3.1 Study Design and Participants
An online self-report cross-sectional multicentre study was conducted from November 2022 to January 2023; we collected data from many different individuals at a single point in time without any external influence[30]. This article was written in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines[31].
Our target population was new nurses working in hospitals in Hunan Province. New nurses were operationally defined as nurses who had entered the workforce within the previous two years[32]. Therefore, registered nurses willing to participate in who had obtained a nursing certificate and had been working in the clinic for ≤ 24 months were included in this study. Trainee nurses, refresher nurses, visiting scholars and those who could not continue participating for various reasons during the survey period were excluded.
3.2 Data Collection and Bias
This study was conducted in the hospitals of Hunan Province, China, at which 29,700 new registered nurses began working from January 2020 to April 2022, representing an increase of 12% across the two-year period[33]. A two-stage convenience sampling method was adopted for the study. First, we selected 10 cities from different geographical regions of Hunan Province (3 in the centre, 2 in the north, 1 in the west, 2 in the east and 2 in the south). Second, we selected 3 tertiary hospitals (hospitals with over 501 beds that provide high-level medical and health services, higher education and scientific research to several regions) and one secondary hospital (101–500–bed regional hospital providing comprehensive health services, teaching and research to multiple communities) from each city. As a result, 10 cities (Changsha, Hengyang, Shaoyang, Yiyang, Loudi, Chenzhou, Xiangtan, Zhuzhou, Yongzhou and Yueyang) and 33 hospitals (24 tertiary hospitals and 9 secondary hospitals) participated. The online questionnaire was produced through an online survey platform called WenJuanXing (https://www.wjx.cn/). We invited hospital administrators to display recruitment posters for our study on their bulletin boards and share them among their work groups; participants were provided with QR codes for self-registration, which they scanned to enter our WeChat group. The research team described the research content, data privacy and survey participation requirements in the WeChat group and then sent the survey link to each participant after obtaining informed consent.
We made the following efforts to address potential sources of bias. At first, participants were selected from central, northern, southern, western, and eastern Hunan provinces to reduce selection bias. The initial response rate in the western region was low but increased after we reposted the recruitment information weekly. Second, to reduce information bias, we required the participants to submit responses to all questions to prevent missing values.
3.3 Sample Size
We calculated the sample size necessary for this cross-sectional survey using the formula N =(Z1−α/2σ/δ)2, where α indicates the significance level, i.e., 0.05, Z1−α/2 is 1.96, σ indicates the standard deviation of the population and δ indicates the allowable error. In a previous study, the mean score and standard deviation of nurses' job embeddedness score was 22.87 ± 4.18[34]; in this study, δ was 22.87 x 2%=0.4574. Therefore, the minimum sample size required was 321; however, considering a 20% potential dropout rate, the final sample size selected was 386.
3.4 Variables and Measures
In this study, the dependent variable is job embeddedness, while the predictive variables are nursing work environment, head nurse leadership and presenteeism. We also chose sex, age, hospital grade, weekly working hours, weekly number of night shifts, monthly salary level and independent occupation as confounding variables in this study. These variables were selected according to Mitchell's theory of job embeddedness [8] and a review study on job embeddedness-related variables[19].
The nursing work environment was measured by the Nursing Work Environment Scale developed by Shao Jing et al[35], which comprises 26 items across 7 dimensions of career development (leadership and management, medical relationship, recognition atmosphere, professional autonomy, basic security and sufficient manpower). The scale adopts a Likert 6-level scoring method, with points 1 and 6 representing "strongly disagree" and "strongly agree", respectively. The sum of the scores of each item reflects the status of the total scale or corresponding subscale. The higher the score is, the better the corresponding environmental characteristics will be. The Cronbach's α coefficient of the scale was 0.946, and the split-half reliability was 0.894. In this study, Cronbach's α coefficient of the working environment scale was 0.982. The principal component analysis was conducted using the variance rotation method to extract 7 common factors from the study samples, accounting for 87.525% of the variance, and the factor loading range was 0.798–0.945.
The job embeddedness of the new nurses was evaluated by the Global Job Embeddedness Scale compiled by Crossley et al[36] and revised by Mei Hua et al[37]. The scale comprises seven single-dimension items, and a Likert 5-level scoring method was adopted: 5 means strongly agree, and 1 means strongly disagree. Items 4 and 6 are reverse-scored on a scale of 7 to 35, with a higher score indicating a higher level of job embeddedness. This scale has been widely used among nurses with a Cronbach's α of 0.845–0.864[34, 38]. In this study, the Cronbach's α coefficient of the scale was 0.775, and the variance rotation method was used to extract a common factor from the sample in the principal component analysis to explain 76.414% of the variance. The loading range of the other factors was 0.666–0.837.
We used the head nurse leadership scale developed by Huang Chunmei et al[39] to evaluate head nurse leadership. This scale includes 6 dimensions: charisma, affinity, foresight, influence, decisiveness and power of control. Forty-four items were scored using a 5-point Likert scale (1 = never, 2 = occasionally, 3 = partially, 4 = often, 5 = always). The higher the total score is, the stronger the leadership of head nurses will be. The Cronbach's α of the scale was 0.988, the retest reliability of the questionnaire was 0.791, and the correlation coefficient between each item and the population was 0.597–0.854. In this study, the Cronbach's α of the scale was 0.994, and 5 common factors were extracted from the principal component variance rotation analysis, with an explanatory variance of 89.446%. The factor loading range was 0.822–0.934.
The Stanford Presenteeism Scale was used to estimate the presenteeism of new nurses; this scale was developed at Stanford University in the United States[24] and translated into Chinese by Zhao Fang et al[40]. The scale comprises 6 items assessed using a Likert 5-level scoring method ranging from 1 to 5 points (completely disagree to completely agree). The higher the score is, the higher the degree of recessive absenteeism will be, and the greater the productivity loss. Cronbach's α of each dimension of the scale ranged from 0.76 to 0.90. In this study, Cronbach's α of the scale was 0.919. The variance rotation method was used in the principal component analysis to extract a working factor from the study sample, explaining 92.035% of the variance. The factor loading range was 0.805–0.950.
3.5 Quantitative and Statistical Analysis
We assigned binary data (sex, independent practice, and hospital grade) values of 0 and 1. We arranged the hierarchical data (weekly working hours, weekly night shifts, and wage level) in order from 1 to 4 and treated them as continuous data in the subsequent statistical analyses.
Descriptive statistics were used for the sociodemographic characteristics and clinical performance levels of the participants. Count data, such as educational level, hospital grade, and whether the nurse worked independently, and ordinal data, such as weekly working hours, monthly income and weekly night shifts, were described as frequencies and constituent ratios. Quantitative data with absolute values of kurtosis and skewness less than 2, such as age, job embeddedness, the nursing work environment, head nurse leadership and presenteeism, were described as the mean and standard deviation [41]. The data analyses were performed using PROCESS (version 3.5). According to the recommendations offered by Hayes [42], we examined the key research questions with sequential mediation model analyses. Before model construction, biserial correlation and Pearson’s correlation coefficients were used to examine the correlations between all variables to discover correlation paths and possible confounding variables. Next, we used model 6 to examine the sequential mediating effect of head nurse leadership and presenteeism on the relationship between variables. To ensure the credibility of the results, we (1) standardized all data; (2) used bias-corrected 5000 bootstrapped confidence intervals to assess the statistical significance of the indirect mediation effects; and (3) controlled for confounding factors.