MEASURING POSITIVE AND NEGATIVE OCCUPATIONAL STATES (PNOSI): STRUCTURAL CONFIRMATION OF A NEW BELGIAN TOOL

————— Marie Barbier is affiliated to the National Fund for Scientific Research, University of Liège; Stéphanie Peters and Isabelle Hansez are affiliated to the Human Resources Valorisation Unit, University of Liège. Data from samples 2 and 3 were collected thanks to grant PS/12/29 – 1999/2003 from the Belgian Science Policy. We wish to thank Fabrice De Zanet, Anne Grisard, Hélène Spitz and Sophie Taeymans for collecting data. Correspondence concerning this article should be addressed to M. Barbier, National Fund for Scientific Research, University of Liège, Boulevard du Rectorat 5 (B32), 4000 Liège. E-mail: Marie.Barbier@ulg.ac.be MEASURING POSITIVE AND NEGATIVE OCCUPATIONAL STATES (PNOSI): STRUCTURAL CONFIRMATION OF A NEW BELGIAN TOOL

The positive psychology perspective has spread to the field of work and organisational psychology. Researchers have been interested not only in negative responses to work demands, but also in positive ones, such as work motivation and engagement. In their so-called "Job Demands-Resources (JD-R) model", Bakker (2004, see also Bakker &Demerouti, 2007;Demerouti, Bakker, Nachreiner, & Schaufeli, 2001) consider both positive and negative well-being at work as different responses developing in different conditions. This model proposes that job characteristics are conceived as either job demands (e.g., high time pressure, workload) or job resources (e.g., participation in decision-making, autonomy). The presence of high job demands would lead to burnout and then to negative outcomes such as absenteeism or health problems (health-impairment process). The presence of job resources would lead to engagement, and then to positive organisational outcomes, such as a low intent to leave the organisation (motivational process).
The theory of affects proposes the existence of two kinds of affect, whereas work psychology proposes the existence of two kinds of individual reactions to the working environment. Even if their topic of interest is different, both ask the same question: What is the relation between the two concepts? While it is widely accepted that two kinds of affect exist (Folkman & Moskowitz, 2000), there is disagreement with respect to the relationship they have with each other. Some authors defend the independence of the two (Warr, Barter, & Brownbridge, 1983), whereas others consider them as opposite poles of a bipolar continuum (Russell & Carroll, 1999). In regard to wellbeing at work, the JDR model considers positive and negative well-being at work to be different constructs that should be measured with different tools or items (Schaufeli, Salanova, Gonzalez-Roma, & Bakker, 2002).
We concurred with the latter perspective when we developed the Positive and Negative Occupational States Inventory (PNOSI) in 2001. This inventory was designed to measure both positive and negative aspects of well-being at work each with different items. De Keyser and Hansez (1996) and Hansez (2008) define negative occupational state as the response of workers facing demands they feel they have to manage, but for which they doubt they have the necessary resources. In contrast, we define positive occupational state as the response of workers facing demands they feel they have to manage and for which they perceive necessary resources are available.
The PNOSI has been used for diagnostic purposes in French-speaking countries. More than 45,000 participants have been tested with the PNOSI.
It has demonstrated good reliability (Cronbach's α higher than .80) in diverse occupational fields such as veterinary practice, farming, informatics, printing, pharmaceutics, and the energy, public, and health care sectors (Dardenne, Coibion, & Hansez, 2007;Hansez & Bertrand, in press;Hansez, Bracci, & Bertrand, 2007;Hansez & Chmiel, in press;Hansez, Schins, & Rollin, 2008). Given its wide use, it is of the utmost importance to establish the structural validity of this instrument. Preliminary factor analyses have confirmed the two-factor structure of the PNOSI (Barbier & Hansez, 2008). Results show that eight items measure positive occupational state, whereas the remaining eleven measure negative occupational state. However, these analyses were exploratory in nature. The first aim of this paper is to test the structural validity of the PNOSI by the means of confirmatory factor analyses.
Our second objective is to test its stability and generalizability across samples of workers active in different areas. We hypothesise that the previously shown two-factor solution would fit the data better than the alternative one-factor solution.
The third aim of this paper is to establish the convergent and discriminant validity of positive and negative states, as measured by the PNOSI, from related constructs such as engagement, commitment, burnout, or workaholism. Engagement at work is defined as a persistent, positive, and satisfying state of mind linked to work, not focused on any particular object, event, person, or behaviour (Schaufeli et al., 2002). Based on this conceptualisation of engagement as an "assumption of 'optimal functioning' at work in terms of well-being" (Hallberg & Schaufeli, 2006, p. 120), we believe positive occupational state to be an intermediate state occurring prior to engagement. We thus hypothesise that there is a positive correlation between positive occupational state and work engagement scores (hypothesis 1a) and a negative correlation between negative occupational state and engagement at work (hypothesis 1b).
Burnout can be defined as a combination of extreme fatigue (which is called exhaustion) and distancing oneself from one's work (Demerouti, Bakker, Vardakou, & Kantas, 2003). According to Demerouti, Bakker, Nachreiner, and Ebbinghaus (2002, p. 426), "whereas short-term effects of strain develop as an immediate reaction to specific work conditions …, burnout will develop only after repeated, prolonged, and unsuccessful confrontation with such conditions". They found both exhaustion and disengagement components of burnout to be significantly correlated with short-term strain effects, such as mental fatigue or monotony (here defined as a worker's reaction and not as a characteristic of the task).
According to Hockey's model of compensatory control (1997) and according to Schaufeli and Bakker (2004), workers facing an imbalance between demands and resources at work can react in a strain coping or passive coping way. In the strain coping way, workers put in greater effort so that goals are achieved at the expense of physiological and psychological costs, leading to exhaustion. In the passive coping way, workers make a downwards adjustment so that health is preserved at the expense of performance, leading to disengagement from work. According to Mackay and Cooper's transactional model of stress (1987), perceiving an imbalance in the situation leads to perceived stress and to coping strategies which have implications for mental and physical health. In both the compensatory and transactional model, coping is described as an antecedent of burnout. Moreover, in the transactional model the stress reaction is simultaneous to coping strategies. Based on the two models, and on our definition of negative occupational state as a reaction to a demands-resources imbalance at work, we consider negative occupational state to be an intermediate state occurring prior to burnout. We thus hypothesise burnout to correlate positively with negative occupational state (hypothesis 2a) and negatively with positive occupational state (hypothesis 2b). Allen and Meyer (1990) conceptualise commitment as a combination of three components. Affective commitment refers to the extent to which a worker identifies with, gets involved in, and feels affectively attached to his/her organisation. Continuance commitment refers to a worker's tendency to stay with his/her organisation to avoid disadvantages linked with quitting. Finally, normative commitment refers to a worker feeling of loyalty and duty to his/her organisation. Thus, affectively committed workers stay because they want to, continuance committed ones stay because they need to, and normatively committed ones stay because they feel they have to. According to Meyer and Allen (1997), affective commitment would be associated with a motivation or a desire to contribute to the organisation, whereas continuance commitment could be associated with frustration and could lead to inappropriate work behaviour. Finally, normative commitment, as a feeling of obligation, would be associated with appropriate work behaviour, but one can question whether such behaviour is performed full heartedly. The authors review studies showing negative correlations between stress and affective commitment, and between stress and normative commitment. Correlations between stress and continuance commitment on the other hand were not significant. In their meta-analysis, Meyer, Stanley, Herscovitch, and Topolnytsky (2002) found desirable work behaviour (such as job involvement) to be positively correlated with affective commitment and (even if less strongly) to normative commitment, and not correlated with continuance commitment. We hypothesise positive occupational state to be positively correlated with affective commitment (hypothesis 3a) and normative commitment (hypothesis 3b), and negatively or not significantly correlated with continuance commitment (hypothesis 3c). We also expect negative occupational state to be negatively correlated with affective commitment (hypothesis 3d), and normative commitment (hypothesis 3e), but not correlated with continuance commitment (hypothesis 3f).
According to Schaufeli, Taris, and Bakker (2006), two types of behaviour define workaholism: working excessively and working compulsively (driven by an inner compulsion and not because of external factors). They have found working excessively, but not working compulsively, to correlate with engagement. They conclude that the inner drive differentiates workaholism from engagement. We believe positive occupational state differs from workaholism because it lacks the compulsive component that characterises workaholic people. A study by Burke (2000) has shown that workaholic people exhibit lower emotional and physical well-being, and more somatic complaints. Finally, Schaufeli, Taris, and van Rhenen (2008) have shown that working excessively is positively correlated with engagement and not correlated with burnout. We hypothesise that positive occupational state correlates positively with working excessively, but does not correlate with working compulsively (hypotheses 4a and 4b respectively). On the contrary, we hypothesise that negative occupational state correlates positively with working compulsively, but does not correlate with working excessively (hypotheses 4c and 4d respectively).

Samples
Eight different samples were used: a calibration sample, a validation sample, and five replication samples (they are defined in the "Analyses" section). A last sample was used to test PNOSI convergent and discriminant validity. Sample distributions as regards to gender, age, and organisational tenure are shown in Table 1.

Calibration and validation samples
Data gathered from a large sample of workers employed in a Belgian electricity production and selling company were randomly split into two halves. In the calibration sample (referred to as sample 1a, N = 2713), 76.89% of the workers are male (n = 2086). A large majority are between 25 and 55 years old (n = 2202, 81.16%) and have worked in the company for 11 years or more (n = 1729, 63.73%). A large majority of validation sample workers (referred to as sample 1b, N = 2714) are male (n = 2098, 77.30%), between 25 and 55 years old (n = 2169, 79.92%), and have worked for the company for 11 years or more (n = 1830, 67.43%).

Replication samples
Data collected in five French and Belgian companies were used as replication samples (referred to as samples 2-6) 1 . Samples 2 and 4 consist of workers employed in public institutions (N = 1345 and N = 1998, respectively). Samples 3 and 5 consist of workers employed in the service sector, specifically a large sale company and an insurance company (N = 3591 and N = 5280, respectively). Sample 6 consists of workers employed in a mental and physical health institution (N = 903).

Validity sample
Data gathered in a Belgian public institution (N = 955) were used as a validity sample (referred to as sample 7). Men and women were equally represented (n = 425, 44.50% and n = 529, 55.39%, respectively) and a majority of workers had worked for this institution for 20 years or more (n = 420, 43.98%).

Wellbeing at work
In developing the PNOSI, we first asked university students (N = 150) to list five words describing negative well-being and five others describing positive well-being. Qualitative analyses showed the words represented four dimensions, namely an emotional, a cognitive, a behavioural, and a somatic dimension. We selected the most frequently cited for each dimension. They were included in the first PNOSI version, which counted 34 items. This questionnaire was administered to employees of a Belgian high school (N = 152). Exploratory factor analyses with Varimax rotation were performed. The results revealed a three-factor solution. The first factor was composed of eleven items related to negative experiences, while the second was composed of eight items related to positive experiences. The third factor was composed of two items related to work performance and was not included in the final PNOSI questionnaire.
The final PNOSI version consisted of 19 first singular pronoun formulated items (see Appendix). Eight were formulated to tap positive occupational state (e.g., "I am very active at work") and the remaining eleven were formulated to tap negative occupational state (e.g., "I feel demoralised by my work"). Workers were asked to indicate how they felt at work during the last seven days. So the PNOSI addressed short term reactions. Answers were given using a Likert scale ranking from 1 ("never or rarely") to 4 ("nearly always or always"). No item had to be reversed before scoring.

Engagement
We used the short version of the Utrecht Work Engagement Scale (UWES-9, Schaufeli, Bakker, & Salanova, 2006), which operationalizes engagement as a combination of vigour (a high level of energy and concentration while working and the willingness to invest effort in one's work), dedication (a feeling of meaning, enthusiasm, pride, and challenge towards work), and absorption (being completely involved in one's work so that time passes quickly and one has difficulty detaching from it). Answers were made on Likert scales ranging from 0 (never) to 6 (every day).

Burnout
We used the Oldenburg Burnout Inventory (OLBI, Demerouti et al., 2003), which consists of two subscales: exhaustion (8 items, e.g., "There are days when I feel tired before I arrive at work") and disengagement (8 items, e.g., "It happens more and more often that I talk about my work in a negative way"). Answers were made on Likert scales ranging from 1 (strongly agree) to 4 (strongly disagree).

Commitment
Affective, normative, and continuance commitment were measured using a French validated version of Allen and Meyer (1990) scales (Stinglhamber, Beintein, & Vandenberghe, 2002). We created a short version by selecting the items with the highest loadings. Our final questionnaire consisted of three items measuring affective commitment (e.g., "I really feel that I belong in this organisation"), three items measuring normative commitment (e.g., "It would not be morally right for me to leave this organisation now"), and four items measuring continuance commitment (e.g., "I have no choice but to stay with this organisation"). Answers were made on Likert scales ranging from 1 (strongly disagree) to 4 (strongly agree).
Workaholism Schaufeli et al. (2008) insist on the importance of using both behavioural (working excessively) and cognitive (working compulsively) measures of workaholism. The working compulsively component of workaholism was measured using seven items from the Drive scale of Spence and Robbin's (1992) Workaholism Battery (e.g., "I feel that there is something inside me that drives me to work hard"). Answers were made on Likert scales ranging from 1 (almost never or never) to 4 (almost always or always). Two open questions measured the working excessively component of workaholism, namely "How many hours per week do you have to work according to your work contract?" and "How many hours per week do you actually work?". We computed the percentage of overwork using the following formulae: [(a-c)/c]*100, with a referring to the number of actually worked hours and c referring to officially worked hours (Schaufeli, Taris et al., 2006).
Questionnaires were administered in French in samples 2, 5, 6 and 7. They were administered in French and Dutch in samples 1a, 1b, 3 and 4. PNOSI was translated from French to Dutch following a translation backtranslation procedure.
Analyses LISREL 8.80 was used for the confirmatory factor analyses (Jöreskog & Sörbom, 2006). Confirmatory factor analysis compares the covariance matrix obtained from the data with a covariance matrix implied by the theoretical factor model. This model states how latent, unobservable constructs are measured by manifest, observable variables, and how these latent constructs are linked with each other. Thus, confirmatory factor analysis (also called "structural equation modelling") gives an evaluation of how well the theoretical model fits the original data: the smaller the discrepancy between the two matrices, the better the fit.
We used a cross-validation strategy in four steps, as recommended by Diamantopoulos and Siguaw (2000). We defined the initial model (Minit) as a two-factor model, the first factor tapping positive occupational state and the second factor tapping negative occupational state. Minit reflected the independence hypothesis of workers' positive and negative reactions to their environment. We defined the alternative model (Malt) as a one-factor model, with all items measuring a global well-being concept. This model reflected the continuum hypothesis of workers' positive and negative reactions to their environment. Both models were first tested and improved on the calibration sample in order to select the best "initial" and "alternative" models. Both were retested on the validation sample. Both Minit and Malt were tested and compared again in replication samples in order to establish their generalizability to other data. The final "best" model was selected based on its fit indices at each comparison step. Our hypothesis was that Minit would better fit the data than Malt.
Several indices were used to assess the fit. The null hypothesis is that the model fits the data perfectly, which is tested by the χ² test. However, given that the value of χ² is influenced by sample size, it has to be complemented by other fit indices. The Root Mean Square Error of Approximation (RMSEA) is generally regarded as one of the most informative fit indices. It shows "how well would the model, with unknown but optimally chosen parameters, fit the population covariance matrix if it were available" (Brown & Cudeck, 1993, p. 137-138). Values less than .08 indicate an acceptable fit. We also used five indices evaluating the discrepancy between the observed and the implied covariance matrices: the Normed Fit Index (NFI), the Non-Normed Fit Index (NNFI), the Goodness of Fit Index (GFI), the Adjusted Goodness of Fit Index (AGFI), and the Comparative Fit Index (CFI). For NFI, NNFI, GFI, and CFI values higher than .90 are indicative of good fit. For AGFI, values higher than .85 are indicative of good fit. Given that the data were ordinal in nature, polychoric correlation and asymptotic covariance matrices were used, as well as the Weighted Least Squares method of estimation.
The reliabilities of the scales that corresponded with the factors were computed. Ideally, Cronbach's α values should be higher than .80 (Henson, 2001). Mean inter-item correlations should be between .20 and .40 (Briggs & Cheek, 1986). Finally, item-total correlations should be higher than .30 (Kline, 2000).
Finally, Pearson correlation coefficients were computed in order to test convergent and discriminant validity. According to Cohen and Holliday (1982), values above .40 are considered moderate, values between .70 and .89 are considered high, and values superior to .90 are considered as very high correlations.

Structural validity
Means, standard deviations, and correlations for all calibration, validation, and replication samples are shown in Table 2. As can be seen, positive occupational state tended to have higher mean values, and was more frequently experienced than negative occupational state. Standard deviations were quite similar in all seven samples (between .53 and .61 for positive occupational state and .48 and .59 for negative occupational state). Apart from Sample 5, the correlation between positive occupational state and negative occupational state scores were between -.21 and -.32, which reflects quite a low, even though significant, association (all p < .01).

Model calibration
Fit indices are shown in Table 3. Both Minit and Malt had significant chisquare values, χ² (151) = 2444.27, p < .00, and χ² (152) = 3054.54, p < .00 respectively. This can be explained by the large sample size. Regarding other fit indices, results were mixed. For Minit, all RMSEA, GFI and AGFI values satisfied the recommended thresholds, while NFI, NNFI and CFI failed. For Malt, only GFI and AGFI values satisfied the recommended thresholds.
Modification indices for Minit suggested adding a path from items 4 ("I feel demoralised by my work") and 5 ("I work in a rush") to positive occupational state. We considered these items to be more related to depression or negative affect (item 4) and to workload (item 5), so we decided to exclude them, and tested the new reduced model (Mfinal). All RMSEA, CFI, GFI and AGFI values satisfied the recommended thresholds, and were superior to that obtained for Minit. Values for NFI and NNFI were near acceptance and were higher than values obtained for Minit.
Given that Mfinal and Malt are not nested models, they had to be compared by the means of the Expected Cross-Validation Index (or ECVI). This index assesses "whether a model is likely to cross-validate across samples of the same size from the same population" (Diamantopoulos & Siguaw, 2000, p. 85). The ECVI is a good indicator of a model's overall error or fit, and models with smaller ECVI values are to be preferred. As can be seen in Table 3, Mfinal showed the smallest ECVI value, and had to be preferred to Malt. As can be seen in Table 4, the reliabilities of both positive occupational state and reduced negative occupational state scales all satisfied the recommended criteria.

Model validation
Both Mfinal and Malt were tested on the validation sample (see Table 3). Chi-square values were statistically significant, χ² (118) = 1196.23, p < .00, and χ² (152) = 2582.52, p < .00, respectively. Again, this can be explained by the large sample size. For Mfinal, all fit indices satisfied the recommended thresholds. Modification indices suggested letting error covariance of items 1 ("I am very active at work") and 3 ("Once I am at work, I feel more focused") correlate. However, given that Minit already showed an acceptable fit, we did not test this model. According to Diamantopoulos and Siguaw (2000), modifying a model showing acceptable fit would lead to an improved, but unstable model that is hard to generalise to other data. For Malt, only GFI and AGFI satisfied the recommended thresholds; all other fit indices failed. Finally, ECVI values were smaller for Mfinal than for Malt. We can thus consider Mfinal as the "best" model for these data. The reliabilities of POSI and reduced NOSI scales all satisfied the recommended criteria, as can be seen in Table 4.

Model replication
Both Mfinal and Malt were tested on all replication samples. Table 3 shows fit indices for both models in all five samples (samples [2][3][4][5][6]. Chi-square values were highly significant in all samples, which again can be explained by the large sample sizes. RMSEA values were satisfactory for Mfinal, but not for Malt (except from samples 3 and 5). NFI, NNFI, and CFI values were satisfactory or near acceptance for Mfinal, but not for Malt. Finally, GFI and AGFI values were satisfactory in all samples, and values were higher for Mfinal than for Malt. Globally speaking, one can say that fit was better for Mfinal than for Malt in all samples. This is confirmed by ECVI values being smaller for Mfinal than for Malt in all samples. Reliabilities of positive occupational state and reduced negative occupational state scales are shown in Table 4. They all satisfy the recommended thresholds.
Model calibration, validation and replication show that a two-factor solution is preferable to an alternative one-factor solution to explain the intern structure of the PNOSI. Moreover, the fit is improved when deleting two items linked with depression or workload.

Convergent and discriminant validity
Means, standard deviations, Cronbach's alpha's and correlations for positive occupational state, negative occupational state, burnout, engagement, commitment and workaholism scales in sample 7 are reported in Table 5. Engagement was positively correlated with positive occupational state, and negatively with negative occupational state, which was in accordance with hypotheses 1a and 1b. However, the correlation with positive occupational state was quite high (r = .78). In regard to burnout, both exhaustion and disengagement correlated positively with negative occupational state and negatively with positive occupational state, which confirmed hypotheses  2a and 2b. However, correlation between negative occupational state and exhaustion was quite high (r = .73). As far as commitment was concerned, positive occupational state was positively correlated with affective commitment (r = .54) and normative commitment (r = .39), whereas the correlation with continuance commitment was not significant. This confirmed hypotheses 3a, 3b and 3c. Negative occupational state was negatively correlated with affective commitment (r = -.27) and normative commitment (r = -.17), which confirmed hypotheses 3d and 3e. However, it is positively correlated with continuance commitment (r = .12), which was contrary to hypothesis 3f. Finally, concerning workaholism, negative occupational state correlated positively with working compulsively (r = .45), but not with working excessively, and positive occupational state was positively correlated with working excessively (r = .20), which confirmed hypothesis 4a, 4c and 4d. Contrary to hypothesis 4b, positive occupational state correlated positively, but slightly with working compulsively (r = .14).

Structural validity
Our results show that a two-factor structure with one positive and one negative dimension of well-being at work fits the data better than an alternative one-factor solution. This finding was replicated across six samples, including more than 16,000 participants. The two factor-structure confirms that positive and negative aspects of well-being at work are different con- structs, as is hypothesised by the JD-R model and by Schaufeli et al. (2002). However, the correlations between the two factors, although small, are highly statistically significant. This shows that the two constructs are not independent. They are distinct, but related constructs.

Convergent validity
With regard to burnout, negative occupational state is moderately and positively correlated with disengagement, and highly and positively correlated with exhaustion. According to Cordes and Dougherty (1993), rather than a particular type of stress, burnout can be conceived as an on-going process with exhaustion occurring first, followed by a distancing from work, and in the longer term by a feeling of being inefficient at work. Given our results, our measure of negative occupational state seems to be more like exhaustion, which is the first step of the burnout process, than like disengagement, which is the second step. This gives support to our conception of negative occupational state being an intermediate state occurring before burnout.
Correlation between positive occupational state and engagement is positive and high, which indicates that the two are very similar constructs. Items from both scales are indeed highly similar. However, both scales were developed independently, and we were unaware of the existence of the UWES when elaborating the PNOSI. If we follow the definition of Schaufeli et al. (2002), which states that engagement is a combination of vigour, dedication, and absorption (as they were defined in the Measure section), we can say that item 1 ("I am very active at work") relates to vigour, whereas items 7 ("My work allows me to excel myself") and 9 ("My work gives me great satisfaction") relate to dedication, and items 3 ("Once I'm at work, I feel more focused") and 18 ("When I'm working I forget my tiredness") to absorption. We should thus consider positive occupational state as we define it as an alternative measurement for engagement at work. Thus, if this study tends to confirm the existence of an intermediate state occurring before burnout, the question remains whether such an intermediate state occurs before engagement. If we rely on the JDR model assumptions, this implies that engagement can be replaced by positive occupational state in the motivational process. Additionally, the energetic process can be improved this way: job demands would lead to negative occupational state, which leads to burnout and then to health problems.

Discriminant validity
Nearly all hypotheses concerning discriminant validity were confirmed. As expected, positive occupational state is positively related to affective and normative commitment, not related to continuance commitment, and negatively related to burnout. As expected negative occupational state is negatively correlated with engagement and with affective and normative commitment. However, contrary to our hypothesis, it is positively correlated with continuance commitment. According to Meyer et al. (2002), it could be due to the fact that a feeling of being 'trapped' in one's organisation would be stressful for workers.
If correlations between negative occupational state and workaholism go the expected way, results concerning workaholism and positive occupational state are less clear. Positive occupational state is positively related to working excessively, which is in accordance with our hypothesis. However, it is also positively correlated with working compulsively. This is contrary to prior studies which found that positive wellbeing differs from workaholism in that it lacks the compulsive component (Schaufeli, Taris et al., 2006). However, the correlation is very low and its significance may be linked to our large sample size.
Positive and negative occupational states are thus differently related to engagement, burnout and commitment, which strengthens findings that positive and negative occupational states are different constructs. Moreover, correlations between positive occupational state and dimensions of burnout, commitment and workaholism range between -.67 and .40, which is considered as moderate (Cohen & Holliday, 1982). Positive occupational state can thus be considered as distinct from burnout, commitment, and workaholism. Correlations between negative occupational state and dimensions of engagement, commitment, and workaholism range between -.17 and -.51. That is considered as low to moderate (Cohen & Holliday, 1982). Negative occupational state can thus be considered as distinct from engagement and commitment.

Limitations and perspectives
One important shortcoming of this study is that negative occupational state items are all negatively worded, whereas positive occupational state items are all positively worded. According to Anastasi (1988), one-side worded scales are not as psychometrically valid as scales that include both positively and negatively worded items. That is because such scales can lead to biased solutions in which positively and negatively worded items are likely to load on separate factors (Doty & Glick, 1998). In consequence, it might be that the two-dimensional structure we obtained is artificial in that it reflects item wording rather than "true" content structure. One important perspective would be to modify wording so that both positive and negative occupational state scales include negative and positive words.
A further point that should be examined is whether all items "behave" the same way with all respondents. For example, if one is interested in negative state at work, it might be that some items are infrequently endorsed by workers (that is, most of them choose the category 'never or rarely') or only by workers who already show a higher level of stress (that is, workers who answer 'nearly always or always' to that item also show higher scores of negative occupational state). Rasch analysis "predicts the probabilities of the responses of a sample of subjects, who vary in the exhibition of a trait, to a test that has items of varying endorsability designed to detect the presence of that trait" (King & Bond, 1996, p. 55). In other words, Rasch analysis indicates how items function according to individual characteristics. This kind of analysis would bring informative and complementary elements to the validation process of the PNOSI.
One should also check for PNOSI structural invariance across sociodemographic variables such as gender, age, occupation, or sector of activity for example. Furthermore, in this study, data collected with French and Dutch versions of PNOSI were mixed in the same sample. In future research, structural analyses should be performed on French and Dutch PNOSI versions separately to address the possible role of language as a confounder. This would also lead to a validation of the Dutch translation of PNOSI.
Finally, further studies should attempt a test of the JDR model using our measures of positive and negative occupational states. More specifically, longitudinal and covariance analyses should test whether positive occupational state can indeed be considered as an alternative measure of engagement in the motivational process. They should also test whether a negative occupational state can be considered as an intermediate state that occurs between job demands and the experience of burnout in the energetic process.

Conclusions
The first aim of our study was to test the intern validity of the PNOSI by the means of confirmatory factor analyses, and to examine the degree to which a two-factor structure should be preferred to alternative solutions. The second aim was to replicate the structure on several samples of workers active in different areas, in order to test its stability and generalizability. The final aim of this paper was to test convergent and discriminant validity with burnout, engagement, commitment and workaholism. Confirmatory factor analyses show that a two-factor solution has to be preferred to the one-factor solution to explain the structure of the PNOSI. Items relate to two distinct yet related dimensions, one tapping positive occupational state, and the other negative occupational state. Moreover, a reduced model, in which two items relating on other dimensions (namely depression and workload) are deleted, is more acceptable than the originally formulated model. Furthermore, the reduced two factor model is stable and generalisable to a wide range of employed samples. Finally, a clear pattern of predicted convergent and discriminant relationships emerges that confirms the predicted interpretation of the scales. We can conclude that the PNOSI has been shown to be a valid tool for measuring negative and positive occupational states.