Study design and data collection
This study carried out a cross-sectional survey in the elderly caring social organizations of Anhui province, China. It was conducted between November 2019 and January 2020. The Biomedical Ethics Committee of Anhui Medical University gave an ethical approval for the study. Participants were enrolled using a multi-stage stratified cluster random sampling method. Seven cities were selected from the sixteen prefecture-level cities in Anhui province (Fuyang, Huainan, Suzhou, Chizhou, Anqing, Lu'an, and Hefei central; the capital of Anhui). One district was then randomly selected from each prefecture-level city forming a total of sixteen districts for this study. In each selected district, a total of 252 elderly caring social organizations were selected and their employees to fill in self-report questionnaires. A total of 605 questionnaires were qualified for analysis.
Measurement of turnover intention
A turnover intention assessment scale was used with modifications to consist 6 items as described by Mickael and Spector  and Li et al. . A four-point Likert scale was used to rate the items, ranging from 1 (never, or strongly impossible) to 4 (always, or very possible) (Table 1). The total score of the six items was computed as the score for turnover intention, ranging from 6 to 24. A high score suggested a greater turnover intention. The scores of 0–6, 7–12, 13–18, and 19–24 were assigned for the low, moderate, moderate to high and high degree of turnover intention, respectively . The scale includes a total of six items, divided into three dimensions: possibility of quitting a current job (turnover intention I, items 1 and 6), motivation to find other jobs (turnover intention II, items 2 and 3), and obtaining the external possibility of work (turnover intention III, items 4 and 5). The scale indicated a good internal consistency with a Cronbach’s alpha coefficient of 0.74.
Measurement of perceived social support
A 12-item self-report measure of social support was used to measure the perceived adequacy of social support for the participants [17, 18]. It used three sub-scales: family support, friend support, and other significant support. The respondents rated each item on a 7-point scale, from 1 (very strongly disagree) to 7 (very strongly agree). A total score was calculated by summing all the responses. Possible scores ranged from 12 to 84. A high score indicated a greater level of perceived social support. The tool showed a Cronbach’s alpha coefficient of 0.83, suggesting good study reliability.
Measurement of other variables
Demographic information and health-related variable data was collected, which include age (in years), gender (male, female), ethnic group (Han, ethnic minority), marital status (married/cohabited, single, never married, divorced, widowed), education (primary school and below, junior high school, college degree and above), have professional qualifications (yes, no), whether to accept business training (yes, no), working period ( in years), satisfaction with current job (very dissatisfied, dissatisfied, fair, satisfied, very satisfied).
Double entry and data validation was done in EpiData3.1. The collected data was analysed using SPSS 20.0 (IBM Corp, Armonk, NY, USA). Count data was expressed in composition ratio while measurement data was expressed in mean (M) ± standard deviation (SD).
For binary variables (gender, ethnic group, marital status, professional qualification certificate, Business training) univariate analysis was conducted using an independent sample t-test The age and working years of employees in elderly caring social organizations was changed into categorical variables. The age of employees was divided into four groups: 30 years and below, 31–50 years old, 51-70 years old and 71and above years. The working years was divided into four groups: less than 1 years, 1-5 years, 5 -10 years and more than 1 years. For multiple categorical variables (age, education level, working years, satisfaction level) we performed a one-way ANOVA (analysis of variance)while Spearman correlation analysis was performed to determine the correlation between turnover intention and social support, the test level was set at α = 0.05.
Multiple linear regression models were done for multivariate analysis to set dummy variables for categorical variables. Set controls for each survey item were established to determine their association with other items using standard and non-standard coefficients, the test level α = 0.05.