Morningness is a trait-like variable which has been extensively studied within chronobiology. Despite the number of existing measures to assess morningness, there is a need for brief instruments that are psychometrically sound to be used in applied settings. Accordingly, the modified version of the Preferences Scale (PS-6) comprises six items and was reported to have adequate reliability and satisfactory validity indicators. In this paper, the psychometric properties of the Portuguese version

more »

... the PS-6 are reported. A total of 700 students attending medical school were recruited and this sample was randomly divided into two groups. In the first group (n = 352) we assessed the internal consistency calculations and conducted a principal component analysis of the posited structure. In the second group (n = 348) we conducted a confirmatory factor analysis (CFA) using structural equation modelling. Overall, the results indicated that the PS-6 has adequate reliability (α = .75) and is constituted by two components: (I) preferred cognitive activities timings and (II) preferred sleepingeating timings, respectively. This factorial structure was then verified through CFA. In conclusion, the Portuguese version of the PS seems suitable for use in research and applied settings such as shift work schedules management. However, the applicability of the PS-6 in other samples and further validity indicators should be both investigated. The use of actigraphy and biological measures should be also collected to enhance the robustness of the PS-6. It is known that circadian rhythmicity is essential for health and well-being. In the chronobiology and chronopsychology domains, morningness has been perhaps the most studied construct (Roenneberg, 2012a). Morningness is a dimension of the chronotype and pertains to the time of day that individuals perform and feel best (Roenneberg, 2012b). Basically, individuals who go to bed and wake up early, and report peak performance in the morning are defined as morningness-type. On the other hand, people who go to bed and wake-up later, and report their performance peak in the afternoon/night period are considered to be evening type (Adan et al., 2012; Roenneberg, 2012a). Most individuals are intermediate or neutral, though a tendency to morningness or eveningness may be present. The literature increasingly suggests morningness is considered a trait comprising a highly heritable load (Roenneberg, 2012b). The impact of biological rhythms on health and well-being is routinely reported. Shift workers, for instance, constitute a group prone to develop a number of dysfunctions related to rhythms desynchronization (Drake & Wright Jr., 2016). Furthermore, Adan et al. (2012) provided a recent review of the literature on chronotype and well-being. The importance of biological rhythms on humans´ lives suggests the need for chronobiology to develop and/or refine instruments that can accurately assess chronotype. Morningness may be objectively evaluated directly through biological measures such as regular recording of temperature variation, or changes in bio-chemical markers such as melatonin. Alternatively, a number of self-report measures have been developed and the more commonly used instruments were recently reviewed by Di Milia et al. (2013). Self-report tools reveal some advantages over the biological methods since they comprise fewer costs, are non-invasive, and are easy-to use measures (Gomes, 2005). Over the last few decades, several instruments concerned with chronotype and morningness type have been developed (Adan et al., 2012). One of them is the 12-item Preferences Scale (PS; Smith et al., 2002). The studies which have used the 12-item PS reported good psychometric properties but there has been disagreement pertaining to its factorial compositions is not clear. Studies by Bohle, Tiller and Brown (2001) and Zickar, Russell, Smith, Bohle and Tilley (2002) posited a unifactorial structure, whereas Smith et al. (2002) proposed a two-factor solution. The PS was also the focus of an investigation by Di Milia (2005) that resulted in a 6-item version of the PS (PS-6) due to the weak measurement properties of some items in the original scale. A strength of the PS is that unlike most self-report measures that refer to timing of activity, the PS-6 makes no reference to timing. A limitation of reference to timing is that activity appears to reflect cultural differences for such behavior and therefore, such measures are not useful in cross cultural research (Caci et al., 2005). Instead, the PS-6 asks participants to consider their preference for activity in terms of much earlier or later compared to other people. The PS-6 is presented as having a two-factor structure and was developed based upon a student sample, and subsequently replicated in a working sample using confirmatory factor analysis. Di Milia (2005) reported the measure had good internal properties and construct validity. A follow up study again demonstrated psychometric robustness of the PS-6 (Di Milia, Wikman, & Smith, 2008). The utility of the PS-6 (Di Milia, 2005) however, has not been further explored in other cultures and/or work settings. The aim of the present study is to extend the few studies that have employed the PS (Smith et al., 2002) and the PS-6 by testing the psychometric properties of the PS-6 in a sample of Portuguese undergraduate students. < .05 was considered to indicate statistical significance. Results (Study 1) Descriptive statistics Regarding the PS-6 total score the overall mean was 17.78 (3.51) [min = 7, max = 30]. Pertaining to results concerning gender (men: M = 17.93; SD = 3.69 / women: M = 17.71; SD = 3.43), no differences were observed [t (350) = .566; p = .572]. No significant association between PS-6 and age was verified (r = -.01; p = .84). Internal Consistency Cronbach alpha for the total scale was .75 which is considered an adequate value (Field, 2013). The corrected item-total correlations ranged from .44 to .58 and this suggested no further improvement to scale reliability (cf. Table 1 ). INSERT TABLE 1 HERE PS-6 factor structure To study the composition of the PS-6 we followed the statistical procedure employed by Di Milia (2005). First, a PCA with Direct Oblimin rotation was performed given that the two-factor components should be correlated (Field, 2013; Tabachnick & Fidell, 2012). In order to check the suitability of our data to a principal component analysis, various assumptions were verified: adequate sample size in terms of case to item ratio (minimum n=10:1); the majority of correlation coefficients (r) above 0.3, absence of multicollinearity and singularity among variables; Kaiser-Meyer-Olkin (KMO) Measure of Sampling Adequacy = .746 (> .60), and a significant Bartlett's Test of Sphericity [χ 2 (15) = 554.146; p < .001]. For components´ extraction, we followed three methods: Kaiser´s eigenvalue criterion > 1, Cattell´s Scree Plot and Parallel Analysis. The Parallel Analysis was run through a SPSS syntax file (O'Connor, 2000). Despite its infrequent use, parallel analysis seems to be the more stringent and valid method to extract components / factors (Horn, 1965). All these methods converged to a same factorial solution. Two components were extracted accounting for 65.9 % of the total variance (cf. Table 2 ). The component I (preferred cognitive activities timings) was constituted by items 1, 3, and 5, whereas component II (preferred sleeping-eating timings) was constituted by the remaining items (i.e., 2, 4, and 6). No cross-loadings were observed and the minimum factor loading was λ = .62. INSERT TABLE 2 HERE Results (Study 2) Descriptive statistics The overall mean score for the PS-6 in the second sample was 17.70 ( 3.53) [min=8, max=30] and again, no gender differences were found (men: M = 17.83; SD = 3.73 / women: M = 17.64; SD = 3.43) [t (346) = .479; p = .63]. A significant association between PS-6 and age was verified (r = .13; p = .01) albeit of small magnitude. Confirmatory Factor Analysis Prior to conducting the CFA we confirmed the data were suitable for the analysis. We checked: adequate sample size; absence of significant outliers and multicollinearity, and normality of the data. Our sample size was > 200 which enabled to perform CFA without major concerns, no significant outliers were found and none of the associations of the variables was r > .85. With regard to normality of the data we found that both univariate skewness (min: -0.55 [item 4]; max: 0.37 [item 1]) and kurtosis (min: -0.46 [item 3]; max: 0.53 [item 6]) were within the normality boundaries frequently considered (i.e., sk <│3│and ku <│7│) (Kline, 2005; Weston & Gore, 2006). Multivariate kurtosis was evaluated through the Mardia´s normalized estimate. The obtained value (9.97) was greater than the recommended estimate and this may be considered a slight violation of the multivariate kurtosis (Byrne, 2010; DeCarlo, 1997). Multivariate outliers were evaluated through Mahalanobis Distance statistic (D 2 ). Possible outliers were identified; however, we decided to retain them in the analysis as they did not significantly influence the overall results. Two possible solutions were tested using CFA: a unidimensional and a two-factor structure according to the literature. Overall, the unidimensional structure showed a poor goodness-of-fit. In terms of local adjustment, the minimum standardized coefficients path was λ = .47 (items 4 and 6) being superior to > .40 as recommended by Tabacknick and Fidell (2012). Regarding global adjustment the principal results were: χ 2 (9) = 78.043; p = .00; χ 2 / df = 8.67; CFI = .86; GFI = .99; RMSEA = .15 [CI 90% = 0.12 -0.18]; ECVI = .29. On the other hand, the two-factor structure achieved an excellent fit to the data. In terms of local adjustment, the minimum standardized coefficients path was λ = .53 (item 6) being superior to > .40 as recommended by Tabachnick & Fidell (2012) (cf. Figure 1). Regarding global adjustment the principal results were: χ 2 (8) = 6.36; p = .61; χ 2 / df = .794; CFI = 1.000; GFI = .994; RMSEA = .000 [CI 90% = 0.000 -0.054]; ECVI = .093. INSERT FIGURE 1 HERE In addition, composite reliability (CR) indices were calculated for PS-6 total score and both factors (i.e., preferred cognitive activities timings and preferred sleepingeating timings). CR values were .83, .74 and .68, respectively. According to the recommendations, CR > .70 is used as a good indicator of the scale´s reliability (Kline, 2005; Tabachnick & Fidell, 2012). Regarding average variance extracted (AVE) which concerns to a well-known measure of convergent validity related to CFA, the values for PS-6 total score and both factors (i.e., preferred cognitive activities timings and preferred sleeping-eating timings) were .46, .50 and .42, respectively. Despite these values are in general below the cut-off of .50, it is probable that this may be related to the few items that compose each factor (i.e., 3 items) in particular and the overall scale