After obtaining permission from Susan Noble Walker (the inventor of the Scale), the original version of HPLP-II was translated into Italian language (Appendix A) by an expert translator and then back-translated into English by another translator with the same language proficiency. Then, two expert psychologists who are proficient in both languages reviewed the items to agree on the final version used in the present study.

The study was designed and performed in accordance with the Declaration of Helsinki. The protocol included a full assurance of anonymity, participation discretion, and the absence of risk, burden, sponsors, conflicts of interest, and incentives for the responding participants. The appropriate ethical committee approved the protocol. Implied consent was obtained, because the questionnaire was administered through an electronic tool (each questionnaire has been filled one time based on IP address), which the students had to specifically and intentionally access on the Internet. Nevertheless, a disclaimer text indicated the aim of the study, and the anonymous and voluntary nature of participation was presented in the header of the web page questionnaire.

The data were collected from an online questionnaire using Research Electronic Data Capture (REDCap), an electronic data capture tool [25] hosted by the university. REDCap is a secure, Internet-based application designed to support data capture for research studies. It provides: 1) an intuitive interface for validated data entry; 2) audit trails for tracking data manipulation and export procedures; 3) automated export procedures for seamless data downloads to common statistical packages; and 4) procedures for importing data from external sources. The online questionnaire used retained the structure of the original questionnaire, and was easily writable and editable using a PC or a tablet or a smartphone. Face validity of the questionnaire was assumed, because it had already been used for studies in the same setting of University students in many different countries.

The target population of the study was university students in southern Italy enrolled in an undergraduate program. A self-selected sample of students was employed, and the participants were recruited by placing posters in the rooms in the student union on campus. Moreover, the students were invited to participate through the student union’s social network webpages. The instrument to collect data was an anonymous, self-reported online questionnaire in which the only requested socio-demographic characteristics were age and gender.

The study was conducted in the middle of the 2015-2016 academic year (around 75 days) to avoid seasonal holidays and stressful periods that are usually concentrated at the beginning of the semester (because of the novelty of the challenge of college) and the end of the semester (because of examinations). In this period, 642 students responded to the request to participate to the survey, of which 517 (80.53%) answered almost all required information in the online questionnaire.

In the final sample under analysis are included only the survey participants (517) who have completed the questionnaire, whereas 125 participants are excluded for the lack of sufficient information. Among them, 181 (35.01%) are men, 314 (60.74%) women, and 22 (4.26%) are subjects who reported an alternative gender identity (e.g. subjects who identified themselves as homosexual) (see Table 1). The respondents have an age ranged from 18 to 35, with a mean age of around 23 (SD = 3.09).

The multidimensional HPLP-II Scale was used to conduct the present study by considering “the theorized dimensions of a health-promoting lifestyle” [4] [9] . The original version of this instrument includes 52 items (Appendix B) categorized into six subscales [9] . The following dimensions were investigated:

・ Health Responsibility (HRESP) [9 items]: taking responsibility for one’s own well-being; attending to and educating oneself with regard to health; using informed consumerism when seeking professional care;

・ Physical Activity (PHACT) [8 items]: regular physical activity, which can be established with a plan or engaged in at random and included as part of one’s everyday life or recreational activities;

・ Nutrition (NUTRI) [9 items]: the choice of a healthy daily diet, consistent with the guidelines provided in the food pyramid;

・ Spiritual Growth (SPGRO) [9 items]: the development of one’s own resources through the ability to change (or grow) that seeks to maximize the potential of well-being by examining the meaning or sense of every target, and works toward achieving numerous goals in life;

・ Interpersonal Relationships (INTRE) [9 items]: communication skills that seek to share one’s thoughts and feelings through verbal and non-verbal messages to build and maintain functional and long-lasting interpersonal relationships;

・ Stress Management (STMAN) [8 items]: the control or reduction of tension; strategies to reduce perceived stress.

The items ask a respondent to indicate how often he/she adopts specific health-promoting behaviors or well-being habits. The items are scored on a 4-point Likert scale: “never” (1), “sometimes” (2), “often” (3), and “routinely” (4).

The original HPLPII [9] was assessed in 712 adults ranging from 18 to 29. Content validity was established by literature review and experts’ evaluation, construct validity was supported by factor analysis that confirmed a six-subscale structure and reliability was measured by α coefficient, which was 0.94 for the total scale and ranged from 0.79 to 0.87 for the subscales.

Minimum [Min], Maximum [Max], Standard Deviation [SD].

Taking into account the peculiarity of socio-cultural factors in each nation, some authors have reported interesting results on the use of this scale in university setting. In Spain, Pérez-Fortis et al. [21] found that a six-factor model of a 44-item scale accounted for 40% of the variance, and the scale had an internal consistency of 0.87. In Jordan, Alkhawaldeh [23] found that the total average score of the HPLP-II Scale for students was 2.4 (Standard Deviation [SD] = 0.4). The highest mean score was for the spiritual growth dimension (Mean = 3.0, SD = 0.7), and the lowest mean score was for the physical activity dimension (Mean = 2.0, SD. = 0.7). In Portugal, Sousa et al. [24] found an adequate fit to a 52-item scale with a six-factor structure in a cross-cultural study. They reported a global Cronbach’s α coefficient of 0.93. In Turkey, Pinar et al. [19] studied adults and found Cronbach’s α coefficients greater than 0.70 for the six dimensions. After performing tests on the latent construct measurement, the final model reduced the number of items from 52 to 48. In Taiwan, Meihan and Chung-Ngok [20] validated the HPLP-II Scale and obtained a total Cronbach’s α coefficient greater than 0.70 for the revised instrument; the values for the six dimensions ranged from 0.71 to 0.91. In Japan, Wei et al. [22] found a mean of 2.5 (SD = 0.29), the highest mean was associated with the interpersonal relationships dimension, and the lowest mean was associated with the health responsibility dimension.

The multidimensional HPLP-II Scale was evaluated for its reliability and validity. Internal consistency was considered to assess the reliability of the results across items within the adopted scale by using Cronbach’s α coefficient. For internal validity, a two-step procedure was performed to confirm the factor structure of the scale by means of exploratory factor analysis (EFA), and confirmatory factor analysis (CFA) [26] , as reported in other studies in different theoretical frameworks (e.g., [27] [28] [29] [30] ).

According to the procedure discussed in Bai et al. [31] and Bozkurt et al. [32] , the total sample of 517 participants was randomly divided into two sample groups (Sample 1 with n = 193, around 40% of the survey participants; Sample 2 with n = 324, around 60% of the survey participants) using the Statistical Package for the Social Sciences (SPSS) random case selection function. This procedure ensured that the EFA sample was not the same as the CFA sample, and then the sample findings could be generalized to the population. Item analysis and EFA were conducted on Sample 1. EFA analysis was performed using principal component analysis (PCA) with Varimax rotation implemented in the SPSS. For Sample 2, a first-order CFA model was used to explore the validity of the factor solution identified in the exploratory step; a second-order CFA model was specified to prove that the items were multidimensional in line with the suggestions presented in Walker et al. (1987) seminal paper. The CFA analyses were conducted with R software using the Lavaan package [33] for the model estimation and the semPlotpackage [34] for the graphical representation of the measurement models. To minimize the loss that occurs in list wise deletion, a pair wise deletion (available-case analysis) was adopted in the exploratory and confirmatory factor analyses, assuming that missing answers were completely random.

Starting from the information provided by the sample of 517 participants, the descriptive statistics for the each subscale and the total scale are presented in Table 2. The mean of the total HPLP-II Scale was 2.28, and the mean of its dimensions were 2.05 for HRESP, 1.97 for PHACT, 2.30 for NUTRI, 2.60 for SPGRO, 2.62 for INTRE, and 2.09 for STMAN. The normality of each item was checked by its skewness and kurtosis (Hair et al. 2010). The results showed that almost all of the items reported values within the +2 to −2 range for skewness and kurtosis; however, for two items [hplp_29 and hplp_41] in the STMAN subscale, the kurtosis value was higher than 3, indicating deviations from the normal distribution. Thus, those items were removed (in our opinion the determinants of this phenomenon are found in the student cultural variation, as we will discuss in the concluding section).

Internal consistency was assessed for the entire instrument as well as for each subscale. The HPLP-II Scale showed a high overall internal consistency: the Cronbach’s α coefficient for the total 52-item scale was 0.90 (see Table 3), which is near the original value of 0.92 for the English language version reported [9] .

Four of the six HPLP-II subscales, HRESP (0.79), PHACT (0.82), SPGRO (0.77), and INTRE (0.79) were associated with values around 0.80, whereas the NUTRI and STMAN subscales had lower values of 0.68 and 0.65, respectively. Corrected item-total correlation was calculated for the total scale and for the six subscales. Step-by-step, the items that decreased the scale’s reliability were deleted. A total of 26 items was deleted from the instrument on the basis of the item analysis findings. More specifically, the following items were not included in further analysis: hplp_09, hplp_21, hplp_33, hplp_45, and hplp_51 from the HRESP subscale; hplp_34 from the PHACT subscale; hplp_14, hplp_32, hplp_38, and hplp_50 from the NUTRI subscale; hplp_18, hplp_48, and hplp_52 from the SPGROW subscale; and hplp_07, hplp_37, hplp_43, and hplp_49 from the INTRE subscale. All the items related to the STMAN subscale were disregarded given their poor inter-item correlations and low α coefficient value. A description of the specific items is presented in Appendix B.

The remaining 26 items were first subjected to EFA. Specifically, PCA using a Varimax rotation with Kaiser Normalization was performed. The scree plot, with Kaiser’s “greater than one” criterion [35] was used to determine the number of factors to be retained. A first solution with six factors was obtained. Nevertheless, the examination of this factorial solution showed that the NUTRI subscale items were split into two factors. Hence, a five-factor solution was estimated by fixing the extracted factors to five. This approach is supported by previous results that explored the dimensionality of the HPLP-II Scale in which the items were included in a unique dimension. The results of the Kaiser-Meyer-Olkin (KMO = 0.80) test and Bartlett’s test of sphericity (χ² = 1777.53, p < 0.001) obtained from this latter solution demonstrate the adequacy of the sampling in which the criteria for conducting a factor analysis were met.

The final five-factor solution (see scree plot in Figure 1 and the items’ factor loadings and communality values in Table 3) shows a clear, interpretable structure accounting for 55.53% of the total variance (specifically, Factor 1: 22.18%; Factor 2: 12.55%; Factor 3: 8.46%; Factor 4: 6.98%; and Factor 5: 5.36%). The first factor captured the PHACT subscale, the second factor captured the SPGRO subscale, the third factor captured the HRESP subscale, the fourth factor captured the INTRE subscale, and the fifth factor captured the NUTRI subscale. The communalities (see Table 3) of most items were greater than or close to 0.50 (except for the item hplp_46 and for other items with values close to 0.40), which indicated that this factor solution adequately accounted for almost all variables [36] .

Standard Deviation [SD]. Health-Promoting Lifestyle Profile Scale II [HPLPII], Health Responsibility [HRESP], Physical Activity [PHACT], Nutrition [NUTRI], Spiritual Growth [SPGRO], Interpersonal Relationships [INTRE], Stress Management [STMAN].

Physical Activity [PHACT]; Spiritual Growth [SPGRO]; Health Responsibility [HRESP]; Interpersonal Relationships [INTRE]; Nutrition [NUTRI].

As a second step, CFA was conducted to confirm whether the emerging five-factor solution was consistent with the model suggested by the PCA in the exploratory analysis.

A first-order factor analysis was performed. All the factor loadings were statistically significant (p < 0.001), ranging from 0.41 (hplp_01) to 0.90 (hplp_10). Correlations between the five factors showed statistically significant coefficients (p < 0.001) for SPGRO and INTRE (0.67) and PHACT and NUTRI (0.51) (see the standardized solutions in the path diagram presented in Figure 2). The fit of model was assessed by considering the χ² test statistic, and some goodness-of-fit [37] [38] [39] : the comparative fit index (CFI), the Tucker?Lewis index (TLI), the root mean square error of approximation (RMSEA), and the standardized root mean square residual (SRMR). In line with the threshold values indicated in most extant structural equation modeling literature [40] [41] , the results suggested that the five-factor solution with the 26-item HPLP-II Scale provided an acceptable fit of model to the data with CFI and TLI values around 0.90, and RMSEA and SRMR values ranging between 0.05 and 0.08. Note that by deleting two items (hplp_16 and hplp_46), from the PHACT subscale that had a factor loading value less than 0.6, the fitness indices of the new measurement model slightly improved.

In addition, a second-order CFA model was estimated in line with the conceptualization advocated in Walker et al. [8] to test the hypothesis that a higher order factor (HPLP-II) represents the pattern of relations among the first-order factor model [42] . In this way, higher-order CFA enables one to test the multidimensionality of the first-order factor solution in the presence of the total score combining all of the items, regardless of the five domains. A diagrammatic representation of the standardized solution of the estimated model is presented in Figure 3; as seen, the covariation among the five first-order factors is explained by their regression coefficients on the second-order factor model. Although all the path coefficients of the HPLP-II Scale secondary factors in relation to the five factors were statistically significant, the values for three of the subscales were around 0.30 or lower (0.25 PHACT, 0.32 NUTRI, and 0.34 HRESP). The values for the path coefficients of INTRE and SPGRO were higher (0.78 and 0.82, respectively). The value of the goodness-of-fit indices of this model had a lower fit to the data than the first-order factor model (see Figure 2 and Figure 3). To compare the two models, the chi-square difference test was used to determine which model was better (Δχ² = 649.09 - 585.34 = 63.75, Δdf = 5, p <0.001). The null hypothesis (the two models are identical), was rejected: the first-order factor model is better than the second-order factor model. Hence, it seems that the higher-order CFA did not support the unidimensionality of the data, and the five dimensions solution of the first-order CFA was confirmed.

Finally, the internal consistencies (Cronbach’s α) and composite reliability (CR, ω coefficients) for the HPLP-II Scale and its subscales in the first-order factor model and the second-order factor model at Level 1 and Level 2 are listed in Table 4. Both criteria show good reliability for the measurement models with Cronbach’s α values greater than 0.70 (except for the INTRE subscale) and composite reliability values greater than 0.06. The average variance extracted (AVE), which is the average percentage of variation explained by the measured items for a latent construct, shows that the values were lower than 0.50 for some of the subscales.