The study was conducted via an online survey website¹. Participants are recruited through community forums. Before participation in the survey, written informed consent was obtained from each participant. A total of 1151 Chinese adults answered the questionnaires. In this sample, the mean age is 26.92 ± 5.59 years ranging from 17 to 57. More detailed demographic information is shown in Table 1. The study was approved by the institutional review board of Shaanxi Normal University.

To measure dispositional gratitude, we used the Chinese version of the GQ that was developed by Leong (2009) using a standard back-translation procedure² and the Chinese GQ made slight adaptation in our study. For example, in our version, the word “appreciate” was translated into , rather than . The GQ includes six items and each item is answered on a seven-point Likert scale ranging from 1 (strongly disagree) to 7 (strongly agree). Consistent with the original research of gratitude (McCullough et al., 2002), internal consistency was satisfactory with Cronbach’s α coefficient of 0.87.

To measure affective well-being, the 14-item Positive Affect and Negative Affect Scale (PANAS; Kuppens et al., 2008) was employed. Participants were asked to indicate how often they felt each effect in the past month on a seven-point Likert scale ranging from 1 = “not at all” to 7 = “almost or lasting.” The scale has good levels of reliability and validity in Chinese populations (Kuppens et al., 2008; Kong and Zhao, 2013). An affect balance score (positive affect minus negative affect score) was calculated as an index of affective well-being. In current study, the Cronbach’s α coefficient was 0.87 for positive affect and 0.85 for negative affect, respectively.

To measure life satisfaction, the Satisfaction with Life Scale (SWLS, Diener et al., 1985) was employed. This scale includes five items and each item is answered on a seven-point Likert type scale ranging from 1 (strongly disagree) to 7 (strongly agree). The SWLS has good levels of reliability and validity in Chinese populations (Kong et al., 2014, 2015a,c). In current study, the Cronbach’s α coefficient was 0.81.

To test whether the Chinese GQ has one-factor structure, a CFA was first performed using Amos 22.0. The Mardia’s coefficients for multivariate kurtosis were > 3, indicating significant multivariate non-normality in our data. Therefore, we used the Bollen-Stine bootstrap procedures (2,000 samples) with Maximum Likelihood (ML) estimation to accommodate the lack of multivariate normality. Because χ² is sensitive to sample size, the following indices were used to evaluate the model’s goodness of fit: non-normal fit index (NNFI), comparative fit index (CFI), standardized root mean square residual (SRMR), and root mean square error of approximation (RMSEA). By convention, the fit was considered acceptable when NNFI and CFI ≥ 0.90, RMSEA ≤ 0.10, SRMR ≤ 0.05 (Hu and Bentler, 1999).

To evaluate measurement invariance (i.e., configural, metric, and scalar invariance) across sex, we conducted multi-group CFAs. Configural invariance refers to invariance of factor structure across sex groups. Metric invariance refers to invariance of factor loadings across sex groups. Scalar Invariance refers to invariance of item intercepts across sex groups. To evaluate differences of the models, we used the change for CFI and RMSEA (ΔCFI and ΔRMSEA) as indices (Gardner and Qualter, 2011; Martinez-Pecino and Durán, 2013). Strong invariance is supported when ΔCFI ≤ 0.01 and ΔRMSEA ≤ 0.015 (Cheung and Rensvold, 2002).

Finally, we tested sex-related latent mean differences. Tests of latent mean differences can provide more precise estimates regarding group differences than t-tests, because traditional t-tests ignore the impact of measurement error (Aiken et al., 1994). The z statistic was used to determine statistical significance between the latent means.

The CFA analysis indicated that all the fit indices of the model, except RMSEA, meet their corresponding criteria, χ²(9) = 177.49, p < 0.001, NNFI = 0.92, CFI = 0.96, RMSEA = 0.13, SRMR = 0.04. Therefore, the proposed model provided an acceptable fit to the data. To improve the model, the modification indices were examined. Item 4 (“I am grateful to a wide variety of people”) and item 5 (As I get older I find myself more able to appreciate the people, events, and situations that have been part of my life history”) reflect gratitude span, so we allowed errors of these two items to correlate. The results found that the model showed a better fit to the data, χ²(8) = 103.14, p < 0.001, NNFI = 0.95, CFI = 0.97, RMSEA = 0.10, SRMR = 0.04. The factor loadings ranged from 0.44 to 0.90 (Figure 1).

First, a multi-group CFA was carried out to test the configural invariance model in which the GQ has a one-factor structure across groups. The results showed that the model (M1) fitted very well, RMSEA = 0.075, CFI = 0.97, and all factor loading were significant (p < 0.001). These results demonstrate that the one-factor model fit the data well in both groups.

Second, following the configural invariance model, we tested a metric invariance model (M2) in which factor loadings are constrained to be equal across groups. The results showed the model fitted well, RMSEA = 0.065, CFI = 0.97. When compared to M1, there were no significant changes, ΔCFI = 0.001 and ΔRMSEA = 0.01 (Table 2). These findings suggest that factor loadings were invariant across the female and male groups.

Finally, we tested a scalar invariance model (M3) in which intercepts and factor loadings are constrained to be equal across the female and male groups. The results showed the model fitted well, RMSEA = 0.059, CFI = 0.97. When compared to M2, there were no significant changes, ΔCFI = 0.001 and ΔRMSEA = 0.006 (Table 2). These findings suggest that intercepts were invariant across the female and male groups.

Taking together, these results suggest that measurement invariance of the GQ holds across sex.

To estimate latent mean differences between different sex groups, we carried out a multi-group CFA. The male group was selected as a reference group and its latent mean is fixed to zero. Latent mean for the female group is freely estimated. The results revealed that females had higher scores than males in the GQ (Z = 2.39, p = 0.02).

The internal consistency of the model was assessed by computing the composite reliability (Raykov, 1997). In comparison with Cronbach’s alpha reliability, composite reliability does not assume that all indicators are equally reliable. If the composite reliability is more than 0.7, the model is satisfactory. In the total sample, the composite reliability for the scale was 0.87. In males the composite reliability for the scale was 0.86, and in females the composite reliability for the scale was 0.88, indicating that the internal consistency of the model is satisfactory.

Because our data were collected at one time from one source, common-method bias might occur. We applied Harman’s single-factor test to examine common method bias. A model with one factor (with all items for the three scales loading on a unique factor) was built. The results revealed that the model did not fit the data well [χ²₍₂₇₅₎ = 7814.55, p < 0.001, CFI = 0.55, GFI = 0.49, SRMR = 0.13, RMSEA = 0.15]. According to these results, we believe that common method bias might not be a serious issue in the current study.

To verify the criterion-related validity of the Chinese GQ, regression analyses were conducted with the total sample to test its relation with life satisfaction and affective well-being. The result revealed that gratitude explained 5.1% of the variance in life satisfaction (β = 0.23, p < 0.001) and explained 3.8% of the variance in affective well-being (β = 0.19, p < 0.001).

Because the significant sex difference in gratitude was found in the previous analysis, we wondered whether there also exists the sex difference in the relation of gratitude with life satisfaction and affective well-being. Therefore, we conducted an exploratory analysis to test whether sex moderates the influence of gratitude on life satisfaction and affective well-being. The results found no significant interaction effect for life satisfaction (β = -0.12, p > 0.05) and affective well-being (β = -0.06, p > 0.05). This indicated that sex did not moderate the influence of gratitude on life satisfaction and affective well-being.

To sum up, this study adds to the existing literature on gratitude in several ways. First, the Chinese GQ had good internal consistency (e.g., composite reliability) and factorial validity, suggesting that researchers can use the GQ to measure gratitude in Chinese adults. Second, the Chinese GQ had good criterion-related validity with measures of life satisfaction and affective well-being and sex did not moderate the effects of gratitude on these two constructs, indicating the males and females with higher gratitude tend to experience a happier life. Third, the Chinese GQ had configural, metric and scalar invariance across male and female groups, which suggests gratitude researchers don’t have to consider the sex matter when measuring gratitude. In addition, through tests of the latent mean differences, females had higher latent mean scores than males. The method considers the influence of measurement errors, so it can reflect the real sex differences.

Despite of these strengths, three limitations need to be considered. First, study participants were adults above 17 years old, and findings cannot be generalized to teenagers. Second, we only tested the sex invariance; therefore, further studies should test measurement invariance across other groups such as age groups. Finally, only the convergent validity of the GQ was tested, so further studies should also test its discriminant validity with other variables.