Prior to data analyses, data were examined to assess the accuracy of data entry, the percentage of missing values, and the assumption of linearity and normality [32], to ensure the quality of the 2 data sets. Data entry accuracy was examined using the range of data [33]. In this study, 1.7% and 10.3% of data were missing from the healthy referents and TB patients, respectively. Both groups had approximately less than 10% of missing data, which indicated a reasonable missing data rate for the following analyses [33].
Because of missing data on both TB patients and healthy referents, the maximum likelihood estimator was applied to obtain an estimation of CFA measurement models (Brown [30]). The normality assumption was examined through skewness and kurtosis, with absolute values less than one. According to Muthen and Kaplan [34], if variables with skewness and kurtosis are close to −1 and +1, estimating the parameter of non-normal variables by using the maximum likelihood method produces acceptable values. The assumption of linearity among pairs of variables was examined through scatterplot inspection [32]. A nonlinear relationship was not detected in the data from TB patients or healthy referents.
Demographic characteristics
Of the 140 patients with tuberculosis, the mean age was 50.13 years (SD = 18.62); 70.7% of the sample was men, and more than half of them were aboriginal Taiwanese (Table 1). The healthy referents in this study were recruited by matching age, sex, and ethnicity proportions to the TB patient group. The mean age of the 130 healthy referents was 47.91 years (SD = 18.94).
Table 1
Demographic characteristics of patients with tuberculosis and healthy referents
Chi-squared tests and t tests were conducted to examine the differences between the TB patient group and the healthy referent group in major demographic characteristics such as age, sex, and education. As expected, no significant differences emerged between the 2 groups in age (t (263) = 0.963, p = .337), sex (χ
2 (1) = 0.650, p = .434), ethnicity (χ
2 (3) = 0.131, p = .988), and marriage status (χ
2 (3) = 3.553, p = .314).
Compared with TB patients, more than half of the healthy referents had a high school or college degree, and relatively higher personal incomes than TB patients. Significant differences existed between the 2 groups in level of education (χ
2 (3) = 38.177, p < .001) and in personal monthly income (χ
2 (2) = 14.621, p < .001). Descriptive statistics are shown in Table 1.
Internal consistency reliability
The Cronbach’s alpha values of the WHOQOL-BREF TW total scale and subscales are presented in Table 2. Regardless of the versions of the WHOQOL-BREF TW total scale, the Cronbach’s alpha values of the WHOQOL-BREF TW total scale were all above .91 for TB patients, healthy referents, and all participants. The alpha values of the WHOQOL-BREF TW subscales ranged from .61 to .82 for the TB patient group, from .53 to .87 for the healthy referent group, and from .58 to .85 for all participants. Except for the social relationship subscale, the alpha values of the WHOQOL-BREF TW total and subscales were all larger than 0.7, the lower acceptable bound for an alpha value [35]. These results demonstrate good internal consistency of the WHOQOL-BREF TW among TB patients and healthy referents.
Table 2
Cronbach’s coefficient alpha values by the status of participants for the WHOQOL-BREF TW total scale and four subscales
Construct validity
The Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy and the Bartlett test of sphericity were used prior to factor analysis, to ensure that the data from both patients with TB and healthy referents were appropriate for conducting factor analysis. The KMO measure of sampling adequacy for TB patients and healthy referents was .879 and .887, respectively, indicating that these 2 samples had a sufficient level of factorability. The Bartlett tests of sphericity for both data were significant at the .001 level, indicating that the correlation matrices were not identical to the factor structure matrices. Both tests revealed that data from TB patients and healthy referents were appropriate for factor analysis [32].
When performing factor analysis, the sample size should be at least 250 to 300 cases [32]. However, in the present study, the number of patients with TB in eastern Taiwan was relatively small. Instead of following the general expected sample size, this research adopted the Gorsuch [36] perspective to evaluate the sufficiency of the sample size in this study. The Gorsuch [36] suggestion on a sufficient sample size for factor analysis is that a ratio of 5 participants per item should be present and that the total sample size should include more than 100 participants. Based on the Gorsuch perspective, the sample sizes of TB patients and healthy referents meant that the data were sufficient for factor analysis.
In the EFA of the TB patient data, factor analysis was conducted by principal component analysis, followed by Oblimin rotation with an eigenvalue above 1. Six conceptually meaningful factors were extracted, explaining 66% of the total variance (Table 3). In this 6-factor model, 3 items (Pain, Medical dependency, and Life enjoyment) were excluded because of low factor loadings (lower than 0.3); thus, only 23 items were included. The 6-factor model was a variation on the original WHOQOL-BREF factor structure.
Table 3
Factor analysis of WHOQOL-BREF TW data from 140 patients with tuberculosis
Factor 1 comprised 7 items belonging to 3 original WHOQOL-BREF domains (physical, psychological, and social relationship domains) and was labeled as the self-confirmation factor to capture the need for people to confirm the meaning of self. Factors 2 and 6 consisted of items belonging to 2 WHOQOL-BREF original domains (environmental and social relationship domains), and were renamed as the social support factor and the accessibility factor. Factor 3 included items from 3 WHOQOL-BREF original domains (environmental, psychological, and social relationship domains), and was renamed as the psycho-social-environmental factor. Factor 4 (Availability) and Factor 5 (Activity) included only some items from the original WHOQOL-BREF environmental and physical domains, and can be regarded as the subscales of these 2 original WHOQOL-BREF domains.
For the healthy referents, 5 factors were extracted using principal component analysis with Varimax rotation, accounting for 62.58% of the total variance. Because of parsimony of the factor structure, Factor 5, which included only one item, was excluded from the final model. Therefore, the final EFA factor model for healthy referents consisted of 4 factors that explained 55.69% of the total variance. In this 4-factor model, items 3 (pain), 4 (medical dependency), 16 (satisfaction with sleep), and 22 (satisfaction with friend support) were excluded because of low factor loadings (lower than 0.3). This resulted in 22 included items (Table 4). The final EFA model was similar to the WHOQOL-BREF 4-factor model, except for the social relationship domain. The first 3 factors consisted of most items belonging to the corresponding original WHOQOL-BREF domains. The social relationship factor, which included only half of the items belonging to the original WHOQOL-BREF social relationship domain, is a more specific definition of the original WHOQOL-BREF domain.
Table 4
Factor analysis of WHOQOL-BREF TW data from 130 healthy referents
In the CFA, the fit indices of the models resulting from EFA operations with the WHOQOL-BREF TW were compared with the fit indices of the original 4-model WHOQOL-BREF for TB patients and healthy referents (Table 5). None of the models fit the chi-square fit index, but performed well for the relative chi-square, with values that ranged from 1.73 to 1.94. This was below the recommended cut-off value of 3 [29]. The 2 EFA models had RMSEA values lower than .08, which indicated no significant errors in either model [31]. However, the 2 WHOQOL-BREF models had large RMSEA values, which indicated that these 2 models might contain significant errors. Although some CFI values were less than the required value of 0.9 [30], most were above 0.8. For the AIC values, the EFA models in this study performed better than the WHOQOL-BREF original models. All the fit indices suggested that these EFA models displayed a better fit for both TB patients and healthy referents from Eastern Taiwan.
Table 5
Fit indices for the EFA models vs. WHO four-factor models of the original WHOQOL-BREF
Convergent validity
Regardless of the sign of a correlation coefficient, Weinberg and Goldberg [37] suggested that Pearson correlation values in the range of .8 to 1.0 are considered strong, in the range of .4 to .6 are considered moderate, and in the range of 0 to .2 are considered weak. Based on this definition, the associations among the WHOQOL-BREF domain scores were moderately correlated and ranged from .56 to .74 for the TB patient group and from .57 to .69 for the healthy referent group (Table 6). The WHOQOL-BREF domain scores were also moderately related to 2 QOL global items--general QOL (G1) and health-related QOL (G2), and ranged from .34 to .69 for the TB patient group (Table 6). For the healthy referent group, the associations among the WHOQOL-BREF domain scores and 2 QOL global items were weak to moderately correlated in the range of .21 to .59 (Table 6). All these correlation coefficients were statistically significant at the .05 significance level.
Table 6
Pearson correlations between the WHOQOL-BREF TW domains, and with the two global items
Discriminant validity
As mentioned, 4 WHOQOL-BREF TW domain scores were highly interrelated (see Table 6), which suggested the use of a multivariate analysis to analyze all these domain scores simultaneously. A series of examinations between the TB patient group and the healthy referent group in the demographic characteristics of participants also revealed significant group differences in both level of education and personal monthly income. These 2 variables were incorporated into the following multivariate analysis.
Using the Wilk Lambda criterion, a MANCOVA test yielded a significant group effect and a significant group*income interaction effect on 4 WHOQOL-BREF TW domain scores, F (4, 200) = 3.365, p = .011 < .05, and F (8, 200) = 2.207, p = .026 < .05, respectively (Table 7). The effect of the covariate (level of education) on the WHOQOL-BREF TW domain scores was also significant, F (4, 200) = 3.495, p = .009 < .05 (Table 7). However, the effect of personal monthly income on these domain scores was not significant, F (8, 400) = 1.550, p = .138 > .05 (Table 7).
Table 7
Multivariate statistics for main effects and interaction effects on the WHOQOL-BREF TW domain scores
Although the results of the univariate analyses indicate that healthy referents had higher scores than TB patients on 3 WHOQOL-BREF domains (physical, environmental, and psychological domains, Table 8). Healthy referents also had higher scores than TB patients in the social relationship domain, but the mean difference between these 2 groups did not reach statistical significance. These analyses revealed the discriminant validity of the WHOQOL-BREF TW on TB patients and healthy referents.
Table 8
Means, standard deviations and F values for the WHOQOL-BREF TW domain scores