(PDF) Using partial least squares struct

The current issue and full text archive of this journal is available on Emerald Insight at: www.emeraldinsight.com/0959-6119.htm IJCHM 30,11 Using partial least squares structural equation modeling in hospitality and tourism 3462 Do researchers follow practical guidelines? Received 20 November 2017 Ahmet Usakli Revised 17 March 2018 10 May 2018 Faculty of Tourism, Gazi University, Ankara, Turkey, and Accepted 11 May 2018 Kemal Gurkan Kucukergin School of Business Administration, Atılım University, Ankara, Turkey Abstract Purpose – The purpose of this study is to review the use of partial least squares-structural equation modeling (PLS-SEM) in the ﬁeld of hospitality and tourism and thereby to assess whether the PLS-SEM-based papers followed the recommended application guidelines and to investigate whether a comparison of journal types (hospitality vs tourism) and journal qualities (top-tier vs other leading) reveal signiﬁcant differences in PLS-SEM use. Design/methodology/approach – A total of 206 PLS-SEM based papers published between 2000 and April 2017 in the 19 SSCI-indexed hospitality and tourism journals were critically analyzed using a wide range of guidelines for the following aspects of PLS-SEM: the rationale of using the method, the data characteristics, the model characteristics, the model assessment and reporting the technical issues. Findings – The results reveal that some aspects of PLS-SEM are correctly applied by researchers, but there are still some misapplications, especially regarding data characteristics, formative measurement model evaluation and structural model assessment. Furthermore, few signiﬁcant differences were found on the use of PLS-SEM between the two ﬁelds (hospitality and tourism) and between the journal tiers (top-tier and other leading). Practical implications – To enhance the quality of research in hospitality and tourism, the present study provides recommendations for improving the future use of PLS-SEM. Originality/value – The present study ﬁlls a sizeable gap in hospitality and tourism literature and extends the previous assessments on the use of PLS-SEM by providing a wider perspective on the issue (i.e. includes both hospitality and tourism journals rather than the previous reviews that focus on either tourism or hospitality), using a larger sample size of 206 empirical studies, investigating the issue over a longer time period (from 2000 to April, 2017, including the in-press articles), extending the scope of criteria (guidelines) used in the review and comparing the PLS-SEM use between the two allied ﬁelds (hospitality and tourism) and between the journal tiers (top-tier and other leading). Keywords Partial least squares, Path modeling, Structural equation modeling Paper type Research paper Introduction International Journal of Empirical studies that utilize covariance-based structural equation modeling (CB-SEM) is Contemporary Hospitality Management widespread in hospitality and tourism. However, in recent years, the composite-based Vol. 30 No. 11, 2018 pp. 3462-3512 partial least squares-SEM (PLS-SEM) has become increasingly popular in social sciences © Emerald Publishing Limited because of its ability to handle non-normal data distributions, small sample sizes, complex 0959-6119 DOI 10.1108/IJCHM-11-2017-0753 models with many indicators and model relationships including both reﬂectively and formatively measured constructs (Hair et al., 2017a). Because of its several advantages, Hair Hospitality and et al. (2011) argue that PLS-SEM can be a silver bullet in many research situations if it is tourism correctly applied. Hospitality and tourism researchers are also taking advantages of lu et al., PLS-SEM and using it in their studies (e.g., Ali et al., 2016; Amin et al., 2017; Dedeog 2016; Fakih et al., 2016; Rasoolimanesh et al., 2017a). CB-SEM and PLS-SEM share the same roots (Jöreskog and Wold, 1982), but they have distinct goals and requirements and thereby are different in their statistical methods (Richter et al., 2016a). Thus, researchers should understand the principles of PLS-SEM, 3463 apply the method correctly and report the ﬁndings properly. Although reviews and assessments of PLS-SEM usage have been frequently carried out in various disciplines, such as accounting (Lee et al., 2011; Nitz, 2016), human resources management (Ringle et al., 2018), international business research (Richter et al., 2016a), information systems (Ringle et al., 2012; Hair et al., 2017b), operations management (Peng and Lai, 2012), strategic management (Hair et al., 2012a) and marketing (Hair et al., 2012b), little attention has been paid for the corresponding assessments of the PLS-SEM in hospitality and tourism. In other words, the use of PLS-SEM in hospitality and tourism research remains largely unknown. To date, there have been only three reviews on the use of PLS-SEM in hospitality or tourism, all having different limitations: a case-study based evaluation of PLS-SEM use by Assaker et al. (2012), an exploratory study in tourism by do Valle and Assaker (2016) and a more recent assessment in hospitality by Ali et al., (2018a). More speciﬁcally, Assaker et al. (2012) are the ﬁrst who evaluated only four studies (selected from one marketing journal and three tourism journals) using a limited number of criteria, which can be considered a case-study based assessment. On the other hand, in their exploratory study, do Valle and Assaker (2016) reviewed a total of 44 studies published in 11 selected tourism journals between 2000 and 2014. Results revealed that the use of PLS-SEM in tourism research was at an early stage of development and problematic issues existed in its usage. However, the authors highlight the small number of articles (44) reviewed in their study as a major limitation as larger number of articles were examined in other disciplines. More recently, Ali et al., (2018a) reviewed a total of 29 PLS-SEM-based papers published in six hospitality journals between 2001 and 2015. It is important to note that many disciplines regularly review the key methods used to ensure rigorous research and publication practices (Hair et al., 2012a, 2012b). Therefore, an updated and expanded assessment of PLS-SEM use in hospitality and tourism seems timely and warranted. Indeed, do Valle and Assaker (2016) concluded that further research, with a larger sample size, was needed for an effective assessment of the use of PLS-SEM in tourism research. In response to this call for a more comprehensive review, the present study aims to ﬁll a gap in the literature and extends the previous reviews on PLS-SEM in several ways. First, this study uses a larger sample size (206 articles published in 19 hospitality and tourism journals) to achieve greater generalizability. Second, to identify all empirical studies that apply PLS-SEM, no speciﬁc dates were chosen to set ﬁnite boundaries; instead, each journal was reviewed beginning from its ﬁrst publication date to the end of April 2017 (including the in-press articles). Thus, it was determined to investigate the issue over a longer time period. Third, for a more comprehensive review, the present study extends the scope of application guidelines (i.e. measures) used by previous research. Finally, previous studies investigated the PLS-SEM use in either tourism or hospitality, indicating that no study has yet investigated the issue from a wider perspective (by combining both hospitality and tourism research) or made a comparative analysis of hospitality and tourism to explore similarities or differences in PLS-SEM use. For this reason, the present study includes all PLS-SEM studies published in both hospitality and tourism journals. IJCHM In brief, the purpose of this study is to systematically review the use of PLS-SEM in the 30,11 ﬁeld of hospitality and tourism and to assess whether the PLS-SEM based papers followed the recommended application guidelines. In other words, the present study identiﬁes whether hospitality and tourism researchers have validly applied the PLS-SEM to their studies. Additionally, the use of PLS-SEM between the two allied ﬁelds (hospitality vs tourism) and between journal rankings (top tier vs other leading) were compared within the 3464 context of the study. Thus, the present study investigates whether “journal type” and “journal quality” have an impact on application and reporting practices or not. Choosing between PLS-SEM and CB-SEM Although the CB-SEM and the composite-based PLS-SEM constitute two types of SEM, the former (i.e. CB-SEM) is undoubtedly the most well-known one. Even the terms SEM and CB- SEM are sometimes incorrectly used synonymously by many social science researchers (Vilares et al., 2010). This is perhaps because of the wide usage of CB-SEM in social sciences. For instance, Ali et al. (2018b) revealed that 93 per cent of all SEM-based papers published in 15 hospitality and tourism journals between 2011 and 2014 used CB-SEM. Moreover, the authors found that most studies applying CB-SEM did not provide justiﬁcations for selecting CB-SEM over PLS-SEM (Ali et al., 2018b). However, each method is appropriate for a different research context, and therefore, researchers should understand the differences between the two methods and clearly emphasize the rationale for the use of one method over the other. There are primarily two criteria that should be considered when choosing between PLS- SEM and CB-SEM, namely, the philosophy of measurement and the aim of the analysis (Hair et al., 2018). First, PLS-SEM and CB-SEM follow different measurement philosophies, indicating that each method treats the latent variables in a different way. While CB-SEM is a factor-based method for CB-SEM, PLS-SEM is a composite-based method for variance-based SEM. More speciﬁcally, CB-SEM uses the common variance of the indicators and therefore considers the latent variables as common factors that explain the covariation between their associated indicators (Sarstedt et al., 2016a). In contrast, PLS-SEM uses the total variance of the indicators to generate linear combinations of indicators, which constitute a composite model approach to SEM (Hair et al., 2017a). Accordingly, when researchers decide to use composites as proxies of latent variables (Rigdon, 2012, 2014; Rigdon et al., 2017; Sarstedt, et al., 2016a), PLS-SEM becomes the method of choice (Hair et al., 2017c). Second, the aim of the analysis should be considered when choosing between PLS-SEM and CB-SEM. In fact, CB-SEM aims to minimize the differences between the estimated and sample covariance matrices, whereas the statistical objective of PLS-SEM is to maximize the explained variance of the endogenous constructs (Hair et al., 2011). Therefore, PLS-SEM is the preferred method when the aim of the analysis is to predict, rather than to conﬁrm (as in CB- SEM) (Hair et al., 2018), even though theory conﬁrmation has also relevance in PLS-SEM (Sarstedt et al., 2017). It is important to note that PLS-SEM has several advantages which have been frequently discussed in the literature, such as working efﬁciently with small sample sizes (Chin, 1998; Reinartz et al., 2009), not requiring normally distributed data (Boßow-Thies and Albers, 2010; Hair et al., 2011), handling complex models with many constructs and many indicators (Chin and Newsted, 1999; Sarstedt et al., 2017), easily incorporating formatively measured constructs (Hair et al., 2017a; Sarstedt et al., 2014) and being more appropriate in a situation of weak theory (Wold, 1985; Henseler et al., 2014). However, formatively measured constructs can also be used with CB-SEM, even though it requires certain model adjustments (Bollen and Davis, 2009; Diamantopoulos and Rieﬂer, 2011). Similarly, CB-SEM may also work with small sample sizes (Lei and Wu, 2012) or its results may be robust Hospitality and against non-normality in the data (Chou et al., 1991; Olsson et al., 2000). Accordingly, tourism Rigdon (2016) discusses such advantages of PLS-SEM (i.e. low sample size, non-normal data, formative measures and exploratory research) as false arguments in favor of the use of PLS- SEM, indicating that justifying the use of PLS-SEM based solely on such arguments is not sufﬁcient. Instead, researchers should provide more profound justiﬁcations when choosing PLS-SEM (Sarstedt et al., 2017), such as the aim of their analysis. Apart from these previously discussed advantages, recent research provides additional reasons for using 3465 PLS-SEM, such as applying PLS-SEM latent variable scores in subsequent analyses (Hair et al., 2017a), using secondary/archival data (Richter et al., 2016b) and mimicking the results of CB-SEM by using consistent PLS approaches (Sarstedt et al., 2017). Recently, some researchers have criticized the use of PLS-SEM with the arguments that it is not truly a latent variable method (Rönkkö et al., 2015), produces inconsistent and biased estimates (Rönkkö and Evermann, 2013), lacks global ﬁt statistics (Rönkkö et al., 2016) and fails to address the measurement error (Antonakis et al., 2010). However, such criticisms represent the misconceptions about PLS-SEM (Henseler et al., 2014; Rigdon, 2016), which ignore the different measurement philosophy that the method relies on and its prediction orientation (Ali et al., 2018a). As a result, Rigdon (2016) considers these criticism as ﬂawed arguments against the use of PLS-SEM. Methodology This review aims at investigating the journal articles that used PLS in the ﬁeld of hospitality and tourism. However, to limit the journal coverage, it was decided to review journals indexed by Web of Science (Social Sciences Citation Index), the research platform maintained by Clarivate Analytics. The rationale behind selecting journals indexed by Web of Science is simply that they include the leading journals in hospitality and tourism. Therefore, Hospitality, Leisure, Sport and Tourism category of the 2016 release of the Journal Citation Reports® was used to obtain a full list of the journals. Of the 44 journals in that category, 22 of them were sport and 3 of them were leisure journals. These sport and leisure journals were excluded. Thus, a total of 19 hospitality and tourism journals (please refer to Table I for full list) were reviewed beginning from its ﬁrst publication date to the end of April, 2017 (including the in-press articles). Keywords used included “PLS” and “partial least squares.” To identify eligible studies for the review, the papers were then examined independently by two researchers proﬁcient in the use of PLS-SEM. In this process, conceptual papers discussing various statistical techniques along with PLS (e.g., Morley, 2012) and a review paper on PLS-SEM (do Valle and Assaker, 2016) were not considered. This resulted in a total of 213 empirical studies that utilize PLS. Two research notes (Marques and Reis, 2015; Oromendía et al., 2015) and a letter (Lopez-Bonilla and Lopez-Bonilla, 2012) were removed as it was unable to make an accurate assessment due to their page limits when compared with full-text articles. Additionally, three studies applying PLS regression (Sui and Baloglu, 2003; Yang et al., 2013; Whitﬁeld et al., 2014) and a paper that uses PLS-SEM only for testing the moderation but not for the measurement and structural models (Lai, 2015) were removed from the sample. Following these criteria, a total of 206 empirical papers were identiﬁed that utilize PLS-SEM (see Table I). The 206 PLS-SEM-based papers were reviewed and assessed by using the compiled application guidelines outlined in previous research (Hair et al., 2012a, 2012b, 2017a, 2017b; Richter et al., 2016a; Sarstedt et al., 2014, 2017). These guidelines include ﬁve key dimensions, namely, reasons for using PLS-SEM, data characteristics, model characteristics, model evaluation and reporting technical aspects (please refer to Table II for full list). IJCHM No. of PLS- 30,11 Journals SEM papers Studies 1 Annals of Tourism 3 Song et al. (2012), Barnes et al. (2014); Kock et al. (2016) Research 2 Asia Paciﬁc Journal of 13 Huang et al. (2009), Kim et al. (2012), Kim et al. (2013); Tourism Research Chang et al. (2014); Guan and Jones (2015); Chen et al. 3466 (2016); Ma and Lai (2016), Nghiêm-Phú (2016), Nikbin et al. (2016), Rasoolimanesh et al. (2016), Shafaei (2016), Fong et al. (2017), Kim et al. (2017) 3 Cornell Hospitality 4 Frías-Jamilena et al. (2012), Aslanzadeh and Keating (2014), Quarterly Beldona et al. (2015), Úbeda-García et al. (2016) 4 Current Issues in 16 Lopez-Bonilla and Lopez-Bonilla (2014), Tan and Kuo Tourism (2014), Blazquez-Resino et al. (2015), Chen and Lin (2015, Loureiro (2015), Santos-Vijande et al. (2015), Toudert and Bringas-Rábago (2015), Vega-Vázquez, et al. (2015), Brochado and Rita (2016), Mohseni et al. (2016), Nikbin et al. (2016), Šeric et al. (2016), Suhartanto (2016), Battour et al. (2017); Nikbin and Hyun (2017), Sanz-Blas et al. (2017) 5 International Journal 18 Castellanos-Verdugo et al. (2009), King (2010), So and King of Contemporary (2010), Kim et al. (2013); Pavlatos (2015), Šeric et al. (2015); Hospitality Hemsley-Brown and Alnawas (2016), Kim et al. (2016), Koo Management et al. (2016); Li and Chang (2016), Martínez-Pérez et al. (2016); Pereira-Moliner et al. (2016); Amin et al. (2017), Chung et al. (2017), Dieck et al. (2017), Garrigos-Simon et al. (2017), Lo et al. (2017), Taheri et al. (2017) 6 International Journal 18 Ku et al. (2011), Loureiro and Kastenholz (2011), Cohen and of Hospitality Olsen (2013), Deng et al. (2013), King et al. (2013), Loureiro Management et al. (2013), Prud’homme and Raymond (2013), Loureiro (2014), Ruizalba et al. (2014), Úbeda-García et al. (2014), Gallarza et al. (2015), Gao and Lai (2015), Kang et al. (2015), Qiu et al. (2015), Agag and El-Masry (2016), Fakih et al. (2016), Chang and Busser (2017), García-Villaverde et al. (2017) 7 International Journal 9 Song et al. (2011), Gomez and Molina (2012), Ku (2014), of Tourism Research Loureiro (2015), Sanz-Blas and Buzova (2016), Forgas-Coll et al. (2017), Oviedo-García et al. (2017), Rasoolimanesh et al. (2017), Seow et al. (2017) 8 Journal of Destination 14 Prayag et al. (2013), Altunel and Erkut (2015), Gomez et al. Marketing and (2015), Tan and Wu (2016), Ali et al. (2017), Campon-Cerro Management et al. 2017, in-press); Chuang (2017), Gomez et al. (2017), González et al. (2017), Hernández-Mogollon et al. (2017), Rodríguez-Díaz and Espino-Rodríguez (2017), Swart et al. (2017), Tan (2017), Xu et al. (2017) 9 Journal of Hospitality 4 Kim et al. (2013), Kang et al. (2014), Wu et al. (2014), Gao and Tourism et al. (2017) Research 10 Journal of Hospitality 2 Eurico et al. (2015), Ali et al. (2016) Leisure Sport and Table I. Tourism Education PLS-SEM studies in 11 Journal of Sustainable 3 Lopez-Gamero et al. (2016), Rasoolimanesh et al. (2017), selected hospitality Tourism Taheri et al. (2017) and tourism journals (continued) Hospitality and No. of PLS- tourism Journals SEM papers Studies 12 Journal of Tourism 0 – and Cultural Change 13 Journal of Travel and 23 Loureiro and González (2008), Mistilis and D’ambra (2008), Tourism Marketing Conze et al. (2010), Loureiro (2010), Battour et al. (2012), 3467 Kim et al. (2012), Regan et al. (2012), Agapito et al. (2013), Cervera-Taulet et al. (2013), Lopez-Bonilla and Lopez- Bonilla (2013), Chiang and Chen (2014), Nikbin et al. (2015), Shahijan et al. (2015), Shu et al. (2015), Ali et al. (2016), Carlson et al. (2016), Dedeo glu et al. (2016), Kim and Preis (2016), Loureiro and Fialho (2016), Taheri (2016), Tan and Chang (2016), Ali et al. (2017), Kim et al. (2017) 14 Journal of Travel 21 Hernández-Maestro et al. (2009); Pike et al. (2011); Gardiner Research et al. (2012), Stienmetz et al. (2012), Assaker et al. (2013), Ayeh et al. (2013), Hernández-Maestro and González-Benito (2013), Gardiner et al. (2014), Kim et al. (2014), Huang et al. (2015), Mohd-Any et al. (2015), Zhang et al. (2015), Díaz- Chao et al. (2016); Park and Tussyadiah (2016), Rasoolimanesh et al. (2016), Simpson and Simpson (2016), Simpson et al. (2016), Agag and El-Masry (2017), Ahrholdt et al. (2017), Phillips et al. (2017), Wang et al. (2017) 15 Scandinavian Journal 4 Escobar-Rodríguez et al. (2016), Ulvnes and Solberg (2016), of Hospitality and Björk and Kauppinen-Räisänen (2017), Folgado-Fernández Tourism et al. (2017) 16 Tourism Economics 1 Mazanec and Ring (2011) 17 Tourism Geographies 1 Kim et al. (2015) 18 Tourism 52 Murphy et al. (2000), Hutchinson et al. (2009), Camarero et Management al. (2010), Alexander et al. (2012), García et al. (2012), Ayeh et al. (2013), Casanueva et al. (2013), Prayag et al. (2013), Roxas and Chadee (2013), Bianchi et al. (2014), Chiu et al. (2014). Escobar-Rodríguez and Carvajal-Trujillo (2014), Lin et al. (2014), Taheri et al. (2014), Amaro and Duarte (2015), Molina-Azorín et al. (2015), Bryce et al. (2015), Fraj et al. (2015), Gomez et al. (2015), Jaafar et al. (2015), Jung et al. (2015), Lai and Hitchcock (2015), Lee and Fernando (2015), Leonidou et al. (2015), Martínez-Martínez et al. (2015), Mohsin et al. (2015), O’Cass and Sok (2015), Ponte et al. (2015), Souto (2015), Vargas-Sanchez et al. (2015), Ali et al. (2016), Barnes et al. (2016), Brown et al. (2016), Buil et al. (2016), Dedeke (2016), Khoshkam et al. (2016), Lee et al. (2016), Matzler et al. (2016), Ram et al. (2016), Simpson et al. (2016), Wells et al. (2016), Zailani et al. (2016), Zhang et al. (2016), Doh et al. (2017), Fong et al. (2017), Lai and Hitchcock (2017), Lin et al. (2017), Molina et al. (2017), Rasoolimanesh et al. (2017), Rasoolimanesh et al. (2017), Ye et al. (2017), Zhang et al. (2017) 19 Tourist Studies 2006 – Notes: The journal list has been retrieved from the “Hospitality, Leisure, Sport and Tourism” category of the Web of Science – Journal Citation Reports®; journals are listed alphabetically Table I. IJCHM Issues Guidelines 30,11 Reasons for using PLS-SEM Provide explicit reasons for using PLS-SEM (e.g. non-normal data, small sample size, theory development, formative measures, etc.) Data characteristics Sample size Ten times rule of thumb: A minimum sample size of ten times the maximum 3468 number of independent variables is required in the outer and inner model (Barclay et al., 1995; Hair et al., 2013) Other issues related to Provide information about sample representativeness and the calculation of sample the required sample size Holdout 30% of original sample (Hair et al., 2010) Distribution Robust when applied to highly skewed data, but skewness and kurtosis should be reported (Reinartz et al., 2009) Missing data Report not only the missing values but also how these values are treated (Hair et al., 2017b) Outliers Report not only the outliers but also how these exceptionally high or low values are treated (Hair et al., 2017b) Model characteristics Description of the outer Include a complete list of indicators used in the study models Description of the inner Provide graphical representation of all structural relationships model Single-item measures Use single-item measures when 1) small sample sizes are present (i.e. n < 50), 2) effect sizes of 0.30 and lower are expected, 3) the items of the originating multi-item scale are highly homogeneous (i.e. inter-item correlations > 0.80, Cronbach’s alpha > 0.90) and 4) the items are semantically redundant (Diamontopoulos et al., 2012) Measurement mode of outer Follow design rules (by Jarvis et al., 2003) and substantiate measurement models mode (by using CTA-PLS) (e.g. Diamontopoulos et al., 2008) Model evaluation Outer model evaluation: Reﬂective Indicator reliability Report standardized indicator loadings ( 0.4 in exploratory research, 0.70 in all other studies) (Hulland, 1999) Internal consistency Report composite reliability rather than Cronbach’s alpha (Henseler et al., 2009). ( 0.6 in exploratory research, 0.70 in all other studies) (Bagozzi and Yi, 1988) Convergent validity Report AVE values ( 0.5) (Bagozzi and Yi, 1988) Discriminant validity Report cross loadings, Fornell and Larcker (1981) criterion (i.e. each construct’s AVE should be higher than its squared correlation with any other construct) or the heterotrait-monotrait (HTMT) ratio of correlations (Henseler et al., 2015) Outer model evaluation: Formative Convergent validity Report the results for redundancy analysis (Chin, 1998) or the use of global Table II. items (Hair et al., 2017a) A list of PLS-SEM Collinearity Report VIF (<5), tolerance (>0.2), or condition index (<30) (Hair et al., 2011) application Indicator contributions Report indicator weights. Additionally, report standard errors, signiﬁcance guidelines used in levels, t-values and p-values the review (continued) Hospitality and Issues Guidelines tourism Inner model evaluation Collinearity Report collinearity statistics for structural path relationships. (Sarstedt et al., 2017) Explained variance – R2 Report R2. Acceptable level depends on the context (Hair et al., 2010) Predictive relevance – Q2 Report cross-validated redundancy measure (Q2) (Sarstedt et al., 2014) 3469 Effect size – f 2 Report f 2; 0.02, 0.15, 0.35 for weak, moderate, and strong effects (Cohen, 1988) Effect size – q2 Report q2; 0.02, 0.15, 0.35 for small, medium, and large predictive relevance (Cohen, 1988) Path estimates Report path coefﬁcients, standard errors, signiﬁcance levels, t-values and p-values Model ﬁt Report SRMR, RMStheta or exact ﬁt test, do not use GoF (Hair et al., 2017a) Reporting technical aspects Resampling procedures Report the resampling techniques such as jackkniﬁng, bootstrapping or blindfolding. Additionally, report the number of bootstrap samples. A bootstrap sample size of 5,000 is recommended (Hair et al., 2017a) PLS Algorithm Report the weighting scheme for inner models and estimation modes for outer models Software Report the software used (Hair et al., 2012b) Covariance/correlation Provide the covariance/correlation matrix for the indicator variables (Hair matrix et al., 2012a) Source: Adapted from Richter et al., (2016a) and extended by using Hair et al. (2012a, 2012b, 2017a, 2017b) and Sarstedt et al. (2014, 2017) Table II. A structured form, consisting of 60 checkpoints, was developed to assess the PLS-SEM papers (see Appendix for the structured form) based on the application guidelines listed in Table II. The review was carried out between July 15 and September 15, 2017. Two independent researchers (coders) with expertise in PLS-SEM separately assessed each paper to minimize the possibility of personal bias. During the evaluation, any disagreements between the coders were clariﬁed through constructive discussion, along with the participation of a third researcher. Unlike previous assessments of the use of PLS-SEM in tourism, the present study evaluated the level of consistency between the two coders by estimating Cohen’s kappa (Landis and Koch, 1977). Cohen’s kappa, a statistical coefﬁcient of inter-rater reliability, ranges from 0 to 1, with larger values indicating higher reliability. In this study, overall inter-coder reliability was found to be 0.88, suggesting almost perfect agreement between the two coders (Landis and Koch, 1977). To investigate whether a comparison of journal types and of journal qualities reveal signiﬁcant differences in PLS-SEM use, all 206 PLS-SEM-based papers were grouped into different groups (i.e. hospitality vs tourism and top-tier vs other leading). To do this, the journals were ﬁrst categorized as either hospitality or tourism journals. During this categorization, ten papers in three journals (i.e. Journal of Hospitality and Tourism Research, JHTR; Journal of Hospitality Leisure Sport and Tourism Education, JoHLSTE; and Scandinavian Journal of Hospitality and Tourism) were excluded as these journals can be categorized as both hospitality and tourism journals. Thus, out of 196 papers included in the hospitality vs tourism categorization, 40 were published in hospitality journals, while the remaining 156 were published in tourism journals. After that, the journals were once again IJCHM grouped into two clusters based on their ﬁve-year impact factors provided by Web of 30,11 Science. Speciﬁcally, journals with a ﬁve-year impact factor of higher than 3.50 were classiﬁed as “top-tier journals” (i.e. Tourism Management, TM; Journal of Travel Research, JTR; Annals of Tourism Research, ATR; Journal of Sustainable Tourism, JOST; International Journal of Hospitality Management; IJHM, International Journal of Contemporary Hospitality Management; IJCHM and Cornell Hospitality Quarterly, CHQ), 3470 whereas those with a ﬁve-year impact factor of lower than 3.50 were labeled as “other leading journals.” Of 206 papers reviewed, 119 of them were published in top-tier journals, whereas 87 papers were published in other leading journals. Independent samples t-test and chi-square test (in situations where it is not appropriate, Fisher’s exact test) were used to assess signiﬁcant changes in PLS-SEM practices between journal types (hospitality vs tourism) and between journal qualities (top-tier vs other leading). PLS-SEM studies in hospitality and tourism: journals and time frame As indicated in Table I, a total of 206 studies that used PLS-SEM were published in the 19 SSCI-indexed hospitality and tourism journals between 2000 and April 2017. In terms of papers per journal, TM (52 studies, 25.2 per cent) published the largest number of PLS-SEM studies, followed by Journal of Travel and Tourism Marketing (23 studies, 11.2 per cent), JTR (21 studies, 10.2 per cent), IJCHM (18 studies, 8.7 per cent) and IJHM (18 studies, 8.7 per cent). On the other hand, Journal of Tourism and Cultural Change and Tourist Studies did not publish a single PLS-SEM study. Figure 1 demonstrates the (cumulative) number of PLS-SEM studies between 2000 and April 2017. It is important to note that Murphy et al. (2000) were found to be the ﬁrst researchers who applied PLS-SEM in the ﬁeld of hospitality and tourism. Interestingly, after this initial study in 2000, no other PLS-SEM-based papers were published until 2008. As clearly shown in Figure 1, the use of PLS-SEM has increased over time. Approximately 32 per cent (66 studies) were published before 2015, compared to 68 per cent between 2015 and 60 250 50 206 200 Cumulave number of studies Number of studies per year 40 159 150 30 107 100 20 66 Figure 1. 47 50 10 The use of PLS-SEM in hospitality and 28 17 tourism over time 7 12 0 1 1 1 1 1 1 1 1 3 0 (from 2000 to April 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 2017) Year April 2017, suggesting that PLS-SEM usage in hospitality and tourism research has Hospitality and accelerated substantially starting from the year 2015. tourism Critical issues in PLS-SEM applications in hospitality and tourism research The results reported here were derived from the assessment of each article based on a list of PLS-SEM application guidelines. These guidelines (i.e. measures) are grouped and discussed under ﬁve main categories, consistent with previous assessments of the use of PLS-SEM (Hair et al., 2012a, 2012b, 2017a, 2017b; Richter et al., 2016a; Sarstedt et al., 2014, 2017), 3471 namely: (1) reasons for using PLS-SEM; (2) data characteristics; (3) model characteristics; (4) model evaluation (outer and inner); and (5) reporting technical aspects. Reasons for using PLS-SEM As previously mentioned, CB-SEM and PLS-SEM are actually two different approaches to the same problem. Although they start from the same set of theoretical and measurement equations, each approach has different assumptions and aims (Henseler et al., 2009). Thus, researchers should clearly explain the rationale behind the selection of PLS-SEM, instead of CB-SEM (Roldán and Sánchez-Franco, 2012). Of the 206 studies analyzed, a total of 70 (34 per cent) did not provide any explicit reasons for applying PLS-SEM. It is important to note that most of these 70 studies provided some information about PLS-SEM, particularly by enumerating its several advantages regarding sample size, data distribution, formative measures, etc. Nevertheless, those researchers have failed to provide an explicit justiﬁcation for choosing PLS-SEM in their studies. For this reason, such studies were not considered as they did not go beyond explaining the PLS-SEM in general. On the other hand, 136 studies (66 per cent) that did provide an explicit reason gave more than one. Furthermore, providing an explicit reason for using PLS-SEM was found to be more prevalent in studies published in top-tier journals (87 studies, 73.1 per cent) than in other leading journals (49 studies, 56.3 per cent). Table III shows the most common reasons for using PLS-SEM. As given in Table III, the most common rationale of using PLS-SEM was the researchers’ focus on prediction and theory development, which 46 studies (22.3 per cent) mentioned. This has become a rule of thumb for choosing between CB-SEM and variance-based SEM. More speciﬁcally, PLS-SEM is ideal for authors conducting exploratory research (i.e. identifying key constructs) with weak theory (Wold, 1985). However, if the goal of a study is theory testing or theory conﬁrmation, CB-SEM should be used (Hair et al., 2017a). Indeed, the results of a large scale Monte Carlo simulation study conducted by Reinartz et al. (2009) indicate that PLS-SEM achieves higher levels of statistical power than CB-SEM. The use of formative measures (43 studies, 20.9 per cent), small sample size (42 studies, 20.4 per cent), non-normal data (33 studies, 16 per cent) and highly complex models (32 studies, 15.5 per cent) were found to be the other most-frequent reasons. All these most common justiﬁcations provided by hospitality and tourism researchers are consistent with the ﬁndings of previous research on the use of PLS-SEM in other disciplines (Hair et al., 2012a; Hair et al., 2012b; Richter et al., 2016a). Moreover, ten studies (4.9 per cent) indicate constructs with fewer items (i.e. single or two) as a motivation for using PLS-SEM. Unlike CB-SEM, no identiﬁcation and convergence problems occur in PLS-SEM when testing single-item IJCHM No. of studies Hospitality Tourism 30,11 reporting Proportion journals journals Top tier journals Other leading Criterion (N = 206) (%) (N = 40) (N = 156) (N = 119) journals (N = 87) Explicit reason (s) provided 136 66 29 100 87* 49* 3472 Speciﬁc reasonsa Prediction and theory development 46 22.3 6 39 28 18 Formative measures 43 20.9 9 30 27 16 Small sample size 42 20.4 10 29 25 17 Non-normal data 33 16.0 9 22 21 12 Model complexity 32 15.5 8 23 25 7 Table III. Constructs with Reasons for using fewer items 10 4.9 2 7 8 2 PLS-SEM in Other 13 6.3 3 10 10 3 hospitality and Notes: *p < 0.05 indicates a signiﬁcant difference between “top-tier journals and other leading journals;” a tourism research Due to multiple reasons provided, total percentage for speciﬁc reasons exceeds 100 measures. A limited number of researchers mentioned other unique reasons for choosing PLS-SEM. These were estimating both outer and inner model simultaneously (four studies, 1.9 per cent), better suited for testing moderation effects (three studies, 1.5 per cent), no identiﬁcation issues (two studies, 1 per cent), high multicollinearity (two studies, 1 per cent), the use of categorical variables (one study, 0.5 per cent) and the nonexistence of improper solutions for second-order models, which sometimes occur in CB-SEM (one study, 0.5 per cent). Data characteristics An overview of data characteristics is presented in Table IV. Although it has been argued that PLS-SEM works well with small sample sizes (Chin and Newsted, 1999; Hui and Wold, 1982; Reinartz et al., 2009), the sample sizes in this review ranged from 55 to 2,760, with a mean of 425 observations (median = 341), exceeding those in other disciplines, such as strategic management (5 per cent trimmed mean = 154.9), strategic marketing (5 per cent trimmed mean = 211.9), international business research (mean = 354) and management information systems (5 per cent trimmed mean = 238.1) (Hair et al., 2012a; Hair et al., 2012b; Richter et al., 2016a, Ringle et al., 2012; respectively). These results clearly indicate that PLS- SEM studies in hospitality and tourism rely on larger sample sizes compared to other disciplines. Even though no statistically signiﬁcant differences were found between “hospitality and tourism journals” and “top-tier and other leading journals,” studies published in tourism journals (mean = 445.4) and in top-tier journals (mean = 446.1) relied on relatively larger sample sizes than studies published in hospitality journals (mean = 348.9) and in other leading journals (mean = 396), respectively. Furthermore, Hair et al., (2012b) argue that very small sample sizes can be problematic for PLS-SEM applications because they fail to capture heterogeneity in the population and thereby results in a larger Results Proportion Hospitality journals Tourism journals Top tier journals Other leading Criterion (N = 206) (%) (N = 40) (N = 156) (N = 119) journals (N = 87) Sample size Mean 425 – 348.9 445.4 446.1 396 Median 341 – 250 354.5 350 340 Range (Min; Max) 2,705 (55; 2,760) – 1,362 (69; 1,431) 2,705 (55; 2,760) 2,705 (55; 2,760) 1,572 (130; 1,702) Sample size calculation 60 29.1 9 49 41* 19* Sampling method Probability 46 22.3 35 116 25 21 Nonprobability 157 76.2 5 37 93 64 No sampling 3 1.5 0 3 1 2 Hold-out (validation sample) 4 1.9 1 2 1 3 Ten times rule of thumb for sample size 204 99 39 155 117 87 Missing data 86 41.7 15 66 53 33 Handling missing dataa Inadequate or no info about treatment 4 4.7 1 3 3 1 Replacement 8 9.3 0 7 6 2 Deletion 69 80.2 14 51 42 27 Both replacement and deletion 5 5.8 0 5 2 3 Outliers 8 3.9 1 7 4 4 Handling outliersb Inadequate or no info about treatment 1 12.5 0 1 0 1 Removed 7 87.5 1 6 4 3 Skewness 17 8.3 3 14 14* 3* Kurtosis 20 9.7 5 15 17* 3* Notes: *p < 0.05 indicates a signiﬁcant difference between “top-tier journals and other leading journals” (no tests for median and range differences; aFor handling missing data, the proportion column indicates the ratio of each missing data treatment method to all studies that report missing data (i.e. N = 86); bFor handling outliers, the proportion column indicates the ratio of each outlier treatment method to all studies that report outliers (i.e. N = 8) characteristics Descriptive statistics Table IV. 3473 Hospitality and for data tourism IJCHM sampling error. Thus, researchers should be careful when performing PLS-SEM with small 30,11 sample sizes. A sample of 100 observations is considered relatively small for PLS-SEM studies (Reinartz et al., 2009; Hair et al., 2012b). In this review, a great majority of studies (200 articles, 97.1 per cent) fulﬁlled the condition of 100 observations or more. Very few studies (six articles, 2.9 per cent) used a sample of 100 observations or less. In the light of the relatively high available sample sizes revealed in this review, 3474 hospitality and tourism researchers should also evaluate the robustness of the PLS-SEM results by using holdout samples. Nevertheless, only 4 of 206 studies (1.9 per cent) included a holdout sample, clearly indicating that much room for improvement is needed in this area. For further cross-validation of the model, researchers can randomly split the entire sample into an estimation sample (75 per cent of the observations) and a holdout sample (25 per cent of the observations) based on the recommendations of Steckel and Vanhonacker (1993), but some authors recommend using a higher rate (i.e. 30 per cent) of original sample as a holdout sample (Hair et al., 2010). Determining the required minimum sample size is an important issue in PLS-SEM applications. As a rule of thumb, Barclay et al. (1995) suggest using a minimum sample size of ten times the largest number of formative indicators used to measure a single construct (i.e. in the outer model) or ten times the largest number of structural paths directed at a particular construct (i.e. in the inner model). In this review, slightly less than one-third of studies (66 studies, per cent 32) included formative measures, whereas 203 studies (98.5 per cent) tested the inner models. First, the latter criterion (i.e. the largest number of structural paths directed at a particular construct) was used in this review to examine the number of studies meeting the ten times rule of thumb. Findings from Table V indicate that the largest number of structural paths directed at a particular construct in the inner model ranged from one to ten, with a mean path relationship of 3.8 (median = 3). Of all 203 studies estimating the inner model, only one study (Kang et al., 2015, one of the three different groups of samples used in the study is 40 per cent below the recommended sample size) did not meet the ten times rule of thumb based on the largest number of structural paths. Then, for the remaining three studies (Pike et al., 2011; Beldona et al., 2015; Ram et al., 2016) that did not estimate structural model, the former criterion (i.e. ten times the largest number of formative indicators used to measure a single construct) was followed. Of these three studies, Pike et al. (2011) did not meet the ten times rule of thumb (i.e. 7.1 per cent below the recommended sample size) based on the maximum number of independent variables in the outer model. Finally, it was found that almost all articles (204 studies, 99 per cent) reviewed in this study meet the ten times rule of thumb for minimum sample size. However, it is important to note that ten times rule of thumb suggested by Barclay et al., (1995) is a rough guideline for minimum sample size requirements which do not take into account reliability, effect size, the number of indicators used and other factors that are likely to affect the statistical power of PLS-SEM. Thus, researchers should be more aware of sample size issues in PLS-SEM applications and determine the required minimum sample size by means of statistical power analyses. This could be done in several ways, such as using Cohen’s (1992) statistical power analysis for multiple regression models, utilizing some speciﬁcally designed programs (e.g. G*Power) that compute power analyses or referring directly to the table listing minimum sample sizes developed for PLS-SEM applications by Hair et al. (2017a). More recently, with the argument that ten-times rule of thumb tends to yield imprecise estimates, Kock and Hadaya (2016) proposed two new methods to calculate minimum sample size estimation in PLS-SEM (available in WarpPLS 6.0), namely, the inverse square root method and the gamma-exponential method. Both methods are found to be fairly accurate in estimating the minimum sample size (Kock and Hadaya, 2016). Other leading Results Proportion Hospitality journals Tourism journals Top tier journals journals Criterion (N = 206) (%) (N = 40) (N = 156) (N = 119) (N = 87) Indicator list 195 94.7 38 148 114 81 Total number of indicators Mean 28.1 – 26.3 28.4 27.6 29 Median 27 – 26 27 26.5 27 Range (Min; Max) 60 (2; 62) – 53 (2; 55) 60 (2; 62) 59 (2; 61) 56 (6; 62) Graphical representation of structural 201 99 39 153 116 85 pathsa Total number of structural paths (n=203) Mean 8.2 – 7 8.3 8.4 8 Median 7 – 6 8 7 8 Range (Min; Max) 23 (1; 24) – 17 (1; 18) 22 (1; 23) 22 (1; 23) 23 (1; 24) The largest number of structural paths directed at a particular construct (n=203) Mean 3.8 – 3.2* 3.9* 3.8 3.8 Median 3 – 3 3 3 3 Range (Min; Max) 9 (1; 10) – 5 (1; 6) 9 (1; 10) 9 (1; 10) 8 (1; 9) Number of latent variables (ﬁrst-order) Mean 7.3 – 6.7 7.4 7.2 7.6 Median 7 – 6 7 6.5 7 Range (Min; Max) 23 (1; 24) – 11 (2; 13) 23 (1; 24) 22 (2; 24) 20 (1; 21) Number of latent variables (higher-order) Mean 1.7 – 1.63 1.8 2 1.4 Median 1 – 1 1 1.5 1 Range 7 (1; 8) – 3 (1; 4) 7 (1; 8) 7 (1; 8) 2 (1; 3) (continued) characteristics for model Descriptive statistics Table V. 3475 Hospitality and tourism 30,11 3476 IJCHM Table V. Other leading Results Proportion Hospitality journals Tourism journals Top tier journals journals Criterion (N = 206) (%) (N = 40) (N = 156) (N = 119) (N = 87) Single-item measures 38 18.4 4 31 26 12 Reasons for single-item measuresb Not speciﬁed 29 76.3 4 23 21 8 Widely used in marketing and tourism 2 5.3 0 2 2 0 Based on previous research 2 5.3 0 1 1 1 Nature of the data analysis 2 5.3 0 2 1 1 Nature of the data set 1 2.6 0 1 1 0 Length and cost of the survey 1 2.6 0 1 0 1 Ease of application 1 2.6 0 1 0 1 Mode of outer models Not speciﬁed 11 5.3 1 8 5 6 Only reﬂective 129 62.6 26 98 73 56 Only formative 5 2.4 0 5 4 1 Reﬂective and formative 61 29.6 13 45 37 24 Notes: *p < 0.05 indicates a signiﬁcant difference between “hospitality journals and tourism journals” (no tests for median and range differences); aFor graphical representation of structural paths, the population size (N) is considered as 203 studies (rather than N = 206) as three studies did not estimate the structural model; b Regarding reasons for using single-item measures, the proportion column indicates the ratio of each reason to all studies that include single-item measures (i.e. N = 38) As shown in Table IV, the results of the present study revealed that the required minimum Hospitality and sample size has been calculated in only 60 studies (29.1 per cent), indicating that no tourism information is usually provided about the sample size calculation in the majority of studies published in SSCI-indexed hospitality and tourism journals. Sample representativeness is another issue that affects not only the solutions provided by PLS-SEM applications but also the results of any research that is based on a sample drawn from a population. Therefore, the sampling method used in each PLS-SEM study was reviewed. Of the 206 PLS-SEM papers analyzed, three studies referred to the census rather than a sample and less than one- 3477 fourth of studies used probability sampling (46 studies, 22.3 per cent), which increase the generalizability of PLS-SEM results. Other issues regarding data characteristics that need to be examined during PLS-SEM applications include missing data, outliers and data distributions. PLS-SEM provides highly robust results as long as missing values are below a reasonable level (i.e. less than 5 per cent missing data per indicator) (Hair et al., 2017a). When identiﬁed, missing data should be dealt with carefully by using replacement options (e.g. mean value replacement, nearest neighbor and expectation-maximization algorithm) or by opting for removing the related observations (e.g. casewise or pairwise deletion). In 86 studies (41.7 per cent), authors stated that their data included missing values. No information has been provided regarding missing data in the remaining 120 studies (58.3 per cent). Of the 86 studies that include missing data, a great majority of them (69 studies, 80.2 per cent) opted for deleting the observations, which reduces the variation in the data and may open the way to biases. In contrast, mean replacement has been seldom applied (eight studies, 9.3 per cent). Similarly, the data should be examined for outliers as they inﬂuence the ordinary least squares regressions in PLS-SEM. Once the outliers are detected, the possible reasons should be investigated. If there is a clear explanation for these extreme values (e.g. data collection or entry mistakes), they should be corrected or removed. If not or if they represent a unique subgroup of the sample, the outliers should be retained (Hair et al., 2017a) because they usually include important information about certain phenomena (Filzmoser, 2005). In only eight studies (4.4 per cent), authors mentioned that the data included outliers. Of the eight studies reporting outliers, seven removed these exceptionally high or low values, usually by referring to the results of the outlier detection methods but without any further justiﬁcations. In one study (Chen et al., 2016), the researchers did not provide adequate information about how they handled the outliers. Even though PLS-SEM is a nonparametric statistical method that does not require normally distributed data, examining the skewness and kurtosis is advised (Baloglu and Usakli, 2017) due to the fact that extremely non-normal data reduce statistical power by inﬂating (bootstrap) standard errors (Chernick, 2008). In fact, less than 10 per cent of studies (for skewness, 17 studies, 8.3 per cent; for kurtosis, 20 studies, 9.7 per cent) reported information about the skewness and kurtosis of the data, with the majority of them being published in tourism and in top-tier journals. Model characteristics Table V summarizes the model characteristics in PLS-SEM applications in hospitality and tourism research. Few studies (11 studies, 5.3 per cent) did not provide the indicator (also known as item or manifest variable) lists used in the models. Furthermore, almost all studies (201 studies, 99 per cent) visually represented the inner model relationships in a table and/or ﬁgure format. The results reveal an average number of 28.1 indicators, 7.3 ﬁrst-order latent variables, 1.7 higher-order latent variables and 8.2 structural paths (i.e. inner model relationships) per model, suggesting that average number of indicators and latent variables used in hospitality and tourism research are similar to the ones reported for management IJCHM (Hair et al., 2012a; an average of 27 indicators and 7.5 latent variables) and marketing (Hair 30,11 et al., 2012b; an average of 29.55 indicators and 7.94 latent variables), but a relatively low level of model complexity was found in PLS-SEM studies used in hospitality and tourism research (based on the average number of inner model relationships) compared with those in management (Hair et al., 2012a; an average of 10.4 structural paths) and marketing (Hair et al., 2012b; an average of 10.56 structural paths). 3478 Although using multiple items to measure a construct increases internal consistency and construct validity (DeVellis, 2017), single-item measures are sometimes used by researchers either because of certain necessities such as when it is difﬁcult to create multiple items as the construct being measured is simple and single-faceted (Poon et al., 2002), when the size of the population is extremely limited and nonresponse is a major issue (Hair et al., 2017a), or because of some practical advantages such as lowering costs, increasing response rates, reducing respondent fatigue and ease of implementation. In fact, there is an ongoing debate in SEM of whether to use single- or multiple-item constructs (Petrescu, 2013 for a review). For instance, some authors argue that single-item measures are inappropriate in terms of psychometric properties compared to multiple-item measures, especially from a predictive validity perspective (Hair et al., 2017a; Kwon and Trail, 2005; Sarstedt and Wilczynski, 2009). Contrary to this conventional wisdom, Bergkvist and Rossiter (2007) found no differences between single- and multiple-item measures, indicating that both have equally high predictive validity. More recently, Diamantopoulos et al. (2012) developed more speciﬁc guidelines for choosing between single- and multi-item measures. According to these guidelines, single-item measures should be used if one of the following four conditions are met: (1) a sample size of less than 50; (2) cross-item correlations less than 0.30; (3) inter-item correlations greater than 0.80 or Cronbach’s alpha coefﬁcients greater than 0.90; and (4) when items are semantically redundant (Diamantopoulos et al., 2012). As indicated in Table V, a total of 38 studies (18.4 per cent) included single-item measures in their models, but none of them followed the guidelines of Diamantopoulos et al. (2012) for choosing them. In fact, a great majority of these 38 studies (i.e. 29 studies, 76.3 per cent) did not provide any reason for the use of single-item measures, while the remaining eight studies included single-item measures usually for pragmatic reasons. Two different measurement models have been discussed in the SEM literature, namely, reﬂective model (also called as principal factor model) and formative model (also called as composite latent variable model). These two measurement models can be used either solely or in combination. Researchers applying PLS-SEM should correctly specify the measurement model used because any misspeciﬁcation in the measurement model inﬂuences both the outer and inner model estimates, thereby providing biased conclusions for the research (MacKenzie et al., 2005; Jarvis et al., 2003). Of the 206 studies analyzed in this review, slightly less than two-thirds of studies (129 studies, 62.6 per cent) were composed of only reﬂectively measured constructs. Regarding formative measurement model, 61 studies (29.6 per cent) applied formative measures with a combination of reﬂective measures, whereas very few studies (5 studies, 2.4 per cent) were based solely on formatively measured constructs. Surprisingly, 11 studies (5.3 per cent) did not explain the mode of measurement model used. Model evaluation Hospitality and As in the traditional CB-SEM, researchers applying PLS-SEM in their studies should follow tourism a two-stage approach for model evaluation: Stage 1 includes outer model evaluation and Stage 2 inlcudes inner model evaluation. Furthermore, one should keep in mind that outer model evaluation differs based on the type of measurement model used (i.e. reﬂective vs formative), indicating that different empirical criteria are used to assess reﬂectively and formatively measured constructs. Figure 2 summarizes the model evaluation process followed in PLS-SEM. 3479 Does the model include reflectively measured constructs? Yes No Stage 1.1. Evaluation criteria (reflective outer models) • Indicator reliability (loadings) • Internal consistency reliability Does the model (composite reliability, Cronbach’s include formatively alpha) measured constructs? • Convergent validity (AVE) • Discriminant validity (HTMT, Fornell-Larcker criterion, cross- Yes loadings) Stage 1.2. Evaluation criteria (formative outer models) • Convergent validity (Redundancy analysis, global-item No measure) • Collinearity (VIF, tolerance, condition index) • Significance and relevance of indicator weights (t-values, p-values, significance levels) Stage 2. Evaluation criteria (inner model) • Collinearity (VIF, tolerance, condition index) • Explained variance (R2) • Predictive relevance (Q2) • Effect sizes (f 2 and q2) • Significance and relevance of path coefficients (t-values, p-values, significance levels) • Model fit (SRMR, RMStheta, exact fit test) Figure 2. Model evaluation in PLS-SEM Source: Adapted from Sarstedt et al. (2014) IJCHM No. of Other 30,11 studies Hospitality Tourism Top tier leading Empirical test criterion reporting Proportion journals journals journals journals in PLS-SEM (n = 190) (%) (n = 39) (n = 143) (n = 110) (n = 80) Indicator reliability 3480 Indicator loadings 168 88.4 32 128 97 71 Indicator loadings (Min; Max) (0.29; 0.993) (0.44; 0.98) (0.29; 0.993) (0.37; 0.99) (0.29; 0.993) Minimum indicator loading 0.7 90 47.4 19 67 54 36 0.5 Minimum indicator loading < 0.7 72 37.9 11 57 38 34 Internal consistency reliability Cronbach’s alpha 115 60.5 22 88 62 53 Composite reliability (CR) 179 94.2 36 135 102 77 Both alpha and CR 112 59 22 85 60 52 Neither alpha nor CR 8 4.2 3 5 6 2 Convergent validity AVE 184 96.8 38 138 107 77 Discriminant validity Cross-loadings 66 34.7 12 48 42 24 Fornell–Larcker criterion 165 86.8 34 124 97 68 HTMTa 17 18.7 2 15 13 4 Table VI. None 17 8.9 4 13 8 9 Evaluation of Notes: No signiﬁcant differences were found between “hospitality journals and tourism journals” and “top- reﬂective outer tier journals and other leading journals” (no tests for median and range differences); aFor the use of HTMT, models the population size (N) is considered as 91 studies published after HTMT was introduced Stage 1.1. Outer model evaluation: reﬂective. Reﬂective outer models are assessed by examining the relevant reliability (i.e. indicator reliability and internal consistency) and validity (i.e. convergent and discriminant validity) measures. Table VI illustrates an overview of the reﬂective outer model assessment in PLS-SEM applications in hospitality and tourism research. In 168 (88.4 per cent) of the 190 studies that included reﬂectively measured constructs, indicator loadings were reported by authors, ranging from 0.29 to 0.993. Speciﬁcally, indicator reliability refers to the contribution of each indicator variable (i.e. communality of an item) in the reﬂective outer model. Regarding indicator reliability, all standardized outer loadings should not only be statistically signiﬁcant but also higher than 0.70 (Henseler et al., 2009). In this review, slightly less than half of studies (90 studies, 47.4) met this critical value. However, outer loadings lower than 0.70 are still prevalent in social sciences (Hulland, 1999) even if they represent relatively weak reliabilities. For instance, nearly 38 per cent (72 studies) met a much lower critical value of 0.50 for indicator loadings. In SEM, researchers generally report both Cronbach’s alpha coefﬁcients and composite reliabilities (Bagozzi and Yi, 2012) for assessing internal consistency. Cronbach’s alpha coefﬁcient assumes that all items are equally reliable. The composite reliability, on the other hand, is not limited by such an assumption (Raykov, 2007). Thus, the composite reliability can be considered more appropriate for the applications of PLS-SEM. Indeed, many, but not all, studies (179 studies, 94.2 per cent) reported composite reliability to establish internal Hospitality and consistency of reﬂective outer models. A further examination of authors’ reliability tourism reporting practices revealed that eight studies (4.2 per cent) did report neither Cronbach’s alpha nor composite reliability, raising an important concern for the assessment of reﬂective outer models in hospitality and tourism research. In addition to reliability, evaluation of reﬂective outer models in PLS-SEM applications include convergent validity and discriminant validity. While the former refers to “the extent to which a measure correlates positively with alternative measures of the same construct” 3481 (Hair et al., 2017a, p. 112), the latter refers to “the extent to which a construct is truly distinct from other constructs by empirical standards” (Hair et al., 2017a, p. 112). The average variance extracted (AVE) is commonly used in SEM to examine convergent validity. More speciﬁcally, AVE estimates higher than or equal to 0.50 are recommended to establish convergent validity (Bagozzi and Yi, 1988), with the argument that a construct exceeding this cutoff value explains more than half of the variance of its indicator variables. Findings from Table VI suggest that only very few (six studies, 3.2 per cent) did not report AVE estimates, indicating that hospitality and tourism researchers are performing well on the assessment of convergent validity in PLS-SEM. Furthermore, a detailed investigation of AVE estimates showed that nearly all studies exceed the acceptable level of 0.50 for convergent validity. Once the reliability and convergent validity of the reﬂective outer model have been established, researchers should assess the discriminant validity. There are three main approaches used to evaluate the discriminant validity of reﬂectively measured constructs. The ﬁrst one is to examine and to report the cross-loadings. In this approach, the outer loading of an indicator on the related construct should be larger than any of its loadings in the other constructs (Chin, 1998). The Fornell–Larcker criterion, which compares the AVE estimates with the inter-construct correlations, is another approach to assessing discriminant validity. To successfully establish discriminant validity, the square root of AVE estimate for each construct should be larger than its correlations with other constructs (Fornell and Larcker, 1981). More recently, Henseler et al. (2015) developed a new criterion, called heterotrait-monotrait (HTMT) ratio of correlations, to evaluate discriminant validity. Based on the results of Monte Carlo simulation studies, the authors conclude that HTMT is superior to both cross-loadings and Fornell–Larcker (1981) criterion. Two threshold values (i.e. 0.85 and 0.90) are suggested for the HTMT criterion, meaning that HTMT correlations exceeding these thresholds indicate a lack of discriminant validity. If the constructs used in a model are conceptually distinct, a threshold value of 0.85 should be considered. However, if they are conceptually very similar, a threshold value of 0.90 could be used (Henseler et al., 2015). As shown in Table VI, Fornell and Larcker (1981) criterion (165 studies, 86.8 per cent) has been commonly used by hospitality and tourism researchers for the assessment of reﬂective outer models, followed by cross-loadings (66 studies, 34.7 per cent). Although HTMT has been recently developed (i.e. the paper was published in 2015), it has been used in 17 studies (18.7 per cent of all 91 papers published after 2015 that includes reﬂective outer models), suggesting that HTMT is being increasingly adopted by tourism researchers (15 studies in tourism journals vs two studies in hospitality journals). Of the two hospitality papers which mention that HTMT has been used, one (i.e. Chang and Busser, 2017) did not use the standard reporting procedures for HTMT or provide any information about the results, making its use questionable in the study. Furthermore, approximately 9 per cent (17 out of 190 studies that include reﬂectively measured constructs) did not report any information regarding the discriminant validity assessment, indicating that much improvement is still needed for hospitality and tourism researchers. IJCHM Stage 1.2. Outer model evaluation: formative. The assessment of formative outer model 30,11 differs from the reﬂective outer model given the fact that traditional procedures used to assess the reliability and validity of reﬂectively measured constructs are not appropriate for formatively measured constructs (Diamantopoulos and Winklhofer, 2001). The evaluation of formative outer model includes the investigation of convergent validity, multicollinearity issues and the relevance and signiﬁcance of indicator weights (Sarstedt et al., 2014). 3482 The ﬁrst step in assessing formative outer models is examining the convergent validity. This can be achieved by conducting redundancy analysis of each construct (Chin, 1998). The redundancy analysis refers to the extent to which a formatively measured construct highly correlates with a reﬂectively measured construct that captures the same concept (Sarstedt et al., 2017). To establish convergent validity, at least 64 per cent of the variance (R2) of the reﬂectively measured construct (which translates into 0.80 path coefﬁcient, from formative to reﬂective) should ideally be explained by the formatively measured construct. At a minimum, on the other hand, an R2 value of 0.50 can also be considered acceptable for convergent validity (Hair et al., 2017a). Another way of assessing convergent validity is to use a single-item (global) measure that captures the formatively measured construct. For instance, a global-item such as “Please indicate your overall image of Istanbul as a vacation destination” can be developed and measured on a scale from 0 (very negative) to 10 (very positive). Using this question as an endogenous construct, the convergent validity of formative measurement of destination image can be tested. However, when using single- item measures, one should consider the arguments regarding the low predictive validity of such items (Sarstedt et al., 2016b). To assess the convergent validity of formative measures, researchers should keep in mind to include reﬂective or single-item measures capturing the No. of studies Hospitality Tourism Top tier Empirical test criterion in reporting Proportion journals journals journals Other leading PLS-SEM (n = 66) (%) (n = 13) (n = 50) (n = 41) journals (n = 25) Convergent validity Global-item 3 4.5 0 3 3 0 Correlation between 1 1.5 0 1 1 0 formative construct and another variable (external to the construct) Multicollinearity VIF 45 68.2 7 36 28 17 Tolerance 3 4.5 1 2 2 1 Condition index 6 9.1 2 4 5 1 None 19 28.8 6 12 11 8 Indicator’s absolute contribution to the construct Indicator weights 40 60.6 8 32 25 15 Signiﬁcance of weights t-values or p-values 19 28.8 1 17 11 8 * * Signiﬁcance level 29 43.9 2 24 17 12 Table VII. None 32 48.5 11* 21* 21 11 Evaluation of formative outer Note: *p < 0.05 indicates a signiﬁcant difference between “hospitality journals and tourism journals” (no models tests for median and range differences) same construct(s) in their ﬁnal questionnaires. Table VII describes an overview of the Hospitality and results regarding formative outer model evaluation for the hospitality and tourism research tourism sample. Overall, 66 of 206 studies (32 per cent) included at least one formatively measured construct. However, only four studies (6 per cent) assessed the validity of formative constructs. Of these four studies, three (4.5 per cent) used global items to evaluate the convergent validity. Following the guidelines of Diamantopoulos and Winklhofer (2001), one study (i.e. Mohd-Any et al., 2015) assessed the external validity of their second-order 3483 formative construct (i.e. customer perceived eValue) by examining its correlation with two other variables (i.e. satisfaction and behavioral intentions) that are external to the formative construct. The second step involves the assessment of collinearity among formative indicators. The multicollinearity is a critical issue for formative model evaluation as it inﬂuences both the estimates and signiﬁcance of outer weights (Centefelli and Bassellier, 2009), thereby producing unstable results (Diamantopoulos et al., 2008). The multicollinearity is determined through variance inﬂation factor (VIF), tolerance value or condition index. As a rule of thumb, a VIF value of 5 or above, a tolerance value of 0.20 or lower and a condition index of 30 or above are indicative of multicollinearity problems among indicators (Hair et al., 2011). Of the 66 studies using formative constructs, 19 (28.8 per cent) did not assess multicollinearity, making nearly one-third of all formative models. On the other hand, those assessing multicollinearity relied primarily on VIF (45 studies, 68.2 per cent), followed by condition index (6 studies, 9.1 per cent) and tolerance value (3 studies, 4.5 per cent). The ﬁnal step in formative model evaluation deals with examining the relevance and statistical signiﬁcance of outer weights. In PLS-SEM, the outer weight represents an indicator’s absolute contribution to the formative construct (Hair et al., 2012b), making it primary statistics that should be reported by researchers. In this review, the outer weights have been reported in 40 (60.6 per cent) of 66 studies using formatively measured constructs. Furthermore, the signiﬁcance of weights should also be included in assessing the formative outer models because researchers typically decide whether to retain or delete an indicator based on the signiﬁcance of weights. Speciﬁcally, an indicator that is statistically signiﬁcant should be retained. However, if an indicator is non-signiﬁcant but it has a loading of 0.50 or above, it could also be retained as long as strong theoretical justiﬁcation or expert opinion is evident (Sarstedt et al., 2014). Of the 66 studies including formative constructs, 32 (48.5 per cent) did not provide any information regarding the signiﬁcance of weights, whereas the remaining 34 (51.5 per cent) reported either t-values (corresponding p-values) for each outer weight or signiﬁcance levels. Moreover, signiﬁcance levels (19 studies, 28.8 per cent) of outer weights have been reported more frequently than their t-values and/or p-values (29 studies, 43.9 per cent) by hospitality and tourism researchers. Stage 2. Inner model evaluation. Having established the reliability and validity of outer (measurement) model, the next stage in PLS-SEM addresses the assessment of inner (structural) model (Figure 2, Stage 2). This involves examining: collinearity; explained variance of endogenous constructs; predictive relevance; effect sizes; signiﬁcance and relevance of path coefﬁcients; and the model ﬁt (Hair et al., 2017a). IJCHM The inner model assessment starts by examining the potential collinearity problems. As 30,11 structural path coefﬁcients are estimated based on ordinary least squares regressions, they might be biased if high collinearity exists among the predictor constructs (Sarstedt et al., 2014). The collinearity is examined using the same measures as in the evaluation of formative outer models. After checking for collinearity issues, the researcher moves to examining the explained variance of endogenous constructs. This can be done by reviewing 3484 the coefﬁcient of determination (R2), which has a range from 0 to 1. Although higher R2 values indicate higher predictive accuracy, the acceptable levels of R2 depend on the context of the study, suggesting that providing rough rules of thumb for R2 values is difﬁcult (Hair et al., 2011). Thus, researchers should interpret the R2 values by considering the previous studies (in our case, previous research in hospitality and tourism). Notwithstanding, Henseler et al. (2009) provide the following rules of thumb for acceptable R2 values: 0.75 (substantial), 0.50 (moderate) and 0.25 (weak). Furthermore, f 2 effect size should also be used to assess the inner model (Chin, 1988). The f 2 effect size refers to “the change in the R2 value when a speciﬁed exogenous construct is omitted from the model” (Hair et al., 2017a, p. 201) and shows whether the omitted exogenous latent variable has a substantial effect on the endogenous latent variable. The following thresholds can be used to interpret the magnitude of f 2 effect sizes (Cohen, 1988): 0.02 (small effects), 0.15 (medium effects) and 0.35 (large effects). Table VIII displays an overview of inner model evaluation in hospitality and tourism research. Because three studies (Pike et al., 2011; Beldona et al., 2015; Ram et al., 2016) did not estimate the structural model, a total of 203 studies has been used in this review to assess the inner model quality of PLS-SEM studies in hospitality and tourism research. The ﬁndings indicate that almost all studies (195 studies, 96.1 per cent) reported R2 values. On the other hand, only 31 studies (15.3 per cent) referred to the f 2 effect size when assessing the inner model. Empirical test No. of studies Hospitality Tourism Top tier Other leading criterion in PLS- reporting Proportion journals journals journals journals Criterion SEM (n = 203) (%) (n = 39) (n = 154) (n = 116) (n = 87) Endogenous R2 195 96.1 38 148 111 84 constructs’ f2 effect size 31 15.3 7 24 22 9 explained variance Predictive Q2 Cross-validated 91 44.8 19 70 56 35 relevance redundancy q2 effect size 1 0.5 0 1 1 0 Path Absolute values 201 99 39 152 115 86 coefﬁcients Signiﬁcance t-values or p-values 131 64.5 26 102 68* 63* of path Signiﬁcance levels 200 98.5 39 151 115 85 coefﬁcients None 1 0.5 0 1 0 1 Model ﬁt GoFa 55 27.2 9 46 31 24 SRMRb 4 2.9 0 4 3 1 RMStheta 0 0 0 0 0 0 Exact ﬁt test 0 0 0 0 0 0 Notes: *p < 0.05 indicates a signiﬁcant difference between “top-tier journals and other leading journals” Table VIII. (no tests for median and range differences); aFor GoF, the population size (N) is considered as 202 studies Evaluation of inner published after it was proposed by Tenenhaus et al. (2004); bFor SRMR, the population size (N) is models considered as 138 studies published after it was proposed by Tenenhaus et al. (2004) In addition to examining the R2 values for endogenous constructs, Stone-Geisser’s cross- Hospitality and validated redundancy (Q2) should also be used in assessing the inner model (Stone, 1974; tourism Geisser, 1974). The Q2 value, which is obtained based on blindfolding procedure, evaluates the predictive relevance of a structural model (Chin et al., 2008). As a rough guideline, Q2 values larger than 0 indicate acceptable level of predictive accuracy for the path model (Sarstedt et al., 2014). As in the case of f 2 effect size, the q2 effect size represents the contribution of an exogenous construct to the Q2 value of an endogenous construct. To interpret an exogenous construct’s predictive relevance on an endogenous construct, the 3485 following values of q2 effect size can be used: 0.02 (small predictive relevance), 0.15 (medium predictive relevance) and 0.35 (large predictive relevance) (Hair et al., 2013). Results from Table VIII show that less than half of the studies (91 studies, 44.8 per cent) reported the Q2 criterion, meaning that hospitality and tourism researchers should improve themselves in this area. Surprisingly, only one study (Matzler et al., 2016) assessed the q2 effect size, although it has been considered a critical criterion in assessing the inner model (Chin, 1998). Subsequently, standardized path coefﬁcients should be reported as they provide evidence for the quality of inner model. While a standardized path coefﬁcient closer to 1 suggests strong a positive relationship, a coefﬁcient closer to 1 represents a strong negative relationship (Sarstedt et al., 2014). As in the case of formative outer model assessment, the signiﬁcance of structural path coefﬁcients should also be evaluated by using resampling techniques. In this review, almost all studies (201 studies, 99 per cent) reported the structural path coefﬁcients, and only one study (0.5 per cent) did not provide any information regarding their signiﬁcance. Similar to the reporting of formative outer weights, hospitality and tourism researchers report the signiﬁcance levels (200 studies, 98.5 per cent) of path coefﬁcients more frequently than their t-values and/or p-values (131 studies, 64.5 per cent). Finally, the inner model assessment should also consider the overall model ﬁt evaluation. Although model ﬁt should be assessed previously (before examining the model parameters), the present study discusses the model ﬁt as a ﬁnal step of inner model evaluation as it is a recent development in PLS-SEM. Additionally, there is considerable debate whether to use the model ﬁt measures in PLS-SEM. These debates are primarily based on the two main statistical modeling perspectives used: explanation versus prediction (Sarstedt et al., 2017). More speciﬁcally, CB-SEM follows an explanatory modeling perspective, indicating that it focuses on “minimizing bias to obtain the most accurate representation of the underlying theory” (Shmueli, 2010, p. 293). To do this, the CB-SEM algorithm minimizes the differences between the estimated and sample covariance matrices (Hair et al., 2011). On the other hand, PLS-SEM follows a predictive modeling perspective that aims to “minimize the combination of bias and estimation variance, occasionally sacriﬁcing theoretical accuracy for improved empirical precision” (Shmueli, 2010, p. 293). To achieve this, the PLS-SEM algorithm maximizes the explained variance of the endogenous constructs (Sarstedt et al., 2017), which makes it difﬁcult for PLS-SEM to develop several global model ﬁt measures such as the ones in CB-SEM (e.g. chi-square statistics, comparative ﬁt index). Nevertheless, a number of model ﬁt measures are available in PLS-SEM but less so compared to CB-SEM (Sarstedt et al., 2017). Within the context of PLS-SEM, Tenenhaus et al. (2004) proposed the ﬁrst global ﬁt index, namely, the goodness-of-ﬁt index (GoF). By referring to the lack of a global ﬁt measure, Tenenhaus et al. (2005, p. 173) argue that “the GoF represents an operational solution to this problem as it may be meant as an index for validating the PLS model globally.” Another version of GoF (called the relative goodness-of-ﬁt index, GoFrel) was developed by Esposito Vinzi et al. (2010). However, Henseler and Sarstedt (2013) empirically revealed that both GoF and GoFrel are inappropriate for model evaluation in PLS-SEM. IJCHM Additionally, Hair et al. (2017a) do not recommend the use of GoF in PLS-SEM due to the 30,11 fact that it is unable to distinguish the valid models from invalid ones. More recently, Dijkstra and Henseler (2015) introduced a chi-square-based model ﬁt measure, known as exact ﬁt test, to make PLS-SEM appropriate for theory testing. Other model ﬁt measures used in PLS-SEM include standardized root mean square residual (SRMR, a commonly used model ﬁt index in CB-SEM), the root mean square residual covariance (RMStheta, which was 3486 introduced by Lohmöller, 1989), normed ﬁt index (NFI; also called as Bentler–Bonett index) and non-normed ﬁt index (also called as Tucker–Lewis index) (Sarstedt et al., 2017). Regarding the applicability of these model ﬁt measures, Henseler et al. (2014) found that SRMR and exact ﬁt test are applicable in the context of PLS-SEM, and also call for further research for the use of RMStheta. On the other hand, the use of NFI is usually not recommended because it improves as the model complexity increases (Hu and Bentler, 1998). Regarding the use of model ﬁt measures in hospitality and tourism research, this review found that slightly more than one-fourth of studies (55 studies, 27.1 per cent) utilized the GoF in assessing the inner model, even though it has been criticized in the PLS-SEM literature. Four studies (2 per cent) referred to SRMR, but no one reported RMStheta or the exact ﬁt test. It is worth noting that researchers should be cautious when using these model ﬁt measures as research examining their use in PLS-SEM is extremely limited (Hair et al., 2017a). Taking into account the prediction-oriented nature of PLS-SEM, researchers should rely on criteria that evaluate the predictive performance of the inner model (Rigdon, 2012, 2014; Sarstedt et al., 2017), rather than reporting only the model ﬁt measures. For instance, researchers can use the new PLS predict approach developed by Shmueli et al. (2016), which extends Stone–Geisser’s cross-validated redundancy (Q2) to assess the predictive performance of PLS models. Reporting technical aspects. Chin (2010) states that researchers applying PLS-SEM should clearly provide information on the following issues: the population of the study and the sampling procedure used, the data distribution, the conceptual model and the statistical results of the study. However, reporting the technical aspects of PLS-SEM is as important as providing information about the aforementioned issues as it helps readers to fully understand the technicalities regarding PLS-SEM and to successfully replicate the research (Hair et al., 2011). In this context, the technical aspects that should be reported in PLS-SEM can be classiﬁed into four categories: (1) the resampling procedures; (2) the PLS-algorithm (i.e. weighting schemes and estimation modes); (3) the software; and (4) the covariance or correlation matrix. Table IX summarizes the technical reporting practices of hospitality and tourism researchers. The resampling methods in PLS-SEM include jackkniﬁng, blindfolding and bootstrapping. Although PLS-SEM relies on resampling procedures, not all studies discussed this issue. In majority of studies (165 studies, 80.1 per cent), the bootstrapping method has been reported. The bootstrapping, proposed by Efron (1979), creates a larger number of subsamples by replacing the original sample. In this review, the number of bootstrap samples ranged from 100 to 5,000 with a mode bootstrap sample size of 5,000 – consistent with the recommended guidelines (Hair et al., 2017a). Furthermore, even though a total of 91 studies reported the Q2 cross-validated redundancy measure (which is calculated based on blindfolding) for the inner Hospitality journals Tourism journals Top tier journals Other leading journals Criterion Results (N = 206) Proportion (%) (N = 40) (N = 156) (N = 119) (N = 87) Jackkniﬁng 0 0 0 0 0 0 Bootstrapping 165 80.1 33 125 95 70 Number of bootstrap samples Mean 2,362 1,510.4* 2,601.1* 2,309.3 2,438 Median 1,000 500 1,000 1,000 1,000 Mode 5,000 500 5,000 5,000 500 Range (Min; Max) 4,900 (100; 5,000) 4,900 (100; 5,000) 4,900 (100; 5,000) 4,900 (100; 5,000) 4,800 (200; 5,000) Blindfolding 34 16.5 4 29 24 10 a Weighting schemes (inner models) Not speciﬁed 189 93.1 37 142 106 83 Centroid 5 2.5 1 4 4 1 Factorial 1 0.5 0 1 0 1 Path 7 3.4 1 6 5 2 Other 1 0.5 0 1 1 0 Estimation modes (outer models) Not speciﬁed 199 96.6 38 151 113 86 Mode A 2 1 1 1 2 0 Mode B 0 0 0 0 0 0 Mode C 4 1.9 1 3 3 1 Other 1 0.5 0 1 1 0 Software used Not speciﬁed 35 17 9 24 18 17 SmartPLS 130 63.1 20 104 69 61 PLS-Graph 18 8.7 6 11 13 5 VisualPLS 2 1 0 2 1 1 WarpPLS 9 4.4 2 6 7 2 XLSTAT-PLS 10 4.9 2 8 10 0 Other 2 1 1 1 1 1 Correlation/covariance matrix 3 1.5 0 3 3 0 Notes: *p < 0.05 indicates a signiﬁcant difference between “hospitality journals and tourism journals” (no tests for median, mode and range differences); aFor weighting scheme, the population size (N) is considered as 203 studies (rather than N = 206) as three studies did not estimate the structural model aspects 3487 Hospitality and tourism Table IX. Reporting technical IJCHM model assessment, only 34 studies (16.5 per cent) commented on the blindfolding procedure. 30,11 Finally, none of the studies reported the use of jackkniﬁng as a resampling method. A potential reason for not using jackkniﬁng might be that it is considered less efﬁcient when compared to bootstrapping (Efron and Tibshirani, 1993). The PLS algorithm estimates ﬁrst the inner weights by an iterative process and then the outer weights. At ﬁrst, the inner weights can be obtained by using one of the three 3488 weighting schemes: centroid scheme (proposed by Wold, 1982), factorial scheme (introduced by Lohmöller, 1989) and path weighting scheme. Esposito Vinzi et al. (2010) state that the centroid scheme is most often used when the indicators of a latent variable are highly correlated, whereas the factorial scheme is more appropriate in cases where such correlations are low. However, the authors strongly recommend the use of path weighting scheme as it is “the only estimation scheme that explicitly considers the direction of relationships as speciﬁed in the predictive path model” (Esposito Vinzi et al., 2010, p. 53). Having obtained the inner weights, the algorithm goes on estimating the outer weights. There are three different modes available for this purpose: Mode A, Mode B and Mode C. The outer weight estimation mode is chosen based on the type of the measurement model. In other words, Mode A is more suitable for reﬂective outer models, whereas Mode B is more appropriate for formative outer models. Mode C integrates both Mode A and Mode B, suggesting that it is used for models that include both reﬂective and formative indicators (Dijkstra, 2010). However, a great majority of studies analyzed in this review did not provide any information about the weighting schemes used for the inner models (189 studies, 93.1 per cent) and estimation modes used for the outer models (199 studies, 96.6 per cent). Unlike conventional weighting schemes (i.e. centroid, factorial and path) and estimation modes (i.e. Mode A, Mode B and Mode C), one study (i.e. Rasoolimanesh et al., 2017b) used the warp3 algorithm of WarpPLS and factor-based algorithm for estimating the inner and outer models to address the criticisms regarding the PLS algorithm. Because PLS-SEM programs have different default settings, it is important for researchers to report the software used (Hair et al., 2012b). Indeed, most studies (171 studies, 83 per cent) reported the software used. Of the 206 studies analyzed, 130 used (63.1 per cent) SmartPLS (Ringle et al., 2005), making it the most commonly used software for PLS applications in hospitality and tourism research. Although not so common as SmartPLS, 18 studies (8.7 per cent) used PLS-Graph (Chin, 2003), 10 used (4.9 per cent) XLSTAT-PLS (Addinsoft, 2011), 9 used (4.4 per cent) WarpPLS (ScriptWarp Systems, 2009), 2 used VisualPLS (Fu, 2006) and 2 used other programs such as ADANCO (Henseler and Dijkstra, 2015) and MatLab. Interestingly, LVPLS (Lohmöller, 1987), the ﬁrst software in this ﬁeld, and the R (a popular statistical computing and graphics software) implementations of PLS were not used by hospitality and tourism researchers. A ﬁnal issue in PLS-SEM applications is to provide the empirical covariance or correlation matrix. Nevertheless, only 3 of the 206 studies (1.5 per cent) reported it for the indicator variables, suggesting that much transparency is required for the improvement of hospitality and tourism research in terms of ethical standards. A comparison of PLS-SEM use between hospitality and tourism journals This section deals with whether a comparison of hospitality journals with tourism journals reveals signiﬁcant differences in the use of PLS-SEM. As noted earlier (in the Methodology section of this paper), ten studies published in three journals (i.e. JHTR, JoHLSTE and SJHT) were excluded during this comparison as these three journals can be categorized as both hospitality and tourism journals. Speciﬁcally, the use of PLS-SEM in 40 studies published in hospitality journals were compared with a total of 156 studies published in Hospitality and tourism journals. Interestingly, very few statistically signiﬁcant differences were found tourism between the two journal types. As shown in Table V, PLS path models in tourism journals included signiﬁcantly (p < 0.05) larger number of structural paths directed at a particular construct (mean = 3.9) than in hospitality journals (mean = 3.2), suggesting that studies in tourism journals have more complex path relationships. In terms of formative outer model assessment, studies in 3489 hospitality journals signiﬁcantly (p < 0.05) fail to provide information about the signiﬁcance of outer weights (11 out of 13 studies, 84.6 per cent) when compared with tourism journals (21 out of 50 studies, 42 per cent) (Table VII). Furthermore, for those providing information regarding the signiﬁcance of outer weights, studies in tourism journals signiﬁcantly (p < 0.05) reported the signiﬁcance levels (24 out of 50 studies, 48 per cent) more often than in hospitality journals (two out of 13 studies, 15.4 per cent) (Table VII). Finally, studies in tourism journals signiﬁcantly (p < 0.05) used larger number of bootstrap samples (mean = 2,601.1) than in hospitality journals (mean = 1,510.4) (Table IX). These results suggest that hospitality researchers should focus more on the signiﬁcance of weights during formative model evaluation and use larger number of samples for the bootstrapping procedure. However, it is important to note that these are very minor differences for the use of PLS- SEM between the two ﬁelds. Indeed, for most other critical issues of PLS-SEM (e.g. reasons for using PLS-SEM, the data characteristics, reﬂective model evaluation and inner model evaluation), no signiﬁcant differences were found between the journal types, suggesting that the use of PLS-SEM between hospitality and tourism researchers is similar to a great extent. A comparison of PLS-SEM use between top-tier and other leading journals The present study also investigates whether any signiﬁcant differences existed in the use of PLS-SEM between top-tier and other leading journals. Taking into account the ﬁve-year journal impact factors (discussed earlier in the methodology section), the PLS-SEM practices in 119 studies published in seven top-tier journals (i.e. TM, JTR, ATR, JOST, IJHM, IJCHM and CHQ) were compared with a total of 87 studies published in other leading journals. The present study revealed few statistically signiﬁcant differences between the journal tiers. Providing an explicit reason for the PLS method was signiﬁcantly (p < 0.05) more prevalent in top-tier journals (87 out of 119 studies, 73.1 per cent) than in other leading journals (49 out of 87 studies, 56.3 per cent), indicating that studies in top-tier journals are more transparent for the selection of PLS-SEM, instead of CB-SEM. However, no signiﬁcant differences were found for the speciﬁc reason categories. In terms of data characteristics, studies in top-tier journals signiﬁcantly (p < 0.05) used a sample size calculation method (41 out of 119 studies, 34.5) and reported the skewness (14 out of 119 studies, 11.8 per cent) and kurtosis (17 out of 119 studies, 14.3 per cent) values more frequently than in other leading journals (p < 0.05). Finally, studies in top-tier journals signiﬁcantly (p < 0.05) more often reported the t-values or p-values in examining the signiﬁcance of path coefﬁcients in the inner model. These ﬁndings suggest that studies published in other leading journals require some improvement (in the areas of providing explicit reasons, reporting data distribution and assessing the structural path coefﬁcients) when compared with those in top-tier journals. As in the comparison of journal types, most of the critical issues analyzed in this review did not yield any signiﬁcant differences between the journal qualities, suggesting that the use of PLS-SEM is similar to a great extent between the top-tier and other leading journals. IJCHM Advanced analyses in PLS-SEM 30,11 In recent years, several advanced analyses have emerged in PLS-SEM (Hair et al., 2018). As these complementary modeling and estimation methods are recently developed and therefore are not used by hospitality and tourism researchers, the present review did not investigate the use of such advanced analyses in PLS-SEM. Nevertheless, these advanced issues in PLS-SEM have been brieﬂy explained below. 3490 An advanced modeling issue in PLS-SEM is the analysis of nonlinear effects (Henseler et al., 2012). When dealing with more complex cause-effect models, researchers should take into account the possible nonlinear relationships between constructs. For instance, it is usually assumed by hospitality and tourism researchers that there is a linear relationship between tourist satisfaction and destination loyalty. However, the relationship between these two constructs may be nonlinear (i.e. a curve rather than a straight line), meaning that increasing tourist satisfaction involves a positive but diminishing effect on destination loyalty. In this case, the size of the effect of tourist satisfaction on destination loyalty depends not only on the magnitude of change in tourist satisfaction but also in its value. PLS-SEM enables researchers to analyze such quadratic effects by including “an interaction term that accounts for the self-interaction of the exogenous construct” (Hair et al., 2018, p. 84). Advanced analyses regarding model assessment in PLS-SEM include conﬁrmatory tetrad analysis (CTA-PLS; Gudergan et al., 2008) and importance-performance map analysis (IPMA; Ringle and Sarstedt, 2016). Because one of the most important threats to the validity of SEM results is the measurement model misspeciﬁcation (Jarvis et al., 2003), researchers should carefully decide whether to specify a measurement model reﬂectively or formatively. Although this should primarily be decided by means of theoretical and conceptual considerations (MacKenzie et al., 2005), CTA-PLS also allows researchers to empirically evaluate whether the data conﬁrm a reﬂective measurement model or a formative measurement model (Gudergan et al., 2008). Furthermore, researchers may miss valuable insights if they do not use an IPMA for the PLS models. Speciﬁcally, “the IPMA contrasts the total effects, representing the predecessor constructs’ importance in shaping a certain target construct, with their average latent variable scores indicating their performance” (Ringle and Sarstedt, 2016, p. 1866). Thus, the IPMA extends the standard reporting of inner model path coefﬁcients and provides a richer discussion of PLS-SEM results (Ringle and Sarstedt, 2016). Other recent developments in PLS-SEM deal with modeling observed heterogeneity. Observed heterogeneity means that “differences between two or more groups of data relate to some observable characteristics” (Hair et al., 2018, p. 137), such as gender, nationality, age or education level. As in CB-SEM, the measurement invariance should be established in PLS models before analyzing observed heterogeneity. However, the CB-SEM’s measurement invariance techniques are not appropriate for PLS-SEM. To overcome this issue, Henseler et al. (2016) developed the measurement invariance of composite models (MICOM) procedure for PLS models, which includes three steps, namely: conﬁgural invariance; compositional invariance; and the equality of composite mean values and variances. After supporting the measurement invariance through MICOM procedure, researchers can make meaningful comparisons between two groups using nonparametric PLS multigroup analysis developed by Henseler et al. (2009). The ﬁnal group of advanced analyses in PLS-SEM focus on uncovering observed Hospitality and heterogeneity, which refers to “differences between two or more groups of data do not tourism emerge a priori from a speciﬁc observable characteristic or combinations of several characteristics” (Hair et al., 2018, p. 138). In addition to observed heterogeneity, researchers should examine the unobserved heterogeneity in their models as a standard practice because it threatens the validity of the results and biases the estimates, which in turn lead to both Type I and Type II errors (Becker et al., 2013). To identify and treat unobserved heterogeneity in PLS models, researchers can use two methods, namely, ﬁnite mixture PLS 3491 (FIMIX-PLS; e.g. Sarstedt et al., 2011b) and prediction-oriented segmentation in PLS (PLS- POS; Becker et al., 2013). It is important to note that Becker et al. (2013) empirically revealed that FIMIX-PLS is limited to uncovering unobserved heterogeneity in the inner model, while PLS-POS can identify unobserved heterogeneity in both the outer and inner models. Conclusions By reviewing a total of 206 empirical studies, the present study critically assessed the use of PLS-SEM in the 19 SSCI-indexed hospitality and tourism journals. The results make important practical and methodological implications to the understanding of the PLS-SEM use in hospitality and tourism research. The major results of this review, which show hospitality and tourism researchers’ overall compliance with the application guidelines, are summarized in Table X. This review conﬁrms that the use of PLS-SEM has increased substantially over time in hospitality and tourism, and the results reveal that some aspects of PLS-SEM are correctly applied by researchers, but there are still some misapplications. Thus, to enhance the quality of research in hospitality and tourism, the present study provides concrete recommendations for improving the future use of PLS-SEM. It is important to note that the more speciﬁc recommendations are thoroughly discussed in the relevant subsections of this study, namely, reasons for using PLS-SEM, data characteristics, model characteristics, reﬂective outer model evaluation, formative outer model evaluation, inner model evaluation and reporting technical aspects. Therefore, this section outlines a larger picture of hospitality and tourism researchers’ PLS-SEM usage (Table X) and provides the following broad recommendations based on this larger picture. Although the majority of studies provided explicit reasons for choosing PLS-SEM, most reasons are related to data and measurement characteristics (e.g. the use of formatively measured constructs, small sample size and non-normal data) which all could be considered as secondary selection criteria. Instead, the primary selection criteria for applying PLS-SEM rather than CB-SEM should be the goal of the study, namely, theory development vs theory conﬁrmation (Richter et al., 2016a). Nevertheless, most exploratory research failed to mention the theory development focus of PLS-SEM as a speciﬁc reason or several theory conﬁrmation studies did use PLS-SEM by incorrectly referring its appropriateness for theory development. Another problematic issue in this area is solely enumerating the several advantages of the PLS-SEM as a rationale for using the method, even if the reported issues are not applicable for the study. For instance, several studies reported that they used PLS- SEM for its ability to handle small sample sizes or non-normal data, even though these studies have relatively larger sample sizes or do not provide any information about the data distribution. In brief, researchers should not consider the PLS-SEM as a universal alternative to its CB counterpart (Hair et al., 2017b). First, researchers should consider a number of key issues regarding their research, including the objective, the data characteristics and the model setup, and then should decide IJCHM No. of studies Total no. of studies 30,11 properly reporting/ (applicable for the Overall Guidelines applying relevant measure) compliance Reasons for using PLS-SEM Providing explicit reasons 136 206 6 3492 Data characteristics Meeting the minimum sample size 204 206 þ Sample size calculation 60 203 – Sample representativeness (based on 46 203 – probability sampling) Holdout sample 4 206 – Skewness 17 206 – Kurtosis 20 206 – Missing data 86 206 6 Handling missing data (based on replacement, 13 86 – rather than deletion) Outliers 8 206 – Handling outliers (based on providing 1 8 – information about the treatment) Model characteristics Indicator list 195 206 þ Graphical representation of structural paths 201 203 þ Providing reasons for single-item measures 9 38 – Specifying the measurement mode of 195 206 þ constructs Model evaluation Outer model evaluation: Reﬂective Indicator reliability (based on minimum 90 190 6 indicator loading 0.7) Internal consistency (based on composite 179 190 þ reliability) Convergent validity (based on AVE) 184 190 þ Discriminant validity (based on Fornell- 165 190 þ Larcker criterion) Discriminant validity (based on HTMT) 17 91 – (after HTMT was introduced) Outer model evaluation: Formative Convergent validity 4 66 – Multicollinearity 47 66 þ Indicator weights 40 66 6 Signiﬁcance of weights (based on t-values, 34 66 6 p-values or sig. levels) Inner model evaluation Table X. Explained variance (R2) 195 203 þ Hospitality and Predictive relevance (Q2) 91 203 6 tourism researchers’ Effect size (f 2) 31 203 – overall compliance Effect size (q2) 1 203 – with PLS-SEM Path coefﬁcients 201 203 þ guidelines (continued) Hospitality and No. of studies Total no. of studies tourism properly reporting/ (applicable for the Overall Guidelines applying relevant measure) compliance Signiﬁcance of path coefﬁcients (based on 202 203 þ t-values, p-values or sig. levels) Model ﬁt (based on SRMR) 4 138 – 3493 (after the applicability of SRMR in PLS-SEM was tested) Model ﬁt (based on RMStheta) 0 203 – Model ﬁt (based on exact ﬁt test) 0 98 – (after exact ﬁt test was introduced) Reporting technical aspects Resampling procedures (based on 165 206 þ bootstrapping) Resampling procedures (based on 34 206 – blindfolding) Weighting schemes (inner models) 14 203 – Estimation modes (outer models) 7 206 – Software used 171 206 þ Covariance/correlation matrix 3 206 – Notes: Overall compliance with PLS-SEM guidelines: , poor compliance: proper application rate # 33%; 6, average compliance: 33% > proper application rate # 66%; þ, good compliance: proper application rate 66% Table X. whether to use PLS-SEM or CB-SEM (Hair et al., 2017a, 2018, for a more detailed discussion on the selection of variance-based SEM or CB-SEM). Finally, researchers should clearly report why the PLS-SEM has been chosen. Except for meeting the ten times rule of thumb (Barclay et al., 1995) for the minimum sample size requirement, hospitality and tourism researchers’ overall compliance with the data characteristics was found to be poor. However, the ten times rule of thumb suggested by Barclay et al. (1995) is a rough guideline that does not consider a number of important factors that are likely to affect the statistical power of PLS-SEM, such as reliability, effect size and the number of indicators in the model. Thus, researchers should not rely on the ten times rule of thumb for the sample size argument. Instead, the minimum sample size should be determined by means of statistical power analyses. This review found that the minimum sample size argument constituted the much of debate for hospitality and tourism researchers. However, more careful thought should also be given to other issues, such as sample size calculation and sample representativeness, which most of the studies did not mention. Another critical issue found in this review is the lack of holdout sample use in PLS- SEM applications. Surprisingly, only 4 of the 206 studies used a holdout sample. For further cross-validation of the model, it is recommended that future researchers should use 30 per cent of the original sample as a holdout sample (Hair et al., 2010). Furthermore, most studies did not report the skewness and kurtosis values. This may be attributed to the fact that PLS- SEM does not require the data to be normally distributed. However, researchers should still examine and discuss the data distribution as extremely non-normal data inﬂates the IJCHM standard errors and thereby produces distorted results (Chernick, 2008). Although 30,11 hospitality and tourism researchers performed relatively well in reporting missing data, little attention has been paid for the outliers. Researchers should keep in mind that PLS-SEM provides highly robust results as long as missing values and outliers are properly handled (Ringle et al., 2012). Therefore, the “pros” and “cons” of deleting or replacing missing values, as well as removing or retaining outliers, should be carefully weighed and justiﬁcations 3494 should be provided for the chosen treatment method. In terms of model characteristics, hospitality and tourism researchers’ overall compliance with PLS-SEM guidelines was found to be good. In particular, almost all studies reported the list of indicators used, provided a graphical representation of structural model relationships in a table or ﬁgure format and speciﬁed the mode of measurement model (i.e. reﬂective, formative or a combination of the two). It is worth noting that less than one-ﬁfth of studies used single-item measures. By taking into account the ongoing debate on single-item measures (as indicated in the data characteristics subsection of this paper), the use of such measures can be considered as a “double-edged sword” in PLS-SEM. Therefore, one should understand the advantages (e.g. low costs, increased response rates and ease of implementation) and disadvantages (e.g. low predictive validity) of single-item measures before applying it in PLS-SEM. Speciﬁcally, 9 out of 38 studies that applied single-item measures provided pragmatic justiﬁcations. However, researchers applying single-item measures should follow the guidelines developed by Diamantopoulos et al. (2012), which none of the studies did. The present study reveals interesting results in terms of measurement model evaluation. Hospitality and tourism researchers performed well in examining the reliability and validity of reﬂective outer models. In contrast, the formative measurement model evaluation in hospitality and tourism gives rise to concern. This could be explained by the more extensive use of CB-SEM in hospitality and tourism. As formatively measured constructs are rarely applied in CB-SEM and requires certain conditions, hospitality and tourism researchers might be unfamiliar with the principles of formative model analysis (Assaker et al., 2012). Thus, researchers applying PLS-SEM should be more aware of formative model evaluation. Regarding reﬂective model evaluation, most studies assessed the internal consistency, convergent validity and discriminant validity (usually by applying the Fornell–Larcker criterion). The only issue that requires improvement in this area is the evaluation of discriminant validity. More speciﬁcally, hospitality and tourism researchers should take further advantage of HTMT, a new approach developed by Henseler et al. (2015), in assessing the discriminant validity of reﬂective models. It has been empirically revealed that HTMT is superior to both Fornell–Larcker criterion and the examination of cross-loadings (Henseler et al., 2015). On the other hand, the formative measurement model evaluation holds much room for improvement. For instance, researchers neglect the convergent validity in formative models; only four studies assessed convergent validity by using global items. The present study recommends researchers to conduct redundancy analysis of formative constructs in the assessment of convergent validity. Furthermore, formative models in hospitality and tourism build on unsatisfactory evaluations of indicator weights. Every PLS-SEM study that includes formative constructs should examine both the signiﬁcance and the relevance of indicator weights. Therefore, researchers are advised to report not only the indicator weights but also the corresponding t-values, p-values or signiﬁcance levels, which provide further information about the measurement quality of formative models. One should keep in mind that the structural model assessment might be considerably biased if the measurement model (either reﬂective or formative) lacks reliability and validity. This review clearly shows that PLS-SEM studies in hospitality and tourism do not Hospitality and sufﬁciently exploit the full range of guidelines available for structural model evaluation, tourism sometimes even misapply them. While a great majority of studies properly reported the explained variance (R2), and commented on the relevance and signiﬁcance of structural path coefﬁcients, other criteria, such as predictive relevance (Q2), effect sizes (both f 2 and q2), and model ﬁt indices, have been neglected, even though methodological research on PLS-SEM has emphasized the importance of such criteria in assessing the quality of inner model (Chin, 1998, 2010; Hair et al., 2012b, Sarstedt et al., 2014, 2017). For instance, the Stone–Geisser’s 3495 cross-validated redundancy (Q2), which shows the predictive relevance of inner model, was mostly not assessed by hospitality and tourism researchers. Although Q2 value is the standard practice used to assess the predictive quality of the structural model in PLS-SEM, future researchers should also apply the recent developments, such as the new PLS predict approach developed by Shmueli et al. (2016), which splits data into training and holdout samples to generate and evaluate different types of predictions from PLS models. Similarly, most hospitality and tourism studies did not report the f 2 effect size, and only one study used the opportunity to interpret the magnitude of predictive relevance by calculating the q2 effect size. Furthermore, only four studies used the SRMR as an appropriate model ﬁt index, whereas more than one-fourth studies reported the GoF of Tenenhaus et al. (2004), which is not recommended as a valid model ﬁt index for PLS-SEM (Henseler and Sarstedt, 2013). Overall, future studies in hospitality and tourism should increase the quality of inner models by reporting the predictive relevance, the effect sizes and the appropriate model ﬁt indices. The present study reveals that hospitality and tourism researchers report the signiﬁcance levels of parameter estimates (e.g. outer weights and path coefﬁcients) more frequently than t-values. However, reporting t-values should become a standard practice in model assessment. Additionally, hospitality and tourism researchers should not only rely on t-values but also analyze and report the bootstrap conﬁdence interval which provides further information on the stability of such parameter estimates. Although several methods are available for constructing conﬁdence intervals, the recent research on PLS-SEM suggest the use of bias-corrected and accelerated bootstrap conﬁdence intervals (Gudergan et al., 2008; Henseler et al., 2009; Sarstedt et al., 2011a; Hair et al., 2017a). The results of this study indicate that reporting some technical aspects of PLS-SEM also leaves much to be desired. As summarized in Table X, hospitality and tourism researchers should pay closer attention to report the blindfolding procedure, the weighting schemes (i.e. centroid, factorial, path weighting) and estimation modes (i.e. Modes A, B and C) used, as well as to provide the empirical covariance and/or correlation matrix. Accordingly, the replicability of research applying PLS-SEM in hospitality and tourism can be increased. It is important to note that the recommendations provided in this review are useful for all researchers using PLS-SEM in their studies, irrespective of discipline, although this review selected hospitality and tourism as a study area. Thus, these recommendations will hopefully provide a better understanding of PLS-SEM use for all disciplines in general and advance the quality of PLS-SEM-based future studies in hospitality and tourism in particular. To achieve this, future researchers can use both the PLS-SEM application guidelines presented in Table II and the check-list provided in the Appendix. The results of the present review are also compared with the previous assessments on the use of PLS-SEM in other disciplines (i.e. information systems, marketing, operations management, strategic management, human resources management, international business research and management accounting) (Table XI). As given in Table XI, researchers in other disciplines performed better than hospitality and tourism researchers in providing explicit reasons for the use of PLS-SEM. On the other hand, hospitality and tourism researchers 30,11 3496 IJCHM Table XI. across disciplinesa SEM review studies Comparison of PLS- Hospitality and/or Tourism Other disciplines Hospitality and Tourism Hospitality Info. systems Marketing Operations mgmt Strategic mgmt HRM Int. Bus. Res. Info. systems Mgmt (Present (Ali et al., Tourism (do Valle (Ringle et. al, (Hair et. al, (Peng and Lai, (Hair et al., (Ringle et al., (Richter et al., (Hair et al., Accounting Criteria Study) 2018) and Assaker, 2016) 2012) 2012b) 2012) 2012a) 2018) 2016a) 2017b) b (Nitzl, 2016) Number of studies analyzed (N) 206 29 44 65 204 42 37 77 43 82 37 Time-period covered 2000-2017 2001-2015 2000-2014 1992-2011 1981-2010 2000-2011 1981-2010 1985-2014 1990-2013 2010-2015 1980-2013 Reasons for using PLS-SEM Providing explicit reasons 136 (66%) 25 (86.2%) 42 (95.5%) 46 (70.8%) n.r. c 30 (71.4%) 32 (86.5%) 65 (84.42%) 39 (90.7%) 79 (96.3%) 33 (89.2%) Data characteristics Sample size Mean 425 332 487 238.1 211.29 d 246 154.9 d 142.5 d 354 333 138 Median 341 382 321 198 159 126 83 145 n.r. –b 105 Range 55; 2760 106; 1500 n.r. 17; 1449 18; n.r. 35; 3926 n.r. 6; 9623 38; 5191 59; 1512 18; 359 Fewer than 100 observations 6 (2.9%) 0 2 (4.5%) 25 (23%) 76 (24.4%) 14 (33.3%) 58 (51.8%) 38 (33.3%) n.r. 9 (11%) n.r. Meeting the minimum sample size 204 (99%) n.r. n.r. 103 (94.4%) 283 (91%) n.r. 93 (83%) 102 (89.5%) 42 (97.7%) 76 (92.7%) 33 (89.2%) Sample size calculation 60 (29.1%) n.r. n.r. 3 (4.6%) n.r. n.r. n.r. n.r. n.r. n.r. n.r. Sample representativeness (based on 46 (22.3%) n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. probability sampling) Holdout sample 4 (1.9%) n.r. n.r. 2 (3.1%) 13 (6.4%) n.r. n.r. 0 1 (2.3%) 0 n.r. Skewness 17 (8.3%) n.r. n.r. 4 (6.2%) e n.r. n.r. 0 0 1 (2.3%) 4 (4.9%) e 2 (5.4%) e Kurtosis 20 (9.7%) n.r. n.r. n.r. n.r. 0 0 1 (2.3%) Missing data 86 (41.7%) n.r. n.r. 10 (15.4%) n.r. n.r. n.r. 8 (7.01%) n.r. 3 (3.7%) 4 (10.8%) Handling missing data (based on 13 (15,1%) n.r. n.r. n.r. n.r. n.r. n.r. 2 (25%) n.r. n.r. n.r. replacement, rather than deletion) Outliers 8 (3,9%) n.r. n.r. 4 (6.2%) n.r. n.r. n.r. n.r. n.r. 4 (4.9%) 4 (10.8%) Handling outliers (based on 1 (12.5%) n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. 4 (10.8%) providing information about the treatment) Model characteristics Indicator list 195 (94.6%) n.r. n.r. 58 (89.2%) n.r. n.r. n.r. n.r. 33 (76.7%) 65 (79.2%) n.r. Total number of indicators Mean 28.1 24.7 24.93 27.4 29.55 n.r. 27 34.7 n.r. 32 n.r. Median 27 22 23 26.5 24 n.r. 19 26 n.r. –b n.r. (continued) Hospitality and/or Tourism Other disciplines Hospitality and Tourism Hospitality Info. systems Marketing Operations mgmt Strategic mgmt HRM Int. Bus. Res. Info. systems Mgmt (Present (Ali et al., Tourism (do Valle (Ringle et. al, (Hair et. al, (Peng and Lai, (Hair et al., (Ringle et al., (Richter et al., (Hair et al., Accounting Criteria Study) 2018) and Assaker, 2016) 2012) 2012b) 2012) 2012a) 2018) 2016a) 2017b) b (Nitzl, 2016) Range 2; 62 12; 78 8; 65 5; 1064 4; 131 n.r. 7; 114 7; 161 n.r. 12; 81 n.r. Graphical representation of 201 (99%) n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. structural paths Total number of structural paths Mean 8.2 8.0 7.55 11.38 10.56 n.r. 10.4 8.76 n.r. 9.1 6.16 Median 7 6 7 8 8 n.r. 9 7 n.r. –b n.r. Range 1; 24 3; 24 0; 16 2; 64 1; 38 n.r. 2; 39 2; 30 n.r. 1; 58 n.r. Number of latent variables Mean 7.8 7.03 6.02 8.12 7.94 n.r. 7.5 7.8 n.r. 7.3 n.r. Median 7 7 6 7 7 n.r. 6.0 7 n.r. –b n.r. Range 1; 24 3; 20 1; 17 3; 36 2; 29 n.r. 2; 31 3; 32 n.r. 2; 25 n.r. Number of latent variables (ﬁrst-order) Mean 7.3 n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. Median 7 n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. Range 1; 24 n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. Number of latent variables (higher-order) Mean 1.7 n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. Median 1 n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. Range 1; 8 n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. Single-item measures 38 (18.4%) 6 (20.7%) 6 (13.6%) 31 (47.7%) 144 (46.3%) n.r. 76 (67.9%) n.r. 20 (46.5%) 17 (20.7%) 12 (32.4%) Providing reasons for single-item 9 (23.7%) n.r. n.r. n.r. n.r. n.r. n.r. n.r. 13 (30.2%) n.r. n.r. measures (continued) 3497 Hospitality and tourism Table XI. 30,11 3498 IJCHM Table XI. Hospitality and/or Tourism Other disciplines Hospitality and Tourism Hospitality Info. systems Marketing Operations mgmt Strategic mgmt HRM Int. Bus. Res. Info. systems Mgmt (Present (Ali et al., Tourism (do Valle (Ringle et. al, (Hair et. al, (Peng and Lai, (Hair et al., (Ringle et al., (Richter et al., (Hair et al., Accounting Criteria Study) 2018) and Assaker, 2016) 2012) 2012b) 2012) 2012a) 2018) 2016a) 2017b) b (Nitzl, 2016) Specifying the measurement mode of constructs Not speciﬁed 11 (5.3%) 0 0 28 (25.7%) 37 (11.9%) n.r. 32 (28.6%) 18 (15.79%) n.r. 0 0 Only reﬂective 129 (62.6%) 18 (62.1%) 27 46 (42.2%) 131 (42.1%) n.r. 12 (10.7%) 47 (41.23%) 26 (60.5%) 72 (87.8%) 29 (78.9%) Only formative 5 (2.4%) 0 1 2 (1.8%) 20 (6.4%) n.r. 12 (10.7%) 0 1 (2.3%) 1 (1.2%) 0 Reﬂective and formative 61 (29.6%) 11 (37.9%) 16 33 (30.3%) 123 (39.6%) n.r. 56 (50%) 49 (42.98) 15 (34.9%) 18 (22%) 8 (21.6%) Model evaluation Outer model evaluation: Reﬂective Indicator loadings 168 (88.4%) 27 (93.1%) 35 (81.4%) 70 (88.6%) 157 (61.8%) n.r. 53 (77.9%) 73 (76%) 31 (72.1%) 70 (85.4%) 28 (75.7%) Indicator reliability (based on 90 (47.4%) n.r. n.r. n.r. n.r. n.r. n.r. n.r. 9 (21.9%) n.r. n.r. indicator loading 0.7) Internal consistency Cronbach’s alpha 115 (60.5%) 14 (48.3%) n.r. 30 (37.9%) 104 (40.9%) n.r. 21 (30.8%) 44 (45.8%) 11 (26.8%) 64 (78%) 18 (48.7%) Composite reliability (CR) 179 (94.2) 23 (79.3%) 43 (100%) 67 (84.8%) 142 (55.9%) n.r. 31 (45.5%) 67 (69.8%) 20 (48.7%) 71 (86.6%) 32 (86.5%) Both alpha and CR 112 (59%) 11 (37.9%) n.r. 22 (27.9%) 69 (27.17%) n.r. 14 (20.6%) 31 (32.9%) n.r. 57 (69.5%) 13 (35.1%) Neither alpha nor CR 8 (4.2%) 3 (10.3%) n.r. 4 (5.1%) 77 (30.3%) n.r. 30 (44.1%) 16 (16.6%) n.r. 4 (4.9%) n.r. Convergent validity (based on AVE) 184 (96.8%) 23 (79.3%) 42 (97.7%) 70 (88.6%) 146 (57.5%) n.r. 29 (42.7%) 75 (78.1%) 33 (80.4%) 73 (89%) 31 (83.8%) Discriminant validity Cross-loadings 66 (34.7%) 4 (17.2%) 9 (21%) 40 (50.6%) 43 (16.9%) n.r. 13 (19.1%) 18 (18.7%) n.r. f 23 (28%) 21 (56.8%) Fornell-Larcker criterion 165 (86.8%) 23 (79.3%) 41 (95.4%) 62 (78.4%) 142 (55.9%) n.r. 13 (19.1%) 57 (59.4%) n.r. f 67 (81.7%) 33 (89.2%) HTMT 17 (18.7%) 0 n.r. n.r. n.r. n.r. n.r. n.r. n.r. f n.r. n.r. None 17 (8.9%) 1 (3.4%) n.r. 7 (8.9%) 100 (39.3%) n.r. 38 (55.9%) 35 (36.5%) 5 (12%) 9 (11%) n.r. (continued) Hospitality and/or Tourism Other disciplines Hospitality and Tourism Hospitality Info. systems Marketing Operations mgmt Strategic mgmt HRM Int. Bus. Res. Info. systems Mgmt (Present (Ali et al., Tourism (do Valle (Ringle et. al, (Hair et. al, (Peng and Lai, (Hair et al., (Ringle et al., (Richter et al., (Hair et al., Accounting Criteria Study) 2018) and Assaker, 2016) 2012) 2012b) 2012) 2012a) 2018) 2016a) 2017b) b (Nitzl, 2016) Outer model evaluation: Formative Convergent validity 4 (6%) 0 n.r. n.r. n.r. 0 n.r. 0 n.r. n.r. n.r. Multicollinearity VIF 45 (68.2%) 3 (27.7%) 11 (64.7%)g 9 (25.7%)g 21 (8.2%)g 4 (21.1%) 1 (1.5%)g 9 (18.3%)g 8 (50%) 15 (88.2%)g 5 (62.5%)g Tolerance 3 (4.5%) 1 (9.1%) n.r. n.r. Condition index 6 (9.1%) n.r. n.r. 0 5 (1.9%) n.r. 1 (1.5%) 0 0 0 n.r. None 19 (28.8%) 7 (63.6%) 6 (35.3%) 26 (74.3%) 121 (84.6%) 15 (79%) 67 (98.5%) 40 (81.6%) n.r. 2 (11.8%) n.r. Indicator weights 40 (68.2%) 2 (18.2%) 17 (100%) 25 (68.6%) 33 (23.1%) 7 (36.9%) 26 (38.2%) 9 (18.4%) 8 (50%) 15 (88.2%) 5 (62.5%) Signiﬁcance of weights (based on t- 34 (51.5%) 2 (18.2%) 14 (82.4%) 25 (57.1%) 25 (17.5%) n.r. 3 (4.4%) 10 (20.4%) 4 (25%) n.r. 6 (75%) values, p-values or sig. levels) Inner model evaluation Explained variance (R2) 195 (96.1%) 24 (82.8%) 41 (97.6%) 105 (96.3%) 275 (88.4%) 36 (85.7%) 90 (80.4%) 94 (82.5%) 41 (95.3%) 80 (94.1%) 35 (95%) Predictive relevance (Q2) 91 (44.8%) 7 (24.1%) 14 (33.3%) 0 51 (16.4%) 4 (9.5%) 3 (2.7%) 8 (7%) 11 (27.5%) 14 (16.5%) 4 (10.8%) Effect size (f 2) 31 (15.3%) 5 (17.2%) 5 (11.9%) 13 (11.93%) 16 (5.1%) 6 (14.3%) 12 (10.7%) 6 (5.3%) 2 (4.7%) 23 (27.1%) 3 (8.1%) Effect size (q2) 1 (0.5%) n.r. 0 0 0 n.r. 0 0 0 5 (5.9%) 0 Path coefﬁcients 201 (99%) 28 (95.6%) n.r. 107 (98.2%) 298 (95.8%) 42 (100%) 107 (95.5) 114 (100%) 41 (95.3%) 70 (82.4%) 37 (100%) Signiﬁcance of path coefﬁcients 202 (99.5) 16 (55.2%) 42 (100%) 107 (98.2% 287 (92.3%) 42 (100%) 107 (95.5) 113 (99.1%) 41 (95.3%) 78 (91.8%) 37 (100% (based on t-values, p-values or sig. levels) Model ﬁt GoF 55 (27.2%) n.r. 6 (14.3%) n.r. 16 (5.14%) n.r. 0 8 (7%) n.r. n.r. n.r. SRMR 4 (2.9%) n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. RMStheta 0 n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. Exact ﬁt test 0 n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. (continued) 3499 Hospitality and tourism Table XI. 30,11 3500 IJCHM Table XI. Hospitality and/or Tourism Other disciplines Hospitality and Tourism Hospitality Info. systems Marketing Operations mgmt Strategic mgmt HRM Int. Bus. Res. Info. systems Mgmt (Present (Ali et al., Tourism (do Valle (Ringle et. al, (Hair et. al, (Peng and Lai, (Hair et al., (Ringle et al., (Richter et al., (Hair et al., Accounting Criteria Study) 2018) and Assaker, 2016) 2012) 2012b) 2012) 2012a) 2018) 2016a) 2017b) b (Nitzl, 2016) Other 0 n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. n.r. Reporting technical aspects Resampling procedures Jackkniﬁng 0 n.r. 7 (15.9%)h 61 (93.4%)h 135 (66.2%)g n.r. 20 (54.1%)h 51 (66.2%)h n.r. 42 (51.2%)h 1 (2.7%) Bootstrapping 165 (80.1%) 12 (41.4%) 22 (52.4%) n.r. 30 (81.1%) Blindfolding 34 (16.5%) n.r. n.r. n.r. 11 (27.5%) n.r. Weighting schemes (inner models) Not speciﬁed 189 (93.1%) n.r. 2 (4.5%)i n.r. 0 n.r. 0 0 n.r. n.r. 0 Centroid 5 (2.5%) n.r. n.r. 0 n.r. 0 0 n.r. n.r. 0 Factorial 1 (0.5%) n.r. n.r. 0 n.r. 0 0 n.r. n.r. 0 Path 7 (3.4%) n.r. n.r. 0 n.r. 0 0 n.r. n.r. 0 Other 1 (0.5%) n.r. n.r. 0 n.r. 0 0 n.r. n.r. 0 Estimation modes (outer models) Not speciﬁed 199 (96.6%) n.r. n.r. i n.r. 0 n.r. n.r. n.r. n.r. n.r. n.r Mode A 2 (1%) n.r. n.r. 0 n.r. n.r. n.r. n.r. n.r. n.r Mode B 0 n.r. n.r. 0 n.r. n.r. n.r. n.r. n.r. n.r. Mode C 4 (1.9%) n.r. n.r. 0 n.r. n.r. n.r. n.r. n.r. n.r. Other 1 (0.5%) n.r. n.r. 0 n.r. n.r. n.r. n.r. n.r. n.r. (continued) Hospitality and/or Tourism Other disciplines Hospitality and Tourism Hospitality Info. systems Marketing Operations mgmt Strategic mgmt HRM Int. Bus. Res. Info. systems Mgmt (Present (Ali et al., Tourism (do Valle (Ringle et. al, (Hair et. al, (Peng and Lai, (Hair et al., (Ringle et al., (Richter et al., (Hair et al., Accounting Criteria Study) 2018) and Assaker, 2016) 2012) 2012b) 2012) 2012a) 2018) 2016a) 2017b) b (Nitzl, 2016) Software used Not speciﬁed 35 (17%) 8 (27.6% ) 8 (18.2%) 27 (41.5%) 104 (51%) 16 (38.1%) 19 (51.4%) 35 (45.5%) n.r. 27 (32.9%) 12 (32.4%) SmartPLS 130 (63.1%) 13 (44.8%) 25 (56.8%) 2 (3.1%) 24 (11.8%) 6 (14.2%) 2 (5.4%) 19 (24.7%) n.r. 32 (39%) 10 (27%) PLS-Graph 18 (8.7%) 6 (20.6%) 6 (13.6%) 35 (53.9%) 64 (31.3%) 19 (45.2%) 10 (27%) 20 (26%) n.r. 13 (15.9%) 15 (40.5%) VisualPLS 2 (1%) 0 2 (4.5%) 0 0 1 (2.4%) 0 1 (1.3%) n.r. 0 0 WarpPLS 9 (4.4%) 0 1 (2.3%) 0 0 0 0 0 n.r. 0 0 XLSTAT-PLS 10 (4.9%) 1 (3.4%) 2 (4.5%) 0 0 0 0 1 (1.3%) n.r. 0 0 Other 2 (1%) 1 (3.4%) 0 1 (1.5%) 12 (5.9%) 0 6 (16.2%) 1 (1.3%) n.r. 0 0 Covariance/correlation matrix 3 (1.5%) n.r. n.r. 54 (83.1%) 10 (4.9%) n.r. 37 (%67.6) 68 (88.3%) n.r. 72 (87.8%) n.r. Notes: n.r. = not reported; aThe percentages are calculated based on the number of either studies or models that is applicable for the relevant measure; bHair et al., (2017b) reviewed the PLS-SEM studies in two journals (i.e. 58 studies in IMDS and 24 studies in MISQ) and reported the results separately for each journal. To make a more meaningful comparison, their results were combined in this table. Note that only the median values cannot be calculated when combining the results; cAlthough the speciﬁc reasons for the use of PLS-SEM are provided, the number of studies not reporting any reason is not available; d5% trimmed mean; e The non-normality of data has been reported in general. No detailed information is available separately for skewness and kurtosis; fRichter et al., (2016a) mentioned that 88% of the studies reported the discriminant validity, but no detailed information was provided for each of the criteria used (i.e. cross-loadings, Fornell–Larcker criterion, HTMT), meaning that the reported percentage (88%) could be based on only the cross-loadings, which is a weaker criterion for assessing discriminant validity. Therefore, the categories are intentionally not combined; gAs it has been reported as “VIF or Tolerance,” the two categories are combined. However, the exact proportion of each category is unknown; hAs the resampling procedures has been reported in general, the exact proportion of each procedure (i.e. Jackkniﬁng, Bootstrapping, Blindfolding) is unknown; ido Valle and Assaker (2016) mentioned that 4.5% of the studies reported the weighting schemes. However, the authors incorrectly referred all the “centroid, factor, or path-weighting scheme for inner model; mode A or B for outer models” (p. 704) as “the weighting schemes.” However, the weighting schemes (centroid, factor or path) and the estimation modes (Mode A or B) are two different things. Therefore, it is not clear whether the reported percentage (4.5%) represent the weighting schemes and/or the estimation modes 3501 Hospitality and tourism Table XI. IJCHM relied on larger sample sizes compared to other disciplines and ranked ﬁrst among all 30,11 disciplines in terms of meeting the ten times rule of thumb. Even though hospitality and tourism researchers’ overall compliance with the remaining data characteristics (i.e. sample size calculation, sample representativeness, the use of holdout samples, reporting skewness, kurtosis, missing values and outliers, and handling missing data and outliers) were found to be poor, there is little or no information to make an accurate comparison with other 3502 disciplines, given the fact that such data characteristics were not reported in most previous research. Regarding model characteristics, hospitality and tourism researchers used a similar number of indicators and latent variables, but the model complexity seems to be relatively low when the average number of inner model relationships are considered. Across all disciplines, the use of single-item measures in models is found to be the lowest in the ﬁeld of hospitality and tourism. In terms of both reﬂective and formative model evaluation, it is important to note that hospitality and tourism researchers performed considerably higher than researchers in other disciplines, but such reporting practices, especially with regard to formative model assessment, should be improved as they are still below standard. Similarly, hospitality and tourism researchers followed the inner model evaluation guidelines more closely than other researchers, except not to use the GoF. Among four disciplines that reviewed the use of GoF, hospitality and tourism ranked ﬁrst, indicating that future hospitality and tourism researchers should not use GoF as it is not able to distinguish the valid models from invalid ones (Henseler and Sarstedt, 2013). Instead of GoF, future hospitality and tourism researchers should consider the use of other model ﬁt statistics (e.g. exact ﬁt test and SRMR) as well as the new PLS predict approach developed by Shmueli et al. (2016). Although the technical aspects have been reported more frequently in hospitality and tourism compared to other disciplines, some aspects, such as reporting weighting schemes and estimation modes and providing the empirical covariance/ correlation matrix for the indicator variables, clearly require much improvement. To ensure rigorous applications of PLS-SEM, the present study offers a step-by-step guidance for researchers. In other words, researchers will easily understand how to report PLS-SEM practices by following the guidelines and recommendations provided in the afore-mentioned ﬁve categories (i.e. reasons for using PLS-SEM, data characteristics, model characteristics, model evaluation, and reporting technical aspects). In addition to providing recommendations for future researchers, the present study has also other implications for journal editors and reviewers in hospitality and tourism. It is worth noting that the present study is the ﬁrst that compares the use of PLS-SEM between hospitality and tourism. The results revealed that PLS path models in tourism journals are signiﬁcantly more complex than in hospitality journals. In addition, the present study found a few other signiﬁcant differences on the use of PLS- SEM between the two ﬁelds, which indicates some room for improvement in hospitality journals. When compared with tourism journals, PLS-SEM studies in hospitality journals signiﬁcantly failed to provide any information about the signiﬁcance of outer weights, reported the signiﬁcance levels in formative model evaluation less often, and used smaller number of bootstrap samples. Furthermore, the present study assessed the impact of journal quality on the use of PLS-SEM in the ﬁeld of hospitality and tourism. It is important to compare the method’s usage between top-tier and other leading journals with the argument that PLS-SEM application and reporting practices may differ based on journal quality given the fact that the method’s use in hospitality and tourism is still in its infancy. The top-tier journals of hospitality and tourism (i.e. Annals of Tourism Research, Cornell Hospitality Quarterly, International Journal of Contemporary Hospitality Management, International Journal of Hospitality Management, Journal of Sustainable Tourism, Journal of Travel Research and Tourism Hospitality and Management) are the most inﬂuential publications that aim to disseminate rigorous tourism empirical research in the ﬁeld. In other words, the present study argues that PLS-SEM reporting practices in top-tier journals are more correctly applied than other leading journals as these top-tier journals have a reputation of enforcing the use of highest methodological standards. In line with this argument, few differences were found between the journal tiers, which points to some room for improvement in other leading journals of hospitality and tourism, also consistent with previous research in which the 3503 usage of PLS-SEM in marketing research (Hair et al., 2012b) and of CB-SEM in tourism research (Nunkoo et al., 2013) were evaluated. When compared with top-tier journals, PLS-SEM studies in other leading journals are signiﬁcantly less transparent for their methodological choice of the method and failed to use a sample size calculation. It should be noted that PLS-SEM is not a panacea for small sample sizes. Thus, researchers should calculate the required minimum sample size for their research, regardless of the method used. In addition, the data distribution (skewness and kurtosis values) and t-values or p-values in examining the signiﬁcance of path coefﬁcients in inner models were signiﬁcantly less reported by PLS-SEM studies in other leading journals. In sum, journals editors and especially reviewers should strongly follow whether or not researchers used the PLS-SEM correctly in their studies, and reported the required information properly. Limitations and future research As with any other research, the present study has certain limitations that should be taken into account when evaluating the results. First, the ﬁndings of this study are derived from the information reported in the published hospitality and tourism articles. In some cases, the authors may properly apply some of the criticized issues in this review but did not include them in their articles due to the page limit policies of journals. However, it is not possible to know this by just reviewing a PLS-SEM-based paper. Thus, every PLS-SEM-based paper should provide sufﬁcient information to readers which helps to fully understand and replicate the study. Second, this review focused only on standard PLS-SEM analyses, indicating that more advanced issues in PLS-SEM were not included, such as nonlinear effects (Henseler et al., 2012), CTA (Gudergan et al., 2008) and IPMA (Ringle and Sarstedt, 2016) in model assessment; measurement invariance of composites (MICOM; Henseler et al., 2016) and multigroup analysis (Sarstedt et al., 2011a) in modeling observed heterogeneity; and FIMIX-PLS (Sarstedt et al., 2011b) and PLS-POS (Becker et al., 2013) in identifying unobserved heterogeneity. These recent developments in PLS-SEM are increasingly becoming standard practices in various ﬁelds, such as marketing, management and information systems (Hair et al., 2018). Therefore, future research (especially methodological papers) introducing how to use such advanced issues will increase hospitality and tourism researchers’ understanding of the latest developments in PLS- SEM. Finally, it was not within the scope of this study to investigate the measurement model misspeciﬁcations in PLS-SEM-based studies. Instead, the articles were reviewed based on the reported mode of measurement models. Because it was found that the formative model evaluation holds much room for improvement, the present study argues that hospitality and tourism researchers are usually unfamiliar with the principles of formative measurement, suggesting that some researchers might incorrectly use reﬂective models although they should use formative measurement models or vice versa (as in the case of marketing and consumer research, Jarvis et al., 2003). Thus, investigating how prevalent the measurement model misspeciﬁcation is in hospitality and tourism would be an interesting future research agenda. IJCHM References 30,11 Addinsoft (2011), “XLSTAT – statistics package for excel”, available at: www.xlstat.com Ali, F., Kim, W.G. and Ryu, K. (2016), “The effect of physical environment on passenger delight and satisfaction: Moderating effect of national identity”, Tourism Management, Vol. 57, pp. 213-224. Ali, F., Kim, W.G., Li, J. and Cobanoglu, C. (2018b), “A comparative study of covariance and partial least squares based structural equation modelling in hospitality and tourism research”, International 3504 Journal of Contemporary Hospitality Management, Vol. 30 No. 1, pp. 416-435. Ali, F., Rasoolimanesh, S.M., Sarstedt, M., Ringle, C.M. and Ryu, K. (2018a), “An assessment of the use of partial least squares structural equation modeling (PLS-SEM) in hospitality research”, International Journal of Contemporary Hospitality Management, Vol. 30 No. 1, pp. 514-538. Amin, M., Aldakhil, A.M., Wu, C., Rezaei, S. and Cobanoglu, C. (2017), “The structural relationship between TQM, employee satisfaction and hotel performance”, International Journal of Contemporary Hospitality Management, Vol. 29 No. 4, pp. 1256-1278. Antonakis, J., Bendahan, S., Jacquart, P. and Lalive, R. (2010), “On making causal claims: a review and recommendations”, The Leadership Quarterly, Vol. 21 No. 6, pp. 1086-1120. Assaker, G., Huang, S. and Hallak, R. (2012), “Applications of partial least squares structural equation modeling in tourism research: a methodological review”, Tourism Analysis, Vol. 17 No. 5, pp. 679-686. Bagozzi, R.P. and Yi, Y. (1988), “On the evaluation of structural equation models”, Journal of the Academy of Marketing Science, Vol. 16 No. 1, pp. 74-94. Bagozzi, R.P. and Yi, Y. (2012), “Speciﬁcation, evaluation, and interpretation of structural equation models”, Journal of the Academy of Marketing Science, Vol. 40 No. 1, pp. 8-34. Baloglu, S. and Usakli, A. (2017), “Summarizing data”, in Sirakaya-Turk, E., Uysal, M.S., Hammitt, W. E. and Vaske, J.J. (Eds), Research Methods for Leisure, Recreation and Tourism, 2nd ed., CABI, Oxfordshire, pp. 243-268. Barclay, D., Higgins, C. and Thompson, R. (1995), “The partial least squares (PLS) approach to causal modeling: Personal computer adoption and use as an illustration”, Technology Studies, Vol. 2 No. 2, pp. 285-309. Becker, J.M., Rai, A., Ringle, C.M. and Völckner, F. (2013), “Discovering unobserved heterogeneity in structural equation models to avert validity threats”, MIS Quarterly, Vol. 37 No. 3, pp. 665-694. Beldona, S., Miller, B., Francis, T. and Kher, H.V. (2015), “Commoditization in the US lodging industry: Industry and customer perspectives”, Cornell Hospitality Quarterly, Vol. 56 No. 3, pp. 298-308. Bergkvist, L. and Rossiter, J.R. (2007), “The predictive validity of multiple-item versus single-item measures of the same constructs”, Journal of Marketing Research, Vol. 44 No. 2, pp. 175-184. Boßow-Thies, S. and Albers, S. (2010), “Application of PLS in marketing: Content strategies on the internet”, in Esposito Vinzi, V., Chin, W.W., Henseler, J., Wang, H. (Ed.), Handbook of Partial Least Squares: concepts, Methods and Applications, Springer, Heidelberg, pp. 589-604. Bollen, K.A. and Davis, W.R. (2009), “Causal indicator models: Identiﬁcation, estimation, and testing”, Structural Equation Modeling: A Multidisciplinary Journal, Vol. 16 No. 3, pp. 498-522. Cenfetelli, R.T. and Bassellier, G. (2009), “Interpretation of formative measurement in information systems research”, MIS Quarterly, Vol. 33 No. 4, pp. 689-707. Chang, W. and Busser, J.A. (2017), “Hospitality employees promotional attitude: ﬁndings from graduates of a twelve-month management training program”, International Journal of Hospitality Management, Vol. 60, pp. 48-57. Chen, X., Goodman, S., Bruwer, J. and Cohen, J. (2016), “Beyond better wine: the impact of experiential and monetary value on wine tourists’ loyalty intentions”, Asia Paciﬁc Journal of Tourism Research, Vol. 21 No. 2, pp. 172-192. Chernick, M.R. (2008), Bootstrap Methods. A Guide for Practitioners and Researchers, 2nd ed., Wiley, Hospitality and Hoboken, NJ. tourism Chin, W.W. (1998), “The partial least squares approach for structural equation modeling”, in Marcoulides, G.A. (Ed.), Modern Methods for Business Research, Lawrence Erlbaum Associates Inc., Mahwah, NJ, pp. 295-336. Chin, W.W. (2003), “PLS graph 3.0”, Soft Modeling Inc., available at: www.plsgraph.com Chin, W.W. (2010), “How to write up and report PLS analyses”, in Esposito Vinzi, V., Chin, W.W., 3505 Henseler, J. and Wang, H. (Eds), Handbook of Partial Least Squares: Concepts, Methods and Applications, Springer, Heidelberg, pp. 655-690. Chin, W.W. and Newsted, P.R. (1999), “Structural equation modeling analysis with small samples using partial least squares”, in Hoyle, R.H. (Ed.), Statistical Strategies for Small Sample Research, Sage, Thousand Oaks, pp. 307-341. Chin, W.W., Peterson, R.A. and Brown, P.S. (2008), “Structural equation modelling in marketing: some practical reminders”, Journal of Marketing Theory and Practice, Vol. 16 No. 4, pp. 287-298. Chou, C.-P., Bentler, P.M. and Satorra, A. (1991), “Scaled test statistics and robust standard errors for non-normal data in covariance structure analysis: a Monte Carlo study”, British Journal of Mathematical and Statistical Psychology, Vol. 44 No. 2, pp. 347-357. Cohen, J. (1988), Statistical Power Analysis for the Behavioral Sciences, 2nd ed., Lawrence Erlbaum Associates, USA. Cohen, J. (1992), “A power primer”, Psychological Bulletin, Vol. 112 No. 1, pp. 155-159. Dedeoglu, B.B., Balıkçıoglu, S. and Küçükergin, K.G. (2016), “The role of tourists’ value perceptions in behavioral intentions: the moderating effect of gender”, Journal of Travel and Tourism Marketing, Vol. 33 No. 4, pp. 513-534. DeVellis, F. (2017), Scale Development: Theory and Applications, 4th ed., Sage Publications, Los Angeles, CA. Diamantopoulos, A. and Rieﬂer, P. (2011), “Using formative measures in international marketing models: a cautionary tale using consumer animosity as an example”, in Sarstedt M., Schwaiger M. and Taylor, C. R. (Eds), Measurement and Research Methods in International Marketing, Emerald Group Publishing Limited, Bingley, Vol. 22, pp. 11-30. Diamantopoulos, A., Rieﬂer, P. and Roth, K.P. (2008), “Advancing formative measurement models”, Journal of Business Research, Vol. 61 No. 12, pp. 1203-1218. Diamantopoulos, A. and Winklhofer, H.M. (2001), “Index construction with formative indicators: an alternative to scale development”, Journal of Marketing Research, Vol. 38 No. 2, pp. 269-277. Diamantopoulos, A., Sarstedt, M., Fuchs, C., Wilczynski, P. and Kaiser, S. (2012), “Guidelines for choosing between multi-item and single-item scales for construct measurement: a predictive validity perspective”, Journal of the Academy of Marketing Science, Vol. 40 No. 3, pp. 434-449. Dijkstra, T.K. (2010), “Latent variables and indices: Herman wold’s basic design and partial least squares”, in Esposito Vinzi, V., Chin, W.W., Henseler, J. and Wang, H. (Eds), Handbook of Partial Least Squares: concepts, Methods and Applications, Springer, Heidelberg, pp. 23-46. Dijkstra, T.K. and Henseler, J. (2015), “Consistent and asymptotically normal PLS estimators for linear structural equations”, Computational Statistics and Data Analysis, Vol. 81, pp. 10-23. do Valle, P.O. and Assaker, G. (2016), “Using partial least squares structural equation modeling in tourism research: a review of past research and recommendations for future applications”, Journal of Travel Research, Vol. 55 No. 6, pp. 695-708. Efron, B. (1979), “Bootstrap methods: another look at the jackknife”, The Annals of Statistics, Vol. 7 No. 1, pp. 1-26. Efron, B. and Tibshirani, R.J. (1993), An Introduction to the Bootstrap, monographs on statistics and applied probability, 57, Chapman and Hall, New York, NY. IJCHM Esposito Vinzi, V., Trinchera, L. and Amato, S. (2010), “PLS path modeling: from foundations to recent developments and open issues for model assessment and improvement”, in Esposito Vinzi, V., 30,11 Chin, W.W., Henseler, J. and Wang, H. (Eds), Handbook of Partial Least Squares: concepts, Methods and Applications, Springer, Heidelberg, pp. 47-82. Fakih, K., Assaker, G., Assaf, A.G. and Hallak, R. (2016), “Does restaurant menu information affect customer attitudes and behavioral intentions? A cross-segment empirical analysis using PLS- SEM”, International Journal of Hospitality Management, Vol. 57, pp. 71-83. 3506 Filzmoser, P. (2005), “Identiﬁcation of multivariate outliers: a performance study”, Austrian Journal of Statistics, Vol. 34 No. 2, pp. 127-138. Fornell, C. and Larcker, D.F. (1981), “Evaluating structural equation models with unobservable variables and measurement error”, Journal of Marketing Research, Vol. 18 No. 1, pp. 39-50. Fu, J.R. (2006), VisualPLS – Partial Least Square (PLS) Regression – An Enhanced GUI for LVPLS (PLS 1.8 PC) Version 1.04, National Kaohsiung University of Applied Sciences, Taiwan, available at: www2.kuas.edu.tw/prof/fred/vpls/ Geisser, S. (1974), “A predictive approach to the random effects model”, Biometrika, Vol. 61 No. 1, pp. 101-107. Gudergan, S.P., Ringle, C.M., Wende, S. and Will, A. (2008), “Conﬁrmatory tetrad analysis in PLS path modeling”, Journal of Business Research, Vol. 61 No. 12, pp. 1238-1249. Hair, J.F., Ringle, C.M. and Sarstedt, M. (2011), “PLS-SEM: Indeed a silver bullet”, Journal of Marketing Theory and Practice, Vol. 19 No. 2, pp. 139-152. Hair, J.F., Ringle, C.M. and Sarstedt, M. (2013), “Partial least squares structural equation modeling: rigorous applications, better results and higher acceptance”, Long Range Planning, Vol. 46 Nos 1/2, pp. 1-12. Hair, J.F., Black, W.C., Babin, B.J. and Anderson, R.E. (2010), Multivariate Data Analysis, 7th ed., Prentice Hall, NJ. Hair, J.F., Hult, G.T.M., Ringle, C.M. and Sarstedt, M. (2017a), A Primer on Partial Least Squares Structural Equation Modeling (PLS-SEM), 2nd ed., Sage, Thousand Oaks, CA. Hair, J.F., Hollingsworth, C.L., Randolph, A.B. and Chong, A.Y.L. (2017b), “An updated and expanded assessment of PLS-SEM in information systems research”, Industrial Management and Data Systems, Vol. 117 No. 3, pp. 442-458. Hair, J.F., Hult, G.T.M., Ringle, C.M., Sarstedt, M. and Thiele, K.O. (2017c), “Mirror, mirror on the wall: a comparative evaluation of composite-based structural equation modeling methods”, Journal of the Academy of Marketing Science, Vol. 45 No. 5, pp. 616-632. Hair, J.F., Sarstedt, M., Pieper, T.M. and Ringle, C.M. (2012a), “The use of partial least squares structural equation modeling in strategic management research: a review of past practices and recommendations for future applications”, Long Range Planning, Vol. 45 Nos 5/6, pp. 320-340. Hair, J.F., Sarstedt, M., Ringle, C.M. and Gudergan, S.P. (2018), Advanced Issues in Partial Least Squares Structural Equation Modeling, Sage, Thousand Oaks, CA. Hair, J.F., Sarstedt, M., Ringle, C.M. and Mena, J.A. (2012b), “An assessment of the use of partial least squares structural equation modeling in marketing research”, Journal of the Academy of Marketing Science, Vol. 40 No. 3, pp. 414-433. Henseler, J. and Sarstedt, M. (2013), “Goodness-of-ﬁt indices for partial least squares path modeling”, Computational Statistics, Vol. 28 No. 2, pp. 565-580. Henseler, J., Ringle, C.M. and Sarstedt, M. (2015), “A new criterion for assessing discriminant validity in variance-based structural equation modeling”, Journal of the Academy of Marketing Science, Vol. 43 No. 1, pp. 115-135. Henseler, J., Ringle, C.M. and Sarstedt, M. (2016), “Testing measurement invariance of composites using partial least squares”, International Marketing Review, Vol. 33 No. 3, pp. 405-431. Henseler, J., Ringle, C.M. and Sinkovics, R.R. (2009), “The use of partial least squares path modeling in Hospitality and international marketing”, in Sinkovics, R.R. and Ghauri, P.N. (Eds), Advances in International Marketing, Vol. 20. Emerald Group Publishing Limited, Bingley, pp. 277-320. tourism Henseler, J., Fassott, G., Dijkstra, T.K. and Wilson, B. (2012), “Analyzing quadratic effects of formative constructs by means of variance-based structural equation modelling”, European Journal of Information Systems, Vol. 21 No. 1, pp. 99-112. Henseler, J. and Dijkstra, T. (2015), Adanco (1.1), Composite Modeling, Kleve, available at: www. composite-modeling.com/ 3507 Henseler, J., Dijkstra, T.K., Sarstedt, M., Ringle, C.M., Diamantopoulos, A., Straub, D.W., Ketchen, D.J., Hair, J.F., Hult, G.T.M. and Calantone, R.J. (2014), “Common beliefs and reality about PLS: comments on Rönkkö and Evermann (2013)”, Organizational Research Methods, Vol. 17 No. 2, pp. 182-209. Hu, L. and Bentler, P.M. (1998), “Fit indices in covariance structure modeling: sensitivity to underparameterized model misspeciﬁcation”, Psychological Methods, Vol. 3 No. 4, pp. 424-453. Hui, B.S. and Wold, H. (1982), “Consistency and consistency at large of partial least squares estimates”, in Jöreskog, K.G. and Wold, H. (Eds), Systems under Indirect Observation: Part II, North Holland, Amsterdam, pp. 119-130. Hulland, J. (1999), “Use of partial least squares (PLS) in strategic management research: a review of four recent studies”, Strategic Management Journal, Vol. 20 No. 2, pp. 195-204. Jarvis, C.B., MacKenzie, S.B. and Podsakoff, P.M. (2003), “A critical review of construct indicators and measurement model misspeciﬁcation in marketing and consumer research”, Journal of Consumer Research, Vol. 30 No. 2, pp. 199-218. Jöreskog, K.G. and Wold, H. (1982), “The ML and PLS techniques for modeling with latent variables: Historical and comparative aspects”, in Jöreskog, K.G. and Wold, H. (Eds), Systems under Indirect Observation: Part I, North-Holland, Amsterdam, pp. 263-270. Kang, J.S., Chiang, C.F., Huangthanapan, K. and Downing, S. (2015), “Corporate social responsibility and sustainability balanced scorecard: the case study of family-owned hotels”, International Journal of Hospitality Management, Vol. 48, pp. 124-134. Kock, N. and Hadaya, P. (2016), “Minimum sample size estimation in PLS-SEM: the inverse square root and gamma-exponential methods”, Information Systems Journal, Vol. 28 No. 1, pp. 227-261. Kwon, H. and Trail, G. (2005), “The feasibility of single-item measures in sport loyalty research”, Sport Management Review, Vol. 8 No. 1, pp. 69-89. Lai, I.K. (2015), “The roles of value, satisfaction and commitment in the effect of service quality on customer loyalty in Hong Kong–style tea restaurants”, Cornell Hospitality Quarterly, Vol. 56 No. 1, pp. 118-138. Landis, J.R. and Koch, G.G. (1977), “The measurement of observer agreement for categorical data”, Biometrics, Vol. 33 No. 1, pp. 159-174. Lee, L., Petter, S., Fayard, D. and Robinson, S. (2011), “On the use of partial least squares path modeling in accounting research”, International Journal of Accounting Information Systems, Vol. 12 No. 4, pp. 305-328. Lei, P.W. and Wu, Q. (2012), “Estimation in structural equation modeling”, in Hoyle, R.H. (Ed.), Handbook of Structural Equation Modeling, Guilford Press, New York, NY, pp. 164-179. Lohmöller, J.B. (1987), Lvpls 1.8, Lohmöller, Cologne. Lohmöller, J.B. (1989), “Latent variable path modeling with partial least squares”, Physica, Heidelberg. Lopez-Bonilla, J.M. and Lopez-Bonilla, L.M. (2012), “Environmental orientation in tourism: the RTEO scale”, Current Issues in Tourism, Vol. 15 No. 6, pp. 591-596. IJCHM MacKenzie, S.B., Podsakoff, P.M. and Jarvis, C.B. (2005), “The problem of measurement model misspeciﬁcation in behavioral and organizational research and some recommended solutions”, 30,11 Journal of Applied Psychology, Vol. 90 No. 4, pp. 710-730. Marques, C. and Reis, E. (2015), “How to deal with heterogeneity among tourism constructs?”, Annals of Tourism Research, Vol. 52, pp. 172-174. Matzler, K., Strobl, A., Stokburger-Sauer, N., Bobovnicky, A. and Bauer, F. (2016), “Brand personality 3508 and culture: the role of cultural differences on the impact of brand personality perceptions on tourists’ visit intentions”, Tourism Management, Vol. 52, pp. 507-520. Mohd-Any, A.A., Winklhofer, H. and Ennew, C. (2015), “Measuring users’ value experience on a travel website (e-value) what value is cocreated by the user?”, Journal of Travel Research, Vol. 54 No. 4, pp. 496-510. Morley, C. (2012), “Technique and theory in tourism analysis”, Tourism Economics, Vol. 18 No. 6, pp. 1273-1286. Murphy, P., Pritchard, M.P. and Smith, B. (2000), “The destination product and its impact on traveller perceptions”, Tourism Management, Vol. 21 No. 1, pp. 43-52. Nitzl, C. (2016), “The use of partial least squares structural equation modelling (PLS-SEM) in management accounting research: directions for future theory development”, Journal of Accounting Literature, Vol. 37, pp. 19-35. Nunkoo, R., Ramkissoon, H. and Gursoy, D. (2013), “Use of structural equation modeling in tourism research: past, present and future”, Journal of Travel Research, Vol. 52 No. 6, pp. 759-771. Olsson, U.H., Foss, T., Troye, S.V. and Howell, R.D. (2000), “The performance of ML, GLS, and WLS estimation in structural equation modeling under conditions of misspeciﬁcation and nonnormality”, Structural Equation Modeling: A Multidisciplinary Journal, Vol. 7 No. 4, pp. 557-595. Oromendía, A.R., Paz, M.D.R. and Rufín, R. (2015), “Relationship versus transactional marketing in travel and tourism trade shows”, Tourism Economics, Vol. 21 No. 2, pp. 427-434. Peng, D.X. and Lai, F. (2012), “Using partial least squares in operations management research: a practical guideline and summary of past research”, Journal of Operations Management, Vol. 30 No. 6, pp. 467-480. Petrescu, M. (2013), “Marketing research using single-item indicators in structural equation models”, Journal of Marketing Analytics, Vol. 1 No. 2, pp. 99-117. Pike, S., Murdy, S. and Lings, I. (2011), “Visitor relationship orientation of destination marketing organizations”, Journal of Travel Research, Vol. 50 No. 4, pp. 443-453. Poon, W.Y., Leung, K. and Lee, S.Y. (2002), “The comparison of single item constructs by relative mean and relative variance”, Organizational Research Methods, Vol. 5 No. 3, pp. 275-298. Ram, Y., Björk, P. and Weidenfeld, A. (2016), “Authenticity and place attachment of major visitor attractions”, Tourism Management, Vol. 52, pp. 110-122. Rasoolimanesh, S.M., Ringle, C.M., Jaafar, M. and Ramayah, T. (2017a), “Urban vs. rural destinations: Residents’ perceptions, community participation and support for tourism development”, Tourism Management, Vol. 60, pp. 148-158. Rasoolimanesh, S.M., Jaafar, M., Kock, N. and Ahmad, A.G. (2017b), “The effects of community factors on residents’ perceptions toward world heritage site inscription and sustainable tourism development”, Journal of Sustainable Tourism, Vol. 25 No. 2, pp. 198-216. Raykov, T. (2007), “Reliability if deleted, not ‘alpha if deleted’: evaluation of scale reliability following component deletion”, British Journal of Mathematical and Statistical Psychology, Vol. 60 No. 2, pp. 201-216. Reinartz, W., Haenlein, M. and Henseler, J. (2009), “An empirical comparison of the efﬁcacy of covariance-based and variance-based SEM”, International Journal of Research in Marketing, Vol. 26 No. 4, pp. 332-344. Richter, N.F., Sinkovics, R.R., Ringle, C.M. and Schlägel, C. (2016a), “A critical look at the use of Hospitality and SEM in international business research”, International Marketing Review, Vol. 33 No. 3, pp. 376-404. tourism Richter, N.F., Cepeda Carrion, G., Roldán, J.L. and Ringle, C.M. (2016b), “European management research using partial least squares structural equation modeling (PLS-SEM): editorial”, European Management Journal, Vol. 34 No. 6, pp. 589-597. Rigdon, E.E. (2012), “Rethinking partial least squares path modeling: in praise of simple methods”, Long Range Planning, Vol. 45 Nos 5/6, pp. 341-358. 3509 Rigdon, E.E. (2014), “Rethinking partial least squares path modeling: breaking chains and forging ahead”, Long Range Planning, Vol. 47 No. 3, pp. 161-167. Rigdon, E.E. (2016), “Choosing PLS path modeling as analytical method in European management research: a realist perspective”, European Management Journal, Vol. 34 No. 6, pp. 598-605. Rigdon, E.E., Sarstedt, M. and Ringle, C.M. (2017), “On comparing results from CB-SEM and PLS-SEM. Five perspectives and ﬁve recommendations”, Marketing Zfp, Vol. 39 No. 3, pp. 4-16. Ringle, C.M. and Sarstedt, M. (2016), “Gain more insight from your PLS-SEM results: the importance-performance map analysis”, Industrial Management and Data Systems, Vol. 116 No. 9, pp. 1865-1886. Ringle, C. Wende, S. and Will, A. (2005), “SmartPLS 2.0 (beta)”, Hamburg, available at: www.smartpls. com Ringle, C.M., Sarstedt, M. and Straub, D. (2012), “A critical look at the use of PLS-SEM in MIS quarterly”, MIS Quarterly, Vol. 36 No. 1, pp. 3-14. Ringle, C.M., Sarstedt, M., Mitchell, R. and Gudergan, S.P. (2018), “Partial least squares structural equation modeling in HRM research”, The International Journal of Human Resource Management, pp. 1-27, available at: https://doi.org/10.1080/09585192.2017.1416655 Roldán, J.L. and Sánchez-Franco, M.J. (2012), “Variance-based structural equation modeling: guidelines for using partial least squares in information systems research”, in Mora, M., Gelman O., Steenkamp A. and Raisinghani, M. (Eds), Research Methodologies, Innovations and Philosophies in Software Systems Engineering and Information Systems, IGI Global, pp. 193-221. Rönkkö, M. and Evermann, J. (2013), “A critical examination of common beliefs about partial least squares path modeling”, Organizational Research Methods, Vol. 16 No. 3, pp. 425-448. Rönkkö, M., McIntosh, C.N. and Antonakis, J. (2015), “On the adoption of partial least squares in psychological research: caveat emptor”, Personality and Individual Differences, Vol. 87, pp. 76-84. Rönkkö, M., McIntosh, C.N., Antonakis, J. and Edwards, J.R. (2016), “Partial least squares path modeling: time for some serious second thoughts”, Journal of Operations Management, Vol. 47-48, pp. 9-27. Sarstedt, M. and Wilczynski, P. (2009), “More for less? A comparison of single-item and multi-item measures”, Die Betriebswirtschaft, Vol. 69 No. 2, pp. 211-227. Sarstedt, M., Henseler, J. and Ringle, C.M. (2011a), “Multi-group analysis in partial least squares (PLS) path modeling: alternative methods and empirical results”, in Sarstedt, M., Schwaiger, M. and Taylor, C. R. (Eds), Advances in International Marketing, Vol. 22, Emerald, Bingley, pp. 195-218. Sarstedt, M., Becker, J.-M., Ringle, C.M. and Schwaiger, M. (2011b), “Uncovering and treating unobserved heterogeneity with FIMIX-PLS: which model selection criterion provides an appropriate number of segments?”, Schmalenbach Business Review, Vol. 63 No. 1, pp. 34-62. Sarstedt, M., Ringle, C.M. and Hair, J.F. (2017), “Partial least squares structural equation modeling”, in Homburg, C., Klarmann, M. and Vomberg, A. (Ed.), Handbook of Market Research, Springer International Publishing, Berlin, pp. 1-40. Sarstedt, M., Diamantopoulos, A., Salzberger, T. and Baumgartner, P. (2016b), “Selecting single items to measure doubly-concrete constructs: a cautionary tale”, Journal of Business Research, Vol. 69 No. 8, pp. 3159-3167. IJCHM Sarstedt, M., Hair, J.F., Ringle, C.M., Thiele, K.O. and Gudergan, S.P. (2016a), “Estimation issues with PLS and CBSEM: where the bias lies!”, Journal of Business Research, Vol. 69 No. 10, 30,11 pp. 3998-4010. Sarstedt, M., Ringle, C.M., Smith, D., Reams, R. and Hair, J.F. (2014), “Partial least squares structural equation modeling (PLS-SEM): a useful tool for family business researchers”, Journal of Family Business Strategy, Vol. 5 No. 1, pp. 105-115. ScriptWarp Systems (2009), “WarpPLS – PLS-based structural equation modeling (SEM) software”, 3510 available at: http://warppls.com Shmueli, G. (2010), “To explain or to predict?”, Statistical Science, Vol. 25 No. 3, pp. 289-310. Shmueli, G., Ray, S., Estrada, J.M.V. and Chatla, S.B. (2016), “The elephant in the room: predictive performance of PLS models”, Journal of Business Research, Vol. 69 No. 10, pp. 4552-4564. Steckel, J.H. and Vanhonacker, W.R. (1993), “Cross-validation regression models in marketing research”, Marketing Science, Vol. 12 No. 4, pp. 415-427. Stone, M. (1974), “Cross-validatory choice and assessment of statistical predictions”, Journal of the Royal Statistical Society, Vol. 36 No. 2, pp. 111-147. Sui, J.J. and Baloglu, S. (2003), “The role of emotional commitment in relationship marketing: An empirical investigation of a loyalty model for casinos”, Journal of Hospitality and Tourism Research, Vol. 27 No. 4, pp. 470-489. Tenenhaus, M., Amato, S. and Esposito Vinzi, V. (2004), “A global goodness-of-ﬁt index for PLS structural equation modeling”, Proceedings of the XLII SIS Scientiﬁc Meeting in Padova, Italy, 2004, CLEU, pp. 739-742. Tenenhaus, M., Vinzi, V.E., Chatelin, Y.M. and Lauro, C. (2005), “PLS path modeling”, Computational Statistics and Data Analysis, Vol. 48 No. 1, pp. 159-205. Vilares, M.J., Almeida, M.H. and Coelho, P.S. (2010), “Comparison of likelihood and PLS estimators for structural equation modeling: a simulation with customer satisfaction data”, in Esposito Vinzi V., Chin W., Henseler J. and Wang H. (Eds), Handbook of Partial Least Squares, Springer Handbooks of Computational Statistics, Springer, Berlin, pp. 289-305. Whitﬁeld, J., Dioko, L.A. and Webber, D.E. (2014), “Scoring environmental credentials: a review of UK conference and meetings venues using the GREENER VENUE framework”, Journal of Sustainable Tourism, Vol. 22 No. 2, pp. 299-318. Wold, H. (1982), “Soft modeling: Intermediate between traditional model building and data analysis”, Banach Center Publications, Vol. 6 No. 1, pp. 333-346. Wold, H. (1985), “Partial least squares”, in Kotz, S. and Johnson, N.L. (Eds), Encyclopedia of Statistical Sciences, Wiley, New York, NY, pp. 581-591. Yang, J., Ryan, C. and Zhang, L. (2013), “Ethnic minority tourism in China–Han perspectives of Tuva ﬁgures in a landscape”, Tourism Management, Vol. 36, pp. 45-56. Appendix. Evaluation form for PLS-SEM studies Hospitality and tourism 3511 IJCHM 30,11 3512 About the authors Dr Ahmet Usakli is Research Associate in the Department of Tourism Management at Gazi University, Turkey. He holds a doctorate from Gazi University and a master’s degree from the University of Nevada, Las Vegas, sponsored by the US Department of State’s prestigious Fulbright Scholarship Program. His research interests revolve around consumer behavior in travel and tourism, destination marketing and management and statistical and methodological aspects of hospitality and tourism research. Ahmet Usakli is the corresponding author and can be contacted at:

[email protected]

Dr Kemal Gurkan Kucukergin is Assistant Professor in the Department of Tourism and Hotel Management at Atılım University, Turkey. His research interests include destination marketing and tourist behavior. The majority of his research involves the use of covariance- and variance-based structural equation modeling. His research interests include destination marketing and tourist behavior. The majority of his research involves the use of covariance- and variance-based structural equation modeling. For instructions on how to order reprints of this article, please visit our website: www.emeraldgrouppublishing.com/licensing/reprints.htm Or contact us for further details:

[email protected]

(PDF) Using partial least squares structural equation modeling in hospitality and tourism: Do researchers follow practical guide