Of the 1090 recruited participants, 1052 (96%; Nmale = 455, Nfemale = 597) completed the questionnaire. The respondents comprised 823 participants (78.2%) without chronic illness and 229 (21.8%) with chronic illness. For this study, "chronic illness" was operationally defined as a medical condition diagnosed by a doctor at least six months before the study, for which either the symptom(s) still persisted or relevant medical treatment continued. In terms of age distribution, there were 667 (63.4%) in the younger group (age < 45 years), 193 (18.3%) in the middle age group (age 45-59 years), and 192 (18.3%) in the older age group (age > 60 years), with an overage mean age of 40.44 years (SD = 18, median = 34). Among the participants, 601 (57.1%) were employed, 125 (11.9%) were students, 236 (22.4%) were retired, and 90 (8.6%) were either casual workers or unemployed.
There were no significant differences in gender, family size, or monthly family income between participants with and without chronic illness. However, age (χ2 = 3.3, p < 0.001), marital status (χ2 = 65.9, p < 0.001), education level (χ2 = 76.9, p < 0.001), and employment status (χ2 = 137.9, p < 0.001) were significantly associated with chronic illness. Specifically, those with chronic illness were older, less educated, and more likely to be divorced and widowed. The participants revealed 35 identifiable physical and mental health conditions according to the International Classification of Diseases (ICD)-10 categories. Of the 229 participants with chronic diseases, 198 (86.46%) had one, 22 (9.61%) had two, and 9 (3.93%) had three or more chronic illnesses. The most prevalent conditions were hypertension (7.89%), diabetes (2.57%), chronic rhinitis (2.57%), and chronic gastroenteritis (2.19%). With regard to monthly family income, 22.0% of the participants received less than CNY(Chinese Yuan) 1500 (broadly described as low income; US$1 ≈ CNY6.50), 53.1% received CNY 1500-4000 (middle income), and 24.9% received more than CNY 4000 (high income).
The responses to each item of the questionnaire were fairly distributed across the full range of the scale, with no evidence of ceiling or floor effects for any item in the total data set (Table ). Missing data was reasonably rare, with a total of only 9.2% of the 1052 participants reporting missing data for one or more of the items. The item with the highest rate of missing values (8.8%) was Item 21 "How satisfied are you with your sexual life?" All other items had a rate of missing values that was well below 0.3%.
Distribution of responses (%) in the mainland Chinese WHOQOL-BREF (N = 1052)
As a measure of the internal consistency of the scale, Cronbach's α was reported for the total subject population and each sub-group (Table ). For the total population, the Cronbach's α was 0.89 for the 26 items adapted from the original standard WHOQOL-BREF instrument and 0.88 for the 28 items that included the two additional questions for mainland Chinese participants. For the original 26-item (> 0.70) instrument, the Cronbach's α was acceptable for the psychological (0.76), social (0.72), and environmental (0.78) domains, but was only marginally acceptable for the physical domain (0.67). For this domain, the Cronbach's α would have increased to 0.71 if Item 3 ('pain and discomfort') and Item 4 ('dependence on medication') had been deleted. The Cronbach's α coefficient for people without chronic illness were much higher than for those with chronic illness (Table ).
Internal consistency reliability (Cronbach's α) of each dimension for different samples
To analyze the two additional items specific to the Chinese instrument, we added them to the relevant domains based on their linguistic and semantic meanings after discussion with the experts who developed the Chinese WHOQOL questionnaires. Item 27 ('Does family friction affect your life?') was included in the social relationship domain, and Item 28 ('How is your appetite?') was added to the physical domain. Cronbach's α showed an increase in the 8-item physical domain that included Item 28, but dropped substantially below 0.70 in the 4-item social domain that included Item 27 (Table ). Table summarizes our findings in comparison with those from Taiwan [25
], Hong Kong [24
], and the original normative sample (23 countries) [9
]. In general, our results were comparable with these studies on other Chinese populations.
Convergent and discriminant validity
In itemized psychometric analyses, to support the convergent and discriminant validity of the items in the instrument, each item should have much higher correlations with items in its own domain and lower correlations with items in the other non-corresponding domains (Table ). As can be seen from Table (item convergence), 20 of the 24 (83%) of the items correlated at least 0.40 with its domain score and thus met the criterion for item convergence. Among the four domains, the items in the physical domain performed the worst and were in line with its lower Cronbach's alpha value. This might reflect the more heterogeneous nature of the items in the domain.
Convergent validity of WHOQOL-BREF
Items correlated more strongly with items in the same (i.e., corresponding) domains than with those in other (i.e., non-corresponding) domains. Thus, for example, items in the physical domain correlated more robustly with the other items in the physical domain (average item corrected-total correlation for own domain = 0.39) than with those in the other three non-corresponding domains (average correlation = 0.29). Similarly, item-domain correlations for the psychological, social relationship, and environmental domains within the same domain were much higher than those for non-corresponding domains (Table ). In general, the above results supported the a priori four-domain structure.
The Pearson correlation coefficients between the domain scores were high, ranging from 0.40 (physical and environmental) to 0.58 (psychological and environmental). The correlations among the various domains of the WHOQOL-BREF for different Chinese populations are summarized in Table [35
A known group comparison method was adopted to provide further support for the discriminant validity of the instrument. Specifically, the QOL was compared between participants with (n = 229) and without (n = 823) chronic illness (Table ). The results showed that the Chinese version of WHOQOL-BREF could discriminate people with and without chronic illness. The participants with chronic physical disorders showed significantly lower scores in the physical, psychological, and social relationship domains (all p < 0.05), but not in the environmental domain, as compared with the participants without chronic illness (Table ). The effect sizes of the comparisons between the participants with and without chronic diseases were 0.55 for physical, 0.15 for psychological, 0.18 for social relationship, -0.04 for the environmental domains, and 0.40 for the general facet on health and overall QOL.
Discriminant validity of WHOQOL-BREF(score range 0-100)
Factorial construct validity
CFA with the commercially available software LISREL (version 8.8) was used to confirm the factorial structure of the WHOQOL-BREF items. When the fit of the data to the a priori model is acceptable, the proposed factor model is said to be applicable to the collected data. For items belonging to different factors in a factorial model, sometimes it is logical and necessary to allow their uniqueness to be freely correlated [38
Results showed that the fit of the data to the four-factor model was only marginally acceptable (Table , II) and would be improved substantially if two pairs of items with matching content had been allowed to be correlated freely (Table , III); Δχ2
(2) = 565, ΔRMSEA = 0.013, ΔNNFI = 0.032, ΔCFI = 0.029. The two pairs of items were Item 5 ('enjoy life?') and Item 6 ('meaningful life') as well as Item 8 ('safe in daily life?') and Item 9 ('healthy physical environment?') The goodness-of-fit statistics with and without the correlated uniqueness were comparable to those reported in the original manual that described the construction of the instruments for various countries (Table , I) [9
]. These results were also comparable with those for the Hong Kong version (CFI = 0.894) [24
] and the Taiwanese version (CFI = 0.886) [25
] of the instrument (Table , IV).
Goodness of fit for confirmatory factor analyses of different samples and multiple-group equivalence conditions
To determine whether the factorial structures for the participants with and without chronic illness were similar, multiple group confirmatory analyses were conducted. The results showed that the fit of the model did not decline much when the loadings, factor covariance, and item uniqueness were forced to be invariant between the two groups (Table , IV). This suggested that the participants with and without chronic illness shared the same factorial structure, loadings, and reliabilities and thus supported the use of the mainland Chinese version of WHOQOL-BREF for these two populations.
Comparison of the mean scores under the multiple-sample structural equation modelling with mean structure [32
] showed that the participants without chronic illness had significantly higher QOL in the physical, psychological, and social relationship dimensions, but not in the environmental dimension; t
(1052) = 4.156, 2.232, 2.347, -0.684, respectively; all p
< 0.05 except for the last comparison. This confirmed the results obtained using the means of the items in each scale reported earlier.
Quality of life outcomes
Table shows the mean and SD of each domain of the WHOQOL-BREF for the total sample in this study as well as the results from the Taiwanese and normative samples [9
]. Zero-order correlations among the demographic variables were computed (Table ). To investigate how QOL was related to the socio-demographic variables, four separate multiple stepwise regression analyses were conducted. The score for each QOL dimension was used separately as the criterion variable and the demographic variables were entered as the predictors. Regression analyses, rather than multiple t-tests or analysis of variance, were adopted so that the effects of gender, for example, could be analyzed while the effects of other demographic variables (e.g. age, marital status) were controlled.
Dimensional scores and regression analyses of demographic variables (standardized beta weights)
The results showed that the males had significantly higher QOL in the psychological domain than the females (Table ). Younger residents had significantly higher physical QOL but significantly lower environmental QOL than older residents. Whereas marital status was not related to QOL, socioeconomic status, measured by levels of education and income, was highly related. Further analyses showed that after controlling for other relevant variables, education had substantially greater positive effects than income. Participants with a higher level of education had significantly higher physical, psychological, social, and environmental QOL than the less educated ones, and wealthier individuals had significantly higher psychological and environmental QOL than poorer ones. Residents without chronic illness had significantly higher physical QOL than those with chronic illness. Furthermore, employed individuals had significantly higher physical and social QOL than those who were unemployed.
Prevalence of low QOL participants
As discussed in the literature review above, we can set cut-off criteria (standard) for low QOL and estimate the proportion of people with low QOL in the sample (Table ). Specifically we set (i) "70% of the maximum score" and (ii) "1 SD below the mean" as the cut-off criteria and calculated the proportion of people below these cut-off scores.
Prevalence of people with low quality of life (all without chronic illness)
The two criteria produced very different figures for the probable prevalence of low QOL in this sample. When the 70% score was used as the cut-off criterion, 54.7-90% of the sample would be considered to have poor QOL, consisting of 57.1-90.8% of men and 62.8-89.4% of women. Using the criterion of < 1 SD, 6.9-15.9% of the subjects had low QOL, consisting of 8.4-15.2% men and 5.0-17% women.