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

This study was conducted in Shandong Province, which is located in eastern China, with a population of about 100 million [29]. In 2016, the gross regional product of Shandong Province amounted to RMB 6802 billion (US$ 1024 billion), ranking it as the third largest economy within China [29]. However, there are only 0.27 million registered nurses in Shandong Province, and the ratio of doctors and nurses is 1:1.10 which is ranked the 16th in China [6].

There are eight medical universities/colleges in Shandong Province [30], and five are provincial independent medical universities/colleges. Considering the different education system, we excluded Shandong University of Traditional Chinese Medicine. Four medical universities/colleges, namely Weifang Medical University, Binzhou Medical University, Jining Medical University, and Taishan Medical College were included. The final year undergraduate nursing students were selected in this study because they were considering various employment choices but had not yet made final decisions [31]. A cluster random sampling method was used, and we aimed to recruit a minimum of 100 respondents from each medical university/college [32]. Accordingly, two or three classes in each university/medical college were randomly selected depending on the number of final year undergraduate nursing students in each class.

Establishing the job attributes and corresponding levels is the first and key step in DCEs and initial identification based on literature reviews. Location, monthly income, training opportunity, promotion opportunities, bianzhi (which refers to the established posts and can be loosely regarded as state administrative staffing) [33], facility quality, workload, management style, and organizational culture have been commonly adopted and suitable for the Chinese health care system [7, 21‐23, 26, 28, 31, 34‐36]. Then, in-depth interviews were conducted with five professional nurses from different hospital levels. As a result, the condition of working night shift was added as the potential attributes. Two focus groups (seven participants per group) were then conducted among the final year undergraduate nursing students from Weifang Medical University. They were asked to discuss which attributes were considered the relative important factors in their job selection until they reached a consensus for the final attributes and their levels. In this process, we used the new attribute of a job which required light working strength to replace both the workload and the condition of working night shift. Meanwhile, facility quality, management style, and organizational culture were generalized into the attribute of work environment. Based on literature reviews, in-depth interview, focus group discussions, and the final review with two specialists on DCEs, six attributes and their levels were finally included within the DCE (Table 1). The final six attributes include a monetary attribute (monthly income) and five non-monetary attributes: location, bianzhi, career development and training opportunity, work environment, and work strength.

A full factorial design will produce 3⁵ × 2 = 486 hypothetical job scenarios. A total of (486 × 485)/2 choice sets will be generated, and those are not feasible for a single individual to choose. Accordingly, 24 choice sets were constructed by using a D-efficient design (assuming zero priors for all attributes) with Ngene DCE design software [37]. In order to lessen the burden of participants, Ngene software was also employed to divide 24 choice sets into two versions. One of the choice sets in each version was included twice as a consistency test but the data of the repeated choice sets were not included in the final analysis. The two versions of questionnaires were randomly allocated to participants. An example of the choice sets was demonstrated in Table 2.

The study has been approved by the Ethics Review Board of the School of Preventive Medicine, Shandong University (Reference No. 20170301). Informed consent was obtained from all participants prior to questionnaire administration.

Responses were coded and double-entered into an electronic database using EpiData version 3.1 (EpiData Association, Odense, Denmark). Statistical analyses were performed using Stata version 13.1 (StataCorp LP, College Station, TX, USA). The general characteristics of the undergraduate nursing students were summarized as means and standard deviations or frequencies and percentages.

The theoretical basis for statistical analysis of DCE data is the random utility model. In this framework, it is assumed that individual n will opt for the option that is associated with the highest utility level. In this study, we analyzed the data using a mixed logit model which has the convenient properties of allowing for the preference heterogeneity and violation of the independence of irrelevant alternative assumption [18, 38]. All attributes were specified as having a random component. Monthly income was coded as a continuous variable, and other attributes were coded as dummy variables. Mixed logit models were used to analyze the variables using mixlogit command [39] and were specified with 500 Halton draws. The utility that individual n obtains from alternative i in choice set t is depicted as:where β_i is a vector of coefficients representing the relative preference weights for each attribute level on average, η_i reflects the degree of heterogeneity among respondents, and ε_nit is unobservable random component.

Willingness to pay (WTP) represents a monetary measure of participants’ valuation for a certain attribute. WTP was the ratio of the coefficient between each attribute level and the monthly income attribute. The positive or negative WTP results indicated the portion of monthly income that participants would be willing to pay or to be compensated for an attribute level. We also conducted a simulation study to predict that the uptake rates of undergraduate nursing students for rural versus city jobs change as the levels of job attributes are changed (policy interventions).

The general characteristics of respondents are presented in Table 3. There were no significant differences between participants who passed or failed the consistency test. The mean age of the analytical sample (n = 445) was 22.3 ± 1.2 years, and 410 (92.1%) students were between ages 21 and 24. Female (91.5%) were overwhelmingly predominant in the participants. Most of the students (69.4%) came from townships or villages, and 327 (73.5%) students had brothers or sisters. More than 90% of participants spent less than RMB 1500 per month, and approximately 60% participants indicated their annual family income as less than RMB 50 000. With regard to future plans, the majority of the students (n = 361) said they wanted to engage in nursing work after graduation.

Results from the standalone ranking of importance of job attributes (which was asked prior to the DCE tasks) is showed in Fig. 1. On average, monthly income was regarded as the most important job attribute, followed by location.

Table 4 shows the main results from the mixed logit model. The mean coefficients of all attributes were statistically significant and were of the expected sign. Among non-monetary attributes, the final year undergraduate nursing students expressed highest stated preferences for a job with light working strength (β = 1.013, P < 0.01), followed by an excellent work environment (β = 0.760, P < 0.01). Although a city location had a positive effect for the respondents (β = 0.339, P < 0.01), it did not appear to be as important as the other attributes.

The results of WTP are shown in Table 5 and are used for relative comparisons. They were willing to pay RMB 2096.2 (US$ 310.1) monthly income for a job with light working strength than a job with heavy working strength. If the work environment improved from poor to excellent, they were willing to forego RMB 1572.4 (US$ 232.6) per month. In addition, they put a value of only RMB 702.4 (US$ 103.9) per month on a work location in the city, as compared to work in a township or village.

The results of mixed logit models and WTP among different subgroups are showed in Table 6. Apart from the attribute of location, the preferences for other job attributes between different subgroups were relatively similar. Students who came from the county or city valued a city location very highly with a willingness to pay of RMB 1594.2 (US$ 235.8) (95% CI, 1037.2–2 185.9), compared to students who came from townships or villages with a willingness to pay of only RMB 366.6 (US$ 54.2) (95% CI, 63.3–662.2). Compared to students whose family income was less than RMB 50 000 (US$ 7396), students with a family income of more than RMB 50 000 (US$ 7396) were willing to pay RMB 611.3 (US$ 90.4) (95% CI, 530.7–709.2) extra monthly income for a job in the city.

Figure 2 shows the varying probabilities of taking a rural (township or village) job versus a city job, with various job conditions; the county job was omitted here but can be calculated based on the regression coefficients reported in the paper. It shows that the baseline probability of taking the rural job is 41.6%, and the probability of taking the city job is 58.4% (all else been equal). If monthly income increased from RMB 2000 (US$ 296) to RMB 5000 (US$ 740), the probability of choosing a rural job would increase to 75.2%. However, raising monthly income by another RMB 3000 (US$ 444) to RMB 8000 (US$ 1183), the probability would increase to 92.8%. The model predicted that a job with light working strength rather than heavy would increase the proportion of students opting a rural job to 66.2%. Improving the work environment from poor to excellent, career development and training opportunity from insufficient to sufficient and offering bianzhi would increase the probability of taking a rural job to 60.4%, 57.8%, and 57.7%, respectively.