Introduction
Health workforce is regarded as a prerequisite for an effective and responsive health system and is also considered to be the key determinant of access to health services (Araújo and Maeda, Reference Araújo and Maeda2013; Mohammadiaghdam et al., Reference Mohammadiaghdam, Doshmangir, Babaie, Khabiri and Ponnet2020). However, many countries are confronted with challenges in training, employing and deploying their workforce (Araújo and Maeda, Reference Araújo and Maeda2013). There are also imbalances in the geographic distribution of healthcare workers (HCWs) within countries (World Health Organization, 2006). Globally, it is estimated that between 51% and 67% of the rural population has limited access to basic health care (WHO, 2019).
Turkey ranks last among the Organisation for Economic Co-operation and Development (OECD) countries in terms of the total number of physicians per capita (Ministry of Health, 2021). Additionally, there is a distribution disparity between rural and urban areas. In order to address this maldistribution, a number of financial and non-financial incentives have been introduced. Another intervention is the mandatory service requiring physicians to work in the public sector for a minimum of 10–20 months depending on their field of service, with restrictions on working in the private sector if not fulfilled. Despite these interventions, the density of physicians in Western Anatolia is twice that of the South-eastern Anatolia region (Ministry of Health, 2019). Considering the demographic and economic structure, health indicators of Turkey have not yet achieved the targeted level. In terms of life expectancy at birth, women’s and children’s health, control of communicable diseases such as tuberculosis and non-communicable diseases, and risk factors, Turkey ranks in the middle among world countries. The geographical region in which people live affects their access to health services in our country (Üner and Okyay, Reference Üner and Okyay2020).
HCWs’ employment decisions are a function of their preferences and expectations. Policies for recruitment and retention of HCWs in underserved areas should include a bundle of incentives. In order to assess HCWs’ preferences and predict the job uptake given a set of job characteristics, discrete choice experiments (DCEs) can be conducted (Araújo and Maeda, Reference Araújo and Maeda2013).
DCE is a quantitative technique that assumes that goods and services can be described by their essential characteristics, and the value of a good or service can be derived from the combination of these characteristics (Ryan et al., Reference Ryan, Bate, Eastmond and Ludbrook2001). In recent years, DCEs have become increasingly utilised in health economics, providing policy-makers with a basis for decision-making. For instance, DCEs have been employed to assess population preferences for vaccination (Adams et al., Reference Adams, Bateman, Becker, Cresswell, Flynn, McNaughton, Oluboyede, Robalino, Ternent, Sood, Michie, Shucksmith, Sniehotta and Wigham2015; Dong et al., Reference Dong, Xu, Wong, Hung, Feng, Feng, Yeoh and Wong2020; Lack et al., Reference Lack, Hiligsmann, Bloem, Tünneßen and Hutubessy2020); primary health care (Kleij et al., Reference Kleij, Tangermann, Amelung and Krauth2017; Lim et al., Reference Lim, Ng, Teh, Ong, Sivasampu and Lim2022); cancer, antenatal and newborn screening programmes (Lee et al., Reference Lee, O’Leary, Umble and Wheeler2018; Vass et al., Reference Vass, Georgsson, Ulph and Payne2019); and tobacco control interventions (Regmi et al., Reference Regmi, Kaphle, Timilsina and Tuha2018). Furthermore, this method has been employed to measure the preferences of health professionals and other stakeholders regarding the provision of health care (Hill et al., Reference Hill, Fisher, Chitty and Morris2012, Reference Hill, Suri, Nash, Morris and Chitty2014; Leigh et al., Reference Leigh, Ashall-Payne and Andrews2020; Koopmanschap et al., Reference Koopmanschap, Stolk and Koolman2010). Another common application of DCEs is to determine the job preferences of HCWs. DCEs provide quantitative information on the relative importance of job characteristics influencing HCWs’ preferences, as well as the trade-offs between these factors and changes in the probability of choices if levels within factors are changed (WHO, 2012).
Aim
To examine (a) the job preferences and affecting individual characteristics (b) the salary students are willing to pay for desired working conditions and (c) to predict the impact of changes in job characteristics on the probability of choosing one job over another for mandatory service of senior medical students.
Methods
This is a cross-sectional analytical study. The target population consisted of the last-grade medical students of a medical faculty (n = 144).
Survey design
In order to ascertain the job characteristics (attributes) and levels, a literature review was conducted, and a semi-structured interview was carried out with 11 students. Six attributes, each with two to four levels were selected (Table 1). We opted to use the term ‘underdeveloped region’ instead of ‘rural area’ because in Turkey the urban–rural classification is based on population size (≤ 20 000 = rural; > 20 000 = urban) and does not accurately reflect the level of development.
Table 1. Job attributes, levels and definitions
HCWs = healthcare workers.
To construct efficient designs, Huber and Zwerina (Reference Huber and Zwerina1996) recommended utilising nonzero priors for parameter estimates. These prior values can be obtained from pilot studies. Uncertainty about priors should be taken into account as true parameter values can differ from assumed ones. The Bayesian design approach, introduced by Sándor and Wedel (Reference Sándor and Wedel2001), assumes a prior distribution of likely parameter values and optimises the design over that distribution (Sándor and Wedel, Reference Sándor and Wedel2001).
Using a large number of attributes in DCEs can increase complexity and cognitive burden, contributing to an increased error variance. To simplify decision-making, some of the attributes’ levels can be held constant in every choice set (Jonker et al., Reference Jonker, Donkers, de Bekker-Grob and Stolk2018). The profiles in such a choice set are called partial profiles. Kessels et al. constructed D-optimal partial profile designs using a Bayesian design algorithm that integrates the D-optimality criterion over a prior distribution of likely parameter values and implemented it in statistical software package JMP (Kessels et al., Reference Kessels, Jones and Goos2011).
We used JMP Pro 14 (SAS Institute, Cary, NC) to generate 12 choice sets, each consisting of 2 profiles for pilot study. At least two attributes were held constant in each choice set. This approach is common in DCEs in health economics, which typically have 16 or fewer choice sets with 4–6 attributes (de Bekker-Grob et al., Reference de Bekker-Grob, Ryan and Gerard2012). Since the participants are physician candidates who have mandatory service obligations, ‘opt out’ or ‘status quo’ alternatives were not included in the design. A pilot study was conducted with 11 students to determine the prior values. Based on this prior information, the final choice design was constructed. Three different versions of the choice design were generated to improve design efficiency. Each version contained 12 choice sets and 24 profiles with different combinations of attribute levels. Additionally, to identify respondents whose preferences violated common rationality, a choice set was inserted between the 12 pairs. This choice set had the same levels for all attributes except for salary. The job offering a higher salary was expected to be chosen. This choice set was not used in the main regression analysis.
The DCE tool also included questions on respondents’ socio-demographic characteristics and attitudes towards mandatory service. All participants received one version of questionnaire online and were asked to select one of the two job scenarios from each choice set. Data were collected during January–March 2021.
Respondents
Sample size was calculated as 84 using Johnson and Orme’s method (Johnson and Orme, Reference Johnson and Orme2003; Orme, Reference Orme1998). We aimed to reach all last-grade medical students of the faculty without selecting a sample to conduct subgroup analysis.
Data analysis
All data from the DCE questionnaires were stored using Microsoft Excel 2016 (Microsoft Corporation, USA). The general characteristics of the students were summarised as median (min–max) or frequencies and percentages. Salary was coded as a continuous variable, and other attributes were dummy coded, with 1 representing their presence in each profile and 0 representing their absence. Following this, the mixed logit (MXL) model was used to estimate participants’ preferences for the different levels of the job attributes using Stata® 15.0 (Stata Corporation, USA) with user-written codes (Hole, Reference Hole2013). The MXL model accounts for the panel data, allowing for multiple observations from each respondent (Hauber et al., Reference Hauber, González, Groothuis-Oudshoorn, Prior, Marshall, Cunningham, IJzerman and Bridges2016). Furthermore, the model accommodates heterogeneity in preferences across the sample by treating coefficients as random. In our study, the salary attribute was specified as fixed to facilitate the calculation of willingness to pay (WTP), while all other attributes were specified as having a random component. To explain the sources of heterogeneity, interactions of gender, hometown, income, having a HCW parent and willingness to perform mandatory service with attributes were included. The model presented in Table 2 includes the interaction terms which were statistically significant. The changes in the probability of choices were calculated using Hole’s ‘mixlpred’ command, in which the levels of attributes were altered. Additionally, the monetary value of attribute levels, namely, WTP and confidence intervals, was estimated using ‘wtp’ command in Stata (Hole, Reference Hole2013).
Table 2. The general characteristics for last-grade medical students
HCW = healthcare worker.
External validity
The questionnaire was also delivered to students from other medical faculties. We performed a 1:1 propensity score matching in IBM SPSS Version 25.0 to include students from other faculties who best matched. The propensity score was calculated based on participants’ gender, age, marital status, income, hometown, having a HCW parent, willingness to perform mandatory service and intention to pursue specialisation. To assess external validity, the results from two groups were compared by calculating the Kappa coefficient (Parady et al., Reference Parady, Ory and Walker2021).
Findings
Of the 144 medical students who were recruited, 107 (%74.3) respondents completed the questionnaire. A total of five (3.5%) respondents failed the rationality test. The estimated models with and without these respondents did not differ significantly, and thus, these respondents were retained in the main analysis. None of the students exhibited a dominant preference, indicating that they all trade off attribute levels.
Table 2 presents the characteristics of the respondents. 56.1% of the participants were female with a median age of 24 years. All of the respondents were single. Only 6.5% of medical students had a rural background (ie, had grown up in a village). Approximately half of the students reported their income status as income equal to expenditure. 20.6% of the students had a HCW parent. While 66.4% of the participants indicated that they would perform mandatory service as general practitioners, almost all of them were planning to pursue specialisation in the upcoming years.
Table 3 presents the MXL model which includes main effect and interaction terms. The MXL model indicates that students exhibited a preference for employment in a hospital or community health centre (CHC) over a military medical centre [β (S.E.) = 0.71 (0.23); P < 0.01, β (S.E.) = 0.88 (0.23); P < 0.001, respectively]. A higher salary [β (S.E.) = 6 × 10−4 (6 × 10−5) per Turkish lira (TRY); P < 0.001], normal workload [β (S.E.) = 1.81 (0.22); P < 0.001] and an ideal working environment [β (S.E.) = 1.27 (0.19); P < 0.001] significantly increased the likelihood a job being selected. Students demonstrated a preference for facilities located in developed regions and closer to their family/friends [β (S.E.) = 0.6 (0.16); P < 0.001, β (S.E.) = 1.47 (0.19); P < 0.001, respectively].
Table 3. Estimation results from the mixed logit model
CHC = community health centre; HCW = healthcare worker; LR, log-likelihood ratio; AIC: akaike information criterion; BIC: Bayesian information criterion.
* P < 0.05.
** P < 0.01.
*** P < 0.001.
The regression results indicate the presence of significant unobserved preference heterogeneity between respondents (as evidenced by the significant standard deviation of the random attribute coefficients). To elucidate the sources of this heterogeneity, a model was also estimated in which participant-specific characteristics were permitted to interact with job attributes. The log-likelihood ratio test [χ2 (df:7) = 50.727, P < 0.001] rejected the null hypothesis that the regression parameters for the MXL model and the MXL model with interactions are equal at 0.5% significance level, indicating that the model fit has improved with an R2 of 0.073. The results of the MXL model with interactions suggest that males place a greater value on the development status of the region more than females. Those with a parent employed in the healthcare sector exhibit a stronger preference for working at a hospital or CHC. There is a greater inclination to work at a CHC and for a higher salary among students willing to perform mandatory service, while others value the development status of the region more. The MXL model with interaction terms also yielded significant derived standard deviations for workload, working environment and proximity to family/friends indicating the existence of unobserved heterogeneity for these attributes.
Table 4 presents the WTP values, which can be described as the salary students would be willing to sacrifice for improvements in job characteristics. The respondents indicated a WTP of 1102.6 TRY (95% CI, 385.1–1820.1 TRY) to work at a hospital and 1372.9 TRY (95% CI, 645–2100.7 TRY) at CHC, 2818.8 TRY (95% CI, 2160.3–3477.3 TRY) for a normal workload, 1968.5 TRY (95% CI, 1401.0–2536 TRY) for an ideal working environment, 2287.5 TRY (95% CI, 1752.8–2822.2 TRY) for a workplace nearer to family/friends and 930.2 TRY (95% CI, 486.5–1373.8 TRY) for a job located in a developed region (the exchange rate from February 2021 of US$1 = 7.07 TRY, 1 Euro = 8.56 TRY) (TCMB Central Bank of the Republic of Turkey Currency Exchange Rates., 2021).
Table 4. Willingness to pay estimates for job attributes
CHC = community health centre.
Figure 1 illustrates the varying probabilities of accepting a position in a developed versus an underdeveloped region, contingent upon the specific job conditions. Ceteris paribus, the probability of accepting a position in an underdeveloped region is 46%, whereas the probability of accepting a position in a developed region is 54%. The probability of choosing a hospital in an underdeveloped region is 51% and that of choosing a CHC in an underdeveloped region is 52%, in comparison to a military medical centre in a developed region. The probability of selecting a job in an underdeveloped region would increase to 57%, 55% and 56%, respectively, if the workload were to be improved, if the location were to be closer to family or friends and if the working environment were to be more favourable. An increase in salary from 6500 TRY to 8000 TRY would result in a 52% probability of choosing an underdeveloped region. The model predicted that introduction of a normal workload with 11 000 TRY per month salary rather than heavy workload with 6500 TRY per month salary would increase the proportion of students opting for a job in an underdeveloped region to 75%.
Figure 1. Probabilities of taking a job located in a developed region versus underdeveloped region with changing job conditions
Indicating the external validity, kappa coefficients for all choice sets were significant, and total kappa score was 0.819 (P < 0.001).
Discussion
This DCE has elicited preferences for job attributes among the senior medical students. All six attributes significantly affected the students’ job choices. They preferred to work at a hospital or CHC, closer to family/friends in a developed region with a higher salary, a normal workload and an ideal working environment. The MXL model estimates revealed the existence of preference heterogeneity in workload, working environment and proximity to family/friends. The gender of the respondents, the presence of a HCW parent and the willingness to perform mandatory service were found to affect the preference weights of certain job characteristics.
Similar with other studies, salary was found to be the most important factor influencing job preferences (Karyani et al., Reference Karyani, Matin, Malekian, Rotvandi, Amini, Delavari, Soltani and Rezaei2020; T. Liu et al., Reference Liu, Li, Yang, Liu and Chen2019; Vujicic et al., Reference Vujicic, Alfano, Shengelia and Witter2010). Additionally, students who were willing to perform mandatory service demonstrated a stronger preference for higher salaries. Given that the majority of students expressed a desire to pursue specialisation training, it can be hypothesised that the motivation for performing mandatory service may be financial. In 2017, the Turkish Ministry of Health conducted a survey to examine the job satisfaction of healthcare staff. The findings of this study indicated that salary is the most important factor influencing job satisfaction and that it is one of the HCWs’ strongest demands to be regulated in the healthcare system (Health Personnel Satisfaction Survey, 2017). In another DCE conducted in Turkey, salary was ranked as the second most important attribute among general practitioners (İşlek and Şahin, Reference İşlek and Şahin2023). According to these findings, providing economic incentives should be a priority.
Workload was the most significant non-monetary attribute influencing job preferences. This finding aligns with previous DCEs, which have demonstrated that HCWs are reluctant to accept heavy workloads and value having adequate leisure time (Rafiei et al., Reference Rafiei, Arab, Rashidian, Mahmoudi and Rahimi-Movaghar2015; Scott et al., Reference Scott, Holte and Witt2020; Sivey et al., Reference Sivey, Scott, Witt, Joyce and Humphreys2012). In the Turkish Healthcare staff job satisfaction survey, half of the respondents claimed to have a heavy workload (Health Personnel Satisfaction Survey, 2017). Islek et al. reported that workload has a significant effect on the job preferences of physicians under the age of 35 years (İşlek and Şahin, Reference İşlek and Şahin2023). It is suggested that young physicians increasingly prioritise work–life balance and believe that they do not have to work as much as previous generations to make a living (Bao and Huang, Reference Bao and Huang2021; Harding et al., Reference Harding, Seal, McGirr and Caton2016; Matthews et al., Reference Matthews, Seguin, Chowdhury and Card2012). Furthermore, the majority of respondents indicated that they planned to pursue specialisation. A normal workload was perceived to mean more free time to study for the residency examination for them. This result is consistent with DCE studies conducted in China, Mozambique and Kenya which found that career development and training were regarded as important attributes of job preferences (Honda and Vio, Reference Honda and Vio2015; S. Liu et al., Reference Liu, Li, Yang, Liu and Chen2018; Takemura et al., Reference Takemura, Kielmann and Blaauw2016).
The proposed study showed that proximity to family/friends of job also had a substantial effect on job preferences, in line with the studies carried out in Australia and Canada (Harding et al., Reference Harding, Seal, McGirr and Caton2016; Matthews et al., Reference Matthews, Seguin, Chowdhury and Card2012; Szafran et al., Reference Szafran, Crutcher and Gordon Chaytors2001). The lack of social support causes depression and burnout among physicians (Kuhn and Flanagan, Reference Kuhn and Flanagan2016). Furthermore, only 4% of the participants stated that they intend to complete their mandatory service. Since this is a relatively short-term and temporary period for the majority of the participants, they might take into account the housing conditions and prefer closer workplaces to their family and friends.
Similar to previously reported DCEs, our respondents had a preference for an ideal working environment (Awases et al., Reference Awases, Gbary, Nyoni and Chatora2004; Zurn et al., Reference Zurn, Dal Poz, Stilwell and Adams2004). This finding also concurs with studies that have identified concerns among medical students regarding their professional competence (Aker and Mıdık, Reference Aker and Mıdık2020; Ergin et al., Reference Ergin, Utku Uzun and Topaloğlu2016; Yalçinoğlu et al., Reference Yalçinoğlu, Kayi, Işik, Aydin, Zengin and Karabey2012). It can be reasonably assumed that newly graduated physicians will expect their employers to provide them with a supportive management structure, as well as the opportunity to consult, refer and collaborate with specialists and more experienced colleagues.
Although the facility had a significant effect on job preference, there was a substantial heterogeneity among the respondents. According to our findings, students with an HCW parent were more likely to work at CHC and hospital. Students who were willing to perform mandatory service were also more likely to work at CHC. The majority of the health workforce is employed in hospitals and CHCs in our healthcare system. All of the medical faculties include rotations to these facilities in their training programme. As a result, students are expected to be more familiar with the working conditions of these facilities. In hospitals, general practitioners are frequently employed in emergency services and are required to cope with the stress associated with night shifts (Ağapınar and Şahin, Reference Ağapınar and Şahin2014). Furthermore, there are no on-calls or night shifts in CHCs, and the risk of malpractice is relatively low. Consequently, CHCs have a higher preference weight than hospitals.
The developmental status of the work location was also valued by the participants. This finding is consistent with other researches that suggest that HCWs tend to prefer centrally located jobs (İşlek, Reference İşlek2021; İşlek and Şahin, Reference İşlek and Şahin2023; Kolstad, Reference Kolstad2011; S. Liu et al., Reference Liu, Li, Yang, Liu and Chen2018; Smitz et al., Reference Smitz, Witter, Lemiere, Eozenou, Lievens, Zaman, Engelhardt and Hou2016). Rural and remote areas are perceived as less desirable due to limited educational opportunities for children, inadequate infrastructure (communication and transportation) and limited career options for spouses (Lehmann et al., Reference Lehmann, Dieleman and Martineau2008; S. Liu et al., Reference Liu, Li, Yang, Liu and Chen2018). Some regulations have been enacted to address this issue. For instance, the duration of mandatory service and the amount of additional payments vary depending on the developmental status of the area (Basic Health Services Law, 1987). Nevertheless, these incentives are insufficient to address the shortage of physicians in underdeveloped areas in our country. The MXL model with interaction terms indicated that males and students who were unwilling to perform mandatory service valued developmental status more than others. As there are differences across the studies, gender is not a consistent predictor for choosing a rural post (Isaac et al., Reference Isaac, Walters and McLachlan2015; Jones et al., Reference Jones, Humphreys and Prideaux2009; Kim et al., Reference Kim, Ngo and Playford2020; King et al., Reference King, Purcell, Quinn, Schoo and Walters2016; Playford et al., Reference Playford, Evans, Atkinson, Auret and Riley2014; Puddey et al., Reference Puddey, Mercer, Playford, Pougnault and Riley2014). Further investigation is required to ascertain the extent to which other factors contribute to this association.
Conclusion
Our study indicates that monetary incentives are crucial to recruiting newly graduated physicians where they are mostly needed. Bundles of both monetary and non-monetary incentives, tailored to individual characteristics, would be more efficient than a single intervention.
In our country, primary healthcare services, catering to both the community and individuals, are primarily provided by general practitioners. Family medicine positions were not included in this study due to their contractual nature. However, students expressed a preference for working in primary healthcare institutions. Nevertheless, nearly all participants expressed a keenness for specialisation. This tendency could precipitate a rapid turnover of physicians, leading to service disruptions. The results of this study offer valuable insights for crafting incentive schemes aimed at attracting and retaining physicians in primary healthcare settings. Similar study frameworks could be devised for specialist physicians and other healthcare professionals across various institutions (family medicine/CHC/provincial health directorate) and fields (communicable diseases/environmental health/vaccination/non-communicable diseases/reproductive health/occupational health). Furthermore, the cost-effectiveness of different incentive schemes can be calculated in future works.
This is the first study using DCE methodology to investigate the job preferences of medical students in our country. Another strength is our utilisation of a pilot survey to create prior values for the coefficients in our experimental design. To control the capability of accurate prediction of our model, we conducted an external validity analysis. It is assumed that respondents apply compensatory decision rules in DCEs. Hence, dominant preferences have been checked.
This study has several notable limitations. Firstly, since this is a single-centre research, the results cannot be generalised to the whole country. Secondly, due to the hypothetical nature of DCEs, there may be disparities between revealed and stated preferences. Thirdly, DCEs imply a certain degree of simplification to limit the number of job attributes and levels. Therefore, many other job characteristics that are likely to affect a HCW’s employment decisions may have been overlooked. It is recommended that policy-makers should validate DCEs’ findings before implementing a specific bundle of interventions (Araújo and Maeda, Reference Araújo and Maeda2013).
Data availability statement
The data that support the findings of this study are available on request from the corresponding author.
Author contributions
All the authors declare that they have made substantial contributions to the study’s conception and design, the data analysis, the article’s drafting and critically scrutinising its content, and the approval of the article’s final draft to be published.
Funding/Support
This research received no specific grant from any funding agency in the public, commercial or not-for-profit sectors. On behalf of all authors, the corresponding author states that there is no conflict of interest.
Ethical considerations
After construction of the final choice design, ethical approval was obtained from Izmir Katip Celebi University (approval number: 2020/13-05 and 2021/02-15). The study was undertaken with permission from the rectorate of the university.
Open access
