A cross-sectional design was used.28 The study population consisted of Dutch nurses and caregivers employed in residential healthcare settings. Current estimates show that approximately 177.000 people are employed as nurses or caregivers, of which 25% work in residential healthcare settings.29 The majority is aged between 50-60 years, and more women (25%) work in this sector than men (12.2%).29

The sampling frame consisted of nurses and caregivers working in residential healthcare settings. Four healthcare settings from different institutions in the northern part of the Netherlands were involved (see Acknowledgements). Also, one healthcare-related Dutch Facebook group was included. This online community is mainly for nurses and caregivers working in residential healthcare settings and hospitals.

Data collection was carried out in March and April 2017 by means of a self-administered questionnaire. Qualtrics software was used to develop the online version. A meta-analysis of techniques to improve participation rates in surveys amongst healthcare workers showed that a response rate of at least 38% could be expected for an online questionnaire.30 The questionnaire contained a variety of subscales measuring SOC, GRRs, learning, and demographic data.

The data were analysed using the statistical package IBM SPSS Statistics 24 and the SPSS add-on PROCESS36 for Windows. The level of significance was set at the recommended .05 level.37

Differences among the organizations regarding sex, level of education, type of contract, and response rates were tested using the Chi-square test. The Kruskal-Wallis H Test, one-way ANOVA, and an independent t-test were used to test for differences regarding age, average hours worked per week, and SOC.

Exploratory factor analysis was performed on the WLPQ items using its three underlying sub-constructs. A preliminary analysis was performed to examine the factorability of the 16 items by checking multicollinearity (determinant>0.00001), the KMO measure of sampling adequacy (>0.5), and the Bartlett's test of sphericity (p<0.05).37 Finally, a scree plot and eigenvalues>1 were used to assess the number of factors to extract.37

Correlation and reliability analyses (Cronbach's alpha) were used for all the study variables. Missing values were excluded by convention pairwise and listwise.37 Multiple linear regression analysis was performed to test whether the three GRRs explain variance in SOC with forced entry.37 This regression analysis was also used to test for moderating effects, after creating centred interaction terms between the predictor(s) (SOC or GRRs) and learning, using hierarchical entry.37 For instance, when testing for the moderating effects of learning on the relationship between SOC and its predictors, GRRs, the interaction terms between learning and the GRRs (e.g. job control * social learning) were added as predictor variables in the regression analysis. Moderating effects were explored by testing the significance of the additional explained variation of the dependent variable(s) (SOC or GRRs) when these interaction terms were added to the model, both individually and in groups. All regression analyses excluded missing values by convention listwise.37

Model 4 (simple mediation) from the add-on PROCESS function was used to examine indirect effects of learning,36 which means that the relationship between the dependent (e.g. SOC) and independent (e.g. job control) variable goes through another mediating variable (e.g. instrumental learning). Simple mediation allows adding one mediator (e.g. instrumental learning) to the regression model at a time, which is a conventional model used for exploring mediating effects.36 The significance of the mediation effect was studied using 10,000 bootstrapped samples, with a random seed number of 5,235.36 Since research showed that SOC increases with age,8 age was included in all regression analyses to control for possible confounding effects.

This study was conducted according to the Wageningen Code of Conduct for Scientific Practice. Participation was entirely voluntary, sustained by an informed consent to ensure anonymity without using any incentives.

Table 1 provides a summary of the demographic data for each of the five groups, including mean SOC and response rates. All response rates were lower than the anticipated 38%, and these rates differed significantly among the four healthcare groups, χ2(3)=47,80, p <0.001. Significant differences were also found regarding age, H(4)=25.82, p <0.001; average number of hours worked per week, H(4)=45.80, p <0.001; and type of contract, χ2(4)=13.45, p=0.009. SOC did not differ significantly among the five groups or between men (M=71.9, SD=12.7) and women (M=70.5, SD=11.2), t(473)=0.653. Since SOC did not significantly differ among the five groups, it was deemed appropriate to combine the five groups into a single dataset for testing the study hypotheses without stratifying, for instance, for sex. This is also in line with the overarching aim of this study, which is to test the model, and not invariances across specified subgroups.

Principal axis factoring analysis was conducted with oblique rotation on the 16 items of the WLPQ (Table 2). Three factors had eigenvalues over Kaiser's criterion of 1 and in combination explained 69.40% of the variance. The scree plot, however, suggested both a three-factor solution and a one-factor solution. The one-factor solution explained 51.21% of the variance. Both the three-factor solution and the one-factor solution were retained.

Regarding the three-factor solution: the items that cluster on the same factors suggest that factor 1 represents conceptual learning, factor 2 represents instrumental learning, and factor 3 represents social learning. The one-factor solution combined these three sub-constructs into a single latent variable designated as learning.

Table 3 displays the means, standard deviations, Cronbach's alphas, and correlations among the study variables. The Cronbach's alphas are good for all the study variables. SOC, the GRRs, and both factor solutions are positively correlated with one another.

Table 4 presents the regression analysis for testing the relationship between SOC and the GRRs (model I), which shows how much variance is explained in SOC by the three GRRs. The three GRRs explain 26% (adjusted R2) of the variance very significantly in SOC. Moreover, each of the three GRRs contributes very significantly after correcting for the other two GRRs and age. Job control seems to be the most important GRRs (standardized β=0.24, p <0.001), followed by social relations (standardized β=0.20, p <0.001), and task significance (standardized β=0.19, p <0.001), thereby supporting hypothesis 1.

Table 4 also shows the moderation analyses (models II, III, and IV). The interaction term(s) between learning and predictor(s) (GRRs or SOC) were added in the multiple regression analyses, and tested for significance. Model II shows that learning (the one-factor solution) moderates the relationship between SOC and its predictor, task significance (standardized β for the interaction term is .11, p=0.012), corrected for the effects of social relations, job control, and age. In turn, Model III shows that learning moderates the relationship between task significance and its predictor, SOC (standardized β=0.13, p=0.005). When its sub-constructs are examined, the regression analysis in model IV shows that instrumental learning moderates the relationship between job control and its predictor, SOC (standardized β=0.14, p=0.002). So, out of the 24 possible moderating effects, only three were found to be significant, thereby giving only a weak confirmation of hypothesis 2.

Simple mediation analyses –examining each mediator variable separately– showed that social learning mediates the relationship between job control and its predictor, SOC, b=0.01, BCa 95%CI [0.001-0.03] corrected for the effects of age. This supports hypothesis 3 (Fig. 1). However, no other indirect effects of learning were found within the relationship between SOC and GRRs.

Figure 1.

Social Learning Mediates the Relationship between SOC and Job Control (n = 427). b: unstandardized regression coefficient; p: probability; BCa: bias corrected and accelerated bootstrap confidence interval.

(0.09MB).

As hypothesized, the present findings show that social relations, task significance, and job control explain variance in SOC. Moreover, each of the three GRRs contributes very significantly after correcting for the effects of the other two GRRs and age. This suggests that the stronger SOC is, the stronger the scores on these three GRRs, and vice versa. The findings align with Antonovsky's9,38 suggestion that a resourceful work environment strengthens employees’ view of the world as comprehensible, manageable, and meaningful.

The results of the analysis in the context of the model used are summarized in Figure 2. The findings indicate that the overarching learning concept moderates the relationship between SOC and task significance corrected for the effects of social relations, job control, and age. This interaction effect suggests that, when learning is higher, the relationship between SOC and task significance is stronger; this is in line with salutogenic theory. Learning can enhance the perception of employees that their job has a positive impact on other people, which is important for the development of their SOC.6

Figure 2.

Building GRRs and SOC in Workplaces through Learning. H: hypothesis; Black arrow: relationship between SOC and GRRs; Blue arrows: moderating effects; Green arrows: mediating effect.

(0.12MB).

The present study shows that instrumental learning moderates the relationship between job control and SOC. This interaction effect indicates that a strong SOC together with the development of skills through courses or learning on the job builds an employee's job control. Social learning mediates this relationship, which suggests that a strong SOC fosters learning through cooperation, feedback, sharing experiences, but also through job challenges like the growth of responsibilities or changes in tasks and functions, all of which also builds an employee's job control. Social relations were not involved in any of the moderating and mediating effects. It should be noted that the moderating and mediating effects of learning were relatively small. Also, conceptual learning on its own did not show any moderating or mediating effects, whereas instrumental and social learning on their own did. This suggests that other factors could play a role in the relationship between SOC and GRRs in workplaces.

The results of this study can be used to develop health and wellbeing of nurses and caregivers to a certain extent. Maintaining and developing health and wellbeing of healthcare workers is essential since the quality of care provided strongly depends on their health and wellbeing.4 To enhance instrumental and social learning in healthcare organizations, a central role should be given to healthcare workers themselves because they know the best what is needed to find a good fit with the work environment.6 Achieving this fit enables them to not only learn through formal activities, but also through cooperation, feedback, sharing experiences, and job challenges, which could strengthen their SOC and build their GRRs.

Although the average age of the total sample was lower than the study population, the percentage of male and female healthcare workers in this particular sector reflects the study population.29 Also, SOC did not significantly differ between them, which indicates that the findings of this study may be generalized to both males and females. However, due to the feminization of the nursing profession,39 learning processes may differ among men compared to their female counterpart. Also, age should be taken into account since the relevance of the GRRs and learning may differ between nurses and caregivers at different ages.

The present study has some limitations that should be taken into account. First, as all measures were self-reported, common method biases may have influenced the results.40 However, constructs such as SOC are difficult to measure using methods other than self-reporting.2,31

A second concern is the cross-sectional study design used to test for mediating effects. Longitudinal designs are required to provide evidence for the mediating roles of variables.41 Since this study indicated that social learning mediates the relationship between job control and SOC, future studies may aim to validate this finding using a longitudinal design.

A third issue is that the original dataset (N=758) was reduced to only those respondents who had completely answered all the SOC and the WLPQ items (N=481). However, the differences presented in Table 1 did not deviate from the differences in the original dataset. Moreover, by excluding the incomplete responses regarding these items, the exploratory factor analysis and reliability analysis were enhanced.42

Another concern is that the response rates were lower than the expected 38%. However, this study reconfirmed the difficulty of achieving a high response rate among healthcare workers.30 Future studies are recommended to already involve nurses and caregivers in the design phase of the survey,43 use a mail survey mode combined with monetary incentives,30 or promise that the outcomes will be used to enhance workplace health promotion policies, which can result in higher response rates compared to an online questionnaire without financial compensation.30

A final point worthwhile discussing is that the Facebook group also included nurses and caregivers from other healthcare settings, such as hospitals. However, the aim of this study was not to test for invariances across specified subgroups. Nevertheless, future studies may aim to validate the findings of this study across specified subgroups, for instance, among healthcare workers in hospitals.