A total of 145 adults participated in this study. The inpatient sample included 28 chronic institutionalized patients (82.14% male) whose age ranged between 31 and 74 years (M = 54.0; SD = 10.96). Schizophrenia was the most frequent diagnosis (57.14% of the sample). Inpatients were recruited who presented fair cognitive and executive functioning (total score of 27/30 or higher in the Spanish validation of the Mini-Mental State Examination, by Lobo, Ezquerra, G´mez-Burgada, Sala & Seve-Días, 1979) and emotional stability, according to their medical record and the staff criteria. The community sample included 117 adults (39.32% male) whose age ranged between 17 and 67 years old (M = 32.17; SD = 14.04) and who met the inclusion criteria mentioned above. The community sample was recruited through online advertisement, and their data were collected through an online survey. In both groups, sampling was carried out according to sample accessibility, and participation was voluntary.

The difference in age and gender between groups resulted in further statistical computations to control for their effects on the target relation.

We employed a correlational and transversal design (Ato García & Vallejo Seco, 2011) to evaluate the relation between PIL and a general sense of health in naturally occurring groups of populations, i.e., community and inpatient adults.

The Human Research Ethics Committee at the university and the hospital reviewed and approved the study. At the hospital, the staff recruited the participants and assisted during questionnaires administration by reading the items and writing the answer when necessary (9.7% of cases). All participants were asked to complete the task in a quiet environment. Completion time was 15 min on average. Participants completed:

Purpose in Life Test (PIL, Crumbaugh & Maholick, 1964). PIL measures purpose in life and/or existential vacuum, following Frankl’s definition of the concept. We used the Spanish validation (Noblejas de la Flor, 2000), including 20 items rated on a 7-point Likert format, from 1 (indicating existential vacuum) to 7 (indicating full purpose in life). The Spanish validation of the PIL is a reliable instrument, with a Cronbach’s alpha of 0.879 (Noblejas de la Flor, 2000). The total score in the PIL ranges between 20 and 140, with higher scores indicating the presence of goals and purpose in life. We note that for the ease of responding in the inpatient sample, we reduced the Likert scale to 5 points; therefore, the total scores in our study ranged between 20 and 100.

General Health Questionnaire (GHQ-28, Goldberg & Hillier, 1979). GHQ-28 measures self-reported general health and is frequently used as an indicator of psychological well-being (Goldberg & Hillier, 1979). We used the Spanish validation of the GHQ-28 (Lobo, Pérez-Echevarría, & Artal, 1986), which shows high test-retest reliability (r = 0.90), sensitivity of 84.6% and specificity of 82% (García-Moratalla, 2008; Lobo et al., 1986). The GHQ-28 includes 28 items grouped in four factors, each of them including 7 items with four possible answers, ranging from 0 (Not at all) to 3 (much more than usual). The four factors are: 1) somatic symptoms or hypochondria; 2) anxiety and insomnia; 3) social dysfunction; and 4) severe depression —the total score range between 0 and 84, with high scores indicating poor self-reported health. In our study, the GHQ-28 was used as a general indicator of self-reported health, so the total score was entered in the analyses (Sterling, 2011).

SPSS version 16.0 was used for coding data, estimating models and performing univariate statistical computations. Univariate (means, standard deviation, skewness, and kurtosis) and bivariate statistics (correlations) were first computed. After that, univariate statistical analyses were conducted to test for differences in gender and age between the inpatient and the community sample. A two-tailed parametric t-test for independent samples, not assuming equal variances, was used to compare mean ages. A two-tailed Pearson chi-squared test and the Fisher’s exact test (given the small sample size) were conducted to compare gender.

Multivariate inferential analyses were conducted using structural equation modeling (SEM) to test the possible influence of population (inpatient vs. community) on the relation between PIL and GHQ. We employed SEM because, as compared to less complex and robust analyses, it allows controlling for measurement errors due to the observed variability between samples. SEM was performed through Mplus 6.01 software (Muthén & Muthén, 1998; 2010) using robust maximum likelihood. To evaluate the fit of the models, the following goodness-of-fit indices were reported: Chi-square test for model fit (χ²), Root Mean Square Error of Approximation (RMSEA), Comparative Fit Index (CFI), Tucker-Lewis Index (TLI) and Standardized Root Mean Square Residual (SRMR). Across analyses, the alpha level was set at 0.05, and when the probability of estimators was lower than 0.01, statistical reliability of 0.01 was informed. Two nested models were compared. Loglikelihood Chi-Square Difference Test was performed to obtain the significance of the comparison of the models.

For statistical analyses comparing the potential difference in magnitude between regression coefficients from independent samples (inpatient vs. community), we used the Fisher r-to-z transformation.

Table 1 shows the descriptive statistics (bivariate correlations, means and standard deviations) and reliability of PIL and GHQ scores. As expected, PIL average score was higher in the community than in the inpatient group, whereas GHQ average score was higher in the inpatient than in the community group. The correlation between PIL and GHQ was negative and significant (p < 0.05) in both groups, being of slightly higher magnitude in the inpatient group.

Statistical analyses revealed that inpatient participants were significantly older than community participants (54.0 and 32.17 years old respectively), t (50.58) = 8.93, p < 0.01. As well, most community participants were female (60.7%), while most inpatients were male (82.1%). Analyses showed that there was a significant relationship between gender and sample, χ²(1) = 16.614, p < 0.01 for Pearson chi-squared test and p <.01 for Fisher’s exact test.

Table 1 Table 1 Correlations, descriptive statistics and reliability for observable variables in community and inpatient samples. PIL = Purpose in Life Test. GHQ = General Health Questionnaire

*p < 0.05

The first model included age, sex and type of population as covariates influencing PIL and GHQ scores. The model was just-identified (the number of free parameters exactly equals the number of known values, with zero degrees of freedom), X² = 0, df = 0, p < 0.01, RMSEA = 0, CFI = 1.00, TLI = 1.00, SRMR = 0. As Table 2 shows, out of the three covariables, only population resulted significantly related with GHQ (β = 0.319, p < 0.05), and expectedly, PIL and GHQ were negatively related (β = -0.56, p < 0.05). This model explained 42% of the GHQ variance (p < 0.01), and 10% of the PIL variance (p > 0.05).

Table 2 Table 2 Path analysis with demographic variables. SE = Standard Error

To further the analysis of the positive and significant contribution of population on GHQ scores, we tested two nested models in which only the variable population was included as a predictor variable (MacKinnon, 2008): model a was unconstrained; and, in model b, the path of population on PIL and GHQ was constrained to zero. The results are shown in Table 3.

Table 3 Table 3 Path analysis: Nested models. SE = Standard Error

The analyses yielded model a as a just-identified model (X² = 0, df = 0, p < 0.01, RMSEA = 0, CFI = 1.00, TLI = 1.00, SRMR = 0). The explained variance was of 42% for GHQ (p < 0.01) and .08% for PIL (p > 0.05). Model b did not show good fit (X² = 16.17, df = 2, p > 0.05, RMSEA = 0.22, CFI = 0.80, TLI = 0.70, SRMR = 0.15). The model explained 38% of GHQ variance (p < 0.01). Further analyses revealed that the confidence intervals of model a and model b overlapped, hence the path difference between models was not significant. A Loglikelihood Chi-Square Difference Test (see Table 4) revealed that model a and model b were significantly different, indicating that PIL and GHQ scores were influenced by type of population.

Table 4 Table 4 Loglikelihood Chi-Square Difference Test

In order to determine whether the relation between PIL and GHQ was different for the inpatient and the community populations, we conducted regression analyses on each sample and a Fisher r-to-z transformation. As Table 5 shows, the regression coefficients were negative and significant in both samples (p < 0.01). Fisher transformation showed that the confidence intervals of the difference contained zero, thus rejecting the hypothesis that the path was different on each population. In other words, the relation between PIL and GHQ was the same in both community and inpatient populations.

Table 5 Table 5 Fisher r-to-z transformation

*p< 0.01