V olume 26 Issue 3 Ar ticle 1 September 2024 A Classification Methodology for Assessing Countries in T erms of A Classification Methodology for Assessing Countries in T erms of T ourism Competitiv eness T ourism Competitiv eness Yiannis Smirlis Univ ersity of Pir aeus, School of E conomics, Business and International Studies, Pir aeus, Gr eece Marilou Ioakimidis National and Kapodistrian Univ ersity of A thens, Depar tment of Business Administr ation, A thens, Gr eece , mioak eim@ba.uoa.gr F ollow this and additional works at: https:/ /www .ebrjournal.net/home P ar t of the T ourism and T ra v el Commons Recommended Citation Recommended Citation Smirlis, Y ., & Ioakimidis, M. (2024). A Classification Methodology for Assessing Countries in T erms of T ourism Competitiv eness. E conomic and Business Re view , 26 (3), 151-167. https:/ /doi.or g/10.15458/ 2335-4216.1339 This Original Ar ticle is br ought t o y ou for fr ee and open access b y E conomic and Business Re view . It has been accepted for inclusion in E conomic and Business Re view b y an authoriz ed edit or of E conomic and Business Re view . ORIGINAL ARTICLE A Classication Methodology for Assessing Countries in Terms of Tourism Competitiveness YiannisSmirlis a ,MarilouIoakimidis b, * a University of Piraeus, School of Economics, Business and International Studies, Piraeus, Greece b National and Kapodistrian University of Athens, Department of Business Administration, Athens, Greece Abstract Tourism industry is important for national economies, and, in this regard, it is vital to monitor its competitiveness. The Travel and Tourism Competitiveness Index (TTCI), developed and reported by the World Economic Forum, serves this purpose by providing a consistent framework, explanatory factors, and corresponding data sets. In this paper we exploit the past data sets of this index, rst, to verify that in several countries, competitiveness in tourism and travel industries hardly changes over time. Next, we identify countries that show consistency in travel–tourism competitive- ness and separate them into classes of best, worst, intermediate, and ambiguous past performance. Building on such a classication, we apply linear discriminant analysis (LDA) as an alternative to the TTCI computational framework in order to compose the new synthetic index TTCI-LDA, which assesses countries’ competitiveness. The analysis of country scores obtained from this index has revealed that ICT readiness and touristic service infrastructure are important for tourism competitiveness. The score thresholds for the best–worst country cases in each class provide additional useful information for management, benchmarking, and policy decision making. Keywords: Travel and Tourism Competitiveness Index, Linear discriminant analysis, Weighting of composite indicators, Classication JEL classication: C38, L83 Introduction T he travel and tourism industry plays an in- creasingly important role in the development of countries’ economies. Therefore, it is vital rst to identify, measure, and aggregate the factors that af- fect countries’ progress in this sector. This need is served and supported by the Travel and Tourism Competitiveness Index (TTCI), hereinafter referred to as TTCI-WEF, developed and rst published in 2007 by the World Economic Forum (WEF, 2019). The last update of TTCI-WEF was reported in 2019 and after that, it has been restructured to form the Travel & Tourism Development Index (TTDI) (WEF, 2022). The TTCI framework is regarded as the most popular, comprehensive, and systematic collection of data related to travel and tourism competitiveness (Salinas-Fernández et al., 2020). For the assessment and benchmarking of the coun- tries, the TTCI original World Economic Forum methodology proposes a conceptual model and a number of factors organized in a hierarchy, on the top level of which there are four main dimensions (subindices) of competitiveness, namely Enabling Environment (e.g., safety and security and human re- sources and labor market), Travel and Tourism Policy and Enabling Conditions, Infrastructure, and Natu- ral and Cultural Resources. At the next lower levels, TTCI has 14 pillars and 90 individual indicators (see Fig. A1 in the Appendix, which briey presents the TTCI-WEF structure). The latest report (WEF, 2019) for the year 2019 evaluates 140 countries (economies), which account for over 98% of world GDP . It is important to emphasize that TTCI-WEF follows the common approach applied to composite indica- tors (CIs), that is, to use the data of the current period Received 25 August 2023; accepted 13 June 2024. Available online 16 September 2024 * Corresponding author. E-mail address: mioakeim@ba.uoa.gr (M. Ioakimidis). https://doi.org/10.15458/2335-4216.1339 2335-4216/© 2024 School of Economics and Business University of Ljubljana. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/ licenses/by-nc-nd/4.0/). 152 ECONOMIC AND BUSINESS REVIEW 2024;26:151–167 to estimate the performance of the countries, ignoring the historical progress that a country may have had. An inspection of TTCI-WEF scores reported for previ- ous years reveals that the best- and worst-performing countries remain approximately in the same ranking positions, and this is hardly improved or worsened from year to year. This is explained by the fact that an exceptional improvement of a country in the travel and tourism industry needs a great deal of economic, administrative–legislative, infrastructural, and politi- cal changes, the result of which is visible and reected in the TTCI in a horizon of several years. This obser- vation raises the idea that countries’ performance in previous years could form the basis for classifying countries in terms of performance, and this classi- cation information could be a starting exploitation point for the reassessment of the TTCI-WEF index. Therefore, for the assessment of countries based on the TTCI-WEF, it is feasible to accept, as initial pref- erence information, a classication of countries in groups of potentially best, worst, and intermediate performance. The concept of using such preference information in the form of a broad classication for the development of CIs has been presented by Smirlis (2020) for the estimation of the Digital Economy and Society Index (DESI). This paper proposes a new methodology for the reassessment of TTCI-WEF: the initial classication of the countries into groups of high, medium, and low performance, combined with a detailed prole of each country in the current year as it is expressed by the scores in the TTCI-WEF pillars, create the neces- sary input to a statistical linear discriminant analysis (LDA). In a next step, intermediate measurements of the LDA procedure are used to calculate proper subindicator weights that could interpret and explain the classication. This approach aims to bring a new idea for constructing CIs and at the same time to deal with the problem of imposing equal importance to indicators/pillars due to a simple arithmetic average formula, which has been a main subject of strong crit- icism in the construction methodology of the original TTCI-WEF index (see next section). The paper has the following structure. Section 1 in- cludes a review of past publications that propose new computational methodologies for the TTCI-WEF and presents a short overview of the proposed LDA-based approach. Section 2 presents the methodological part of constructing the proposed TTCI-LDAindex. In this, based on a weighted sum formula for the aggregation of the pillars, we present the LDA formulations for the estimation of the weights. Section 3 presents the data analysis of the past TTCI-WEF scores, the imple- mentation of LDA, the estimations of the weights, and the comparison of the proposed TTCI-LDA country scores to those of the original TTCI-WEF. Section 4 concludes the method and discusses the results. 1 Literature review and methodology overview CIs measure multidimensional and complex con- cepts and phenomena (Greco et al., 2019). Their construction is based on individual subindicators that measure various dimensions and comply with a the- oretical framework and an underlying model of the concept to be measured (Nardo et al., 2008). In the process of the construction, weighting and aggregation are among the main issues to be con- sidered. The weighting determines the assignment of an explicit importance to the subindicators, while the aggregation refers to the mathematical operations for combining the values of indicators into a sin- gle summary (Nardo et al., 2008). For the weighting problem, a plurality of methods has been proposed, among them the data-driven, originated by multi- variate statistics factor/principal component, cluster, correspondent, canonical correlation analysis, and so forth (Greco et al., 2019; Nardo et al., 2008). Particularly in the eld of travel and tourism com- petitiveness, the work of Mendola and Volo (2017) describes methodological foundations to build CIs and evaluates the currently available CIs. Among them, TTCI-WEF is a noteworthy contribution, aim- ing to evaluate the set of factors that enable the sustainable development of travel and tourism (WEF, 2019). This index has offered a consistent frame- work and a credible and accurate data set since 2007 (Abreu-Novais et al., 2016). It has been extensively used for research in the travel and tourism sector, allowing direct comparison of countries (Dwyer et al., 2014; Kayar & Kozak, 2010). In its recent 2019 report, TTCI-WEF evaluates 140 economies by using a hierarchical structure (see Fig. A1—Appendix) composed of 4 subindices, 14 pillars, and 90 low-level indicators. In detail, subindex A—Enabling Environment—includes 5 pil- lars and captures the general conditions necessary for operating in a country. Subindex B—T&T Pol- icy and Enabling Conditions—is analyzed in 4 pil- lars and includes indicators for policies and strate- gic aspects that impact the travel–tourism industry. Subindex C—Infrastructure—has 3 pillars that mea- sure the availability and quality of physical infrastruc- ture in each economy. Finally, subindex D—Natural and Cultural Resources—, which has 2 pillars, cap- tures the principal “reasons to travel.” The calculation of the TTCI-WEF score of each country is performed on a bottom-up basis. First, ECONOMIC AND BUSINESS REVIEW 2024;26:151–167 153 the response data from the World Economic Forum’s Executive Opinion Survey are used to grade the coun- tries’ performance on the low-level indicators on a scale of 1 (worst) to 7 (best). Then, the scores in pil- lars are calculated as an arithmetic average of their constituent low-level indicators’ scores; the scores in subindices are derived as averages of their con- stituent pillars, and nally the TTCI-WEF total score results from the average score in the 4 subindices. This arithmetic average formula used in TTCI-WEF im- plicitly denes weights that account for the number of subindices and pillars. For example, at the upper level, which includes 4 subindices, the weight for each subindex score is 1=4D 0:25; at the lower level, the weight of the pillars in subindex A (Enabling Envi- ronment) is 0:25=5D 0:05 as it includes 5 pillars, while the weight of the pillars in subindex D (Natural and Cultural Resources) will be 0:25=2D 0:125 as only two pillars exist in this subindex. In this manner, TTCI- WEF assigns equal weight values to the pillars in every subindex and equal weights to all subindices. Conceptually, this is interpreted as attributing equal contribution of subindices or pillars to tourism com- petitiveness, independently of their context. Although TTCI-WEF represents a widely accepted approach to measuring tourism competitiveness, it has been criticized (Croes & Kubickova, 2013) for both its conceptual model of tourism competitive- ness and the index composition methodology and consequently the reliability of the measurement. At the conceptual level of the TTCI-WEF, Ring (2011) restructured the pillars and subindicators to form four additional models. Then, by comparing the de- rived scores, she examined how pillars inuence the countries’ competitiveness. One of the results is that Natural Resources and Environmental Sus- tainability, considered by some other studies as the main factors of attractiveness in tourist destinations, have not been proven signicant. Kunst and Ivandi ´ c (2021) ascertain the methodological shortcomings of TTCI-WEF by using the Mediterranean countries as a sample and observe that the variation of scores does not match signicantly with international ar- rivals and inbound tourism expenditures taken as proxies of performance-related tourism activity. The authors suggest the application of equal weights to pillars and subindicators along with the addition of new indicators. As far as the issue of the TTCI-WEF methodol- ogy is concerned, a number of publications criti- cize its above-mentioned equal weights arrangement (Crouch, 2007; Rodríguez-Díaz & Pulido-Fernández, 2020) as unrealistic and not representing the ac- tual tourism and travel competitiveness (Wu et al., 2012). Furthermore, the arithmetic average formula for the aggregation of the components entails that the higher the number of indicators used to calcu- late the score of the upper-level CI, the lower their weight value is, and thus their importance is reduced. To address this drawback, several attempts to recon- struct the TTCI-WEF have been published in the past years, focusing on the problem of reestimating the weights. From a methodological viewpoint, it is pos- sible to distinguish between publications employing multivariate statistics and those that use multicri- teria decision making or linear programming. For the multivariate statistical approaches, Mikuli´ c et al. (2015), by comparing past studies, highlight the re- quirement of strong correlation among the pillars and subindicators. Moreover, Koži´ c and Mikuli´ c (2014), comparing Croatian coastal destinations, juxtapose three different procedures for weighting sustainabil- ity and raise the argument that the application of factor analysis is questionable. In this group, indica- tive publications are as follows. Lin and Huang (2009) propose grey relational and sensitivity analysis in order to evaluate the tourism-competitive potential in Asian countries. Mazanec and Ring (2011) ex- amine the predictive power of the TTCI data by applying different computational methods (partial least squares path modelling, PLS regression, mixture modelling, and non-linear covariance-based struc- tural equation modelling) and conclude that different unobserved factors and complicated relations affect tourism competitiveness. Lan et al. (2012) combine an expectation–maximization (EM) clustering algo- rithm with an articial neural network structure to group countries into three classes and thus obtain an objective weighting system for the 14 pillars of TTCI. The resulting weights indicate high importance in six pillars, namely Tourism Infrastructure, Ground Transport Infrastructure, Air Transport Infrastruc- ture, Cultural Resources, Health and Hygiene, and ICT Infrastructure. Mili´ c and Jovanovi´ c (2019) con- tribute to the problem of weighting TTCI pillars by employing factor/principal component analysis to obtain the weights of each pillar and therefore new country rankings. Salinas-Fernández et al. (2020) esti- mate new weights for the TTCI pillars by introducing a DP 2 distance-based method to derive a synthetic indicator that linearly aggregates the distances of each country relative to the least desirable situation. Then they apply factor analysis to explore the underly- ing dimensions. In terms of pillars’ signicance, this approach concludes that ICT Readiness and Prioriti- zation of Travel & Tourism exert the greatest inuence in determining the nal index, while Natural Re- sources have the least inuence. Litavcová and Sí ˇ c (2021) use a quantile regression approach to examine how the four pillars of the T&T Policy and Enabling 154 ECONOMIC AND BUSINESS REVIEW 2024;26:151–167 Conditions contribute to the TTCI-WEF scores. In the multicriteria-decision-making group of publications, the work of Pulido-Fernández and Rodríguez-Diaz (2016) introduces two reference points for each pillar, an aspiration and a reservation level, and, by building on these, estimates a weak index and a strong index, which are nally combined into a composite one used for the country rankings. Gomez-Vega and Picazo- Tadeo (2019) used multicriteria and data envelopment methods to estimate new weights for the TTCI-WEF. Gomez-Vega and Picazo-Tadeo (2019) propose a re- gression and bootstrapping method based on the benet of the doubt principle of the data envelopment analysis modelling approach (Despotis, 2005) to esti- mate the weights of the pillars endogenously. In this paper, we propose LDA as the main method- ological technique to estimate new weights for the TTCI-WEF pillars, following the stream of multivari- ate statistical approaches. Discriminant analysis has not been very common for aggregating factors in CIs, particularly in tourism. In other elds, indica- tive works are that of Gupta et al. (1994), who used this method to calculate relative weights of human rights indicators to evaluate countries in terms of hu- man rights abuse and violations, and that of Iwuagwu and Nwosu (2021), who use LDA as a basis to model the Human Development Index. We construct the new TTCI-LDA index by considering the 14 pillars of the TTCI-WEF structure as variables contributing independently to travel and tourism competitiveness. Based on the country scores in these pillars, we cal- culate the new TTCI-LDA scores for the countries by using a weighted sum formula, with the pillar weights to be derived directly from the LDA statis- tical procedure. The steps followed in the TTCI-LDA construction process are: (i) classication of countries in terms of their past performance scores, (ii) application of the LDA, estimation of new weight values for the pillars, and calculation of the new country scores, (iii) validation of the scores obtained. In step (i), we analyze the past TTCI-WEF data for the years 2007–2017, and by employing a sta- tistical procedure (percentile distribution thresholds, hypothesis testing and condence interval estima- tions), we select those countries that show consistency in travel–tourism competitiveness and separate them into classes of best, worst, and intermediate past performance. The rest of the countries, showing vari- ability in their past TTCI-WEF scores and ambiguous rankings, are left unclassied. In step (ii), the four-class categorization of the coun- tries (best, intermediate, worst, and ambiguous past performance) together with the country scores in the 14 pillars for the recent year 2019 are considered in- put to a LDA procedure from which the new pillar weights derive. By denition, these weight values re- ect the impact and contribution of each pillar to the separation of classes. The new pillar weight values are then used to calculate the total score for all coun- tries, including those previously left unclassied due to ambiguous performance. In step (iii), the necessary validation with external data sources is performed. 2 Estimation of the weights using Linear Discriminant Analysis To formulate the problem of aggregating and weighting of subindicators in order to construct a composite indicator, we assumed that in the gen- eral case, n countries ADfa 1 ; a 2 ;:::; a n g( jD 1;:::; n) have to be assessed on X 1 ; X 2 ;:::; X m subindicators so that the total performance of country a j is de- rived from the m-dimensional level of achievement of achievement vector (x 1 j ; x 2 j ;:::; x m j ). The common approach for the aggregation of the m subindicators is to employ the weighted average formula (1) I j D m X iD1 w i x i j (1) In formula (1) the weights w i ; iD 1;:::; m are scaling positive variables. In our indicator approach, the 14 pillars (mD 14) were considered as the constituent subindicators, and the weights w i ; iD 1;:::; 14 were estimated by ap- plying an LDA statistical procedure, as is explained latter in this section. Then, the countries’ scores I j ; jD 1;:::; 140 derived directly from (1), as the pillar scores x 1 j ; x 2 j ;:::; x 14 j were known from the original TTCI- WEF calculations. However, it is worth noting that formula (1) is actually the same as the one used by the original TTCI-WEF and that the corresponding weight values are those estimated by its arithmetic average calculation (see column 6, Table 4). Additionally, we assume that it is feasible to classify a number of countries in terms of their perfor- mance and use this classication as initial preference information. Let C 1 ;C 2 ;:::;C K denoted K in num- ber classes of decreasing level of performance, with the best performing countries belonging to class C 1 , the worst performing to class C K , and the classes C 2 ;:::;C K 1 including countries of intermediate lev- els of total performance. Note that several countries may have been left unclassied, as there may have been no strong indication of their level of per- formance. These countries comprised class C ? . For ECONOMIC AND BUSINESS REVIEW 2024;26:151–167 155 simpler modelling, we accept that all the dened classes are non-empty (C 1 ;C 2 ;:::;C K ;C ? 6D;), that a country should belong to only one class or be left unclassied ( C 1 \ C 2 \:::\ C K \ C ? D;) and that no country is left without initial classication mem- bership C 1 [ C 2 [:::[ C K [ C ? D A. The order of the classes, due to the common, positive weight values in (1), implies a monotonic relation on the total scores I j , that is, I j 1 I j 2 ;8a j 1 2 C k 1 ; a j 2 2 C k 2 , where class C k 1 is superior to class C k 2 . In the case of TTCI-WEF, which was under consid- eration, there were three dened classes ( KD 3), set as C 1 best, C 2 intermediate, and C 3 worst. This parti- tioning derived from the past TTCI-WEF data for the years 2007–2017 in the statistical process of step (i) of our methodology, as explained in Section 3. The initial classication of the countries was to be further exploited by LDA to estimate new weights of the pillars. LDA, introduced by Fisher (1936), is a popular multivariate statistical method to model, an- alyze, and predict classications of observations. The mechanism of LDA arranges linear transformations on the data set to dene K 1 in numbered linear functions F 1 ; F 2 ;:::; F K 1 , called discriminant functions. These adopted the initial classication suggested and achieved the best possible separation of countries. The rst function, F 1 , is the most powerful in ex- pressing the differentiation of the classes, while the next, F 2 ;:::; F K 1 , have decreasing importance. Ev- ery discrimination function F k is associated with an eigenvalue, l k , which denotes the amount of vari- ance between the classes explained by this function. The normalization of the eigenvaluesl 1 ;l 2 ;:::;l K 1 in the ratio 8 k D l k P K 1 iD1 l k can represent the relative contribution of the discriminant function F k to the total discriminating power of the model (Hair et al., 2010). In addition to the discriminant functions and the eigenvalues, a typical LDA procedure also re- ports the structure coefcients , which, for the case under consideration, denote the correlation between the discriminant functions and the pillars, regarded as variables. We considered r ik to be such a structure coefcient between the i th pillar X i and the k th discrim- inant function F k . The combination of the correlation coefcients r ik and the ratios 8 k creates the potency index PI i in formula (2). PI i D K 1 X kD1 r 2 ik 8 k (2) The potency index (2) was initially proposed by Perreault et al. (1979) as an aggregative measure that summarizes information across different func- tions. Hair et al. (2010) denes the potency index as the “composite measure of the discriminatory power of an independent variable when more than one discriminant function is estimated. Based on the dis- criminant loadings it is a relative measure used for comparing the overall discrimination provided by each step independent variable across all signicant discriminant loadings”. It has been used in conjunc- tion with LDA in different elds (e.g., Njoku, 2013; Shobha & Siji, 2018) to indicate the order of impor- tance of the variables used. In the case of TTCI-LDA, the potency index expresses how signicantly pillar i participates in the separation of classes. In this man- ner, the weight values needed for the calculation of total performance of the countries in (1) can derive from the normalization formula (3) of PI i . w i D PI i P m iD1 PI i (3) The denition of the weights in (3) implies that P m iD1 w i D 1, so each weight w i will explain, in terms of percentage, to what extent pillar i contributes to the countries’ classication. LDA also has predictive power to assign the unclas- sied countries in C ? to the previously dened classes C 1 ;C 2 ;:::;C K , thus obtaining a full classication. This was achieved rst by calculating the class centroids as the within-class average scores of the discriminating functions and then by calculating the probability of a country to belong to a class by evaluating its dis- tance from the corresponding class centroid (Huberty & Olejnik, 2006). It is important to point out that for the reliability of the analysis, LDA imposes strong assumptions on the data: the size of the smallest group must be larger than the number of predictor variables (size of the prob- lem); the predictor variables should approximately follow the normal distribution within each class (mul- tivariate normality); the variance/covariance struc- ture of indicators should be the same among classes (homoscedasticity); the correlation (multicollinearity) between indicators should be insignicant, and the indicators’ scores have to be independent of each other (independence) (Huberty, 1994; Watson, 1982). Such assumptions limit the application of LDA in practical problems somewhat. Quadratic discrimi- nant analysis (Friedman et al., 2009) does not impose strict statistical properties such as the equal covari- ance matrices between classes but does not allow linear form for the discriminant functions. 3 Classication of countries and estimation of the new TTCI-LDA index In this section, we apply the steps (i)–(iii) described in Section 1 in order to reassess countries in terms of 156 ECONOMIC AND BUSINESS REVIEW 2024;26:151–167 Table 1. One sample t-test at level of signicance aD 5% against the estimated thresholds. Class Condition Hypotheses C 1 Top 10% Average score greater than t 1 D 5:23 H 0 : ¯ X t 1 H 1 : ¯ X < t 1 (1-tail) C 2 Average Average score approximately equal to t 1 D 4:29 H 0 : ¯ XD t 2 H 1 : ¯ X6D t 2 (2-tail) C 3 Bottom 10% Average score lower than t 3 D 2:89 H 0 : ¯ X t 3 H 1 : ¯ X > t 2 (1-tail) tourism competitiveness. First, based on TTCI-WEF scores for the years 2007–2017 (TCdata360, 2019), we develop a statistical process to detect those countries that had consistent performance in these years and distinguish them in classes of high-, intermediate-, and low-level total performance. Then a classication of the countries is performed, and according to the weighting method presented in the previous section, the new TTCI-LDA country scores are estimated. 3.1 Classication of countries in terms of their past performance scores The denition of classes is based on the TTCI-WEF country total scores in the past years 2007, 2009, 2011, 2013, 2015, and 2017. Table A1 and Fig. A2 in the Appendix present the basic descriptive statistics and the relative box-plot graphs, respectively, for that pe- riod. The inspection of these reveals that there is no signicant variation of TTCI scores among these years, except for a small decrease observed for the recent period 2015–2017. The three classes of performance are dened as follows. The best performing class, C 1 , includes the countries of the 90th percentile of the TTCI-WEF scores; the worst performing class, C 3 , the countries belonging to the 10th percentile, and C 2 the coun- tries with an intermediate level of performance. Let t 1 be the maximum 90th percentile value of past scores, t 3 the minimum 10th percentile value, and t 2 the av- erage 50th percentile value, that is t 1 D maxfP t 90 ; tD 2007; 2009; 2011; 2013; 2015; 2017g t 2 D averagefP t 50 ; tD 2007; 2009; 2011; 2013; 2015; 2017g t 3 D minfP t 10 ; tD 2007; 2009; 2011; 2013; 2015; 2017g (4) (the notation P t q stands for the qth percentile of the distribution of TTCI scores at year t). From the corresponding data, the resulting thresh- old values are t 1 D 5:23; t 2 D 4:29, and t 3 D 2:89. These values were further utilized in hypothesis tests to in- dicate the initial class membership of each country. Table 1 presents the classes, the condition set, and the associated hypothesis test. Positive test results (acceptance of the null hypoth- esis H 0 at signicance level 95%, p-value> .05) were conrmed for 58 countries in total, of which 16 were classied in class C 1 , 25 in class C 2 , and 17 in class C 3 . These countries are listed in Table 2. The other 82 out of 140 total ( 59%) assumed to have unknown group membership and assigned to class C ? . These, either had a rejection of the null hypothesis H 0 or appeared with insufcient data. The classication result of Table 2 is compara- ble with the WEF report 2019 (WEF, 2019), which identies Spain, France, Germany, Japan, the United Table 2. List of countries assigned to classes C 1 ;C 2 ;C 3 . Class C 1 Class C 2 Class C 3 Number of countries 16 25 17 Countries: Austria Bahrain Montenegro Angola Australia Brazil Oman Bangladesh Canada Bulgaria Panama Burundi France Chile Poland Burkina Faso Germany Costa Rica Qatar Cameroon Hong Kong SAR Dominican Rep. Russian Fed. Chad Japan Hungary Seychelles Guinea Luxemburg Israel Slovak Rep. Haiti Netherlands Jamaica Slovenia Lesotho New Zealand Jordan Tunisia Malawi Singapore Latvia Turkey Mali Spain Lithuania Uruguay Mauritania Switzerland Mauritius Mozambique Sweden Nigeria United Kingdom Pakistan United States Sierra Leone Yemen ECONOMIC AND BUSINESS REVIEW 2024;26:151–167 157 States, United Kingdom, Australia, Italy, Canada, and Switzerland as the top 10 countries. The same report certies the main underlying concept of the study for score stability in successive years by stating, “the top 10 TTCI scorers remain the same ::: Spain is the top performer for the third consecutive report:::”. In an alternative but equivalent statistical reformu- lation of the previous process, the countries’ total scores in years 2007, 2009, 2011, 2013, 2015, and 2017, considered as values of a random variable X, may be regarded as a small sample (nD 6) retrieved from a Student-t approximated distribution. In this manner, it is possible to calculate a 95% condence interval for the unknown average TTCI-WEF score ¯ X for each country and test it against the previously dened thresholds t 1 D 5:23, t 2 D 4:29, and t 3 D 2:89. Fig. 1 exhibits such intervals for the countries included in the 2019 assessment with vertical lines. The dot inside each vertical line represents the average score in years 2007–2017. The class membership of the countries can be tested visually in this chart by inspecting whether the horizontal lines of the thresholds intersect the ver- tical lines corresponding to countries’ score ranges. For example, the condence interval for Austria is es- timated to [4.92, 5.56], and as this includes the class 1 threshold t 1 D 5:23, it classies Austria in the rst high class. Pakistan with an estimated interval score [2.78, 3.40] is placed in the third class due to the value of the t 3 D 2:89 threshold. China with expected scores in [4.37, 4.63] is left unclassied as this interval lies be- tween and does not include the class 1 and 2 threshold values. Regarding Fig. 1, a few more comments are pos- sible. First, the relatively large condence interval observed in countries such as Albania, Barbados, Denmark, Egypt, Finland, Iceland, Luxemburg, Swe- den, Tunisia, and so forth is a sign of high variability. On the opposite side, small ranges in condence inter- vals in countries such as Argentina, Brazil, Colombia, France, Germany, South Africa, Sri Lanka, and Spain are an indication of consistent behavior in past years. 3.2 Application of the LDA, estimation of new weight values for the pillars, and calculation of the new country scores The application of LDA to the data set of the 58 classied countries (Table 1) using the 14 pillars as predictor factors was feasible as all statistical prop- erties assumed were satised. First, the size of the smallest class, C 1 , was equal to 16, which was larger than the number of pillars (14). Second, the multivari- ate normality of the pillars was tested and conrmed by using quantile–quantile (Q–Q) graphs (not in- cluded here for the economy of the presentation). Table 3. Class centroids coordinates. Class F 1 F 2 C 1 5.94 1.68 C 2 1.44 1.60 C 3 7.71 0.78 Third, the Box’s M test performed marginally con- rmed the equity of the covariance matrices at a signicance level of 95% as it was associated with a p-value of .059, which marginally enabled its ac- ceptance. Fourth, a Wilks’ Lambda test indicated the signicance of the pillars ( pD .00) for the discriminat- ing problem, and nally, the conceptual scheme and consequently the denition of pillars in TTCI implied that they were independent, representing different dimensions of travel and tourism competitiveness. The multicollinearity test (IBM, 2021) between the 14 pillars, regarded as predictor variables for the classication, showed values of VIF (variance ina- tion factor) between 1.65 and 9.53, that is, less than 10, which is an empirical threshold value to indi- cate serious multicollinearity requiring correction. The maximum VIF value has been detected in A.5 ICT Readiness marginal correlation. Following the analysis of LDA, due to the three classes dened, there were two discrimination func- tions, F 1 ; F 2 . The rst function, F 1 , explained 93.1% of the vari- ance; the eigenvalue was estimated as l 1 D 29:623, and the canonical correlation was equal to 0.984. The second function, F 2 , explained 6.9% of the variance; the eigenvalue was estimated as l 2 D 2:203, and the canonical correlation was equal to 0.828. The estimated class centroids in terms of the two discrimination functions’ normalized scores are re- ported in Table 3. Based on the class centroids, the membership prob- abilities calculated for the unclassied C ? countries predicted their classication as follows: 9 countries were assigned to class C 1 , 44 countries to C 2 , and 29 countries to C 3 . There were no misclassication errors reported. The classication of the countries in terms of the two discriminating functions is graphically presented in the charts of Fig. 2. The rst chart presents the initially classied countries in Table 2, the unclassi- ed countries (which appear with grey symbols), and the estimated class centroids, shown in Table 3. The second chart of Fig. 2 presents the nal classica- tion result, after the implementation of LDA and the prediction of class membership for the unclassied countries. The rst outcome of LDA is the class membership prediction for the 82 countries that were not included 158 ECONOMIC AND BUSINESS REVIEW 2024;26:151–167 Fig. 1. Chart of 95% condence intervals for country TTCI scores in 2007–2017. ECONOMIC AND BUSINESS REVIEW 2024;26:151–167 159 Fig. 2. Graphical representation of classes in terms of the discrimination functions, before and after the discriminant analysis. Fig. 3. Mean score of TTCI 2019 pillars for countries’ classes; 1—high, 2—medium, and 3—low competitiveness. in the rst assignment (Table 2). The third column of Table A2 in the Appendix presents the predicted class, thus providing a full classication in classes of high, medium, and low level of tourism competi- tiveness. This classication further enables to explore the differences between these three groups. By cal- culating the average country score in all fourteen pillars (Table A3 in the Appendix) for the whole data set (including the countries with predicted classes), it is feasible to determine similarities and differ- ences. Fig. 3 presents the mean scores in the pillars graphically. The predominating pattern of Fig. 3 is that, in al- most all pillars, class 1 surpasses class 2, which in turn surpasses class 3. The exception is B3—Price Competitiveness, in which class 1 has lower mean performance. This evidence is mentioned in the WEF report (WEF, 2019), which indicates “the top 25% (countries) tend to greatly outscore the global aver- age on all pillars apart from Price Competitiveness”. This is explained by the fact that countries in the highest class (1) are mainly advanced economies with strong business environments, good safety conditions and healthcare standards, superior human resources and labour market circumstances, increased level of ICT services and readiness, and so forth, comprising rather expensive destinations compared to the coun- tries of the other two classes. According to Fig. 3, a signicant difference in mean scores between class 1 and classes 2–3 is detected in D2—Cultural Resources and Business Travel and C1—Air Transport Infrastructure. Furthermore, coun- tries of class 3, compared to those of classes 1–2, are inferior in terms of competitiveness in A3—Health 160 ECONOMIC AND BUSINESS REVIEW 2024;26:151–167 Table 4. LDA intermediate calculations and estimated weights. TTCI-LDA Function 1 Function 2 Potency normalized TTCI-WEF Pillar l 1 D 29:623 r 1k l 2 D 2:186 r 2k index PI i weight w i weight w i Difference A1—Business environment .243 .204 .0570 .0483 .050 .002 A2—Safety & security .195 .074 .0316 .0268 .050 .023 A3—Health & hygiene .381 .294 .1435 .1216 .050 .072 A4—Human resources and labor market .192 .191 .0362 .0306 .050 .019 A5—ICT readiness .501 .052 .2844 .2409 .050 .191 B1—Prioritization of travel & tourism .55 .150 .0671 .0568 .063 .006 B2—International openness .265 .178 .0510 .0432 .063 .020 B3—Price competitiveness .235 .077 .0140 .0119 .063 .051 B4—Environmental sustainability .091 .3 .0202 .0171 .063 .046 C1—Air Transport infrastructure .136 .214 .1341 .1136 .083 .031 C2—Ground & port infrastructure .364 .385 .0784 .0664 .083 .017 C3—Tourist service infrastructure .284 .202 .2143 .1815 .083 .099 D1—Natural resources .476 .095 .0183 .0155 .125 .110 D2—Cultural resources & business travel .134 .195 .0304 .0257 .125 .099 & Hygiene, A5—ICT Readiness, C1—Air Transport Infrastructure, and C3—Tourist Service Infrastructure due to their lower level of development. Table 4 presents the intermediate calculations and the weights of the pillars (rst column) estimated by this method. The second and third column of Table 4 present the structure coefcients as reported by the application of the LDA procedure; the fourth column lists the scores of the potency index calculated for each pillar from formula (2); the fth column the weight values estimated by formula (3), and, for com- parison reasons, the sixth column lists the weights assigned by the original TTCI-WEF methodology. The last column of Table 4 presents the difference between the values of the fth and sixth columns, that is, the initial TTCI-WEF weight and the proposed nor- malized weight value, according to the TTCI-LDA. Positive values show that the pillar was assigned greater weight value in the proposed TTCI-LDA ver- sion of the index (e.g. D1—Natural Resources), while negative values show a lower weight value (e.g. A5— ICT Readiness). According to the results presented in Table 4, the most decisive factors that distinguish countries and explain their differentiation in terms of travel– tourism competitiveness are those that appear with the highest potency index, PI i , and related with the highest absolute values r 1k and r 2k . These are A5—ICT Readiness, C3—Tourist Service Infrastruc- ture, A3—Health & Hygiene, and C1—Air Transport Infrastructure. For them, the difference in the last column of Table 4 is negative, denoting higher TTCI- LDA weight values, compared to the corresponding TTCI-WEF. These four pillars were detected as the most signicant for the countries to develop their competitiveness and improve their position in the countries’ ranking. Note that the above-mentioned pillar priority is comparable with other past publi- cations. Rodríguez-Díaz and Pulido-Fernández (2020) indicated that ICT Readiness is signicant for tourism competitiveness, and the work of Lan et al. (2012) mentions that C1—Air Transport Infrastructure and C3—Tourist Service Infrastructure have a signicant inuence in their proposed index. On the opposite end, the lowest weight values were assigned to B3—Price Competitiveness (.0119), D1—Natural Resources (.0155), B4—Environmental Sustainability (.0171), and D2—Cultural Resources & Business Travel (.0257). This result is similar to that reported in the work of Ring (2011), who identied pillars D1—Natural Resources and B4— Environmental Sustainability as least signicant. No- tably, two Enabling Conditions—A2—Safety and Se- curity (.0268) and A4—Human Resources and Labor Market (.0306)—also had comparatively low weight values. The latter result suggests that while much of the tourism industry is decient in the HRM areas of working conditions, training, and pay (Baum, 2007), there is not much difference in these key human re- lations practices between more and less competitive countries. The new estimated weight values differ from those of the original TTCI-WEF index. Due to the equal contribution of the four subindicators A, B, C, and D, the pillars at the immediate lower level share the same weight values. This value is smaller in subindi- cators with a greater number of pillars and larger with a smaller number. For example, in subindica- tor D—Natural and Cultural Resources, the pillars D1—Natural Resources and D2—Cultural Resources & Business Travel, being only two, have the greatest weight value of .125, while in subindicator A— Enabling Environment, the pillars’ weight is only .050. Under this view, pillars D1 and D2 seem to have ECONOMIC AND BUSINESS REVIEW 2024;26:151–167 161 Fig. 4. Box plots to compare TTCI-WEF and TTCI-LDA scores in the three country classes. the greatest contribution to travel–tourism competi- tiveness, an issue that is not justied by the objective, data-driven estimation of our methodology. Based on the weights in Table 4, Table A2 in the Appendix presents, for the whole country set, the calculated scores and ranks of both the original TTCI- WEF and the new TTCI-LDA indices, ordered by the predicted class and then alphabetically. Statistical measurements show that the new proposed TTCI- LDA scores have a signicant correlation with the original TTCI-WEF scores. Indeed, the R-square value between the two index scores is equal to .895; the Pearson correlation coefcient is equal to .946, and Kendall’s Tau (Kendall, 1976) for the proximity of the ranking positions is equal to .827. The close relation between TTCI-WEF and TTCI- LDA scores can also be observed from the box-plot graph of Fig. 4. According to that, the country scores in each class are placed in similar ranges. Particularly for the countries in C 1 and C 2 with the best and in- termediate performance, the TTCI-LDAscores appear increased, while in C 3 (worst performing countries) the new scores show, in average terms, a very small decrease. The observed dispersion between the two scores can also be veried by the fact that the stan- dard deviation and range for TTCI-LDA are 0.968 and 3.64 respectively, compared to the corresponding val- ues 0.713 and 3.02 for TTCI-WEF. The best and worst country in each class, represented by the whiskers in the box plot in Fig. 4, together with the associated scores in parentheses, are presented in Table 5. Besides the similarity between TTCI-WEF and TTCI-LDA scores, in a number of countries, the score difference is signicant. For example, India’s score of 4.421 in TTCI-WEF has worsened to 3.935 in TTCI- LDA due to the different weighting scheme and the Table 5. First and last ranked countries in classes. Best country Worst country Class in class in class Class 1—best Switzerland (5.835) Belgium (5.219) performance Class 2—intermediate Malta (5.178) Paraguay (3.545) performance Class 3—worst Botswana (3.578) Congo (2.187) performance low performance of this country in pillars A5—ICT Readiness and C3—Tourist Service Infrastructure, which have been revealed as very signicant. The same applies to Congo and Malawi, for which their scores 2.675 and 2.927 have been reduced to 2.187 and 2.515, respectively. Unlike the country cases with worsened performance, a number of countries have shown signicant score improvement. This is the case for Israel, for which a 25.2% score increase from 3.984 to 4.988 has been observed. Iceland has had a 23.28% score increase from 4.499 to 5.547, which places it in the rst class. Kuwait has had a 23.6% score increase from 3.419 to 4.227. 3.3 Validation of the resulting scores This step validated and explained the derived TTCI-LDA scores. The validation was performed by employing an external, different index, the Global Competitive Index (GCI) developed by the World Economic Forum. This index measures annually the countries’ level of productivity against relative factors, organized into 12 pillars: Institutions, In- frastructure, ICT Adoption, Macroeconomic Stability, Health, Skills, Product Market, Labor Market, Finan- cial System, Market Size, Business Dynamism, and Innovation Capability. GCI is considered to have a link with the TTCI-WEF in the sense that a high level of travel and tourism competitiveness is expected to lead to increased general competitiveness and pro- ductivity. The relation of the two indices is mentioned in WEF’s Global Competitiveness Report (Schwab, 2019) and conrmed by Jovanovi ´ c et al. (2014), who identied a high correlation between TTCI and GCI, suggesting that “the increase in the tourism compet- itiveness of the country enables an increase in its overall competitiveness”. A correlation test between TTCI-LDA and GCI performed on the common country data set (140 countries) in the same period, 2019, indicated a signif- icant connection (Pearson correlation coefcient .955). The scatter plot in Fig. 5 of the countries’ scores in the indices shows a signicant linear connection indicating that the resulting TTCI-LDA scores follow the observed link between TTCI-WEF and GCI scores. 162 ECONOMIC AND BUSINESS REVIEW 2024;26:151–167 Fig. 5. Comparison graph of TTCI-LDA and GCI 2019 country scores. Furthermore, Fig. 5 provides visual conrmation that the separation of classes in terms of GCI score is dis- tinct. 4 Conclusion This paper brings an alternative methodology for weighting CIs by using LDA and calculating the new weights endogenously. This approach is new and be- longs to the category of methods that use multivariate statistical analysis. The new index weights derive from the intermediate measurements and results of the typical LDA procedure and particularly from the potency index by using a simple calculation formula. Furthermore, this paper applies the above-mentioned method to the TTCI-WEF data set to provide new weights for pillars, which give different priorities compared to the original TTCI-WEF. The new TTCI-LDA scores enable a global classi- cation of the countries in such a way that the overall score discrimination is increased and that a signicant correlation to original TTCI scores is retained. The pri- ority of pillars that emerge as either decisive or least signicant factors for the discrimination are compa- rable with other related publications. Furthermore, the new TTCI-WEF allows for direct country compar- isons and benchmarking. The score thresholds and the best–worst country cases in each class provide additional information for policy decision making. In this regard, the best country in each class can be regarded as a benchmark for the other members in the same class, and its policies and practices should constitute short-/mid-term goals to achieve. However, the data analysis part depends on several critical points. The rst point is whether the initial data enable the application of LDA. LDA imposes strict statistical assumptions on the data, which are often not met in various problems and applications. Another issue concerns the initial classication. In many applications, this derives from initial prefer- ence information provided by a decision maker or an expert. In the case of TTCI-WEF, we introduced a statistical procedure that exploits past TTCI scores and, based on their level and variation, provides the required three-class country segmentation (countries at the top, middle, and bottom ranking positions). This approach is objective and statistically adequate but relates the countries’ classication to the original calculating method of the TTCI index. Moreover, it depends on the number of years for which the data are available and requires a minimum number of coun- tries to appear in each class. The proposed methodology has the potential to be applied to other similar CIs’ cases, for example, to the United Nations’ Human Development In- dex (HDI; United Nations Development Programme, 2024), commonly used in research, studying vari- ous aggregating and weighting methods. In the case of HDI, the original report indicates empirical cut- off points to distinguish countries in classes of low, medium, high, and very high human development. This initial classication suggestion, in conjunction with the performance indicators assumed in the HDI framework, can be used as input to an LDA procedure to obtain a new weighting scheme, according to our methodology. References Abreu-Novais, M., Ruhanen, L., & Arcodia, C. (2016). Destina- tion competitiveness: what we know, what we know but shouldn’t and what we don’t know but should. Current Issues in Tourism, 19(6), 492–512. https://doi.org/10.1080/13683500 .2015.1091443 Baum, T. (2007). Human resources in tourism: still waiting for change. Tourism Management, 28(6), 1383–1399. https://doi .org/10.1016/j.tourman.2007.04.005 Croes, R., & Kubickova, M. (2013). From potential to ability to compete: towards a performance-based tourism competitive- ness index. Journal of Destination Marketing and Management, 2(3), 146–154. https://doi.org/10.1016/j.jdmm.2013.07.002 Crouch G. I. (2007). Measuring tourism competitiveness. Research, theory and the WEF Index. Proceedings of the ANZMAC 2007, 3–5 December, Dunedin, New Zealand. Despotis, D. K. (2005). A reassessment of the human development index via data envelopment analysis. Journal of the Opera- tional Research Society, 56(8), 969–980. https://doi.org/10.1057/ palgrave.jors.2601927 Dwyer, L., Kneževi´ c, L., Mihali, T., & Koman, M. (2014). Integrated destination competitiveness model: Testing its validity and data accessibility. Tourism Analysis, 19(1), 1–17. https://doi.org/10 .3727/108354214X13927625340073 Fisher, R. (1936). The use of multiple measures in taxonomic prob- lems. Annals of Eugenics, 2(10), 179–188. https://doi.org/10 .1111/j.1469-1809.1936.tb02137.x Friedman, J., Hastie, T., & Tibshirani, R. (2009). The elements of sta- tistical learning (2nd ed.). Springer. Gomez-Vega, M., & Picazo-Tadeo, A. J. (2019). Ranking world tourist destinations with a composite indicator of competi- tiveness: to weigh or not to weigh? Tourism Management, 72, 281–291. https://doi.org/10.1016/j.tourman.2018.11.006 ECONOMIC AND BUSINESS REVIEW 2024;26:151–167 163 Greco, S., Ishizaka, A., Tasiou, M., & Torrisi, G. (2019). On the methodological framework of composite indices: a review of the issues of weighting, aggregation, and robustness. Social In- dicators Research, 141(1), 61–94. https://doi.org/10.1007/s11205 -017-1832-9 Gupta, K. D., Jongman, A. & Schmid, P . A. (1994). Creating a com- posite index for assessing country performance in the eld of human rights: proposal for a new methodology. Human Rights Quarterly, 16(1), 131–162. https://doi.org/10.2307/762414 IBM. (2021). Collinearity diagnostics. https://www.ibm.com/docs/ en/spss-statistics/28.0.0?topic=scores-collinearity-diagnostics Iwuagwu, C., & Nwosu, M. (2021). Modelling human development index using discriminant analysis. Progress in Natural Science, 1(1), 145–165. Jovanovi´ c, S., Jankovi´ c Mili´ c, V ., & Krsti´ c, B. (2014). Homo- geneity analysis of south-eastern European countries accord- ing to tourism competitiveness performances. Economic Re- search/Ekonomska Istraživanja, 27(1), 207–220. https://doi.org/ 10.1080/1331677X.2014.952113 Hair, J., Anderson, R., Tatham, R. & Black, W. (2010). Multivariate data analysis (7th ed.). Prentice Hall. Huberty, C. J. (1994). Applied discriminant analysis. John Wiley & Sons. Huberty, C. J., & Olejnik, S. (2006). Applied MANOV A and discrimi- nant analysis (2nd ed.). John Wiley & Sons. Kayar, Ç. H., & Kozak, N. (2010). Measuring destination competi- tiveness: an application of the Travel and Tourism Competitive- ness Index (2007). Journal of Hospitality Marketing & Management, 19(3), 203–216. https://doi.org/10.1080/19368621003591319 Kendall, M. G. (1976). Rank correlation methods (4th ed.). Grifn. Koži´ c, I., & Mikuli´ c, J. (2014). Research note: measuring tourism sustainability: an empirical comparison of different weight- ing procedures used in modelling composite indicators. Tourism Economics, 20(2), 429–437. https://doi.org/10.5367/te .2013.0283 Kunst, I., & Ivandi´ c, N. (2021). The viability of the travel and tourism competitiveness index as a reliable measure of desti- nation competitiveness: the case of the Mediterranean region. European Journal of Tourism Research, 27, Article 2704. https:// doi.org/10.54055/ejtr.v27i.2124 Lan, L., Wu, W., & Lee, Y. (2012). Exploring an objective weighting system for travel & tourism pillars. Procedia—Social and Be- havioral Sciences, 57, 183–192. https://doi.org/10.1016/j.sbspro .2012.09.1173 Lin, C. T., & Huang, Y. L. (2009). Tourism competitiveness evalua- tion in Asian countries. Applying Grey Relational Analysis and Sensitivity Analysis. The Journal of Grey System, 21(3), 269–278. Litavcová, E., & Síˇ c, J. (2021). Travel and Tourism Competitiveness Index 2019—Quantile regression approach on 2nd sub-index. Journal of Management and Business: Research and Practice, 13(2). https://doi.org/10.54933/jmbrp-2021-13-2-2 Mazanec, J., & Ring, A. (2011). Tourism destination competitive- ness: second thoughts on the World Economic Forum reports. Tourism Economics, 17(4), 725–751. https://doi.org/10.5367/te .2011.0065 Mendola, D., & Volo, S. (2017). Building composite indicators in tourism studies: measurements and applications in tourism destination competitiveness. Tourism Management, 59, 541–553. https://doi.org/10.1016/j.tourman.2016.08.011 Mikuli´ c, J., Koži´ c, I., & Kreši´ c, D. (2015). Weighting indicators of tourism sustainability: a critical note. Ecological Indicators, 48, 312–314. https://doi.org/10.1016/j.ecolind.2014.08.026 Mili´ c, V ., & Jovanovi´ c, S. (2019). Sensitivity analysis of travel and tourism competitiveness index to changes in compo- nent weighting. Economics of Sustainable Development, 3(2), 1–7. https://doi.org/10.5937/ESD1902001J Nardo, M., Saisana, M., Saltelli, A., Tarantola, S., Hoffman, A., & Giovannini, E. (2008). Handbook on constructing composite indi- cators: Methodology and user guide. OECD Publishing. https:// doi.org/10.1787/9789264043466-en Njoku, J. (2013). Defacement of bank going concern public secrecy. International Journal of Critical Accounting, 5(3), 229–249. https:// doi.org/10.1504/IJCA.2013.055477 Perreault, W. D., Jr., Behrman, D. N., & Armstrong, G. M. (1979). Alternative approaches for interpretation of multiple discrimi- nant analysis in marketing research. Journal of Business Research, 7(2), 151–173. https://doi.org/10.1016/0148-2963(79)90004-3 Pulido-Fernández, J., & Rodríguez-Diaz, B. (2016). Reinterpreting the World Economic Forum’s global tourism competitiveness index. Tourism Management Perspectives, 20, 131–140. https:// doi.org/10.1016/j.tmp.2016.08.001 Ring, A. (2011). Alternative models within the framework of the World Economic Forum’s Travel & Tourism Competitiveness Index— Suggestions for theory development (Travel and Tourism Research Association: Advancing Tourism Research Globally No. 42). https://scholarworks.umass.edu/ttra/2011/Oral/42 Rodríguez-Díaz, B., & Pulido-Fernández, J. I. (2020). Analysis of the worth of the weights in a new travel and tourism competi- tiveness index. Journal of Travel Research, 60(2), 267–280. https:// doi.org/10.1177/0047287519899982 Salinas-Fernández, J. A., Serdeira-Azevedo, P ., Martín-Martín, J. M., & Rodríguez-Martín, J. A. (2020). Determinants of tourism destination competitiveness in the countries most visited by international tourists: Proposal of a synthetic index. Tourism Management Perspectives, 33, Article 100582. https://doi.org/ 10.1016/j.tmp.2019.100582 Schwab, K. (2019). The global competitiveness report 2019. World Economic Forum. https://www3.weforum.org/docs/ WEF_TheGlobalCompetitivenessReport2019.pdf Shobha, K., & Siji K., (2018). Determinants of credit gap among the borrowers of Malur taluk in Kolar district of Karnataka, International Journal of Engineering Development and Research, 6(3), 735–739. https://www.ijedr.org/papers/IJEDR1803121 .pdf Smirlis, Y. (2020). A trichotomic segmentation approach for esti- mating composite indicators. Social Indicators Research, 150(2), 393–410. doi.org/10.1007/s11205-020-02310-1 TCdata360 (2019). World Economic Forum Travel & Tourism Compet- itiveness Index. https://prosperitydata360.worldbank.org/en/ dataset/WEF+TTCI United Nations Development Programme. (2024). Human Devel- opment Index (HDI). https://hdr.undp.org/data-center/human -development-index#/indicies/HDI Watson, C. (1982). Approaches for the interpretation of multiple discriminant analysis in organizational research. The Academy of Management Review, 7(1), 124–132. https://doi.org/10.2307/ 257257 World Economic Forum. (2019). The travel & tourism compet- itiveness report 2019. https://www.weforum.org/reports/the -travel-tourism-competitiveness-report-2019 World Economic Forum. (2022). Travel & Tourism Development Index 2021: rebuilding for a sustainable and resilient fu- ture. https://www.weforum.org/reports/travel-and-tourism -development-index-2021/ Wu, W-W., Lan, L. W., & Lee, Y-T. (2012). Critiquing the World Economic Forum’s concept of destination competitiveness: a further analysis. Tourism Management Perspectives, 4, 198–206. https://doi.org/10.1016/j.tmp.2012.08.008 164 ECONOMIC AND BUSINESS REVIEW 2024;26:151–167 Appendix Fig. A1 exhibits the high-level structure of TTCI in subindicators and pillars. The numbers in parentheses denote the number of low-level indicators that are associated with every pillar. Fig. A1. The 4 subindices and 14 pillars of TTCI. Table A1. Descriptive statistics of TTCI-WEF scores in years 2007–2017. Year of WEF report N Mean SD Min Q1 Median Q3 Max 2007 123 4.24 0.68 2.88 3.67 4.52 4.8 5.66 2009 133 4.08 0.69 2.52 3.54 4.41 4.55 5.68 2011 139 4.09 0.71 2.56 3.49 4.38 4.59 5.68 2013 140 4.12 0.72 2.59 3.56 4.39 4.67 5.66 2015 141 3.74 0.68 2.43 3.22 3.98 4.25 5.31 2017 136 3.82 0.69 2.44 3.28 4.05 4.38 5.43 Fig. A2. Box-plot diagram for TTCI total score in years 2007–2017. ECONOMIC AND BUSINESS REVIEW 2024;26:151–167 165 Table A2. Classication, countries scores and ranks. Country Initial class 1, 2, 3 Predicted class TTCI-WEF TTCI-LDA Rank TTCI-WEF Rank TTCI-LDA Australia 1 1 5.1415 5.5725 7 9 Austria 1 1 4.9538 5.6812 11 6 Belgium 1 4.5471 5.2197 24 26 Canada 1 1 5.0513 5.5452 9 12 Denmark 1 4.5811 5.3340 21 21 Finland 1 4.5183 5.2823 28 25 France 1 1 5.4032 5.5813 2 8 Germany 1 1 5.3882 5.7313 3 3 Hong Kong SAR 1 1 4.8119 5.5429 14 13 Iceland 1 4.4996 5.5470 30 11 Ireland 1 4.5383 5.2873 26 24 Italy 1 5.0856 5.2926 8 23 Japan 1 1 5.3716 5.7083 4 4 Korea, Rep. 1 4.7806 5.4840 16 14 Luxembourg 1 1 4.5556 5.4300 23 19 Netherlands 1 1 4.7915 5.4524 15 18 New Zealand 1 1 4.7459 5.4567 18 16 Norway 1 4.5924 5.4543 20 17 Portugal 1 4.8936 5.4579 12 15 Singapore 1 1 4.7578 5.5475 17 10 Spain 1 1 5.4401 5.6836 1 5 Sweden 1 1 4.5626 5.3141 22 22 Switzerland 1 1 5.0159 5.8350 10 1 United Arab Emirates 1 4.4349 5.3896 33 20 United Kingdom 1 1 5.1921 5.6585 6 7 United States 1 1 5.2539 5.7450 5 2 Albania 2 3.5846 4.0523 86 78 Argentina 2 4.1522 4.4329 50 61 Armenia 2 3.7096 4.2684 79 68 Azerbaijan 2 3.7993 4.2626 71 69 Bahrain 2 2 3.9069 4.7474 64 44 Bolivia 2 3.4959 3.5660 90 99 Bosnia and Herzegovina 2 3.2813 3.8517 105 91 Brazil 2 2 4.4564 4.3632 32 63 Brunei Darussalam 2 3.7834 4.4260 72 62 Bulgaria 2 2 4.2113 4.8608 45 38 Cambodia 2 3.3942 3.5602 98 100 Cape Verde 2 3.5508 4.0337 88 81 Chile 2 2 4.1001 4.5384 52 57 China 2 4.8759 4.6094 13 52 Colombia 2 4.0088 4.0942 55 77 Costa Rica 2 2 4.2682 4.7467 41 45 Croatia 2 4.5258 5.0692 27 30 Cyprus 2 4.2162 5.1336 44 29 Czech Republic 2 4.3267 5.0648 38 32 Dominican Republic 2 2 3.7753 4.1850 73 75 Ecuador 2 3.8645 4.0353 70 80 Egypt 2 3.8973 3.9654 65 85 El Salvador 2 3.2321 3.6291 108 97 Estonia 2 4.1961 5.0673 46 31 Georgia 2 3.8763 4.5145 68 59 Greece 2 4.5454 5.1436 25 28 Guatemala 2 3.3930 3.6446 99 95 Honduras 2 3.4569 3.5506 94 101 Hungary 2 2 4.1936 4.8186 48 41 India 2 4.4211 3.9353 34 86 Indonesia 2 4.2700 4.2344 40 71 Iran, Islamic Rep. 2 3.5427 3.7125 89 94 Israel 2 2 3.9841 4.9881 57 33 Jamaica 2 2 3.7493 4.3043 76 64 Jordan 2 2 3.5888 4.1656 84 76 Kazakhstan 2 3.6696 4.2007 80 73 (continued on next page) 166 ECONOMIC AND BUSINESS REVIEW 2024;26:151–167 Table A2. (Continued) Country Initial class 1, 2, 3 Predicted class TTCI-WEF TTCI-LDA Rank TTCI-WEF Rank TTCI-LDA Kuwait 2 3.4196 4.2274 96 72 Latvia 2 2 4.0408 4.8234 53 40 Lebanon 2 3.3821 3.8920 100 89 Lithuania 2 2 3.9751 4.7479 59 43 Malaysia 2 4.5135 4.8514 29 39 Malta 2 4.3582 5.1788 35 27 Mauritius 2 2 4.0095 4.6366 54 51 Mexico 2 4.6894 4.5405 19 56 Moldova 2 3.2898 3.8084 103 92 Mongolia 2 3.4700 3.7530 93 93 Montenegro 2 2 3.8914 4.6589 67 49 Morocco 2 3.8954 4.1920 66 74 Namibia 2 3.6672 3.9011 81 88 Nicaragua 2 3.4944 3.6313 91 96 North Macedonia 2 3.3578 3.9686 101 84 Oman 2 2 3.9776 4.5254 58 58 Panama 2 2 4.1937 4.5543 47 55 Paraguay 2 3.2318 3.5458 109 102 Peru 2 4.1670 4.2533 49 70 Philippines 2 3.7519 3.9801 75 83 Poland 2 2 4.2322 4.7466 42 46 Qatar 2 2 4.1346 4.9557 51 36 Romania 2 3.9892 4.4962 56 60 Russian Federation 2 2 4.3172 4.7240 39 47 Saudi Arabia 2 3.8752 4.6897 69 48 Serbia 2 3.6277 4.2763 83 67 Seychelles 2 2 3.9295 4.7731 62 42 Slovak Republic 2 2 3.9733 4.5693 60 54 Slovenia 2 2 4.3464 4.8943 36 37 South Africa 2 3.9721 4.0486 61 79 Sri Lanka 2 3.7261 3.8620 77 90 Taiwan, China 2 4.3323 4.9770 37 34 Thailand 2 4.4971 4.9583 31 35 Trinidad and Tobago 2 3.5832 4.3042 87 65 Tunisia 2 2 3.5868 4.0211 85 82 Turkey 2 2 4.2227 4.6532 43 50 Ukraine 2 3.7235 4.2836 78 66 Uruguay 2 2 3.7658 4.5989 74 53 Vietnam 2 3.9145 3.9295 63 87 Algeria 3 3.1477 3.3618 116 106 Angola 3 3 2.7367 2.6375 134 131 Bangladesh 3 3 3.1004 3.1513 120 114 Benin 3 3.0212 2.8321 123 124 Botswana 3 3.4772 3.5787 92 98 Burkina Faso 3 3 2.7799 2.6315 132 133 Burundi 3 3 2.6604 2.3428 137 138 Cameroon 3 3 2.8978 2.6848 128 130 Chad 3 3 2.5232 2.2959 139 139 Congo, Democratic Rep. 3 2.6750 2.1873 136 140 Côte d’Ivoire 3 3.1140 3.1970 119 113 Ethiopia 3 3.0239 2.6997 122 128 Gambia, The 3 3.2278 3.2955 111 109 Ghana 3 3.1489 3.2342 115 111 Guinea 3 3 2.9217 2.7886 126 126 Haiti 3 3 2.7612 2.6345 133 132 Kenya 3 3.6285 3.4650 82 104 Kyrgyz Republic 3 3.2316 3.4396 110 105 Lao PDR 3 3.4153 3.5359 97 103 Lesotho 3 3 3.0192 2.9708 124 120 Liberia 3 2.6098 2.4291 138 136 Malawi 3 3 2.9279 2.5215 125 135 Mali 3 3 2.8064 2.8808 130 122 (continued on next page) ECONOMIC AND BUSINESS REVIEW 2024;26:151–167 167 Table A2. (Continued) Country Initial class 1, 2, 3 Predicted class TTCI-WEF TTCI-LDA Rank TTCI-WEF Rank TTCI-LDA Mauritania 3 3 2.6859 2.8012 135 125 Mozambique 3 3 2.9129 2.5744 127 134 Nepal 3 3.3469 3.2671 102 110 Nigeria 3 3 2.8180 2.7562 129 127 Pakistan 3 3 3.0969 3.1468 121 115 Rwanda 3 3.2494 3.1441 107 116 Senegal 3 3.2619 3.3377 106 107 Sierra Leone 3 3 2.7840 2.6986 131 129 Swaziland 3 3.1248 2.9897 118 119 Tajikistan 3 3.2839 3.3330 104 108 Tanzania 3 3.4316 3.0465 95 117 Uganda 3 3.1937 2.8595 112 123 Venezuela 3 3.1314 3.2156 117 112 Yemen 3 3 2.4180 2.4027 140 137 Zambia 3 3.1617 2.9609 113 121 Zimbabwe 3 3.1533 3.0232 114 118 Table A3. Mean score of TTCI 2019 pillars for countries’ classes 1—high, 2—medium, and 3—low competitiveness. TTCI Pillar Class 1 Class 2 Class 3 A1—Business Environment 5.436 4.6234 3.8306 A2—Safety & Security 6.075 5.4391 4.5797 A3—Health & Hygiene 6.2116 5.7483 3.0963 A4—Human Resources and Labor Market 5.4465 4.6926 3.5554 A5—ICT Readiness 6.0968 5.1905 2.6767 B1—Prioritization of Travel & Tourism 5.316 4.9734 3.3563 B2—International Openness 4.2986 3.4677 2.2302 B3—Price Competitiveness 4.5473 5.3042 5.3753 B4—Environmental Sustainability 4.992 4.4044 4.0835 C1—Air Transport Infrastructure 5.1555 3.2842 1.7168 C2—Ground & Port Infrastructure 5.2747 3.8508 2.2867 C3—Tourist Service Infrastructure 5.7683 4.7828 2.2806 D1—Natural Resources 4.0199 2.9971 2.4172 D2—Cultural Resources & Business Travel 4.1163 1.9711 1.2932