B measured the latent constructs using adapted items from previously validated reward, reward perception and retention intention scales.^9,11,13,17 The items were adapted to fit the Ghanaian public healthcare context.

Total rewards were measured as a higher-order construct reflecting financial rewards and non-financial rewards. Financial rewards captured salary and allowances, while non-financial rewards captured promotion and career development opportunities. Reward adequacy measured perceived fairness, sufficiency and fit between rewards and doctors’ workload, responsibility and contribution. Reward expectation measured doctors’ expectations of future salary progression, allowances, promotion and career development. Medical doctor retention was measured through intention to remain in the current health system or organisation. All scale items were measured on a five-point Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree).

A pilot test was conducted with 20 doctors who were not included in the final study sample. The pilot assessed clarity, completion time, contextual relevance and item wording. Minor wording revisions were made to improve clarity without changing the conceptual meaning of the validated items.

Data were screened before model estimation. Cases with extensive missing data were removed. For retained cases, item-level missingness was checked and was minimal; therefore, complete-case analysis was applied for the final 400 responses. Response patterns were inspected for duplicate submissions and straight-line responses. Descriptive statistics, skewness and kurtosis were examined to assess distributional characteristics, while boxplots and standardised scores were used to inspect potential outliers. Because PLS-SEM is variance based and does not require multivariate normality, it was considered appropriate for the data and the model’s objective.¹⁵

Data were analysed using Statistical Package for the Social Sciences (SPSS) version 26 (IBM Corp., Armonk, New York, United States [US]) for descriptive statistics and SmartPLS version 4 (SmartPLS GmbH, Monheim am Rhein, Germany) for partial least squares structural equation modelling (PLS-SEM). The analysis followed a two-stage procedure. Firstly, the measurement model was assessed to determine whether the latent constructs were measured reliably and validly. Indicator loadings of 0.70 or higher were preferred although loadings between 0.60 and 0.70 were retained where construct reliability and content validity were adequate. Internal consistency was assessed using Cronbach’s alpha and composite reliability, with values of 0.70 or higher considered acceptable. Convergent validity was evaluated using average variance extracted (AVE), with AVE of 0.50 or higher indicating that the construct explained at least half of the variance in its indicators.

Discriminant validity was assessed using the Fornell–Larcker criterion and the heterotrait-monotrait ratio (HTMT). For the Fornell–Larcker criterion, the square root of each construct’s AVE was expected to exceed its correlations with other constructs. Heterotrait-monotrait ratio values below 0.85 were interpreted as evidence of discriminant validity. Multicollinearity was evaluated using the variance inflation factor (VIF), with VIF values below 5.00 considered acceptable.

Secondly, the structural model was assessed by examining path coefficients, standard deviations, t-statistics, p-values, coefficient of determination (R²) and mediation effects. Bootstrapping with 5000 resamples was used to assess the statistical significance of direct and indirect effects. The hypothesised paths were: Total rewards to retention, total rewards to reward adequacy, total rewards to reward expectation, reward adequacy to retention and reward expectation to retention. Mediation was assessed by comparing the total effect, direct effect and bootstrapped specific indirect effects through reward adequacy and reward expectation.

Model fit was considered using the standardised root mean square residual (SRMR), while explanatory power was interpreted using R² values for endogenous constructs. Given the predictive and theory-testing objective of the study, the main interpretation focused on reliability, validity, path coefficients, R² and bootstrapped indirect effects, consistent with PLS-SEM guidance.

Ethical clearance to conduct this study was obtained from the Topview Medical Centre Research and Ethics Oversight Committee (TMC–REOC) with reference number TMC/ERC/2026/01. Participation was voluntary. Written informed consent was obtained electronically before the questionnaire could be completed. Participants were informed about the study’s purpose, confidentiality, anonymity, voluntary participation and the right to withdraw before submission. No names, staff identification numbers or facility-specific identifiers were collected in the dataset used for analysis.

A total of 400 valid responses were included in the analysis. The demographic profile showed variation in sex, age, years of experience and facility level, supporting interpretation across different categories of doctors in the public health system. Table 1 summarises the participant characteristics.

Table 1 shows that the sample comprised 400 medical doctors. Males constituted the majority of respondents (n = 234, 58.5%), while females represented 41.5% (n = 166). Most participants were aged 26–35 years (n = 193, 48.2%), followed by those aged 36–45 years (n = 153, 38.2%), indicating that the sample was largely composed of early- to mid-career doctors.

Regarding professional experience, most respondents had less than 5 years of experience (n = 152, 37.9%) or 5–10 years of experience (n = 140, 35.0%). This suggests that a substantial proportion of the sample was still within the early stages of their medical careers, making reward adequacy and retention concerns particularly relevant. At the facility level, over half of the respondents worked in primary-level facilities (n = 209, 52.3%), followed by district and secondary hospitals (n = 105, 26.2%) and tertiary hospitals (n = 86, 21.5%).

Overall, the demographic profile reflects a predominantly early- to mid-career medical workforce drawn from different levels of Ghana’s healthcare system, providing an appropriate basis for examining total reward, reward adequacy, reward expectation and physician retention.

The measurement model showed acceptable reliability and validity, consistent with established PLS-SEM criteria for reflective measurement assessment.^15,16 Cronbach’s alpha and composite reliability values were above the 0.70 threshold, supporting internal consistency reliability. Average variance extracted values were above 0.50, supporting convergent validity. Variance inflation factor values were below 5.00, indicating that multicollinearity was not a concern. Table 2 presents the reliability, validity and collinearity results.

Discriminant validity was supported by the Fornell–Larcker criterion and HTMT. The square root of AVE for each construct exceeded its correlations with other constructs, and all HTMT values were below the recommended 0.85 threshold. These results indicated that the constructs were empirically distinct and suitable for structural model assessment.

In particular, TotalRD demonstrated satisfactory reliability and convergent validity, confirming the adequacy of the total rewards measurement. The retention intention (RET) dimension also showed strong construct validity, with high internal consistency and acceptable AVE, supporting its suitability for assessing retention intention in the structural model.

The structural model explained a meaningful proportion of variance in the endogenous constructs, as shown in Table 3. Total rewards explained substantial variance in reward adequacy, indicating that doctors’ perceptions of adequacy were closely related to the reward system. The variance explained in reward expectation was smaller, suggesting that future-oriented expectations may be influenced by other institutional factors beyond current rewards. The model also explained a meaningful proportion of variance in retention intention, showing that reward-related perceptions contributed to doctors’ stated intention to remain.

The direct path from total rewards to retention intention was positive but not statistically significant (β = 0.048, t = 0.663, p = 0.507), so H1 was not supported. Total rewards were positively and significantly associated with reward adequacy (β = 0.689, t = 17.674, p < 0.001), supporting H2 and aligning with evidence that employees’ evaluation of reward fairness and sufficiency is central to workforce outcomes.^5,11 Total rewards were significantly associated with reward expectation (β = −0.254, t = 4.068, p < 0.001), supporting H3 in statistical terms but indicating a negative direction.

Reward adequacy was positively and significantly associated with retention intention (β = 0.623, t = 11.773, p < 0.001), supporting H4 and reinforcing the argument that adequate and fair rewards can strengthen employees’ intention to remain.^6,9 Reward expectation was positive but not statistically significant (β = 0.045, t = 1.078, p = 0.281), so H5 was not supported. These findings indicate that doctors’ current perception of reward sufficiency was more important for retention intention than expectations of future rewards, Table 4 and Figure 1.

The mediation results showed that reward adequacy significantly mediated the relationship between total rewards and retention intention. The total effect of total rewards on retention was significant (β = 0.465, p < 0.001), but the direct effect became non-significant after reward adequacy and reward expectation were included (β = 0.048, p = 0.507) (Table 5). The specific indirect effect through reward adequacy was significant (β = 0.429, t = 8.884, p < 0.001), supporting H6 and indicating full mediation.

The indirect effect through reward expectation was not significant (β = −0.011, t = 0.893, p = 0.372), so H7 was not supported. This result indicates that future-oriented reward expectations did not explain the relationship between total rewards and retention intention in this model.

For primary health care delivery, the findings imply that retention strategies should prioritise fairness, sufficiency and transparency in reward systems. High doctor turnover can weaken continuity of care, reduce trust between patients and providers and increase pressure on remaining staff. Improving perceived reward adequacy may therefore contribute indirectly to service continuity and system resilience.

At the policy level, national and facility-level managers should move beyond periodic salary adjustments and evaluate how doctors perceive the full reward package. Practical strategies include transparent criteria for allowances, clear promotion pathways, consistent access to career development, timely communication of reward policies and routine assessment of perceived reward adequacy. These strategies may be particularly relevant in LMIC contexts where fiscal constraints limit continuous increases in monetary rewards.

A strength of the study is its focus on the perceptual mechanism through which total rewards influence retention, using PLS-SEM to assess both measurement quality and structural relationships. The sample size of 400 also provided adequate statistical power for estimating the model. However, the cross-sectional design limits causal interpretation. The results should therefore be interpreted as evidence of hypothesised theoretical associations rather than proof of causality. The use of self-reported data may also introduce common method bias although anonymity and electronic self-administration may have reduced social desirability pressure. Finally, the model focused on reward-related factors and did not include other potential determinants of retention, such as leadership, workload, organisational support or migration intention.

Future studies should use longitudinal designs to assess whether changes in reward adequacy predict actual retention over time. Qualitative research could also explore how doctors define adequate rewards in different facility levels and career stages. Further studies may test broader models that include workload, burnout, organisational support, leadership and migration intention alongside reward adequacy.