Considering the frequent occurrence of disasters due to natural or human causes, particularly attention must be given to rescue workers dealing with critical incidents or disasters, since chronic emotional demands and coping with stressful situations harms their health and the services provided by their institution (Queirós et al., 2013; Rabjohn, 2013; International Labour Organization, 2016). Therefore, these professionals can develop negative emotional states, psychological disorders or mental diseases (e.g., anxiety, loneliness, panic attacks, depression, post-traumatic stress disorder, vicarious trauma), but particularly, they can experience chronic job stress and become vulnerable to burnout (Cieslak et al., 2014; Katsavouni et al., 2016; Krok, 2016). Several studies have emphasized the relationships between burnout and work engagement (Schaufeli et al., 2002a; Maslach and Leiter, 2008; Leon et al., 2015; Leiter and Maslach, 2017; Schaufeli and De Witte, 2017), suggesting that promoting work engagement can prevent burnout (Argentero and Setti, 2011), thus ameliorating rescue workers' mental health and helping their stress management (Nishi et al., 2016; O'Neill and Rothbard, 2017). In fact, work engagement seems to be an important psychological resource for rescue workers, since it may protect them from the risk of developing negative health effects (Setti and Argentero, 2014), depression, sleep disturbances, relational conflicts, burnout, compassion fatigue or vicarious trauma (Lenthall et al., 2009; Kumar, 2011; Ray et al., 2013), or presenteeism (Burton et al., 2017).

For several decades, Psychology has focused on pathology and psychological illness, which has led to a restricted view of human behavior that has consequently contributed to the depreciation of resources, qualities and skills of human beings as promoters of individual growth and psychological well-being (Diener et al., 1999). However, the recent emergence of Positive Psychology has counteracted this focus, valorizing the study of the positive part of human experience (Luthans, 2002), aiming to increase the quality of life, to promote individual qualities, resources, and characteristics that increase psychological well-being and growth (Seligman and Csikszentmihalyi, 2000). This approach within Positive Psychology explores building positive qualities, instead of focusing on repairing psychological illness. One of the theoretical concepts that arises within this perspective is work engagement (Schaufeli et al., 2002a), focused on the positive aspects of behaviors, the well-being and the optimal functioning of each person (Schaufeli and Salanova, 2007).

Although work engagement is a relatively new construct, it has been viewed as being comprised of slightly different aspects, such as physical, cognitive, emotional and mental expression by employees within the exercise of their roles (Kahn, 1990). Distinctions between organizational and job engagement have appeared regarding employees' different roles suggesting that engaged workers invest and dedicate themselves deeply into their tasks, which are viewed as motivated situations (Saks, 2006). Other authors (Maslach and Leiter, 1997, 2017; Maslach et al., 2001) have considered work engagement as a concept diametrically opposed to burnout, being defined as an energetic state of involvement in vital personal activities, but quite different from job satisfaction or organizational commitment (Hallberg and Schaufeli, 2006).

While the discussion about burnout and work engagement as different constructs or as opposite poles of the same construct continues (Schaufeli et al., 2002b; Macey et al., 2009; Cole et al., 2012; Leon et al., 2015; Leiter and Maslach, 2017; Schaufeli and De Witte, 2017), we adopt Schaufeli and colleagues' perspective on work engagement as a positive motivational, cognitive-affective, and emotional persistent state (Schaufeli and Salanova, 2007; Schaufeli, 2017b), divided into three factors or dimensions: vigor, dedication and absorption. Vigor is related to persistence, energy and mental resilience during work, while dedication implies being involved in the work and living with enthusiasm, inspiration, and pride, and finally, absorption means to be concentrated and immersed in one's work, with the perception that time flies (Schaufeli et al., 2002b). Thus, workers with high levels of work engagement see themselves as full of energy when facing their professional activities (Albrecht, 2010).

Work engagement is also related with other important occupational health and organizational outcomes, including in-role and extra-role behavior (Bakker et al., 2004). It is positively correlated with performance (Cooper and Quick, 2017; Reijseger et al., 2017); more specifically, a direct and positive relation has been found with job performance and a direct and negative relation with turnover intention (Karatepe and Avci, 2017; Kim, 2017), proactive behaviors and employee intrapreneurship (Gawke et al., 2017), job crafting and personal resources (Xanthopoulou et al., 2009; Vogt et al., 2016), general job resources (Schaufeli et al., 2009), structural and social resources (Harju et al., 2016), and job, career and life satisfaction (Karatepe and Karadas, 2015).

In line with its three-dimensional definition, the work engagement measure was divided into three sub-scales, in a psychometric instrument called Utrecht Work Engagement Scale (UWES), a self-report instrument that includes the vigor, dedication and absorption factors (with six, five and six items respectively). Although the UWES presents good psychometric properties. it is common to find very strong correlations between the factors (Chughtai and Buckley, 2013; Agarwal, 2014), leading some studies to prefer the single factor measure (Shimazu et al., 2008; Seppälä et al., 2009; Klassen et al., 2012), while others (Shimazu et al., 2008; Alok, 2013) tested UWES as a second-order construct (Edwards, 2001). Kulikowski (2017) conducted a literature review with 21 studies using Confirmatory Factor Analysis (CFA) of UWES; six of these recognized the three-factor version as superior to the one-factor solution, six studies found the opposite, eight studies found both structures as equivalent, and one study failed to confirm both structures. However, usually, the CFAs present good results with the three-factor structure. Moreover, various studies found invariance between different occupational and racial groups and countries (Storm and Rothmann, 2003; Xanthopoulou et al., 2012), and relative longitudinal invariance for the short version (Seppälä et al., 2009). This shorter version with nine items is divided into the same three dimensions, with three items per dimension (Schaufeli et al., 2006).

The UWES has been used in several studies with rescue workers and translated into many languages (Schaufeli and Bakker, 2003). Several studies have shown that the structure of the UWES remains unchanged among different occupational groups. Table 1 presents a synthesis of 30 studies between 2003 and 2017, conducted mainly in Europe but also in America, Asia, Africa and Oceania, with professionals with occupation statuses similar to the ones of our study, revealing that work engagement levels are moderate to high among these rescue workers.

Considering previous studies with UWES among these professionals, our research question was to investigate if UWES is an adequate instrument to assess rescue workers' engagement. Thus, this study aims to estimate work engagement levels of rescue workers (namely comparing nurses, firefighters, and police officers) and to assess the validity evidence related with the internal structure of the Portuguese versions of the UWES-17 and UWES-9, namely to evaluate their dimensionality (comparing the fit of the three-factor first-order model of the UWES-17 and UWES-9, and of a possible UWES-9 second-order model), measurement invariance between the three occupational groups, and reliability of the scores, according to the Standards for Educational and Psychological Testing framework (American Educational Research Association et al., 2014). According to these aims, we have as hypotheses: (H1) UWES-17 presents validity evidence that supports its usage among rescue workers; (H2) UWES-9 presents validity evidence that supports its usage among rescue workers; (H3) UWES-9 presents measurement invariance that allows for comparing different occupational groups; (H4) UWES-9 reveals different work engagement levels between the three different occupational groups of rescue workers.

A sample of 3,887 Portuguese rescue workers completed the UWES-17 (except 428 workers that only completed the UWES-9), with data collected from several research projects (seven studies) developed at the Psychosocial Rehabilitation Laboratory from FPCEUP, between 2010 and 2016. Data were from different Portuguese districts, but with all occupations contributing to the data during the same year. The average age was 33.35 years old (SD = 9.84), 50% male, 44% married (and 50% single), 62% at least had a college degree. Regarding occupations, 50% were nurses, 39% voluntary firefighters (according to the Portuguese firefighters' organization, they can be voluntary or professional firefighters; this study collected data from voluntary firefighters, a larger group than professional firefighters) and 11% police officers.

A sociodemographic questionnaire was administered, with items such as age, sex, marital status, school level and occupation. Despite other questions specific to each rescuer occupation, these questions were asked to all groups. To assess work engagement, we used the UWES (Table 2) developed by Schaufeli et al. (2002b) in its Portuguese version (Marques-Pinto and Picado, 2011).

Participants were invited to voluntarily participate after formal authorization from each organization where they were working at and filled out printed versions of the UWES anonymously. The study was approved by the Ethics Committee of University of Porto, Portugal, and written informed consent from all research participants was obtained.

To study the dimensionality of the UWES versions, a CFA was conducted to verify if the three-factor structure proposed by the UWES' authors presented an adequate fit for this study's sample. We used as the goodness-of-fit indices the TLI (Tucker Lewis Index), NFI (Normed Fit Index), χ^2/df (ratio chi-square and degrees of freedom), CFI (comparative fit index) and the RMSEA (root mean square error of approximation). The fit of the model was considered good for χ^2/df smaller than 5, CFI, NFI and TLI values above 0.95 and RMSEA values below 0.08 (Hoyle, 1995; Boomsma, 2000; McDonald and Ho, 2002; Marôco, 2014; Byrne, 2016).

To analyze convergent validity evidence, the average variance extracted (AVE) was estimated as described in Fornell and Larcker (1981). Values of AVE ≥ 0.5 were considered indicative of the constructs' convergent validity evidence, for both versions of the UWES per the proposal by Hair et al. (2009).

Discriminant validity indicates that the items that represent a dimension are not strongly correlated with other dimensions (Fornell and Larcker, 1981; Marôco, 2014). For two factors x and y, if AVEx and AVEy ≥ ρ² xy (squared correlation between the factors x and y) there is evidence of discriminant validity (Marôco, 2014).

The reliability of the UWES scores was investigated using internal consistency estimates, Composite Reliability (CR), the ordinal Cronbach's alpha coefficient (α) and the ordinal omega coefficient (ω) for each factor and for the total scale. The second-order factor reliability was also calculated with the omega coefficient (semTools Contributors, 2016).

The measurement invariance corresponds to the situation where links between the items and the construct studied are invariant for different contexts, such as countries, religions, times, or languages (Gregorich, 2006; Little, 2013; Brown, 2015; Hayduk, 2016). This should be tested on the construction or adaptation plan of any psychometric instrument (Edwards et al., 2017). This analysis will show if in different measurement conditions we are measuring the same attributes. Without evidence for this, the basis for drawing inferences will be severely compromised, and, as a result, the findings of possible differences between samples of different groups cannot be unambiguously interpreted (Horn and McArdle, 1992).

Establishing measurement invariance consists of the estimation of an increasingly constrained set of structural equation models, and then evaluating if the differences between those models are significant (van de Schoot et al., 2012). To detect whether the same three-factor model holds in each group of professionals, a group of nested models with indications of equivalence is needed (Marôco, 2014). For the three-factor first-order version of the UWES-9, the first model to be tested is one where only the factor loadings are equal across groups, but the thresholds are allowed to differ between groups; this is called metric invariance and assesses if the professionals in the different occupational groups attribute the same meaning to the three work engagement factors. Next, to assess scalar invariance, a model in which the thresholds and loadings are forced to be the same is tested; it indicates that the three work engagement latent variables (in other words, the loadings), and that the levels of the thresholds are equal for the three different kinds of professionals (van de Schoot et al., 2012). As a result, the occupational groups can be compared on their scores for the vigor, absorption and dedications variables. The next step tests a model in which the residual variances are also forced to be equal across the three groups of professionals, achieving the so-called full uniqueness Measurement Invariance; if obtained, the explained variance for all items is the same for the occupational groups. This means the three factors are measured identically across professional groups. Finally, a model in which the latent means are the same for the different groups is tested. This approach is based on the recommendations of Millsap and Yun-Tein (2004), who specifically discuss the measurement invariance for ordered-categorical measures using the theta-parameterization. Most researchers still use multi-group confirmatory maximum likelihood factor analysis of a Person covariance matrix irrespective of the nature of the measures (Koh and Zumbo, 2008). The approach used in this study tries to be more precise and adequate given the kind of measures adopted (Hirschfeld and von Brachel, 2014).

For the second-order model, a different approach (with more models) is needed. A group of seven models were tested, based on the recommendations of Chen et al. (2005) to test the invariance in second-order models, and in the recommendations of Millsap and Yun-Tein (2004) due to the ordered-categorical nature of our measures. The following models with increasing constraints were compared: (a) configural invariance, (b) first-order factor loadings, (c) second-order factor loadings, (d) thresholds of measured variables, (e) intercepts of first-order factors, (f) disturbances of first-order factors, and (g) residual variances of observed variables. Invariance is supported if the ΔCFI < 0.01 between constrained and free models is presented (Cheung and Rensvold, 2002) and if Δχ² test comparing the fit of the constrained vs. free models is not statistically significant (Satorra and Bentler, 2001).

All statistical analyses were performed with R (R Core Team, 2017) and Rstudio (RStudio Team, 2017). The descriptive statistics were obtained with the skimr package (Arino de la Rubia et al., 2017). The CFA analysis was conducted with the lavaan package (Rosseel, 2012) using the Weighted Least Squares Means and Variances (WLSMV) estimation method. The reliability estimates and measurement invariance (for the first-order model) were calculated with the semTools package (semTools Contributors, 2016), the measurement invariance for the second-order model was analyzed with lavaan (Rosseel, 2012), the Mardia Kurtosis (Mardia, 1970) was assessed with the psych package (Revelle, 2017), and the doBy package (Højsgaard and Halekoh, 2016) for the percentiles. The comparisons of the raw levels of the factors (estimated as the mean of the items for each factor) between the occupational groups were tested using one-way ANOVA with pairwise comparisons using the stats package (R Core Team, 2017). The effect sizes were calculated using Partial Eta Squared from the lsr package (Navarro, 2015).

Considering rescue workers' occupations, Table 8 presents percentiles and mean levels of each work engagement dimension, and a comparison between nurses, firefighters, and police officers. For the hypothesized differences between occupational groups, some statistical significant differences in work engagement were found between the three groups (Table 9), being highest among firefighters and lowest among nurses. Percentile analyses reveal the exact value for each category and occupation, reflecting the tendency revealed by the mean analysis of each sample.

Considering the psychometric properties of data gathered with the UWES in the present sample, our hypotheses were confirmed since UWES-17 and UWES-9 presented validity evidence that supports its usage among rescue workers (respectively H1 and H2), and UWES-9 presented measurement invariance that allows for comparing occupational groups (H3). In fact, the data presented better psychometric properties for the UWES-9 and supported a second-order latent factor for the three-factor structure due to strong correlations between the first-order dimensions. This solution is supported by other international studies, presenting better goodness-of-fit indexes for the shorter version of the instrument (Nerstad et al., 2010; Hernandez-Vargas et al., 2016). The convergent validity evidence and reliability of the scores were good, although the discriminant validity was not acceptable, leading to the proposal of a second-order latent factor. There have been several reports that discuss the dimensionality of work engagement as assessed by UWES (see Table 1). Our data suggests that work engagement can be defined as a second-order factor that is reflected in the workers' vigor, dedication, and absorption. Thus, we suggest that work engagement can be estimated as a single value in research contexts and in clinical settings facilitating the use of the UWES. A second-order latent factor has advantages over a single first-order factor for the UWES as proposed by several researchers (see Table 1) since it allows for the estimation of a single value for work engagement while retaining the three first-order factors proposed (Schaufeli and Bakker, 2003).

Regarding H4 (UWES-9 can evaluate different occupational groups of rescue workers), and considering rescue workers' occupation specificities, work engagement levels varied; despite being moderate to high, they were higher among firefighters and lower among nurses. This can be explained by the fact that Portuguese firefighters are traditionally volunteers (paid or not), being highly engaged in their tasks and frequently assuming physical risks to help in rescue situations (e.g., the tragedy of the Portuguese forest fire during June 2017 at Pedrógão Grande where a firefighter died, and several were severely burned; or the Portuguese forest fire during the summer of 2013 at Caramulo where eight firefighters died). According to Portuguese data on firefighters, in 2014 there were 42,000 firefighters, only 6,000 of whom were professionals and the remaining were volunteers, despite all of them being trained by the Portuguese Firefighters' National School. Moreover, some studies explain that aid workers, such as firefighters, can present a sense of coherence, a concept developed by Antonovsky (1987). Related to the salutogenic theory, it is defined as the “ability of the persons to understand what happens around them, to what extent they were able to manage the situation on their own or through significant others in their social network, and the ability to find meaning in the situation” (Eriksson and Lindström, 2005, p. 460). This can help someone to better cope with traumatic events life adversities (Veronese and Pepe, 2014; Veronese et al., 2016). The study of 464 firefighters by Dudek and Koniarek (2000) revealed that post-traumatic stress symptoms were associated with a low level of sense of coherence, especially if they already have post-traumatic stress disorder. Additionally, firefighters, like police officers, share some personality traits like sensation seeking (Perrott and Blenkarn, 2015), that increase their motivation to pursue certain occupations. Nurses presented the lowest work engagement values, maybe due to the external demands their occupation is facing, namely lack of human or material resources, conflicts with patients or being aggressed by them or their families, etc. (Lin et al., 2016; Marôco et al., 2016). Indeed, recent austerity measures adopted in countries such as Portugal have had significant negative effects on the availability of healthcare system resources, including reduced staffing and employment conditions (Dussault and Buchan, 2014). Regarding the work engagement dimensions, it is important to notice that they are differentiated according to each rescue occupation, since vigor is related to physical energy, used more by firefighters and police officers, while dedication is clearly higher among firefighters who voluntarily seek this activity. Finally, absorption must be interpreted according to each task, which was higher among firefighters and police officers, especially when they are in rescue operations where time “flies,” while nurses have more rigid protocols to obey, accomplishing schedules and some rigid tasks. Results are similar to other studies where rescue workers present moderate to high work engagement levels (see Table 1), with nurses having lower work engagement than police officers (Schaufeli et al., 2006), and a lack of studies about firefighter's work engagement (curiously the one found was with a Portuguese sample).

This study has some limitations, namely having a convenience sample collected using participants' voluntary participation. In the future, respecting voluntary participation, we would like to attract more participants, trying to have equal representation among the occupations. In fact, in this sample the percentages were not homogeneous, with nurses representing half of the sample, and police officers comprising only 11% of the sample. Firefighters were also included, but not other rescuers such as medical emergency staff, civil protection technicians, social workers or psychologists, who, for example, during the summer of 2017 in Portugal, participated in rescue tasks for the population due to the tragic forest fires. Furthermore, we did not have any source of evidence-based on relations to other variables, which can provide external validity evidence of UWES when compared with other concurrent or divergent measures (American Educational Research Association et al., 2014). This was a consequence of collecting data from each occupational group separately or during a different year, according to different research projects developed in the same University research unit, which were addressing different topics, having in common the measurement of work engagement with UWES. In the future, we would like to increase the sample, with replication efforts to collect data among more diverse rescue occupations. Also, we intend to have more homogeneity between rescue workers' samples sizes, having thus a stronger national representativeness, which can be obtained using web-based versions (Ilieva et al., 2002) of UWES (with other measures) through, for example, LimeSurvey (Limesurvey GmbH, 2017). The advantages of web-based surveys can also include increased delivery speed, control and monitoring, automatic data validation, and decreased data entry error (Falletta and Combs, 2002). Finally, we would like to combine work engagement measurement with other variables such as burnout, trauma or job satisfaction, as frequently studied in numerous research (e.g., Argentero and Setti, 2011; Ângelo and Chambel, 2015; Espinoza-Parra et al., 2015).

The study has some theoretical implications, enabling a better understanding of work engagement dimensions related with each occupational group specificities (e.g., vigor is higher among firefighters and police officers who use more physical energy). Additionally, the recent Schaufeli (2017a) publication of the general engagement scale (UGES), with 9 items evaluating daily activities and not only work/job activities will broad the evaluation of engagement and its relationship with psychological variables such as positive and negative affect, satisfaction with life, personality traits (Schaufeli, 2017b), or sense of coherence and trauma (Dudek and Koniarek, 2000; Veronese and Pepe, 2017). Thus, work engagement seems to be an important concept to study among workers and organizations, considering that organizations are facing more competition, innovation, and constant changes, and also a decrease of resources (Graffigna, 2017). Engaged workers are thus an important element for organizations' productivity or to the care/services they provide to citizens.

Regarding practical implications, this validation of a Portuguese version of UWES can contribute to the improvement of rescue workers' quality of life, since these results can alert individuals and institutions about the need to explore work engagement level to buffer against negative physiological health effects (Lenthall et al., 2009; Kumar, 2011; Ray et al., 2013; Setti and Argentero, 2014; Burton et al., 2017). Moreover, reference values for the Portuguese firefighters, nurses and police officers' populations make it possible to establish comparisons with other occupational groups of rescue workers (e.g., physicians, social workers). Taken together, the findings of the present study have important ecological implications for Portuguese policymakers and human resource managers who can structure rescue workers' work conditions, or develop intervention programs to promote their work engagement, and therefore help prevent burnout and other mental health conditions that affect these professionals' well-being.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.