A convenience sample of 963 Italian employees [69.9% female (n = 673)] was surveyed in 2022. Participants were employed in a variety of occupational sectors although a large proportion of participants were schoolteachers (n = 456). Participants' mean age was 40.433 (SD = 10.611). The sample was recruited from training events addressing an occupational stress prevention program.

Respondents took part in the study on a voluntary basis. Participation involved no compensation. Respondents were guaranteed full confidentiality, in compliance with privacy rights described in current Italian law (Law Decree DL-196/2003). Informed consent was obtained from all subjects. The study met the ethical standards of the institutional review board of LUMSA University (Rome, Italy; Prot. N. 6/2021).

We ran our factor analyses with Mplus 8.6 (32). We first examined the ODI's factorial structure within an ESEM bifactor analytic framework (18). We relied on a partially specified target rotation (PSTR). Compared to “classical” confirmatory factor analysis (CFA), the PSTR does not fix nontarget loadings at 0; instead, nontarget loadings are “encouraged” to get as close to 0 as possible, allowing factorial complexity to be modeled. Consistent with Bianchi and Schonfeld's (8) findings on the ODI's factorial structure, we extracted two specific factors (or bifactors) in addition to the general Occupational Depression factor. The ODI's “anhedonic-somatic” items (Items 1, 3, 4, 5, 7, and 8) were directed toward the first specific factor; the ODI's “dysphoric” items (Items 2, 6, and 9) were directed toward the second specific factor. We used an orthogonal PSTR—the bifactors were not allowed to correlate. We approached the ODI items as ordinal and used the weighted least squares—mean and variance adjusted—estimator. To ascertain how the general factor accounted for the common variance extracted, we computed the Explained Common Variance (ECV) statistic (20). An ECV index exceeding 0.80 is considered to signal essential unidimensionality. We relied on the omega and omega hierarchical (omegaH) coefficients to scrutinize the ODI's reliability and the general factor's correlation with the observed total scores. We further inquired into the factorial structure of the ODI using “classical” CFA. We tested a one-factor model—we set all ODI items to load on a single factor.

In a final effort to assess the ODI's dimensionality, we estimated the scale's homogeneity (or scalability) within a Mokken scale analytic framework (19, 33). We conducted the analysis with the Mokken package version 3.0.6 (34) in R version 4.0.3 (35). Homogeneity refers to the extent to which a scale's items hierarchically align on a single dimension. The hierarchy concerns item difficulty, i.e., the likelihood for an item to be endorsed by respondents. In the context of psychopathology items, item difficulty is equivalent to symptom severity. In the ODI, we expect, for instance, the fatigue/loss of energy item to be less “difficult” (i.e., more frequently endorsed) than the suicidal ideation item because suicidal ideation represents a much more severe symptom than fatigue/loss of energy. Homogeneity is indexed by H coefficients. As per commonly applied rules of thumb (36), a scale's homogeneity is regarded as weak if 0.30 ≤ H < 0.40, moderate if 0.40 ≤ H < 0.50, and strong if H ≥ 0.50; a scale-level H coefficient below 0.30 suggests that the scale of interest cannot be regarded as unidimensional. Pairwise H coefficients should be >0. Item-level H coefficients should be >0.30. In addition to computing H coefficients, we relied on the automated item selection procedure (AISP), a method for evaluating scale formation. The AISP enables us to identify subscales and deviating or unscalable items (37). We computed Cronbach's alpha, Guttman's lambda-2, and the Molenaar-Sijtsma statistic as additional reliability indicators.

We investigated the measurement invariance of a unidimensional model across sexes (male/female), age groups (based on a tercile split), and occupations (teachers/other professions) focusing on: (a) configural invariance—the equivalence at the level of model forms; (b) metric invariance—the equivalence at the level of factor loadings; and (c) scalar invariance—the equivalence at the level of item thresholds (38). We relied on conservative standards for flagging measurement invariance violations: 0.005 for ΔRMSEA and ΔSRMR; and −0.005 for ΔCFI and ΔTLI (38, 39).

We examined the criterion validity and nomological network of the ODI based on Pearson and Spearman correlations as well as Welch's analysis of variance (ANOVA). Welch's ANOVA is a robust test of equality of means that allows investigators to cope with homoscedasticity violations.

Regarding the ODI, our ESEM bifactor analytic structure showed a satisfactory fit: RMSEA = 0.047; CFI = 0.997; TLI = 0.990; SRMR = 0.014; χ²(12) = 38.023. Factor loadings are displayed in Figure 1. The mean factor loading on the general factor was 0.743 (SD = 0.047). The ECV index indicated that the general factor accounted for 85.3% of the common variance extracted. Omega was 0.941 and OmegaH, 0.846. We found a correlation of 0.920 between the general factor and the observed total scores. Comparing OmegaH with Omega, we found that most of the reliable variance in observed total scores could be attributed to the general factor (0.846/0.941 = 0.899), assumed to reflect individual differences in occupational depression.

Consistent with our ESEM bifactor analytic findings, a one-factor confirmatory analytic model showed an acceptable fit: RMSEA = 0.078; CFI = 0.980; TLI = 0.973; SRMR = 0.052; χ²(27) = 186.737. Factor loadings are displayed in Figure 2. The mean factor loading was 0.761 (SD = 0.044). Omega was 0.926.

A one-factor confirmatory analytic model exhibited measurement invariance across sexes, age groups, and occupations (Table 2). Regarding sexes, RMSEA never increased and TLI never decreased; CFI never decreased by more than 0.001 and, in a similar vein, SRMR never increased by more than 0.001. Regarding age groups, RMSEA never increased and TLI never decreased; CFI remained virtually identical as constraints were added; SRMR never increased by more than 0.003. Regarding occupations, RMSEA never increased by more than 0.003 and TLI never decreased by more than 0.002; CFI never decreased by more than 0.005; SRMR never increased by more than 0.002 as constraints were added.

Results of our Mokken scale analysis are presented in Table 3. The ODI exhibited strong homogeneity. The scale-level H coefficient reached 0.548 (95% confidence interval: 0.514, 0.582), with a standard error of only 0.017. The pairwise H coefficients largely exceeded the zero threshold and the item-level H coefficients were well above the 0.300 threshold. The AISP, used with increments of 0.005, signaled a single scale involving all ODI items up to a threshold of 0.475. The most frequently endorsed item was the fatigue/loss of energy item (Item 4) and the least frequently endorsed item was the suicidal ideation item (Item 9). Cronbach's alpha, Guttman's lambda-2, and the Molenaar-Sijtsma statistic had values ≥0.878 (Table 3).

Occupational depression correlated negatively with general wellbeing, Pearson r = −0.392 (p < 0.001), Spearman ρ = −0.408 (p < 0.001), and positively with financial strain, Pearson r = 0.185 (p < 0.001), Spearman ρ = 0.176 (p < 0.001). The correlation between occupational depression and age was small and statistically nonsignificant, Pearson r = −0.036 (p = 0.260), Spearman ρ = −0.046 (p = 0.152). Descriptive statistics for these variables are available in Supplementary material 2.

Welch's ANOVA revealed positive associations of occupational depression with a 12-month history of depressive disorder, antidepressant intake, 12-month sick leave, 6-month physical assault at work, 6-month verbal abuse at work, and lack of money for leisure activities; occupational depression showed no links to participants' 12-month promotion and participants' sex (see Table 4 for a summary of the results). As per Cohen's (40) interpretation grid, the associations of occupational depression with a 12-month history of depressive disorder, antidepressant intake, 6-month physical assault at work, and 6-month verbal abuse at work were large or close to large in magnitude (Cohen's ds ranging from 0.724 to 1.082).

ESEM bifactor analysis, common-practice CFA, and Mokken scale analysis consistently indicated that the ODI's Italian version meets the requirements for essential unidimensionality, thus justifying the use of the instrument's total score. Moreover, we found the ODI to exhibit high reliability on the basis of five different indicators—omega, omegaH, Cronbach's alpha, Guttman's lambda-2, and the Molenaar-Sijtsma statistic. Our findings are consistent with the results of previous ODI studies that employed the measure in its English, French, and Spanish versions (8, 13–17). Our measurement invariance analysis supports the use of the ODI for comparisons between (a) men and women, (b) individuals across adulthood, and (c) individuals from different occupational domains.

Occupational depression was negatively linked to general wellbeing and positively linked to a 12-month history of depressive disorder, current antidepressant intake, 12-month sick leave, 6-month physical assault at work, 6-month verbal abuse at work, lack of money for leisure activities, and financial strain in the household. The links that we observed were generally medium to large in size. Our findings demonstrate the criterion validity of the ODI and further illuminate the nomological network of occupational depression. Our results are consistent with findings emanating from research on job stress and antidepressant medicine (41, 42), workplace bullying and health (43), and economic stress and employee wellbeing (44, 45).

All in all, our findings are consistent with the research on the ODI in English-, French-, and Spanish-speaking samples. The validity and reliability of the ODI in those samples can be extended to Italian-speaking samples. As previously mentioned, the present study provides new information. It links the ODI to past episodes of depression, the use of antidepressants, and general wellbeing. It also connects the ODI to economic stress, sick leave, and workplace violence.

At least four limitations to this study are noteworthy. First, although our sample was relatively large (N = 963) and included individuals displaying various ODI scores (reflective of various levels of symptom severity), its representativeness is unclear (e.g., in terms of sex, age, and health status). As a consequence, our estimate of occupational depression's prevalence should not be generalized to the Italian working population. The implementation of methods such as random sampling, which promotes sample representativeness, is very costly and frequently unworkable (e.g., because the population of interest cannot be accurately circumscribed or exhaustively contacted) (46). Unsurprisingly, such methods have rarely been used in clinical and occupational health sciences.

Second, we relied exclusively on self-reported measures, within a cross-sectional design. We note, however, that several of our self-reported measures addressed “objective” events likely to be readily identified by respondents (e.g., sick leave and job promotion over the previous year). Moreover, many of our items were retrospective in nature, a characteristic that optimizes the informativeness of cross-sectional designs (47).

Third, we relied on single-item measures to assess several of our variables of interest. Although multiple-item measures are generally considered more robust, there is evidence that single-item measures represent an acceptable measurement approach for many constructs in organizational science (48).

Fourth, our study did not reexamine the overlap between burnout and depression. Fortunately, this issue has been addressed extensively in past research (4, 46, 49–52), including ODI-based research (17, 21). The advantages of relying on the construct of (occupational) depression have been discussed on many occasions (12, 22, 23, 53–55).