Porter et al. (1974, p. 604) defined organizational commitment (OC) as “the strength of an individual’s identification with and involvement in a particular organization.” Porter and colleagues’ definition refers to an attitudinal or affective understanding of OC. Besides this affective view on OC, other perspectives of OC evolved in the literature. For instance, Becker’s (1960) side-bet theory described OC from a cognitive and calculating perspective, which explains why people continue to work for a specific company. Such side-bets include actions or circumstances besides the usual occupation (e.g., time spent in an organization, personal effort). Other factors than occupational interests attach the person to the company due to the high costs of leaving (Becker, 1960).

Wiener and Vardi (1980) stated a third perspective on OC driven by the attempt to investigate outcomes of OC. They argued that not only the consequences of leaving but also normative beliefs toward leaving might be essential to explain people’s commitment to an organization.

The three-component model (Meyer & Allen, 1991), the most popular OC model, combines the normative conceptualization with the affective and calculative/continuance one. However, some empirical issues arose when testing the three-component model (for a review, see Solinger et al., 2008). First, a high obliqueness of affective and normative commitment made one of the perspectives nearly redundant (e.g., Allen & Meyer, 1990; Meyer et al., 2002; Meyer et al., 2012). A factor analysis of Allen and Meyer’s (1990) questionnaire measuring all three perspectives showed that the affective commitment factor extracted the most variance across all items. The high eigenvalue of this factor consisted not only of high primary factor loadings of the affective items but also of noticeable cross-loadings of items pertaining originally to the other factors. Second, meta-analyses indicated that affective commitment showed the best concurrent and predictive validity for external criteria such as job satisfaction (Cooper-Hakim & Viswesvaran, 2005; Mathieu & Zajac, 1990; Meyer et al., 2002). Therefore, researchers often define OC as affective/attitudinal commitment.

The scientific interest in OC is still very high; this is reflected in a large number of citations of Porter et al.’s (1974) and Meyer and Allen’s (1991) initial articles on theoretical OC models, which were cited about 12,000 times and about 20,000 times in Google Scholar at the time of this writing (December 2022). Moreover, OC is related to several performance-related, behavioral, and affective job outcomes, and physical and mental health in general, which underlines its practical relevance.

In particular, previous research found that OC negatively predicted turnover (intentions) (Gaudine & Thorne, 2012; Lapointe et al., 2011; Neininger et al., 2010; Vandenberghe et al., 2004; Vander Elst et al., 2014). Strong correlations (r ≈ .53) occurred in previous research (Mathieu & Zajac, 1990; Tett & Meyer, 1993) between OC and job satisfaction (i.e., “the extent to which an employee has a positive affective orientation or attitude towards their job,” Cramer, 1996, p. 389) and physical and psychological health (Lapointe et al., 2011; Schalk, 2011; Vander Elst et al., 2014). Moreover, there was evidence for a strong positive correlation with the person-job fit (i.e., “the relationship between a person’s characteristics and those of the job or tasks that are performed at work,” Kristof‐Brown et al., 2005, p. 284), for instance in the study of Weiwen and Chaoping (2021). Medium to strong positive correlations occurred when investigating job characteristics such as training (i.e., if a participant had a workplace or voluntary coaching to improve their job skills or not), promotion (i.e., the opportunity for advancement vs. stagnation/career plateau), and job security (i.e., the respondent’s perception of their tenure safety). Kooij et al. (2010), for example, reported correlations between r ≈ .25–.30 for the relationship of OC with those three constructs. Moreover, previous studies pointed to more or less substantial correlations between OC and job stress (i.e., “an individual’s reaction to work environment characteristics that appear emotionally and physically threatening to the individual,” Jamal, 2005, p. 225). Correlations in studies ranged from r ≈ .10 to r ≈ .30 (Abdelmoteleb, 2019; Jamal, 2011). Furthermore, negative medium to strong correlations arose when investigating the association between OC and discrimination or harassment (i.e., the “systematic and prolonged exposure to repeated negative and aggressive behaviour of a primarily psychological nature, including non-behaviour and acts of social exclusion,” Nielsen & Einarsen, 2012, pp. 309–310). Anita et al. (2020), as well as Hasan et al. (2021), found a strong negative correlation between OC and work-life balance (i.e., “the extent to which an individual is equally engaged in—and equally satisfied—with his or her work role and family role,” Greenhaus et al., 2003, p. 513). Kooij et al. (2010) measured flexibility as the adaptability of the work arrangement and schedule and reported a strong positive association with OC (r = .35). While Thorsteinson (2003) did not find differences between full-time and part-time employees, evidence for differences regarding OC existed for job-level (white-collar vs. blue-collar) with a medium effect size of r ≈ .20 (Mathieu & Zajac, 1990). Focusing on demographic data, previous research showed medium positive correlations (r ≈ .20) of OC with age (Cohen, 1993; Cohen & Lowenberg, 1990; Mathieu & Zajac, 1990) and no relation to gender (Cohen & Lowenberg, 1990; Mathieu & Zajac, 1990).

Meta-analyses (Cooper-Hakim & Viswesvaran, 2005; Lapierre, 2001; Mathieu & Zajac, 1990; Wallace, 1993) reported strong correlations (r ≈ .45 to .54) between OC and occupational commitment (i.e., “a psychological link between a person and his or her occupation that is based on an affective reaction to that occupation,” Lee et al., 2000, p. 800).

To measure OC, researchers typically utilized or adapted the Organizational Commitment Questionnaire (OCQ; Mowday et al., 1979) or the Affective Commitment Scale (ACS) (Allen & Meyer, 1990). However, both scales consist of at least 9 (OCQ) or 6 (ACS) items in their shortest forms. So, the ISSP scale presented herein measures OC even more economically by using only three items. These items are related to the involvement in and the identification with a company as well as commitment-related behavior.

We ran all analyses with R (version 4.1.1) using the following packages: car (version 3.0-11; Fox et al., 2007), foreign (version 0.8-81; R Core Team, 2020), lavaan (version 0.6-9; Rosseel, 2012), lme4 (version 1.1-27.1; Bates et al., 2007), psych (version 2.1.6; Revelle & Revelle, 2015), semTools (version 0.5-5; (Jorgensen et al., 2016), sjstats (version 0.18.1; Lüdecke, 2017), and sirt (version 3.10-118; Robitzsch & Robitzsch, 2020). The R Code is available online (Urban, 2022). For confirmatory factor analyses (CFA), we handled missing values on items of the ISSP OC scale using full-information maximum likelihood (FIML). We applied robust maximum likelihood (MLR) to correct for any deviations from multivariate normality. To evaluate the size of the correlations between the OC and other variables, we followed the suggestions of Gignac and Szodorai (2016) by interpreting correlations of r = .10 as small, r = .20 as medium, and r = .30 as large.

Item generation and selection

The ISSP used the three items of the ISSP OC scale as part of the ISSP work orientations II module in 1997 (Harkness, 2001). Furthermore, the items correspond to items of the Mowday et al.’s (1979) OCQ (i.e., their Items 1 and 6) and the adapted version by Cook and Wall (1980) (i.e., their Item 7). For the German version, a committee merged and modified five independent translations of one professional translator, two survey researchers, and two students, considering modifications of experts and results of pretesting (following the suggestions of Harkness, 2001). In the 1997 survey, OC formed one scale together with two items measuring the identification with one’s job/career (i.e., occupational commitment). Since ISSP 2005, the three items of OC and the two items of occupational commitment constitute their own scales (Jutz & Scholz, 2017; Scholz & Faaß, 2007).

Samples

We used the most recent ISSP survey (ISSP 2015), which is available on the ISSP website, for data analyses. The data collection took place from 2015 to 2017 and took 1 to 16 months (M = 4.51, SD = 3.39) per individual country. The following survey modes were used: face-to-face interview (PAPI), computer-assisted interviews (CAPI, CASI), web questionnaire (CAWI), or telephone interview. Participants of the survey were at least 18 years old, except for Finland (15+ years), Estonia (15+ years), Japan (16+ years), South Africa (16+ years), and Suriname (21+ years). The researchers used random sampling to collect data in 38 countries: Australia, Austria, Belgium, Chile, China, Croatia, Czech Republic, Denmark, Estonia, Finland, France, Georgia, Germany, Great Britain, Hungary, Iceland, India, Israel, Japan, Latvia, Lithuania, Mexico, New Zealand, Norway, Philippines, Poland, Russia, Slovak Republic, Slovenia, South Africa, Spain, Suriname, Sweden, Switzerland, Taiwan, Turkey, United States, Venezuela. The ISSP 2015 variable report includes a detailed overview of the data collection. As we conducted all analyses separately for the different countries, we applied no weighting technique.

The original ISSP sample consisted of N = 53,203—only participants working for pay at the time of the survey participation could provide their OC. Furthermore, we used cases with complete data on all three items of the ISSP OC scale for the sample and item statistics, as well as the validity analysis, reducing the sample to N = 26,100. We coded the response category Can’t choose as a missing value. The remaining sample consists of 51.54% male and 48.46% female participants with a mean age of M = 42.88 years (SD = 12.95). Table 2 shows the sample characteristics, including gender, age, and educational level with respect to the different countries.

Table 2

Sample Size (N), Gender, Age, and Educational Level for Each Country and Across All Countries

Note. We used only valid cases on the respective variables to calculate the percentual values. *Low = no formal education, primary/elementary education, or lower secondary (does not allow entry to university), Middle = upper secondary or post-secondary (allows entry to university, other upper secondary programs), High = Lower or upper-level tertiary (Bachelor, Master, Ph.D.)

Item analyses

In this section we investigated the factorial structure of OC as well as descriptive item parameters such as means, standard deviation, and percentage of maximum possible.

Pre-analyses

First, we checked the factorial structure of the ISSP OC scale with a single-group CFA in each country. When only considering the three items of the initial ISSP OC scale, the scale is just identified. We cannot evaluate the model fit. To assess the model fit, we need more degrees of freedom. Thus, we additionally added the occupational commitment scale to our initial OC model to increase the degrees of freedom and to be able to evaluate the model fit. Moreover, we fixed the latent means to zero and the unstandardized factor loading of the respective first item to one for identification purposes. We compared a first order-correlated factor model with two factors OC and occupational commitment (from now on called the correlated two-factor model), with a one-factor model in which the items of OC and occupational commitment load on one single factor. Due to similar item wordings across factors, we added two residual covariances. The first residual covariance was between Item 3 of the ISSP OC scale (“I would turn down another job that offered quite a bit more pay in order to stay with this organization”) and Item 1 of the occupational commitment scale (“Given the chance, I would change my present type of work for something different”). The second added residual covariance occurred between Item 2 of the ISSP OC scale (“I am proud to be working for my firm or organization”) and Item 2 of the occupational commitment scale (“I am proud of the type of work I do”). We excluded India and the Philippines from the analysis because the model did not converge for India and negative correlations of the occupational commitment items in the Philippine sample. To evaluate the model fit, we employed the criteria of Hu and Bentler (1999) for the CFI (CFI ≥ .95 good) and SRMR (SRMR ≤ .08 good) and the specifications for the RMSEA (RMSEA ≤ .05 good, RMSEA ≤ .08 acceptable, RMSEA ≥.10 misfit) of Browne and Cudeck (1992).

Overall, the correlated two-factor model showed an acceptable fit for most countries (CFI > .963, SRMR < .029). Only the RMSEA was too large (> .10) to indicate the model as acceptable for Georgia and Israel. The one-factor model showed a bad global fit in most countries (17 countries CFI < .950, 29 countries RMSEA > .100). Only for Chile and Venezuela, relative fit indices (AIC, BIC, adj. BIC) pointed to an advantage for the one-factor model. The absolute fit of the one-factor model was good for these two countries (CFI > .994, RMSEA < .044, SRMR < .017). Concluding, we accepted the correlated two-factor model for 32 of 38 countries. We conducted all following analyses only with these 32 countries.

Next, we investigated essentially tau-equivalent two-factor models for the OC (and occupational commitment) items. Therefore, we compared the correlated two-factor model with and without essential tau-equivalence, showing a worse fit of the essentially tau-equivalent model (2 countries CFI < .950, 9 countries RMSEA > .100, 2 countries SRMR > .080), Thus, we rejected the essentially tau-equivalent  model and used the congeneric model as a basis for further analyses.

Item parameters

We computed means, standard deviations and Percentage of Maximum Possible (POMP) scores of the ISSP OC scale. Table 3 presents the results for Germany, the USA, and the total sample including all countries.

We included the specific statistics for each country in Table A1 of the Appendix. To shortly summarize the descriptive statistics, all items showed sufficient variance. The POMP scores were highest for Item 2 (range over countries: 54.34–84.03), followed by Item 1 (range over countries: 44.73–77.76), and lowest for Item 3 (range over countries: 34.58–57.59).

Table 3

Means, Standard Deviations, and Percentage of Maximum Possible (POMP) of the Manifest Items and Scale Scores for Germany, the USA, and Across all Countries

Note. For ease of interpretation, we inverted scale scores to range from 1 = strongly disagree to 5 = strongly agree. Respective samples sizes are N = 962 (Germany), N = 925 (USA), and N = 22,690 (Total)

Moreover, we checked skewness and excess kurtosis of the items and the scale for violations of the normal distribution assumption. The highest absolute skewness values were for Item 1 |skewness| = 1.17 (Taiwan), Item 2 |skewness| = 1.19 (Suriname), Item 3 |skewness| = 0.67 (Estonia), and for the scale |skewness| = 0.50 (Switzerland). For excess kurtosis, the highest absolute values were for Item 1 |excess kurtosis| = 1.76 (South Africa), Item 2 |excess kurtosis| = 2.20 (Suriname), Item 3 |excess kurtosis| = 1.29 (Suriname), and for the scale |excess kurtosis| = 0.60 (Switzerland). As the |skewness| was constantly lower than 2, and the |excess kurtosis| was constantly lower than 4, no severe violations of normality occurred (West et al., 1995). We attached exact values for skewness and excess kurtosis in Table A2 of the Appendix.

As an indicator for the selectivity, we used fully standardized factor loadings of the accepted correlated two-factor model in a multi-group CFA across all countries. The model showed an acceptable fit (χ²(64) = 176.192 p < .001, CFI = .996, RMSEA = .052, SRMR = .013). The factor loadings for Item 1 ranged from λ = .45 (Norway) to λ = .80 (Turkey), for Item 2 from λ = .78 (Mexico) and λ = .97 (Czech Republic), and for Item 3 from λ = .18 (Mexico) and λ = .72 (Finland). Thus, all factor loadings (besides the one on Item 3 for Mexico) are above λ > .30. Hence, all items measured OC to a meaningful amount in every country (Floyd & Widaman, 1995). Table A3 of the Appendix demonstrates the fully standardized factor loadings per country. The standardized factor loadings of the correlated two-factor model presented in Figure 1 are the result of a CFA across all countries respecting the nested data structure by using country as cluster.

Objectivity

Trained interviewers administered most surveys in each country (72.54%) (see ISSP 2015 variable report). For the participants who took part via self-assessment or web questionnaire, the standardized format and written instructions of the survey assure the objectivity of application. The objectivity of evaluation and interpretation is given due to the written instructions, the ordered and labeled categories, the coding scheme for missing/ambiguous answers, the scoring procedure, and the country-specific data provided in the Appendix (Table A1–Table A5).

Reliability

We calculated Cronbach’s alpha (Cronbach, 1951) as an estimate for internal consistency (for a review, see McNeish, 2018). Nonetheless, we also computed McDonald’s omega (McDonald, 1999) as a more robust estimator of reliability as it does not require the assumption of tau-equivalence to hold. Table 4 shows the average reliability estimate and the respective ones for the ISSP OC scale in each country. We estimated McDonald’s omega utilizing the correlated two-factor model mentioned above in multi-group CFA across countries (though extracting only the reliability for the ISSP OC scale).

Omega ranges between ω = .56 and ω = .76, with a mean of ω = .69. To compare this to the reliability of the original OCQ (α ≈ .90) (Mowday et al., 1979), we applied the Spearman-Brown prophecy formula (Brown, 1910; Spearman, 1910). Due to the original OCQ consisting of 15 items, we used 5 as the correction factor in the prophecy formula, revealing a result of ω_corrected = .92. Thus, the items of the ISSP OC scale can measure OC as reliable (α ≈ .90) as the items of the original, typically used OCQ questionnaire of Mowday et al. (1979).

Table 4

Reliabilities for Each Country and Across All Countries

Note. alpha = Cronbach’s alpha, omega = McDonald’s omega, AVE = average variance extracted

Validity

Content validity

We assume content validity of the ISSP OC scale as the three items match the definition of affective commitment (Porter et al., 1974).

Nomological network

We used manifest items and scale scores to investigate convergent, divergent, and criterion validity. The correlations should be interpreted as a lower bound of the true correlation. In detail, we computed multilevel models (i.e., random-intercept and random-slopes models) to estimate mean correlations across countries. More specifically, we predicted OC with the respective variable of the nomological network. We standardized both OC, the continuous dependent variable (DV), and the variables of the nomological network, the independent variables (IV), to anticipate convergence problems. For binary IVs, we report unstandardized as well as standardized mean differences. To compute the average unstandardized and standardized mean differences across all countries, we used random-intercept and random-slope models. To compare the effects of the binary predictors with the effects reported in the literature, we converted the standardized mean difference to a correlation (Borenstein et al., 2009). For the continuous variables of the nomological network, the fixed slope is directly comparable to the correlations of previous studies.

Table 5 shows the multilevel correlations between OC and continuous variables of the nomological network, while Table 6 shows the unstandardized and DV standardized effects for OC and binary variables. We present Pearson’s product-moment correlations for each country in Table A4 of the Appendix, and unstandardized and DV standardized effects for each country in Table A5 of the Appendix. Further, we attached the items measuring the constructs of the nomological network in Table A6 of the Appendix .

Table 5

Multilevel Correlations of OC and Continuous Variables of the Nomological Network

Note. Effect = multilevel correlation, SE = standard error of the effect, N = sample size, SD = standard deviation of the random slope, lower = lowest country specific correlation, upper = highest country specific correlation.

Table 6

Multilevel Effects of OC and Binary Variables of the Nomological Network

Note. Negative values indicate higher OC for the group coded 0 and vice versa. SE = standard error of the standardized effect, N = sample size, SD = standard deviation of the random slope, lower = lowest country specific standardized effect, upper = highest country specific standardized effect. ^a 0 = no training, 1 = training; ^b 0 = not discriminated, 1 = discriminated; ^c 0 = not harassed, 1 = harassed; ^d 0 = no, 1 = yes; ^e 0 = women, 1 = men

As mentioned in the theory section, meta-analytic findings suggest a strong positive correlation among OC and occupational commitment of (r ≈ .45 – .54; Cooper-Hakim & Viswesvaran, 2005; Lapierre, 2001; Mathieu & Zajac, 1990; Wallace, 1993), job satisfaction (r ≈ .53; Mathieu & Zajac, 1990; Tett & Meyer, 1993), person-job-fit (r ≈ .50; Weiwen & Chaoping, 2021). We replicated these strong positive associations (OC with occupational commitment: r = .53; OC with job satisfaction: r = .52), although the correlation for OC and person-job-fit was slightly lower but still strong and positive (r = .34). Regarding work-related problems, we found a small negative correlation (r = –.11) between OC and job stress which is in line with previous studies (r ≈ –.30 to –.10; Abdelmoteleb, 2019; Jamal, 2011). We found slightly lower negative effects (r ≈ –.12) for OC with discrimination / harassment (r ≈ –.30 to –.20; Nielsen & Einarsen, 2012), and contradictory results (r = .04) for OC and work-life-balance (r ≈–.40; e.g., Anita et al., 2020). The latter may be attributed to a selective sample of studies (female bank employees and managerial employees) examining OC and work-life-balance (Anita et al., 2020; Hasan et al., 2021). For job characteristics such as training, promotion, job security, and flexibility, Kooij et al. (2010) found medium to strong positive correlations of r ≈ .25 or larger. Our data replicated these medium to strong positive results (r ≈ .22 to .32), with lower effects for OC with training (r ≈ .10). Job status (full-time vs. part-time employees) had no association with OC (r = –.07) as reported by earlier studies (Thorsteinson, 2003). For supervising (white-collar vs. blue-collar) and OC our positive effect (r ≈ .16) did not reach the reported medium positive effect of r ≈ .20 (Mathieu & Zajac, 1990). Examining the relationship of OC with demographic data, previous studies reported medium positive correlation (r ≈ .20) with age (Cohen, 1993; Cohen & Lowenberg, 1990; Mathieu & Zajac, 1990) and no effect for gender (Cohen & Lowenberg, 1990; Mathieu & Zajac, 1990). We only replicated the expected effects for gender by finding negligible effects for the association of OC with both of these variables (r = .05 for gender and r = .08 for age).

Descriptive statistics (scaling)

We included reference values for all items and the total ISSP OC scale in Table A1 of the Appendix.

Further quality criteria

Economy

The scale is economical due to its short duration (< 1 min; authors’ estimation).

Measurement invariance across countries

To test for cultural fairness, we applied multi-group CFA. Moreover, we tested four levels of measurement invariance to check for cultural fairness (Vandenberg & Lance, 2000). As these levels build up successively on each other, we only checked for the next level of measurement invariance if we accept the lower one. The four levels include the comparability of the measurement model (configural invariance), factor loadings (metric invariance), intercepts (scalar invariance), and uniqueness (strict invariance). We evaluated the fit of the configural model applying the previously mentioned criteria regarding the global model fit (Browne & Cudeck, 1992; Hu & Bentler, 1999). For the higher levels we assessed the fit relative to the respective lower level. Therefore, we applied the criterion of ΔCFI equal or smaller than –0.01 of Cheung and Rensvold (2002) or Chen (2007). As described above, we cannot evaluate the fit of the congeneric unidimensional configural model as it is just identified. Thus, we examined measurement invariance using the correlated two-factor model of OC with occupational commitment.

We investigated the 32 countries, where the correlated two-factor model held.. Thus, the sample size shrunk to N = 24,816. We utilized the same procedures (MLR, FIML, identification) as the previous CFAs. We only successively restricted the parameters of the ISSP OC scale and not of the occupational commitment scale, when checking for measurement invariance. 7 shows the results of the measurement invariance tests.

As expected after the exclusion of non-fitting countries, the configural model showed an acceptable fit (χ²(72) = 228.07, p < .001, CFI = .995, RMSEA = .058, SRMR = .014). According to the mentioned criterion (Chen, 2007; Cheung & Rensvold, 2002), the constraints of the metric model did not lead to a considerable decrease in fit (ΔCFI = –.008). However, the scalar model did not hold (ΔCFI = –.089). Note, that for the scalar model the latent means were able to vary freely. The lack of invariance does not allow a comparison of latent means, which is usually desirable in cross-cultural studies. However, the explained measurement invariance approach (multi-group CFA) often fails to reach the scalar level when applied to many groups. Thus, we examined approximative measurement invariance using the alignment method (Asparouhov & Muthén, 2014; Muthén & Asparouhov, 2018). Based on the configural model, this approach rotates the unconstrained loadings and intercepts to minimize non-invariance without decreasing model fit. This is comparable to the factor rotation of an explorative factor analysis, where factor loadings change, but the explained variance stays the same. As an indicator for invariance, an R² measure is helpful. This R² specifies how well the variation of factor means and variances, fixed to 0 and 1 in the configural model for identification, can explain the variance of the parameters (i.e., loadings and intercepts). Thus, a high R² value implies a high degree of measurement invariance. We extracted the respective factor loadings and intercepts of the three items measuring organizational commitment within the configural two-factor model as input for this investigation. The effect sizes for approximative invariance were R² = .984 for the loadings and R² = .994 for the intercepts, indicating good invariance. However, investigating the invariance of the parameters, only the factor loading of Item 3 (i.e., “I would turn down another job that offered quite a bit more pay in order to stay with this organization”) of Lithuania differed from the respective loadings of the other countries. We detected no differentiations for the intercepts so that they are comparable across countries. Thus, we accepted approximate scalar invariance considering the high R² values. This enables researchers to compare latent variances, covariances, and means across 32 countries (Vandenberg & Lance, 2000).

Table 7

Fit Indices for Measurement Invariance Across Countries

Note. N = 24,816

Measurement invariance across organizational employees and self/family business employees

We conducted multi-group CFA to test for measurement invariance across organizational employees (N = 20,640) and self/family business employees (N = 3,658) because previous research indicated that organizational employees and self-employees could differ in their factor loadings (Felfe et al., 2008). Again, we utilized the same procedures (MLR, FIML, ΔCFI, successive invariance testing, identification) as for the previous measurement invariance examination. First, we checked the measurement model separately for both groups. As described above, we tested a one-factor model of the OC and occupational commitment items, as well as correlated two-factor model with and without tau-equivalence. The one-factor model did not fit the data well in both groups (Organizational employees: CFI = .950, RMSEA = .143, SRMR = 0.034; self/family business employees: CFI = .933, RMSEA = .160, SRMR = .044). For the correlated two factor model, both models showed an acceptable fit, without (Organizational employees: CFI = .996, RMSEA = .049, SRMR = .011; self/family employees: CFI = .984, RMSEA = .096, SRMR = .023) and with essential tau-equivalence (Organizational employees: CFI = .980, RMSEA = .078, SRMR = .037; self/family employees: CFI = .976, RMSEA = .083, SRMR = .032). Therefore, the following analyses of measurement invariance were based on a one-factor model of OC (i.e., without the items of occupational commitment) assuming essential tau-equivalence. The configural and metric model in a multi-group CFA are equal when assuming essential tau-equivalence. We accepted configural/metric invariance due to the acceptable absolute fit (χ²(4) = 223.030, p < .001, CFI = .979, RMSEA = .037, SRMR = .037). Furthermore, scalar invariance also held (ΔCFI = –.004), but strict invariance did not hold (ΔCFI = –.016). After freely estimating the disturbance term of Item 3 (i.e., “I would turn down another job that offered quite a bit more pay in order to stay with this organization”), partial strict invariance held (ΔCFI = –.010). Additionally, we set the latent means of both groups to zero to test if the latent means were equal across groups. The constraint on latent means showed a decrease in fit comparing the partial strict model with and without it (ΔCFI = –.088) and, therefore, a higher level of OC for self/family employees. We included all fit statistics in Appendix Table A7.

Acknowledgement

All authors have contributed equally. Therefore, the authors share the first authorship.

• Julian Urban, GESIS – Leibniz Institute for the Social Sciences, P.O. Box 12 21 55, 68072 Mannheim, Germany; e-mail:  Julian.Urban@gesis.org
• Isabelle Schmidt, GESIS – Leibniz Institute for the Social Sciences, P.O. Box 12 21 55, 68072 Mannheim, Germany; e-mail: Isabelle.Schmidt@gesis.org
• Katharina Groskurth, GESIS – Leibniz Institute for the Social Sciences, P.O. Box 12 21 55, 68072 Mannheim, Germany; e-mail: katharina.groskurth@gesis.org

Data sources

ISSP surveys using the ISSP OC scale:

• International Social Survey Programme: Work Orientations II - ISSP 1997, ZA3090
• International Social Survey Programme: Work Orientations III - ISSP 2005, ZA4350
• International Social Survey Programme: Work Orientations IV - ISSP 2015, ZA3090