The Work Design Questionnaire: Spanish version and validation

The translation of the WDQ into Spanish was accomplished through the translation/back-translation procedure recommended by Brislin (1980) . The researchers first translated the WDQ from English to Spanish and then a bilingual professional translator with experience in the Business Administration field back-translated the Spanish version into English. Following the translation from English to Spanish, we compared the original questionnaire to the back-translated English version and differences were resolved through discussion among authors; the professional translator was not aware of the study purpose. The wording of items was aimed to reflect general forms rather than specific idioms and expressions of the Spanish language as it is spoken in different countries. An initial version of the questionnaire was tested with a group of 18 workers from different occupational levels; once the questionnaire was administered, an interview with these workers was conducted in order to identify problems with the language expressions or wording. The resulting Spanish questionnaire used in the validation is presented in the Appendix.

The sample was collected in Colombia, a country classified as an upper-middle-income economy with a GDP per capita (purchasing power parity) of US$ 9,125 (USA, US$ 43,063), an average economic growth from 2007 to 2011 of 4.5%, an annual employment growth of 3.5%, and 46.4% of its total work force being wage and salaried workers (USA, 93.2%); 77.2% of these salaried workers are distributed across four main economic sectors: wholesale and retail service (26.4%), health and social work (19.9%), agriculture, forestry, and fishing (17.5%), and manufacturing (13.4%); this same four economic sectors represent the 40.5% distribution of salaried workers in the USA ( International Labour Organization, 2013a, 2013b ).

Eight hundred forty-one Colombian employees participated in the study; however, 10 questionnaires were not usable due to participants not responding the WDQ section, leaving 831 useful questionnaires. The mean age for all participants was 34.9 years (range: 18-70, SD = 11 years); the mean tenure for all participants was 6.05 years (range: 1-51, SD = 7.48 years); 43% of workers were females (98.5 valid percent), 7.7% of respondents had completed education only at a high school/diploma level, 69.9% had completed undergraduate level (university, technical, or technological education), and 16.2% had completed postgraduate level or higher (93.8 valid percent).

Data were collected from three different samples. Sample 1 consisted of 279 full-time employees working for an organization that manufactures pumps, compressors, and valves (83.5% men; age: M = 31.2 years, SD = 8.2 years; education: M = 14.4 years, SD = 3.5 years); this sample d both blue and white collar workers. Sample 2 consisted of 89 full-time administrative employees working for a university (37% men; age: M = 39.3 years, SD = 8.8 years; education: M = 14.3 years, SD = 5.2 years). Sample 3 consisted of 463 full-time employees working for different organizations (44% men; age: M = 36.2 years, SD = 12.3 years; education: M = 16.2 years, SD = 4.1 years). Sample 3 was obtained in the context of an organizational behavior course with junior-level business administration students. These students d the work of a family member or acquaintance (job incumbent) who has worked full time for at least one year, and administrated the WDQ to the job incumbent (the students received specific training on the application of the questionnaire from one of the authors). This particular sampling strategy was employed so data could be collected on a wide range of jobs following the strategy used in the original work of Morgeson and Humphrey (2006) and is used when the goal is to sample a wide range of different jobs (e.g., Raymark, Schmit, & Guion, 1997 ). Workers in all samples filled a paper-and-pencil version of the WDQ and were informed about the confidential use of all information provided. The procedures were approved by the ethics review board of Pontificia Universidad Javeriana before the study began.

The distribution of the whole sample, depending on the economic sector in which job incumbents work, can be seen in Table 1 . This distribution s 17 out of 21 economic activities considered in the International Standard Industrial Classification of All Economic Activities (ISIC) (United Nations, 2008 ). However, it is important to note the under-representation of the agriculture, forestry and fishing sector, which has a very important role in the Colombian economy. The explanation for this under-representativeness is because sample was collected in an urban context and the access to these workers is difficult. Nevertheless, apart from this sector, the sample reflects the general formal labor market composition in Colombia.

Occupational classification . We d five broad occupational categories (i.e., professional, non-professional, health, commercial, and manufacturing occupations) to test the occupational focused hypothesis ( H2a, H2b, H2c, H 2d). For professional and nonprofessional classification, job incumbents self-reported if their occupation was professional (their work activities a university degree) or nonprofessional (their work activities do not a university degree). For the other categories (health, commercial, and manufacturing occupations), job incumbents self-reported the economic sector in which their organization was located using the ISIC classification ( Table 1 ). For sample 1 (manufacture organization), all workers were classified in the manufacture sector; for sample 2 (university), all workers were classified in the education sector; for sample 3 (different organizations), business administration students helped the job incumbents to report this information. The “health” category was composed of the jobs within the “human health and social work activities” of the ISIC classification, whereas the “non-health” category was composed of the jobs in the remaining occupations. The “commercial” category was composed of the jobs within the “wholesale and retail service” of the ISIC classification. Finally, the “manufacturing” category was composed of the jobs within the “manufacturing” group in the ISIC classification.

A one-way ANOVA was conducted in order to verify any possible differences in demographic variables among the samples. For age, tenure, and education, significant differences were detected: age, F(2, 802) = 26.02, p ? .01 (sample 1, M = 31.2 years, sample 2, M = 39.3 years, sample 3, M = 36.2 years); tenure, F(2, 811) = 42.88, p ? .01 (sample 1, M = 3.1 years, sample 2, M = 10.3 years, sample 3, M = 7 years); education, F(2, 778) = 19.21, p ? .01 (sample 1, M = 14.4 years, sample 2, M = 14.3 years, sample 3, M = 16.2 years). Because of these results, analyses that use full-sample will be performed using these three variables as controls on the regression model of hypotheses 4a and 4b.

Work design. We used the WDQ developed by Morgeson and Humphrey (2006) that is a self-reporting measure that s 77 items using a 5-point Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree). Cronbach's alpha of the original version of the WDQ ranged from .64 (ergonomics) to .95 (task variety and physical demands), with a mean alpha = .86.

Job satisfaction was measured using the 4-item job satisfaction scale developed by De Witte (2000) in its Spanish version used by Isaksson (2007) . The scale uses a 5-point Likert scale ranging from 1 ( nothing important) to 5 (extremely important ). The Spanish version Cronbach's alpha = .81. An example item is: “La mayoría de los días estoy entusiasmado/a con mi trabajo.”

Perceived performance was measured through the 6-item perceived performance scale developed by Abramis (1994) in its Spanish version used by Isaksson (2007) . This scale asks respondents to think about their previous week at work and to rate how well they performed on six tasks, namely decision-making, performing without making mistakes, goal attainment, effort, taking initiatives, and taking responsibility. Five response categories were used, ranging from 1 ( very badly) to 5 (very well ). The Spanish version Cronbach's alpha = .79. An example item is: “¿En qué medida ha realizado satisfactoriamente la siguiente tarea? - Trabajar sin cometer errores.”

To test hypothesis 1, we used confirmatory factor analysis (CFA) conducted with AMOS 21 ( Arbuckle, 2012 ). We applied a maximum likelihood estimation method; in order to run this analysis, no missing data will be allowed; taking into account that the missing data pattern was completely at random ( Schafer & Graham, 2002 ) and that the missing data percentage was less than 2% per variable, a mean imputation was done using SPSS. Maximum skewness was -1.74 and maximum kurtosis was 3.83, which is among the recommended bounds for skewness |2.0| and kurtosis |7.0| ( Hancock & Mueller, 2010).

Consistent with Morgeson and Humphrey (2006) , we report various fit indices: ? 2/df ratio, comparative fit index (CFI), standardized root-mean-square residual (SRMR), and root-mean-square error of approximation (RMSEA). For ? 2/df ratio, a value of 2.0 indicates good fit. For CFI, values higher than .90 indicate good fit. For SRMR, values lower than .08 indicate good fit. For RMSEA a value of .05 indicate good fit. We used the chi-square difference ( ??2 ) to compare the models, and accepted the more parsimonious model if it was not significantly different from a more complex model. We tested five different models of work design using CFA techniques:

• •

  4-factor model examines the four broad categories of work characteristics (i.e., task, knowledge, social, and work context characteristics).

• •

  18-factor model examines the work characteristics without any divisions (i.e., autonomy and interdependence as unique factors).

• •

  19-factor model separates interdependence into its received and initiated components.

• •

  20-factor model separates autonomy into its three components, which s autonomy in work scheduling, decision making, and work methods.

• •

  21-factor model separates both interdependence and autonomy into the identified components.

To test hypotheses 2a, 2b, 2c and 2d we used t -tests to compare means of two occupational segments: professional and non-professional workers, health and manufacturing workers, commercial and manufacturing workers. To test hypothesis 3a and 3b we performed bivariate Pearson correlations; finally, to test hypothesis 4a and 4b, we used hierarchical regression analyses aiming to compute incremental validity, which determined the variance accounted for knowledge, social, and work context characteristics in job satisfaction and perceived performance beyond the variance explained by task characteristics.

In order to test the first hypothesis, that the Spanish WDQ version will fit a 21-factor structure, we run a set of five different CFAs. The results of our CFAs are presented in Table 3 , broken down for each sample. First, for the full sample, the 4-factor model showed poor fit, as the fit indexes were all off the accepted levels. Second, the 18-, 19-, and 20-factor solutions showed adequate fit, with the SRMR and RMSEA reaching adequate levels, whereas the CFI was slightly low and the ? 2 /df was slightly high. Finally, we tested the 21-factor solution; this model was the best fitted model overall, with the lowest ? 2/df ratio, SRMR, RMSEA and the highest CFI. This model was significantly better than the 18-factor model (? 2 change = 627, df change = 57, p < .001), the 19-factor model (? 2 change = 424, df change = 39, p < .001) and the 20-factor model (? 2 change = 204, df change = 20, p < .001). These same results patterns are present in the separate analyses for each sample ( Table 3 ); thus, like the original WDQ, the 21-factor model, which separates interdependence in two factors and autonomy in three factors, fits our data the best.

Additional evidence for a 21-factor structure is provided when the structure of task and social characteristics are studied separately. For task characteristics, a comparison between 5- and 7-factor model was developed (separating or integrating the three dimensions of autonomy). The 7-factor model performed better in ? 2 /df ratio in comparison to the 5-factor model (? 2 change = 372, df change = 11, p < .001), SRMR = .08, RMSEA = .07, CFI = .93). For social characteristics, a comparison between a 4- and a 5-factor model was developed (separating or integrating the two dimensions of interdependence); the 5-factor model performed better in ? 2 /df ratio than the 4-factor model (? 2 change = 184, df change = 4, p < .001), SRMR = .06, RMSEA = .07, CFI = .92). In sum, these data fully support hypothesis 1.

Once we confirmed the 21-factor measurement model, it is important to assess the extent to which the items of a specific factor converge or share a high proportion of variance (convergent validity). In addition, it is also important to assess the extent to which a factor is truly distinct from other factors both in terms of how much it correlates with other factors and how distinctly items represent only this single factor (discriminant validity).

To evaluate convergent validity, two methods were used: (a) assessment of standardized factor loadings of observable variables and (b) average variance ed (AVE) for each factor. For the first method, a comparison of factor loadings of each item was conducted, loadings estimates should be significant and with a factor loading of .50 or higher for the associated item; for the AVE method, “the average variance ed is calculated as the mean variance ed for the item loading on a [factor] and is a summary indicator of convergence” ( Hair, Black, Babin, & Anderson, 2010 , p. 687) and values greater than .50 are considered adequate.

On the other hand, to evaluate discriminant validity, a comparison between the AVE and the maximum shared squared variance (MSV) of each factor was carried out; the MSV represents the maximum shared squared variance found when comparing for any two factors (in this case, each factor was compared with all other 20 factors); with the square of the correlation estimate between these two factors, the AVE estimates should be greater than the MSV; this is because “a latent [factor] should explain more of the variance in its item measures that it shares with another construct” ( Hair et al., 2010 , p. 688); when the MSV values are lower than the AVE, the measure has good discriminant validity.

The results for convergent validity indicate that the standardized factor loadings of the WDQ indicate average estimate of .74, with just 4 out of 77 items below .50 (items 15, 18, 25, and 66). 1 1

Full standardized factor loadings are available from the first author.

For practitioners, a broader range of work design potentials is possible beyond the traditional five job characteristics of the JDS; however, it is important that practitioners be fully aware of potential cultural influences in the work design practice. In the case of Latin-America countries, a deep collectivism value can be found, and in Colombia there is an important role of collectivism and rejection to individualism ( Ogliastri, 2007 ). In an article from the GLOBE project ( Dorfman, Javidan, Hanges, Dastmalchian, & House, 2012 ), the authors report that the autonomous dimension (tendencies to act independently without relying on others) is strongly negative related to institutional collectivism (degree in which organizational practices encourage and reward collective distribution of resources and collective action); this relation is important in order to redesign (increase) autonomy in positions where high institutional collectivism is present.

With regard to research, the WDQ is a tool that allows to investigate the impact of different work configurations on organizational and personal outcomes and let open a research line of the influence of different mediators and moderators in the relation between work characteristics and personal and organizational outcomes (e.g., cultural characteristics).

Two major drawback limited the present research: first, the presence of some level of common method bias, which implies that the results must be interpreted with caution, even though the CLF test results indicated that variance due to common method is between the acceptable limits. This is consistent with previous research on common method variance, which has concluded that while common method bias may be present, it may not always significantly affect the results and conclusions drawn from the data ( Crampton & Wagner, 1994; Doty & Glick, 1998 ). The second limitation was the sample method ion; although we used three different samples and almost all occupational groups of the ISIC were considered, some groups were sub-represented (e.g., arts, entertainment, and recreation); besides, it is also important to consider that half of the labor market in Colombia is informal and the conclusions of this study can apply only to the workers that are inside the formal labor market (50% of the Colombian total labor force). Future research should consider the structure of work characteristics in jobs d in the informal labour market. They represent in developing countries an important amount of the total economy and deserve a better analysis and understanding.