1 INTRODUCTION

The healthcare delivery system is known to undergo rapid changes because of the evolution of the complexity of patient conditions and the fast growth of knowledge. To adapt to the career challenge, students must learn to be responsible for their continued learning throughout their personal and professional careers, thus becoming lifelong learners. Self-directed learning (SDL), usually tied to the demand for lifelong learning (Kaulback, 2020), is a crucial goal for nursing education worldwide (Cadorin et al., 2017; Mentz & Oosthuizen, 2016; Wong et al., 2021). Teaching students to become self-directed learners is highly recommended by higher education institutions and continuing education agendas. Many studies have explored problems or factors related to SDL (Chukwuedo et al., 2021; Wong et al., 2021). Self-directed learning was also found to be significantly associated with teaching competencies (Chen et al., 2012), critical thinking (Kang et al., 2020), self-efficacy (Turan & Ko, 2018) and even leadership development (Cleary et al., 2020). Nevertheless, a tool cannot reliably measure data without good psychometric properties and provide strong scientific evidence for the self-directed learning literature.

2 BACKGROUND

The definition of SDL varies in the literature. SDL has been described as a process, an ability, a trait, a learning style, and a teaching method for educators (Cadorin et al., 2017; Dehnad et al., 2014). Some emphasize the learners' control over the planning and implementation of learning, such as Knowles (1975) and Tough (1979), while some focus on the potential characteristics of self-directed learners (Guglielmino, 1977). According to Knowles (1975), self-directed learning is a process in which individuals initiate the process based on their learning needs, identify their learning goals and required resources and select and apply appropriate strategies to achieve desired learning results. Thus, self-directed learning is a type of learning that requires taking more action than just being a passive receiver of knowledge. The “self” is the one who manages the learning process (Mentz & Oosthuizen, 2016). Individuals are expected to initiate the activities, curious to learn new things and enjoy learning (Alfaifi, 2016). Studies also have reported that self-directed learners tend to be more independent, disciplined, initiative, curious, self-motivated in learning and have a socio-professional commitment (Mentz & Oosthuizen, 2016; Shirazi et al., 2021).

Numerous tools have been developed to measure SDL, including the SDL Readiness Scale for Nursing Education (SDLRSNE), the self-rating scale of SDL (SRSSDL), the SDL Inventory (SDLI) and the SDL readiness scale (SDLRS). The SDLRSNE was developed by Fisher et al. (2001) and contained 40 items divided into three components: self-management, desire for learning and self-control. The construct validity was evaluated by confirmatory factor analysis. Despite 11 items being identified as redundant, the authors suggested that the 40-item SDLRSNE be used until further research refinement (Fisher & King, 2010). However, a Japanese nursing study could not validate the reliability and validity of the SDLRSNE in their population (Fujino-Oyama et al., 2016). The SRSSDL was developed by Williamson (2007). The tool consisted of 60 items with the following five components: awareness (12 items), learning strategies (12 items), learning activities (12 items), evaluation (12 items) and interpersonal skills (12 items). A sample of Italian health professionals validated the instrument, and an exploratory factor analysis produced an eight-factor solution. As a result, the number of SRSSDL_-ITA items was reduced from 60 to 40 (Cadorin et al., 2013). The concurrent validity of SRSSDL and SDL ability was supported (Cadorin et al., 2016). The SDLI was developed by Cheng et al. (2010) and contained 20 items in four components: learning motivation (6 items), planning and implementing (6 items), self-monitoring (4 items) and interpersonal communication (4 items). The construct validity was verified using confirmatory factor analysis (Cheng et al., 2010).

SDL readiness (SDLR) refers to the extent to which an individual has the abilities, attitudes and personality traits required for SDL (Guglielmino, 1977) and the degree to which a self-directed learner can take personal control and accept their responsibility to learn (Fisher et al., 2001). Guglielmino's (1977) SDLR scale (SDLRS) is one instrument most commonly used to measure SDLR, which consists of 58 items categorized into eight components, namely openness to learning opportunities, effective learner, initiative and independence in learning, informed acceptance of responsibility for their learning, love of learning, creativity, positive orientation to the future, ability to use basic study skills and problem-solving skills. The construct validity was confirmed by factor analysis. However, other studies have failed to confirm its factor structure (Field, 1991; Fisher et al., 2001). Deng (1995) developed the Traditional Chinese version of the SDLRS (TC-SDLRS) based on concepts of self-directed learning readiness for the business continuing education students in junior colleges. The TC-SDLRS consists of 55 items in six components: effective learning, love of learning, learning motivation, active learning, independent learning and creative learning. Satisfactory construct validity and internal consistency were provided. Chen and Liang (2009) revised the TC-SDLRS and used it in their Taiwanese undergraduate sample; however, only four subscales were included in the final version. Cronbach's alphas for the four scales were 0.64 (creative learning), 0.66 (love of learning), 0.69 (independent learning) and 0.70 (active learning), respectively. We needed to use the scale in our study (Chen & Liu, 2021), but the preliminary psychometric findings of the scale were not satisfactory. An instrument with sound psychometric properties is essential to build up the research evidence in the scientific community. Additionally, it takes more time to fill out a 55-item scale. A short version of the scale can improve the efficiency of the test. Thus, the purpose of the study was to revise the scale items and identify the least number of items needed to explain the original latent factors, using the technique of exploratory factor analysis (EFA) and confirmatory factor analysis (CFA).

3 THE STUDY

4 RESULTS

The mean age was 20.35 (SD = 1.91) years for the EFA group and 20.14 (SD = 0.65) years for the CAF group. Most of the participants were women (EFA = 89.7% vs. CFA = 96.2%). The majority of the sample in the EFA group was 4-year BSN students (46.8%), followed by RN-BSN students in a regular daytime class (53.2%). Most of the students in the CFA group were in the RN-BSN programme during the daytime (60.6%). Twenty-three per cent were on-the-job RN-BSN students, and the remaining 16.4% were 4-year BSN students. The content validity was examined by three experts for clarity and appropriateness. The wordings of the scale were also revised for those items with Cronbach's alpha <0.65. The content validity index was also calculated. All the values were ≥0.85, which implied a satisfactory content validity.

The factor structures of the TC-SDLRS from 263 participants were first assessed using a PCA. All items have been entered into the model. Next, we examined the items of each factor. Ten items were deleted because of cross-loading over three factors, and the factor loadings of another three items were below 0.3, or the item-to-total correlation was below 0.20. Eventually, 13 items were deleted. Another PCA was performed on the remaining items, which produced a six-factor solution that explained 52.10% of the total variance. All items were loaded on the predetermined latent factor, which showed that the items could represent the underlying constructs intended to be measured. The calculated KMO was 0.888, and Bartlett's test of sphericity was statistically significant (χ² = 4,869.20, p < .001). The eigenvalues of factors were >1, and all factor loadings were >0.35. The original six factors remained. As a result, 55 items were reduced to 42 items.

A CFA was used to validate the latent structure validity with 396 participants. Five models were tested. Fit statistics for model 1 were acceptable, but factor loadings of four items (item 48, item 29, item 52, item 54 and item 38) were low. Items were removed one by one to ensure necessity. After eliminating the item, fit indices were significantly improved (AIC = 1847.78, NFI = 0.95, GFI = 0.81 and RMSEA = 0.066). In the Lagrange analysis, the three large values of MI and EPC were between item 40 and item 41 (MI = 41.98, EPC = 0.20), item 25 and item 26 (MI = 42.15, EPC = 0.13) and between item 1 and item 5 (MI = 40.08, EPC = 0.11). Three error terms between these items were individually correlated one after another. The revisions resulted significantly improved the model fits (AIC = 1,629.94, NFI = 0.95, GFI = 0.84 and RMSEA = 0.059). A 37-item six first-order model produced the best-fit statistics. The six first-order factors were re-specified into a second-order model (Figure 1). Despite an acceptable fit of the second-order factor to data (AIC = 1,645.618.11, NFI = 0.95, GFI = 0.83 and RMSEA = 0.059), the factor loading of independent learning was 0.23 (t = 3.75, p = .038). However, removing independent learning from the second-order model did not improve the fit indices of the model (Table 1). The target coefficient was 0.98 (Brown, 2015). The six first-order factors and the five first-order one second-order factors demonstrated comparable and acceptable fits. Hoelter's Critical N was 204.66, which indicated that the sample size for data analysis was enough.

Most TC-SDRLS items have factor loadings above 0.50, with ranges of 0.59–0.84 for love of learning, 0.56–0.73 for active learning, 0.57–0.76 for effective learning, 0.64–0.77 for independent learning, 0.59–0.73 for learning motivation and 0.48–0.70 for creative learning. The estimates of all items were statistically significant and were loaded on a pre-destinated latent factor. Therefore, convergent validity was supported (Brown, 2015). The internal consistency for the revised TC-SDLRS was examined. All Cronbach's alpha (α = 0.71–0.88) and composite reliability (CR = 0.76–0.86) for the total scale and the subscales exceeded the recommended values satisfactory, indicating that the internal consistency between items was sufficient (Table 1). The 37-item six first-order model was further validated by EFA and explained 53.30% of the total variance. The love of learning (6 items) explained 26.54% of the variance. Other factors, namely active learning (8 items), effective learning (6 items), independent learning (7 items), learning motivation (5 items) and creative learning (5 items), contributed 8.51%, 5.85%, 5.53%, 3.54%, and 3.33% of the variance, respectively. The calculated KMO was 0.886, and Bartlett's test of sphericity was statistically significant (χ² = 4,108.05, p < .001). The communality of all items was above 0.34. The eigenvalues of factors were >1, with factor loadings ranging from 0.33 to 0.74 (Table 2). After the EFA, concurrent validity and predictive validity were performed. Significant correlations between the scores of the CTDI and TC-SDRLS were found; thus, concurrent validity of the TC-SDLRS was supported (Table 3). The TC-SDLRS score was also significantly associated with midterm (r = .333, p = .036) and final grades (r = .321, p = .043) of the students' physical assessment course. Predictive validity was also supported. Correlation analyses between the short and original versions also reached statistically significant (r = .22–.81).

5 DISCUSSION

The TC-SDLRS aims to measure an individual's readiness for attitudes and skills related to SDLR such as individual ability to initiate learning autonomously, task control and a deep desire for learning (Dehnad et al., 2014; Fisher et al., 2001). The items were revised following the original concepts identified by Deng (1995) to maintain the scale's comprehensive profile. The group of experts also agreed with the item determination. The CFA findings showed that all items were loaded on the predetermined factor, indicating that the items may represent the underlying constructs. Correlations between the original factor and the short version of the TC-SDLRS were significant. The factorial structure was also validated by the findings of EFA and CFA as a six-factor solution structure. The similarity of the underlying constructs was supported. The model showed the best fits of indices. The six factors were the love of learning, active learning, effective learning, independent learning, learning motivation and creative learning. These factors are consistent with the characteristics of highly self-directed learners who are more motivated to learn and accept their learning responsibilities (Mentz & Oosthuizen, 2016; Shirazi et al., 2021). Self-directed learners are enthusiastic and curious about learning, set their own learning goals and learn independently. Active learners are more effective than passive ones. Learners may use different resources to facilitate their learning, strive to meet the requirements and use creative approaches to solve problems (Buthelezi & Phahamane, 2016; Deng, 1995).

A higher-order factor analysis was applied for theory testing and to clarify the patterns of relationships among first-order factors (Brown, 2015). Statistical fits revealed that the six first-order and second-order factors of the TC-SDLRS were acceptable, indicating that the observed variables could be aggregated in this model to represent the underlying concept of SDLR. The factor loading of independent learning was 0.32 in the second-order model, which maybe because of the reverse worded items of observed variables. Adding reverse worded items on a scale is a common strategy to reduce or eliminate acquiescence bias, a respondent tendency, whereby participants provide a specific answer regardless of its content (van Sonderen et al., 2013; Vigil-Colet et al., 2020). Our findings, similar to the previous findings in the literature (Suárez-Álvarez et al., 2018; Vigil-Colet et al., 2020) showed that a combination of positive and reversed items might risk the scale's unidimensionality. The study for scale development should consider the potential effect of inattentive or acquiescent answering resulting from reversed items or lengthy scale and balance the effect of acquiescence bias and linguistic barriers. Although the factor loading of independent learning was 0.32, the path was statistically significant. The latent factor did not reduce test reliability or worsen the model fits. Conversely, even with a minor degree of freedom, the statistic fitness worsens after the deletion of the factor. Therefore, independent learning can be used as a single sub-score or counted into a total score, especially in a repeated measurement design, because the method may lessen confounding effects by reducing the sources of error and the variance of test scores (van Sonderen et al., 2013).

Concurrent validities were supported by significant associations between the TC-SDLRS and critical thinking disposition scores. Consistent with previous study results (Kang et al., 2020; Turan & Ko, 2018), the findings demonstrated that higher levels of the SDLR were associated with a more critically oriented thinking disposition. Critical thinking disposition was also a significant predictive factor of SDL (Turan & Ko, 2018). The process of SDL is closely related to individual intellectual characteristics and thinking skills. Individuals may learn more effectively if they are aware of their cognitive processes and can successfully plan their learning or daily life. The TC-SDLRS score was also significantly associated with the midterm and final grades of the students' physical assessment course. The course curriculum was designed using SDL to teach students assessment skills (Chen & Liu, 2021), and the TC-SDLRS score was collected at the beginning of the course. Thus, the predictive validity was supported.

SDL is significantly associated with lifelong learning (Chukwuedo et al., 2021), an essential characteristic of professional development. Readiness for SDL can be developed (Noh & Kim, 2019), and the ability to learn self-directly in one situation or environment cannot be generalized to other contexts (Buthelezi & Phahamane, 2016; Fisher et al., 2001; Turan & Ko, 2018). Teaching students to learn self-directly is difficult because the SDLR is highly personalized. Some learners may be highly self-directed, whereas others may strongly prefer specific and direct instruction when learning (Chen & Liu, 2021). Implementing self-directed teaching strategies may lead to negative feelings, such as anxiety and frustration, if students are not adequately prepared (Zhang, 2020). To improve students' competence in SDL, teachers and learners must be familiar with the concept and the possess basic skills to implement it. Evaluating learners' readiness for SDL could enable faculty to understand the extent of SDL learning capacity and design appropriate teaching activities for more favourable outcomes.

6 LIMITATIONS

Despite the rigorous methodology used to validate the reliability and validity of the scale, generalization of the study results may be limited. The study was a survey design, and a self-report bias cannot be excluded. Furthermore, a convenience sample was recruited from two universities in Taiwan, and selection bias may arise. Finally, the SDLR is an individualized feature. Future studies using a multigroup comparison in different contexts or cultures could assess measurement invariance of the scale to ensure equivalent interpretation. As a result, the utility of the scale may be improved.

7 CONCLUSION

The psychometric properties of the short version of the TC-SDLRS were examined. Validity was examined using content validity, EFA and CFA. Cronbach's alpha and composite reliability assessed reliability. The concurrent and predictive validity of the TC-SDLRS were also acceptable, as the C-SDLRS score was significantly associated with the CTDI and can predict students' academic scores. The internal consistency of the scale was also satisfactory. The study findings provided evidence supporting that the TC-SDLRS has sound psychometric properties. Furthermore, compared with the original version, the short version of the TC-SDLRS has fewer items and is more manageable for a respondent to fill out. According to the CFA findings, independent learning can be used as a single sub-score or counted into a composite score, especially when the study is a repeated measurement design. The scale could help educators diagnose learners' readiness for SDL, incorporate teaching activities regarding learning how to learn and foster students' capacity for SDL to meet future challenges.

AUTHOR CONTRIBUTIONS

Shiah-Lian Chen and Jun-Yu Fan contributed to study conception and design, data collection, drafting of the article and critical revision of the article. Shiah-Lian Chen contributed to data analysis and interpretation.

All authors have agreed on the final version and meet at least one of the following criteria [recommended bythe ICMJE (http://www.icmje.org/recommendations/)]:

• substantial contributions to conception and design, acquisition of data or analysis and interpretation of data;

• drafting the article or revising it critically for important intellectual content.

FUNDING INFORMATION

This work was supported by the Ministry of Science and Technology [grant numbers: MOST 104-2511-S-025 -001 -MY2], Taiwan, ROC. This work was also supported by a grant from the Administration Center of the Mdical Research Department, Chang Gung Memorial Hospital, Taiwan, ROC: Grant No. BMRPB80, which was awarded to the corresponding author.

CONFLICT OF INTEREST

No conflict of interest has been declared by the authors.

ETHICS STATEMENT

The study was approved by the Research Ethics Committee of the China Medical University and Hospital, Taichung, Taiwan (CRREC-104-103).