Research design and setting
This study employed a cross-sectional survey methodology, utilizing a convenience sample comprising 98 Nursing students drawn from various academic years at a prominent college located in central of the country. Recruitment and data were collected from December 2022 to September 2023. Data collection was facilitated via a structured questionnaire administered to participants who had either completed a nutrition course (n = 57) or had not yet enrolled in or completed such a course (n = 41).
The Health Promotion course in nutrition is a newly added course that is offered in the later years of study (semesters 6–7/8). The primary goal of this course is to build upon the knowledge gained in previous classes and apply it to specific populations. We built the learning outcomes including assignments according to Bloom’s taxonomy [
15]. students in this course are expected to move beyond simply acquiring knowledge and understanding. They are encouraged to analyze, synthesize, and evaluate the course material from a clinical perspective. This higher level of critical thinking enables students to explore concepts such as competence and self-epistemic authority, which play a crucial role in shaping their professional identity as nurses.
Participants
All nursing students enrolled at the college were invited to partake in the research, with 120 students consenting and volunteering to participate. However, 22 of these volunteers failed to complete the majority of the questionnaire items, resulting in the exclusion of their responses from the analysis. Of the 98 students who were included in the analyses, 28 were first-year students (28.6%), 38 were second-year students (38.8%), 11 were third-year students (11.2%), and 21 were fourth-year students (21.4%).
Handling missing data
To handle missing data, we performed multiple imputation using SPSS (version 23.0). The pattern and extent of missing data were first analyzed, revealing that 2.5% (150 missing data values and 5926 complete data values) of the data were missing at random across the four questionnaires (professional autonomy, professional authority, self-epistemic authority, sense of meaning).
We used the Fully Conditional Specification (FCS) method to generate five imputed datasets, which is within the recommended range for reliable imputation. The imputation model used predictive mean matching and included all variables from the primary analysis to ensure that the relationships among the data were preserved. Each imputation underwent 10 iterations to ensure convergence.
Diagnostic checks, including trace plots, were used to assess the convergence of the imputation process. Additionally, we compared the distributions of imputed values with observed values to ensure the quality of imputations. For the analysis, we used Rubin’s rules to combine parameter estimates and standard errors across the five imputed datasets.
Statistical Analysis, All statistical analyses were conducted using JASP version 0.18.2, which utilizes R version 4.3 as its underlying engine. Descriptive statistics of means and SDs were used to describe age and the four measures (professional identity, professional autonomy, self-epistemic authority, and sense of meaning) and Pearson correlation was used to describe the relationships between them. Frequencies and percentages were used to describe qualitative background characteristics (gender, religion, marital status, academic year, and whether they had participated in and completed a nutrition course). Independent samples t-tests were performed to compare the two cohort groups (had completed a nutrition course vs. had not) in the study’s four measures.
To investigate the mediation effects, we employed the Generalized Least Squares (GLS) method, implemented in JASP version 0.18.2. The GLS method was chosen due to its ability to handle potential issues of heteroscedasticity and autocorrelation in the data, providing more efficient and unbiased parameter estimates compared to Ordinary Least Squares (OLS) regression. GLS adjusts for variations in error variances and covariances, thereby offering a robust approach for estimating the relationships among the independent variable, mediator, and dependent variable. This method enhances the precision and reliability of the mediation analysis, ensuring that the effects are accurately captured, and the assumptions of the regression model are appropriately met.
To ensure the validity of our statistical analyses, we conducted assumption checks for Pearson correlations, t-tests, and mediation analysis. For Pearson correlations, we assessed linearity, homoscedasticity, and the absence of significant outliers using scatter plots and standardized residual plots. For t-tests, we tested normality using the Shapiro-Wilk test and inspected Q-Q plots, while Levene’s test was used to confirm homogeneity of variances for independent t-tests. In mediation analysis, we verified linearity and normality of residuals through residual plots and the Kolmogorov-Smirnov test, and then assessed homoscedasticity via standardized residual plots. Multicollinearity was checked using variance inflation factors (VIF), ensuring all VIF values were below 10. All assumptions were satisfactorily met, validating the robustness of our findings. Mediation model estimates and path coefficients were adjusted for gender and age.