Work characteristics and occupational health: the WDQ

Work characteristics are relevant to occupational health to the extent that they can represent risk factors, which is to say, any aspect of the work environment where exposure significantly increases the probability of an individual suffering an adverse health outcome. This may take the form of a disorder, syndrome, disease or injury, and will also impact on the performance of the worker and the organization. When these risk factors derive from the way in which work is organized and the relationships that exist within an organization, we are referring to psychosocial risk factors (e.g., ^{Cox & Griffiths, 1996}; ^{Leka & Cox, 2008}). These risk factors are counterbalanced by protective factors that reduce the probability of the occurrence of the dysfunctional or pathological response. Depending on the case, whether a factor is a risk factor or a protective factor will depend on various aspects, such as intensity, frequency or the type of relationship between the variables, whether linear or non-linear.

^{Morgeson and Humphrey (2006)} tried to measure work characteristics and conducted a study in the USA with 540 incumbents holding 243 distinct work positions and developed the WDQ, composed of three categories of work characteristics. The first category (i) refers to motivational characteristics, which include the factors task and knowledge characteristics, given that job enrichment is improved when high levels of these characteristics are present. Task characteristics refers to the work involved in any given work position and the way in which the required tasks are carried out, and include the following variables: work scheduling autonomy, decision-making autonomy, work methods autonomy, task variety, task significance, task identity and feedback from job. The WDQ also includes knowledge characteristics, which refer to the knowledge, skill and ability demands on the job, and include the following variables: job complexity, information processing, problem solving, skill variety and specialization. (ii) The second category is composed of social characteristics, which refer to relationships with others, and include: social support, initiated interdependence, received interdependence, interaction outside the organization, and feedback from others. Finally, (iii) the third category is composed of work context characteristics, which refer to the physical environment of the workplace, and include the variables: ergonomics, physical demands, work conditions and equipment use.

Empirical analysis of the constructed questionnaire showed that the 21-factor version was the WDQ that presented the best psychometric properties after analyzing the following five factorial structures: (i) 4 factors, where items are grouped into the four general categories of task, knowledge, social and work context characteristics; (ii) 18 factors, where autonomy and interdependence are each considered as a single factor rather than separated into work scheduling, decision-making and work methods autonomy, and into initiated and received interdependence; (iii) 19 factors, where autonomy is considered a single factor rather than separated into three factors, as in the previous case, and interdependence is divided into initiated and received interdependence; (iv) 20 factors, where autonomy is divided into work scheduling, decision-making and work methods autonomy, and interdependence is considered a single factor; and (v) 21 factors, where autonomy and interdependence are divided into their respective components, as described above. ^{Bayona et al. (2015)} and ^{Fernández et al. (2017)} obtained similar results with their corresponding Spanish versions of the WDQ, with the 21-factor structure obtaining the best results. However, more recent research by ^{Borges-Andrade et al. (2019)} on Brazilian workers, and by ^{Peiró et al. (2020)} on Colombian workers, found that the 18-factor structure was the most suitable version. Given that a shorter version of the questionnaire would facilitate its application, we will therefore study the possibility of reducing the length of the WDQ while maintaining its adequate psychometric properties (Objective 1).

With respect to criterion-oriented validation of the WDQ (Table 1) we have found several studies (e.g., ^{Bargsted et al., 2019}; ^{Bayona et al., 2015}; ^{Gorgievski-Duijvesteijn et al., 2016}; ^{Humphrey et al., 2007}; ^{Morgeson & Humphrey, 2006}; ^{Stegmann et al., 2010}) where we can see evidences between the WDQ and occupational health variables, specifically job satisfaction, burnout, and well-being.

Table 1: Found and expected relations between the dimensions of the WDQ and occupational health outcomes for hypotheses H₁-H_3. 

Taking all of the above into consideration, we expect to find the following relationships in this study with the short form of the WDQ, expressed in hypotheses H1-H3, and summarized in Table 1:

Gender and work conditions

It is vitally important to take gender into account when analyzing working conditions, as the failure to do so risks overlooking any differences that may exist in vulnerability and exposure for women and men. These differences derive from the implementation of occupational safety and health standards that are primarily geared towards men, and a failure to take into account that, due to gender roles, women and men may not perform the same tasks (e.g., ^{Osca & López-Araújo, 2020}), or perform tasks in the same way. Gender division of work can be explained by gender roles theories, as the Role Congruity Theory of Prejudice Toward Female Leaders (^{Eagly & Karau, 2002}), and the Status Incongruity Hypothesis (^{Rudman et al., 2012}). These theories posit that women have been historically consigned to feminized positions (e.g., ^{Castaño et al., 2019}). Feminized work would appear to be more related to employment in positions with less responsibility, more repetitive tasks, and a heavier family burden; while masculinized work is related to positions of greater responsibility that require more overtime and greater overload (e.g., ^{Cifre et al., 2015}). In this regard, the study by ^{Rivera-Torres et al. (2013}) on work stress showed that men are more often affected than women by quantitative demands, such as having to work rapidly and under greater time pressure; while women are more often affected by qualitative, intellectual, emotional and social support demands. At the same time, different risks may be present when women and men occupy positions that are stereotypically considered inappropriate for their gender (e.g., ^{García-Izquierdo et al., 2018}).

In short, this all points to the fact that women and men may suffer different consequences for their health and well-being. Indeed, these differences have been corroborated in several meta-analyses (e.g., ^{Batz-Barbarich et al., 2018}; ^{Purvanova & Muros, 2010}) in which women's emotional exhaustion and job dissatisfaction scores were also found to be higher than those of men.

Taking all the above into consideration, in this study we expect to find that:

We will also analyze the invariance of the shorter version of the WDQ in order to ensure that it provides equivalent measurement irrespective of gender (Objective 2). This will contribute to the consideration of the gender perspective in occupational health future research. In addition, we will explore gender differences with respect to the work characteristics analyzed in the shorter version of the WDQ (Objective 3).

Procedure

A protocol was developed prior to commencing the research that was subsequently assessed by an independent ethics review committee¹. Having approved the protocol, the researchers conducting the study trained several interviewers. Trained interviewers asked diverse companies and workers for participation. Participation criteria were: to be working at least six months during the last two years in Spain, and not to be self-employed nor to be a student. We requested the collaboration to fill an online questionnaire of a number of incumbents and then to distribute that link to other workers selected by means of non-probabilistic snowball sampling. The questionnaire included socio-demographic variables and various scales. All participants acknowledged informed consent before completing the questionnaire.

Respondent anonymity and the research purpose were emphasized throughout the process in order to avoid social desirability effects and reduce evaluation apprehension (e.g., ^{Podsakoff et al., 2012}).

Participants

A total of 566 people participated in the study. However, 66 responses were discarded for the following reasons: (i) 39 self-employed respondents completed the questionnaire, despite the fact that the questionnaire is intended for salaried employees; (ii) four participants were discarded because they indicated that they had no work experience beyond their final student work placement; and (iii), the contributions of 23 participants were discarded because they failed to correctly follow the instructions for answering two items (e.g., "Please indicate option (X) for this question"), which were included to avoid low-quality data (^{DeSimone & Harms, 2017}).

The final sample of 500 participants consisted of 234 men (46.8%) and 266 women (53.2%), with a mean age of 37.59 years (SD = 12.704). Of these, 61.2% had received a university education. In terms of work context, the average number of employees in the organizations to which the participants belonged was 252.19 (SD = 967.437), with 50% of the participants belonging to organizations with 30 or fewer employees, and around 10% to organizations with 300 or more employees. 75% of the participants were from the private sector, and the remaining 25% from the public sector. Furthermore, as can be observed in Table 2, the incumbents were drawn from a wide variety of sectors, which facilitates the consistency of results with respect to the development and validation versions of the WDQ in previous studies. Participants were grouped and assigned to fourteen categories based on the twenty nine sectors of the National Classification of Economic Activities (CNAE) in Spain. It should also be noted that organizational experience (M = 111.814 months, SD = 178.744) and on-the-job experience (M = 73.43 months, SD = 102.739) was generally high, which ensured that responses to the WDQ were reliable. A total of 62.6% of the participants had a permanent contract, 31% had a temporary contract and 6.4% had a work or service contract. With regard to type of working day, 67% worked a continuous working day and the remaining 33% worked a split shift.

Table 2: Number and percentage of participants by sector. 

Instruments

In addition to questions on socio-demographic data, the online questionnaire also included the following scales:

Work characteristics: measured using the 77 items of the Spanish version of the WDQ developed by ^{Bayona et al. (2015}). The scale is composed of 21 factors, as previously explained in the section on the WDQ: work scheduling, decision-making and work methods autonomy (i.e., “The job allows me to make my own decisions about how to schedule my work”), task variety (i.e., “The job involves a great deal of task variety”), task significance (i.e., “The job has a large impact on people outside the organization”), task identity (i.e., “The job involves completing a piece of work that has an obvious beginning and end”), feedback from job (i.e., “The job itself provides me with information about my performance”), job complexity (i.e., “The tasks on the job are simple and uncomplicated”; reverse scoring), information processing (i.e., “The job requires me to analyze a lot of information”), problem solving (i.e., “The job requires me to be creative”), skill variety (i.e., “The job requires the use of a number of skills”), specialization (i.e., “The job requires a depth of knowledge and expertise”), social support (i.e., “I have the chance in my job to get to know other people”), initiated and received interdependence (i.e., “The job requires me to accomplish my job before others complete their job”), interaction outside organization (i.e., “The job involves a great deal of interaction with people outside my organization”), feedback from others (i.e., “I receive a great deal of information from my manager and co-workers about my job performance”), ergonomics (i.e., “The job involves excessive reaching”; reverse scoring), physical demands (i.e., “The job requires a great deal of muscular endurance”; reverse scoring), work conditions (i.e., “The work place is free from excessive noise”), equipment use (i.e., “The job involves the use of a variety of different equipment”; reverse scoring). The Cronbach's alpha reported by the authors ranges from .57 (ergonomics) to .95 (physical demands). Participants responded using a five-point Likert-type scale, from 1 (strongly disagree) to 5 (strongly agree).

Job satisfaction: measured using the 12 items of the S10-S12 Job Satisfaction Questionnaire (e.g., "The support you receive from your superiors") developed by ^{Meliá and Peiró (1989)}, who reported a Cronbach's coefficient of .88. Participants responded using a 7-point response scale, from 1 (very dissatisfied) to 7 (very satisfied).

Burnout: measured using five items from the emotional exhaustion dimension taken from the Spanish version of the Maslach Burnout Inventory General Survey (e.g., "I feel used up at the end of a workday") adapted into Spanish by ^{Salanova et al. (2000)}, who found a Cronbach's coefficient of .85. Emotional exhaustion have been seen as the core of burnout and is the best reliable burnout dimension (e.g., Salanova et al., 2000; ^{Schaufeli & Dierendonck, 1993}). Moreover, the content of this scale has a greater relationship with labour health issues than the other dimensions (depersonalization and reduced sense of personal accomplishment) which are more related with job performance. Participants responded using a 7-point response scale from 0 (never) to 6 (always).

Well-being: measured using the 12 items of the Spanish version of the General Health Questionnaire (e.g., "Felt constantly under strain") by ^{Goldberg and Williams (1996)}, which indicated a Cronbach's Alpha of .86. Considering this instrument as unidimensional can be used for differentiating psychological morbidity (anxiety combined with depression). Participants responded using a 4-point response scale, from 0 (not at all) to 3 (much more than usual).

Data analysis

The dimensionality of the WDQ was analyzed through confirmatory factor analysis (CFA), using Mplus software (version 8.1, ^{Muthén & Muthén, 2012}). The result of the Kolmogorov-Smirnov test with Lilliefors correction was significant (p < .05), so the normality of the data could not be confirmed. We therefore used categorical weighted least squares, since these provide robust solutions (e.g., ^{Castaño & García-Izquierdo, 2018}), and conducted bootstrapping analyses. In addition, the common method bias was verified by means of the Harman's one-factor test, described in ^{Podsakoff et al. (2003}), in which all items loaded on a single factor.

As a strategy to reduce the WDQ, two researchers took part in a two-phase inter-rater agreement process aimed at reaching an agreement as to which items should be eliminated. In the first phase, each researcher indicated the items they judged appropriate to eliminate according to the following criteria (e.g., ^{Azañedo et al., 2017}; ^{Lloret-Segura et al., 2014}): the reliability of the dimensions; the loadings of the items; and the repetitiveness of the items for each dimension. In the second phase, the researchers were able to make the appropriate modifications until total or close-to-total agreement was reached, in accordance with criteria established by ^{Landis and Koch (1977}). The reliability and factorial structure of the shorter version of the instrument were then analyzed, as well as the correlations with the original instrument (e.g. ^{DeBode et al., 2013}).

The following comparative and adjustment indexes were used to test for goodness of fit (e.g., ^{Browne & Cudeck, 1993}): (i) comparative fit index (CFI) and Tucker-Lewis index (TLI), where values of 0.90 to 0.95 suggest an acceptable fit and values above 0.95 suggest a good fit; (ii) root mean square error of approximation (RMSEA), where values of 0.05 and below indicate a good model fit, values of 0.05 to 0.08 a moderate fit, and values of 0.10 and above a poor fit; and, finally (iii), the χ2/degree of freedom ratio, where values of between one and three suggest a good fit.

In addition, the invariance of the WDQ across gender was also analyzed. A sequential analysis was conducted for this purpose, using a configural invariance model, and applying progressive restrictions towards a metric and scalar invariance model (e.g., ^{Elosúa, 2005}; ^{Ramis et al., 2015}). Indicators of invariance were defined as incremental changes in CFI, TLI and RMSEA not exceeding |0.01| (e.g., ^{Chen, 2007}; ^{Cheung & Rensvold, 2002}).

Finally, analyses of descriptive statistics, Cronbach's reliability index, t-tests and Pearson's correlations were conducted using SPSS software (version 20). The calculator developed by ^{Diedenhofen and Musch (2016)} was used to compare Cronbach's reliability indices, and the calculator developed by ^{Ellis (2009}) was used to calculate the effect size. We took d= .20 as a small effect, d= .50 as a medium effect, and d= .80 as a large effect in t-tests analyses (e.g., ^{Cohen, 1988}). In addition, G*Power software version 3.1.9.2 by ^{Faul et al. (2009}) was used to determine r= .144 as the Pearson's minimum correlation coefficient with 500 participants which is sensitive to effects to 90% power (α = .05, two-tailed), meaning that this research would not be able to reliably detect correlations smaller than that value. Missing data were then removed by means of listwise deletion.

Confirmatory factor analysis and reduction of the WDQ

The results in Table 3 show that the fit for the single factor and 4-factor structure was poor, while the best fit was obtained for the structures with 18 or more factors. Being parsimonious (in order to select lower number of factors and similar RMSEA), we took as a starting point the 18-factor structure (RMSEA = 0.049, 90% CI (0.047, 0.050), χ2 = 5889.538, df = 2696, χ²/df = 2.185, TLI = 0.951, CFI = 0.954) in order to reduce the number of items for each dimension of the WDQ. Taking into account the criteria for reliability, factor loadings, content, and number of items per dimension described in the Data Analysis section, the first phase of the inter-rater agreement process for reducing the number of items in the WDQ resulted in a Kappa index of .54. Given that this agreement index was moderate, the second phase of the agreement process involved the pooling of the researchers' responses until total agreement was reached (Kappa index of 1) on the WDQ items to be eliminated. Specifically, in this second phase, it was agreed to eliminate 25 items, which therefore left a WDQ with 52 items. Table 4 contains a list of the 25 items eliminated from the WDQ (the original 77 items can be consulted in ^{Bayona et al., 2015}) in the second phase of constructing the shorter 52-item version of the WDQ, which we have named WDQ18-S. Finally, Table 3 shows the fit indices, where it can be observed that the 18-factor 52-item structure of the WDQ18-S presents adequate results with the items selected after the reduction process (RMSEA = 0.058, 90% CI (0.056, 0.061), χ² = 3011.491, df = 1121, χ2/df = 2.686, TLI = 0.944, CFI = 0.953).

Table 3: Results of the CFA of the WDQ and its shortened version (WDQ 18-S). 

Table 4: Eliminated items as a result of the inter-rater agreement process used to establish the WDQ18-S. 

The WDQ18-S was subsequently analyzed in order to verify the extent to which it continued to measure the constructs of the original WDQ18. For this purpose, the 18 dimensions of the 52-item WDQ18-S were correlated with the corresponding 18 dimensions of the original 77-item version. In addition, an analysis was carried out to determine whether there were significant differences between the reliability indices of the dimensions of the two versions. Table 5 shows high correlations exceeding .9, between the dimensions of the original WDQ and the WDQ18-S. It should also be noted that in 10 of the 18 dimensions of the WDQ there were statistically significant differences between the comparative reliabilities of the two versions, with a lower value obtained for the shorter version, although this lower value was less than 10% for 9 of the 10 dimensions: autonomy, task variety, task significance, task identity, feedback from job, skill variety, social support, initiated and received interdependence and feedback from others. However, although there was a more than 10% decrease with respect to the original version, in the case of the remaining specialization dimension, the reliability index was higher than .75.

Table 5: Correlations and reliability comparison between the dimensions of the original 18-factor WDQ and the WDQ18-S. 

Measurement invariance analyses were also conducted considering gender. As shown in Table 6, the fit indices obtained for the configural model allowed us to accept the equivalence of the instrument for the two groups. Subsequently, the metric invariance model also presented a good fit, and there was no increase in the indices with respect to the less restrictive model. Finally, given that the scalar invariance model also presented good fit indices, and following the established criteria, the absolute increase in the CFI, TLI and RMSEA did not exceed 0.01, the scalar invariance model could also be accepted.

Table 6: Gender invariance results of the WDQ18-S. 

Criterion-oriented validity

Table 7 shows the descriptive statistics of the dimensions of the WDQ18-S and the criterion variables.

Table 8 shows the correlations between the dimensions of the WDQ18-S and the other variables. With regard to the significant correlations between the dimensions of the WDQ18-S and health variables, we highlight those above r= .144. This is the coefficient sensitive to effects with 90% power (α = .05, two-tailed) for our sample size, that is: (i) job satisfaction correlated significantly and positively with: autonomy, task significance, task identity, feedback from job, specialization, social support, feedback from others, ergonomics, physical demands and work conditions; (ii) emotional exhaustion correlated negatively with autonomy, task significance, feedback from job, social support, ergonomics, physical demands, and work conditions; and (iii) well-being correlated positively with autonomy, task variety, task significance, task identity, feedback from job, social support and ergonomics. We need to remind that physical demands and ergonomics are reverse scored.

Table 7: Descriptive statistics of the WDQ18-S and criteria. 

Table 8: Pearson's correlations between the dimensions of the WDQ18-S, occupational health variables and sociodemographic variables. 

Moreover, Table 9 shows the analyses of mean differences for the dimensions of the WDQ18-S and occupational health outcomes by gender. Statistically significant differences were found showing a small effect size for the following dimensions of the WDQ18-S: specialization (p = .014, d = 0.219) and initiated and received interdependence (p = .008, d = 0. 259), with higher scores for men than for women; while in physical demands (p = .004, d = -0.274), work conditions (p = .002, d = -0.417) and equipment use (p = .002, d = -0.538), higher scores were found for women than for men, reflecting a small to medium effect.

Table 9: T-test for equality of means between WDQ18-S and occupational health variables depending on gender. 

Limitations and future research

It is important to mention the limitations of the present study, as well as lines of future research. Firstly, the fact that this is a cross-sectional study makes it difficult to establish causal relationships between work characteristics and occupational health outcomes, and the results may be affected by the problem of common variance. It would therefore be interesting to carry out longitudinal studies to analyze the causal relationships between organizational factors, their consequences, and their evolution over time, in order to establish the interventions that most closely correspond to reality.

Secondly, the fact that the participants belong to different organizations prevents detailed data analysis that takes into account other types of organizational variables that may have affected the results obtained. In this regard, it would be particularly useful to have a sample of each relevant organizational sector in Spain that is large enough to allow a comparative analysis. This broadening of the sample by sector would also enable an exhaustive analysis of the effects of occupational segregation across gender, and the possible impact of gender stereotypes on occupational health.

Thirdly, although our sample size is not so big, we must emphasize the use of robust analysis techniques for reducing these potentials adverse effects. Thus, our results can be of interest due sector variety and gender balanced sample composition as well. Nevertheless, we highlight the need to study these job characteristics in more detail, in order to analyze their relationship with other important organizational outcomes such as job performance.