The strong worldwide interest in the quality of life (QOL) concept and the heterogeneity of its definition led the WHO, through the World Health Organization Quality of Life group (WHOQOL group), to develop the most comprehensive definition of QOL in the scientific literature [1
]. This group defines QOL as “a state of complete physical, mental and social well-being and not merely the absence of disease…” [2
]. This concept favors a transcultural, multidimensional, subjective view of QOL (self-evaluation) and considers correlating factors such as physical and mental health; independence; social relations; personal convictions and beliefs; and the environment [3
As a result of the work undertaken by the WHOQOL group, the international WHOQOL-BREF questionnaire [4
] was developed for QOL assessment. The international WHOQOL-BREF is a shortened version of the WHOQOL-100 [5
]), which has been translated/validated in many countries, including Portugal. The WHOQOL-BREF(PT) (the European Portuguese version of the international WHOQOL-BREF) preserves the WHOQOL-100’s 24 facets and the General Health Facet (GHF); it is cross-cultural and thus can be applied to individuals living in different contexts [6
The WHOQOL-BREF(PT) is generic and includes the domains of Physical Health (seven items), Psychological (six items), Social Relationships (three items), and the Environment (eight items) as well as the GHF (which includes two general items: how respondents rate their QOL and how they rate their satisfaction with their health) [7
In translating/validation the WHOQOL-BREF(PT), a team of Portuguese researchers belonging to the WHOQOL group adopted two samples from the entire Portuguese population. Both samples included people aged 18 years and over [7
]. The healthy sample included 315 citizens who had no chronic disease and did not take any type of medication, and the clinical sample included 289 citizens with different medical backgrounds coming from three public health units in Coimbra City.
Although the psychometric properties of the WHOQOL-BREF(PT) were studied for the Portuguese population in [7
], the elderly population was poorly represented in the sample. Since the aging and QOL of older people is a concern, we reevaluate the psychometric properties of the WHOQOL-BREF(PT) exclusively for the elderly population. We use a sample of 351 people aged 65 years or over residing in a community setting (in their homes or in family members’ or friends’ homes) in the Baixo Alentejo Region (BAR). This Portuguese region was chosen because (i) it faces a delicate and worrying situation of population aging with heterogeneous sociodemographics; (ii) it is a rural area located in the mid-southern part of Portugal’s main territory; (iii) it has very low population density, as geographic distances between villages range from 25 to 120 km; and (iv) the public transportation network is scarce, which presents difficulties in the mobility of older people using their own means. Finally, we compare the psychometric properties/qualities we obtain with those observed by Canavarro et al. [7
], which represents the publication of all the research work developed by the Portuguese team of researchers belonging to the WHOQOL group that translated and validated the international WHOQOL-BREF for use in Portuguese language.
The 24 facets show no substantial violation of normality because |Sk| ≤ 0.934 and |Ku| ≤ 1.301. The initial CFA model shows a poor overall quality of fit (see Table 2
). To obtain a better model fit, the measurement errors between certain facets were correlated, as suggested by MI values, with the assumption that all respective items involved presented some similar content. After all possible adjustments were made to the initial model, we obtained the best-fitting model (a lower MECVI = 2.093; values presented at the top of Figure 1
and Table 2
). Regarding the standardized factor loadings (0.391 ≤ λ
≤ 0.938, p
< 0.001 and average = 0.626), approximately 75% present values λ
≥ 0.5, 13% present values between 0.5 and 0.45 (near 0.5), and only 12% present values lower than 0.45 (the lowest λ
= 0.391 for item F19.3). According to Canavarro et al. [7
], values of λ
≥ 0.3 are admissible for this type of construct. With respect to FC, only the Social Relationships domain (CRSocial Relationships
= 0.590) presents a value lower than 0.7 (with CR ≥ 0.70 recommended as a positive threshold for CR in Marôco [13
]): (i) CRPhysical Health
= 0.880; (ii) CRPsychological
= 0.849; and (iii) CREnvironment
= 0.761. In the reliability analysis based on the α value, the Social Relationships domain presents the lowest value (αSocial Relationships
= 0.580), with the remaining domains presenting 0.774 ≤ α ≤ 0.876. Factorial validity is guaranteed since the items (facets) are aligned with what each specific latent factor measures. With respect to convergent validity, only the Physical Health domain presents an acceptable value of AVEPhysical Health
= 0.532. We observe a nearly acceptable value of AVEPsychological
= 0.488, a weak value of AVESocial Relationships
= 0.326 and a very weak value of AVEEnvironment
= 0.291 (with AVE ≥ 0.50 recommended as a positive threshold for AVE in Marôco [13
] in the case of exploratory investigations). Finally, the adjusted model presents no discriminant validity since the results of all possible expressions (AVEi
^2)) are false.
As an alternative solution to the high correlations between latent factors shown in the model of Figure 1
(all with p
< 0.001), we performed a hierarchical model (Figure 2
) that included a high-order factor called QOL (second-order factor), as recommended for this type of model in Marôco [13
]. The values at the top of Figure 2
indicate that this second-order model shows a reasonable adjustment (similar to that presented at the top of Figure 1
), with the second-order factor being the QOL measure expressed through various items (observed variables) and associated with each of the four domains. The correlations between QOL and the four domains are all high and statistically significant (p
< 0.001): (i) ρPsychological
= 0.95; (ii) ρ = 0.91 for both the Social Relationships and Environment domains; and (iii) ρPhysical Health
= 0.85. CR and α for the QOL factor achieve very good values of 0.947 and 0.927, respectively.
The WHOQOL-BREF(PT) includes the GHF, as described in the Background section. We proceeded with a third model (see Figure 3
) to verify the correlational effect between the latent factors GHF and QOL. The top part of Figure 3
shows that the adjustment indexes present a similar quality to the second- and first-order models. The correlation between GHF and QOL is high (ρ = 0.88; p
< 0.001), which allows us to infer that the two factors have concurrent validity.
summarizes the descriptive statistics for the four domains, GHF and QOL (24 items), of the WHOQOL-BREF(PT).
The translation and validation performed by the Portuguese team of researchers belonging to the WHOQOL group comprising Canavarro et al. [7
] was based on a sample of individuals aged 18 or older. In this paper, a sample of elderly people (65 years and older) was used instead. Therefore, comparisons strictly based on age groups are difficult to perform. However, there are no major differences between the psychometric qualities found in Canavarro et al. [7
] and those described here. A large set of psychometric properties were extracted based on the approach developed in this paper, as described below.
The reliability analysis based on Cronbach’s α reported in Canavarro et al. [7
] is considered reasonable by the authors. The results in our study are similar to those in that study. The Social Relationships domain presents the lowest value in both studies (α = 0.64 in Canavarro et al. [7
] and 0.580 in the present study), and the remaining domains present values ranging from 0.78 ≤ α ≤ 0.86 (in Canavarro et al. [7
]) and 0.774 ≤ α ≤ 0.876 (in the present study). In terms of each factor’s reliability, the lowest values for Social Relationships are usually justified by the small number of items included within this domain [3
Pearson’s correlations reported in Canavarro et al. [7
] are all lower than those shown in Figure 1
, namely, between the Psychological and Physical Health domains (0.55 and 0.82, respectively), between the Psychological and Social Relationships domains (0.56 and 0.89, respectively), between the Psychological and Environment domains (0.57 and 0.86, respectively), and between the Social Relationships and Environment domains (0.50 and 0.89, respectively). This allows us to state that in our elderly sample, the domains are more interrelated than those in the sample reported by Canavarro et al. [7
In Canavarro et al. [7
], there are no results on the use of a CFA model. Therefore, a comparison with certain results presented in this study cannot be performed. The second-order factor (QOL, see Figure 2
) was predicted only by the authors of this paper and not by the authors of the original translation/validation of the questionnaire but only in this paper.
The WHOQOL-BREF(PT) showed factorial validity in the present study (no factor loadings are reported in Canavarro et al. [7
]) and presented high correlations between first-order factors but low AVE values. Therefore, the construct failed in discriminant validity because according to Marôco [13
], the AVE of the CFA model is less than the square of the correlation between the latent factors involved. Notably, the discriminant validity reported in Canavarro et al. [7
] was considered successful. However, discriminant validity in [7
] was evaluated as the capability of the WHOQOL-BREF(PT) questionnaire to discriminate between individuals in the normal population and individuals with a medical pathology with respect to all domains and the GHF of QOL. Therefore, the two approaches, i.e., discriminating between sample elements in [7
] and extracting domains through CFA in this study, are completely different; therefore, a comparison of the two studies’ results is not possible.
Based on CFA (a model not developed in [7
]), a factor score weight (fsw
) is extracted for each item and is used to compute each factor score. In the original translation/validation work, each factor score (four domains and GHF) was computed as the average item value (1 to 5) of the items included in the respective domain using a weight value of 1 for each item. However, we propose the computation of factor scores (four domains, GHF and a general QOL) based on the extracted fsw
(a unique weight for each item). Figure 4
shows the two computation cases, with the average score of the entire sample calculated using the WHO strategy (black cylinders) and using the fsw
values extracted from CFA (light-gray cylinders). Comparing the six cases, four domains, GHF and QOL generalized score (24 items), there is no significant difference between the two approaches in the global analysis of the sample results. However, analyzing individual scores for each participant separately, we detect differences (maximum positive differences, i.e., MaxDiff
, and negative differences, i.e., MinDiff
, both in dark-gray cylinders, as well as the standard deviations of the differences found, i.e., StdDiff
, in dark-gray cylinders), which are greater for the Physical Health and Psychological domains and weaker than the remaining factor scores.
With respect to the third model (see Figure 3
), the correlation between QOL and GHF is strong (ρ = 0.85, p
< 0.001), suggesting that GHF is a factor that could be used as a QOL generalized measure.