Job preferences of undergraduate pharmacy students in China: a discrete choice experiment

Developing attributes and levels is a key step in DCEs [33, 34]. Both a literature review and qualitative studies were conducted to ensure that the attributes and levels included were the most meaningful for respondents. First, we identified 9 attributes from a literature review, including monthly income, work environment, work location, bianzhi (which refers to the established posts and can be loosely regarded as state administrative staffing), workload, training opportunity, career development opportunity, management style, and welfare [23‐27]. Next, we consulted 3 clinical pharmacists, who provide clinical pharmacy services in hospitals and have rich work and research experience. Based on their suggestions on the importance and feasibility of attributes and levels, we removed two attributes (the workload and bianzhi), added in one new attribute (the years to the promotion), and adjusted the levels of monthly income attribute. In addition, we also consulted 3 researchers with experience in conducting DCEs on human resources for health, and they suggested that two attributes (welfare and career development opportunities) could be removed. Then, we conducted 6 in-depth interviews with students from Shandong University of Traditional Chinese Medicine and Weifang Medical University. Based on their feedback, the work environment attribute was removed because most students thought it was not important. Finally, a focus group discussion (with 7 undergraduate pharmacy students) from Shandong University was conducted. Participants were asked to discuss the remaining 5 attributes and levels until they reached a consensus on the final version of the attributes and levels. They were also provided with an opportunity to add in additional attributes they considered important but not on the list. One additional attribute (and its corresponding levels), i.e., the work unit (which refers to the nature of the workplace), was added at the end. The final 6 attributes and levels are shown in Table 1.

Of the 6 attributes, 2 are two-level, 3 are three-level, and 1 is four-level. A full factorial design will produce 432 (= 2² × 3³ × 4¹) hypothetical scenarios and 93,096 (=  (432 × 431)/2) pairwise choice tasks. A D-efficient design was used to generate 24 manageable choice sets using DCE design software Ngene 1.1.2 (Choice-Metrics, Sydney, Australia) [35]. To reduce the response burden of respondents, the 24 choice sets were further divided into two blocks. An opt-out was included in the second-stage question after each DCE task to allow for unconditional choices [36]. In the first step, respondents were asked to choose the job they preferred from two hypothetical jobs, and in the second step, they were asked to answer whether they would take the job if it were available in real life. An example of a DCE choice set is shown in Table 2. To check for internal consistency, one choice set was duplicated. Hence, in total, every respondent answered 13 DCE questions. Respondents who failed the consistency test were excluded from the main analyses following the previous literature [37, 38].

Final-year undergraduate pharmacy students were chosen as the targeting respondents in this study given that the questions were highly relevant for these students, as they would be on the job market soon. A multistage cluster sampling design was used. Firstly, we selected 6 provinces to represent eastern (Hebei, Shandong and Jiangsu), middle (Henan) and western (Shaanxi and Ningxia) China according to the level of economic development and the geographical location. Secondly, a representative university offering pharmacy degree was selected in each province according to the type of universities, including 3 general universities (Shandong University, Henan University, Xi’an Jiaotong University) and 3 medical universities (Hebei Medical University, China Pharmaceutical University, Ningxia Medical University). Finally, 1 to 2 graduation classes of students majoring in pharmacy were randomly selected from each school according to the sample size requirements of each region. The sampling map is shown in Fig. 1.

The sample size for DCE is not straightforward and depends on many factors, such as the number of attributes and levels, the number of choice tasks, and the accuracy of expected results [39]. A rule of thumb based on the number of attributes and levels is commonly used to estimate the sample size [40]. Accordingly, it was calculated that the sample size required for this study should be more than 83 respondents (500*4/ (2*12) = 83). Combined with the sample size requirement of de Bekker-Grob [41] (i.e., how to calculate the sample size for healthcare-related DCE studies) and the minimum sample size requirement proposed by Pearmain et al. [42], the sample size should be no less than 100 respondents. A review of domestic studies on discrete choice experiments in human resources for health found that about 70% of the studies had a sample size of less than 500 people. Therefore, we decided that the sample size in eastern, middle and western China should be more than 100, and the total sample size should be no less than 500.

Before the formal survey, we conducted a pre-test among final-year undergraduate pharmacy students at Shandong University. In the pre-test, we tested whether the attributes and corresponding levels were reasonable and easy to understand. Minor adjustments were made based on the pre-test. Finally, a face-to-face anonymous survey was conducted from April to July 2017.

The questionnaire consists of two parts: section one collects personal background information, and section two is the DCE. Prior to the survey, we obtained oral informed consent from all respondents. This study was approved by the Ethics Review Board of the School of Preventive Medicine, Shandong University (Reference No. 20170301).

The DCE data were analysed using a conditional logit (CL) model (which assumes a homogeneous preference among respondents) and a mixed logit (MIXL) model (which allows for potential preference heterogeneity). Based on the random utility framework, the utility function can be expressed as:

where U_njt refers to the utility obtained by respondents n by choosing alternatives j in choice scenario t. U_njt consists of the observable component V_njt and the unobservable component ε_njt. The observable component is equal to the attribute level vector X_njt multiplied by the coefficient vector \(\beta ^{\prime}_{n}\)_, and the unobservable component is a random error term. Except for monthly income, which was treated as a continuous variable for the calculation of willingness to pay (WTP), all other attributes were coded as dummy variables [43]. WTP is calculated by \(- \frac{{\mathop \beta \nolimits_{{\text{q}}} }}{{\mathop \beta \nolimits_{m} }}\), where β_m is the monthly income coefficient and β_q is the coefficient for attribute levels q [44]. In this context, the WTP shows the relative monetary value that pharmacy students place on different job characteristics, which will facilitate our understanding of the relative importance of nonmonetary attributes in DCEs.

In MIXL, coefficients of attribute levels are usually assumed to follow a normal distribution (described based on a mean coefficient and a standard deviation). The mean coefficients reflect the relative preference weights, and the standard deviation reflects the extent of preference heterogeneity [45, 46]. The choice between CL and MIXL was guided by model fit statistics, including log-likelihood ratio tests, Akaike information criterion (AIC) and Bayesian information criterion (BIC). The distance between the best and worst preference weights within each attribute can be used to compare the relative importance of different attributes to the respondents. Subgroup analyses were also conducted. After estimating the regression coefficients, a series of policy simulations were conducted to predict the probability of job choices given the changes in attributes levels, the results of which would be of interest to policymakers. Descriptive statistics were also presented. All statistical analyses were conducted using Stata software version 15.1 (StataCorp LP, College Station, TX, USA).

Among the attributes, monthly income was the most important attribute to the respondents, followed by work location, whereas training opportunities and years to promotion were less important. For example, among 500 respondents, more than one-third regarded monthly income as the most important attribute, and nearly one-third considered training opportunities to be the least important attribute. The relative importance of attributes from analysing the ranking question is shown in Fig. 2.

Both the CL and MIXL models were initially applied for data analysis. Based on information criteria (i.e., the log-likelihood ratio test, AIC and BIC), the MIXL estimates were preferable and are reported in Table 3, while the CL estimates can be found in Additional file 1: Table S1. According to the MIXL estimates, all attribute levels were statistically significant except management style. On average, respondents preferred jobs in the city, with high monthly income, to work in public institutions, with sufficient training opportunities and fewer promotion years. The statistically significant SDs indicate that preference heterogeneity existed among respondents in all significant attributes. See Table 4 for details.

We found a clear preference among respondents regarding work location. Respondents were willing to pay 2889 yuan (US$427) for a job in the city over a job in the county or village. If the training opportunities increased from insufficient to sufficient, respondents were willing to pay 1166 yuan (US$172) per month. Furthermore, they were willing to pay almost the same money to work in public institutions or to shorten the promotion years from 5 to 2 years.

The changes in the probability of taking a job in rural areas (township or village) are shown in Fig. 3. We set the monthly income of 3000 yuan (US$444), private enterprises, insufficient training opportunities, and 5 years to promotion as the baseline scenario. The initial probability of taking a job in the city was 0.869, whereas the probability of taking a job in rural areas was only 0.131. In the policy simulation analyses, holding all else the same, when the monthly income increased from 3000 yuan (US$444) to 9000 yuan (US$1332) and other factors remained constant, the probability of working in rural areas increased to 0.884. The effect of a single noneconomic incentive for taking a rural job was not very obvious. For the multiple incentives, the combination “③ + ⑤ + ⑥ + ⑦” was the most effective, and the probability of choosing a job in rural areas was 0.942.

The detailed results of the subgroup analyses are shown in Additional file 1: Tables S4–S8, and the WTP results for the subgroups are shown in Table 5. We found that the respondents who were only children or were from the city were more likely to work in the city (based on the calculated WTP values, 95% CI). In addition, female students preferred to work in the city compared with male students. Open management was statistically significant for males but not for females, and students in eastern and central China placed more value on public institutions. Regarding training opportunities, students in the east valued them more, as they were willing to pay 1306 yuan (US$193) to obtain sufficient training opportunities. Students in the middle of the country paid more attention to years to promotion and were willing to pay 1683 yuan (US$249) to reduce the years to promotion from 5 to 2 years.