Use of mul -criteria decision analysis to classify and regulate interstate bus services in Brazil

Recebido: 7 de junho de 2018 Aceito para publicação: 10 de junho de 2019 Publicado: 31 de dezembro de 2019 Editor de área: Helena Beatriz Cybis ABSTRACT Interstate bus services in Brazil can occur between nearby or distant municipali es. In the former situa on the bus service should be classified as ‘semi-urban’; it is delegated by Federal concession that requires a bidding process and must be provided by urban buses. In the la0er situa on the bus service should be classified as ‘road’; it is directly authorized by the public authority and must be provided by coaches. Brazilian law disnguishes between ‘nearby’ and ‘distant’ ci es when the service occurs with ‘urban’ or ‘regional’ characteris cs, but does not specify parameters for that dis nc on. This paper presents a mul -criteria method used to be0er classify and regulate those bus transport services. The method was applied to categorize 11 interstate markets of which 4 were considered to be ‘semi-urban’. In conclusion, the study served as a basis to support a government decision.


INTRODUCTION
In Brazil, road passenger transport services can occur within the same Brazilian state or between different states. In the irst case, the services are under state/Federal District responsibility and, in the second, they are under Federal responsibility (Nunes, 2010). There are two types of interstate passenger transport services: 'semi-urban' and 'road'. Although both of them are interregional transport services under Federal responsibility, they have several operational and regulatory differences.
The interstate 'semi-urban' bus service is the one which, while maintaining urban characteristics, traverses the limits of Brazilian states (in Brazil, the Federal District is treated as a state). For that, according to national regulations, it is necessary to use city buses with TRANSPORTES | ISSN: 2237-1346 94 controlled bus fares. The 'semi-urban' regulatory model is based on exclusive concessions and requires a bidding process to select a company to perform the operation (Nunes, 2010;Schiller, 2018). In such cases, the fares are usually cheaper and the frequency of buses is usually higher, both in comparison to a 'road' service (Brandão Filho et al., 2016). On the other hand, any company that meets certain prerequisites can operate the 'road' interstate bus service (Schiller, 2018). This service should be offered on interstate routes with regional transport characteristics, using coaches and charging non-regulated fares (Brazil, 2001;Schiller, 2018). The fares are usually more expensive and the frequency of buses is usually lower than 'semi-urban' services (Nunes, 2010). Also known as 'long distance coach services', the 'road' bus service has been delegated by a nonexclusive authorization model (free competition environment) since 2015, due to the deregulation process (Brazil, 2001;Schiller, 2018). Table 1 displays the differences between 'road' and 'semi-urban' interstate bus service. Before 2013, the bus services were classi ied as 'road' whenever the extension of the line exceeded 75 km (Brandão Filho et al., 2016). However, that distance should not be the unique criterion to determine whether the service is 'semi-urban' and it is also not a good parameter for measuring the 'urban characteristics' (Ojima, 2007). For instance, between Poços de Caldas city (in Minas Gerais State) and Divinolândia city (in São Paulo State) there are 40 km without urban transport characteristics. Consequently the operation is currently conducted using coaches. On the other hand there is the example of the bus services between Brasıĺia (in the Federal District) and Luziânia (in Goiás State), cities 60 km apart, that are served by urban buses (Figure 1). Because of inadequate use of distance as the unique criteria, many transport services have been wrongly classi ied. As a result, some markets operated by road services have been requesting to switch to semi-urban services, and vice versa. That required studies to develop a new classi ication method that could support the decisions of the public authority. According to Brazilian Law 010.233/2001 (Brazil, 2001), the National Land Transport Agency (ANTT) is responsible for determining whether a bus service should be classi ied as 'semi-urban' or 'road'.
To try to solve the problem, a preliminary method was proposed by Brandão Filho et al. (2016), which considers territorial continuity and the integration of municipalities (Figure ). However, this method requires a 'technical analysis' that has not yet been presented. A methodological complement is needed to guide any such technical analysis.
To complement the paper presented by Brandão Filho et al. (2016), the main objective of this study is to develop a method to measure the degree of 'urban characteristics' in order to classify the interstate bus transport services and assist the sector regulation (Brazil, 2001). In the case of a bad classi ication, there could be serious operational and regulatory problems, such as expensive bus fares or insuf icient bus frequencies (Nunes, 2010

. Mul>-Criteria Decision Analysis
Classic Operational Research (OR) has developed some positive paradigms in the search for good solutions to certain problems (Bortoluzzi et al., 2017). With one such paradigm, depending on the complexity of the decision to be made, decision-makers can propose different values and preferences in an attempt to solve those problems (Marttunen et al., 2017). For that reason, the Multi-Criteria Decision Analysis (MCDA) approach appears as the methodological alternative to be used (Marttunen et al., 2017). In essence, the MCDA consists of a systematic evaluation of alternatives in terms of multiple and often con licting objectives (Marttunen et al., 2017). In some cases, it is necessary for decision-makers to increase their degree of understanding of the decision to be made, especially when presenting their different points of view (PV), in order to ind the best solution in a 'contructivist' way. When that occurs, the technique is known as Multi-Criteria Decision Analysis -Constructivist (MCDA-C) .
The irst stage begins the identi ication of aspects of the object to be studied. The problem structuring (second stage) seeks to de ine the title of the problem and to determine the family of Fundamental Points of View (FPV). In this paper, in order to de ine the FPV criteria it is necessary to search for speci ic literature on urban mobility, land-use, and spatial con iguration (Adelt et al., 2018;Banister, 2008;Da Silva et al., 2008;Davidson et al., 2007;Gouvêa, 2010;Waddell, 2011).
At the stage of Multi-criteria Model Structuring (third stage), initially, decision makers de ine the focus on which they intend to act and what aspects they consider to be essential for their analysis. Subsequently, that materializes in the form of 'evaluation axes', which enable the transition from a cognitive map to a multicriteria model Ensslin et al., 2001). In the last stage (Evaluation of Potential Actions), the performance of the actions is evaluated. For that, an additive aggregation formula is used, which makes it possible to aggregate the local performances of the actions in a global evaluation Ensslin et al., 2001).

Transport with urban characteris>cs
Regarding the Problem Structuring stage (second stage) and Multicriteria Model Structuring stage (third stage), in the context of transport with urban characteristics, the structure to be developed must consider social and technological interactions, taking into account the interplay of agents' choices and situational constraints (Adelt et al., 2018). That is because urban mobility involves relatively complex aspects, such as land-use and spatial con iguration (Da Silva et al., 2008;Davidson et al., 2007;Gouvêa, 2009;Ojima, 2007).
According to Banister (2008), considering a sustainable mobility, a multicriteria analysis must involve a demand of people, according to their interests, associated with a minimization of their travel time (such as increasing bus frequency). Moreover, supposing that demand could be based on activities, travel is generated according to the daily actions performed by families and people (Davidson et al., 2007;Ojima, 2007). As Davidson et al. (2007) clarify, an activity-based platform implies that modeled travel is derived within a general framework of the daily activities undertaken by households and persons, such as work or school and personal maintenance (e.g., health services, food shopping). Finally, transport systems are sensitive to policy changes over time (Nunes, 2010;Waddell, 2011), so that any changes affect travel demands (Davidson et al., 2007). This means that the FPVs to be considered must deal with: (i) social aspects; (ii) political-institutional aspects, and; (iii) technical and operational aspects.

MATERIAL AND METHODS
To achieve the proposed objectives, researchers developed a set of measures to be applied to public management groups. The selected group was composed of specialists in transport regulation from the National Land Transport Agency (ANTT) of Brazil, including its managers and superintendent. In an attempt to avoid only portraying the Agency's positions, one of the concerns in the choice of these participants was the need for the specialist to have experience external to ANTT and, preferably, contact with the regulated market.
In general, this research found that the difference between 'semi-urban' and 'road' bus services depends on social, political, institutional, technical and operational aspects (Adelt et al., 2018;Banister, 2008;Davidson et al., 2007;Ojima, 2007). In that kind of complex situation, MCDA makes it feasible to evaluate several alternative models or variants based on many quantitative/qualitative criteria (Quadros e Nassi, 2015). According to Macharis and Bernardini (2015), there are numerous Multi-Criteria Decision techniques in use in transport activities and that includes passenger and freight transportation. Other papers that have used the multi-criteria techniques have cited their use in transportation planning (De Luca, 2014) and mobility/logistics (Macharis et al., 2010, Ferretti e Degioanni, 2017 or for ranking railway projects (Mandic et al., 2014), or to evaluate preferential options among modes of transport (Sayers et al., 2003), or to guide a furniture industry production manager (Bortoluzzi et al., 2017).
This study used a constructivist analysis (MCDA-C), grounded in a learning pattern, such as that used by Ensslin et al. (2000) or Bortoluzzi et al. (2017). The structure proposed is multidimensional and integrates mental maps with the Measuring of Attractiveness by means of a Categorical-Based Evaluation Technique (MACBETH) (Bana e Vansnick, 1994, 1999;Bana e Costa et al., 2012;Bortoluzzi et al., 2017). In the ield of transport, Da Silva et al. (2008) used this same method to establish the key elements of a sustainable urban mobility approach in Brazil, and Costa and Granemann (2017) developed a technique, using MCDA-C, to evaluate the quality of semi-urban interstate bus services.
Following the guidelines of Ensslin et al. (2000), the MCDA-C technique used in this paper is detailed in two main stages: • Stage 1: Structuring Preparation stage that develops a better communication environment among the many actors involved (Ensslin et al, 2000). Step to elaborate a multidimensional and quantitative model, establishing values for each FPV depending on their contribution to the study (Ensslin et al., 2000).
(i) develop the value functions for each FPV; (ii) establish weights (compensation rates); (iii) evaluate the system impact pro ile and verify the behavior of the model; (iv) interpret the results.

MULTICRITERIA CHARACTERIZATION OF INTERSTATE BUS PASSENGER TRANSPORT SERVICES 4.1. Structuring phase
To construct the mental map and the value tree, during the structuring phase six decision makers were identi ied: four ANTT regulatory specialists with professional experience external to the Agency, one manager and one superintendent (both from ANTT but with market network experience). Although the number of decision makers may have been a limiting factor, together and based on a literature review (Adelt et al., 2018;Banister, 2008;Davidson et al., 2007;Nunes, 2010;Ojima, 2007;Waddell, 2011), they identi ied three FPV capable of in luencing interstate bus service regulation: (i) social aspects; (ii) political-institutional aspects; (iii) technical and operational aspects. Those FPV were decomposed into 14 EPV, divided into 8 Sub-Elemental Points of View (Sub-EPV), as displayed in Figure 3. Table 2 presents the full set of FPV, EPV and Sub-EPV. It is important to note that the decision makers made these subdivisions also considering the views of agents external to ANTT (such as users and entrepreneurs) obtained in work meetings and public consultations.

Social aspects
As displayed in Table 2, to specify the social aspects of each city involved in the study, 3 Elemental Points of View (EPV) were considered. For this analysis, the speci ic values of each indicator were compared among the cities evaluated. This comparison aims to analyze the types of interdependence of one city with another, which are: health service interdependence, school interdependence, and work interdependence. If there is a great interdependence of these indicators between two cities, there is a tendency to have urban transport characteristics between them (Adelt et al., 2018;Davidson et al., 2007;Ojima, 2007).

Political-institutional aspects
For the political-institutional aspects, the study analyzed the availability (or existence) of public policies for urban mobility in the city. Whenever there are polices like these, the urban organization is better structured (Waddell, 2011). Another way to consider the existence of 'political-institutional aspects' is when an external entity (such as a bus company or a municipal government) requests ANTT for a public transport operation between those cities. In this respect, there is probably an urban cluster, indicating urban characteristics in the region.

Technical and operational aspects
The de inition of 'technical and operational aspects' is not a simple task. This analysis involves factors, such as: urban infrastructure; existence of clandestine transportation (without authorization of the public authority); bus line extension; or fare prices. There are situations that indicate a trend towards 'urban transport characteristics', such as: the existence of clandestine urban transportation; good availability of urban infrastructure; shorter extension of the bus line or relatively low fare prices, considering inter-municipal transport characteristics.
If there is a regular offer of interstate bus passenger service in the city, it is feasible to describe the degree of urbanity using the bus occupancy rate and frequency (Banister, 2008;De Luca, 2014). In the same way, if there are additional services (i.e., package service/ transport of luggage under orders) or bigger vehicles in use (i.e., coaches with toilets) it means that the TRANSPORTES | ISSN: 2237-1346 100 transport should be classi ied as 'not urban'. Supposing that there is no interstate transport service in the region, then it is relevant to observe whether urban transport (public municipal transport) is available inside the analyzed city or not.

Evalua>on Stage
The study used the Unique Synthesis Criterion Approach to construct the descriptor. This approach creates functions that combine and then maximize preferences of alternatives over attributes, constructed according to the collaborator performances (Guitouni e Martel, 1998;Ensslin et al., 2000;Ishizaka and Siraj, 2017). In this study, each transport market was represented by a function composed of all attributes that in luence 'semi-urban' or 'road' classi ication, accompanied by their respective weights. This function is analogous to that presented in Ensslin et al. (2000) and in Bortoluzzi et al. (2017), as displayed in the following equation: M is the group of markets that will be analyzed; P(m) is the overall score of m Є M; m is the market composed of two cities; FPVc is the Fundamental Points of View linked to criteria c; wc is the weight (or 'compensation rate') for the criterion c which impact on each FPVc; (PFPVc(m)) is the local score. In this paper, wc (c = 1, … 19), which indicates each EPV or Sub-EPV analyzed. Although this study performed the complete analysis, just the FPV 1 (social aspects) are presented in detail for better representation. The FPV 2 and 3 are treated analogously.

Construction of value functions
To construct the value functions MACBETH was used, as developed by Bana e Costa and Vansnick (1994). According to Ensslin et al. (2000), to use the MACBETH it is necessary to question the decision maker about the attraction difference between two elements of analysis.
Usually, the answer is one of the following: • Null; • Very weak attraction difference; • Weak attraction difference; • Moderate attraction difference; • Strong attraction difference; • Very Strong attraction difference; • Extreme attraction difference.
If necessary, two ordinal scales can be combined (i.e.: 'weak-moderate' or 'moderate-strong'). When this happens, the MACBETH software uses diffused linear programming, according to the model requirements.
The impact level of the criterion ('Level 1' -high in luence -or 'Level 3' -low in luence) and the scales types were developed according to the decision maker's preference. The scale values were set to make higher values represent 'semi-urban' transport characteristics, while lower values represented 'road' bus service transport.
Regarding the EPV 1.1 ('interdependence of health service'), when there are two relatively close cities with severe distinction between the availability of their health services, a low of people is created between those cities (Ojima, 2007;Pires et al., 2010), indicating a 'semi-urban' or 'metropolitan' transport characteristic (Gouvêa, 2009). This occurs for several reasons, such as trips for health treatment or doctor's appointments. In the context of rural residential areas, this approach is known as 'public service facilities integrity' and is part of 'social equilibrium' (Ma et al., 2019). To calculate the interdependence of health service, the numbers of hospital beds per inhabitant in each city were used as reference, compared with quartile position in the national ranking. The Brazilian Institute of Geography and Statistics (IBGE) gives the national ranking.
The criterion used to de ine the function values was the interquartile gap. The greater the interquartile gap positioning for each city, the greater the in luence of hospital services is between them. For instance, if the city 'X' is positioned in the irst quartile (worst health structure) and the city 'Y' is positioned in the fourth quartile (better health structure), there is considered to be 'high' in luence between both cities, because there will be a low of people from 'Y' to 'X'. However, if both cities are in the fourth quartile, the in luence will be low. The value functions used for this criterion were 'high', 'medium' or 'low' in luence between cities, according to the Figure (  To measure the 'existence of basic health care services' and the 'existence of emergency services': 'yes', was used if one city has the service and the other does not; or 'No', if both cities have or do not have those services. In regard to 'school interdependence' and 'work interdependence', they were structured using quartiles in a similar way to the 'interdependence of health service'. However, in this respect, the number of students (or workers) that study (or work) outside their city of residence was used, according to IBGE classi ication (Figure). All the other value functions for the model (FPV, EPV and Sub-EPV) were determined in similar ways.  Figure 5. Value functions for 'school interdependence' and 'work interdependence' Note: the arrows indicate the positioning/gap between cities A matrix was developed for semantic judgments using the MACBETH software and considering the value functions as reference. After the judgments, the scale value is given by the software (Bana e Costa and Vansnick, 1999). The EPV 1.1 (Sub-EPV 1.1.1, 1.1.2 and 1.1.3) are displayed in Table 3, and the EPV 1.2/EPV 1.3 are displayed in Table 4. All the other matrixes (FPV 2 and 3) were determined analogously.

Establishment of weights
The weights ('compensation rates') were determined according to the decision makers' choices, especially considering the relevance of each criterion according to the experts' experience. In this case, for example, the experts considered that the political-institutional aspects have less relevance than the social or technical-operational aspects for the following reasons: • Some existing public policies are not well de ined (Da Silva et al., 2008), while others may have been induced by speci ic interests -according to Nunes (2010), some politicians are supporters or owners of bus companies; • Some regional public policies may not correspond to the reality of the municipality (Gouvêa, 2009).
• The mere existence of a public policy does not guarantee its effectiveness (Pires et al., 2010).
• The social aspects are treated as equally relevant to those technical-operational aspects, since they are complementary in terms of transport demand/offer (Davidson et al., 2007). The weights of each EPV and Sub-EPV were also assigned in an analogous way. Each criterion was ranked by level, assuming the given weight. Those weights were normalized to a percentage value obtained from the semantic judgment matrix. This technique followed the procedure proposed by Bana e Costa e Vansnick (1994), based on MACBETH methodology.  Tables 5 and 6 displays the normalized compensation rates in cases where there is, and there is no, a regular offer of interstate bus passenger services. In both cases, the FPV 1 (Social Aspects) and FPV 3 (Technical and Operational Aspects) each represent 40% of all the inal weight of the model. The remaining 20% were attributed to FPV2.

Threshold calibration: 'semi-urban' or 'road'?
Having identi ied the weights and value functions of each variable, it was necessary to de ine a threshold value to classify the market as 'semi-urban' or 'road'. The threshold value delimits the border between 'semi-urban' and 'road' transport services, according to the overall score of each market -P(m). The higher the overall score of the market, the more 'semi-urban' it will tend to be. So, if this overall score value is above the threshold, the market will be considered as 'semi-urban'. There is an indeterminate zone around the threshold. If the overall score is in this zone it means that the market is not precisely classi ied and requires further study. Figure 6 illustrates that situation.
To calculate the threshold value, 10 known markets were selected, ive already operated as 'semi-urban' and ive already operated as 'road' services between cities short distances apart (less than 100 km). Those markets were chosen because there the public authority has no doubts about their classi ication in view of their social and technical-operational characteristics. Just to test the model's behavior, the study included the market between Rio de Janeiro city and São Paulo city which is unquestionably a 'long distance' market and a 'road' service. In other words, it is inconceivable to transport passengers between Rio de Janeiro and São Paulo (440 km apart) in urban buses. Table dispays the selected markets.    The overall score for each market was calculated using the proposed method and the data collected. The threshold value was considered as the average value obtained between the maximum overall score for 'road' market and the minimum overall score for 'semi-urban' market. The indeterminate zone was the difference between those maximum and minimum values. Table 8 sets out the overall score for each market and the threshold values.

Market analysis
The public authority listed 11 interstate bus services markets that were uncertain about their current categorizations (markets operated by road services that have been requesting to switch to semi-urban services, and vice versa). Those uncertainties were generated by several causes, such as user, company or municipal requesting.
Considering the need to technically categorize these 11 unknown interstate bus services markets (which have regular offer of interstate bus passenger service), each of them was inserted in the developed model for analysis. The overall score of each of them was obtained using the proposed method (Table 3).

RESULTS
To represent the result, all the 'road service', 'unknown', and 'semi-urban service' overall scores were ranked in the same table (Table 4). For better comprehension, each analyzed market was referenced by a letter (from letter 'A' to letter 'V'), and displayed graphically in Figure. According to these results, the markets 'G' (Guaçuı/Natividade), 'H' (S. J. da Boa Vista/ Poços de Caldas), 'I' (Poços de Caldas/Divinolândia), 'J' (Brasıĺia/Formosa) and 'K' (Espıŕito Santo do Pinhal/Andradas) could be classi ied as 'road' according to the proposed method. In this respect, the public authority will directly authorize them for an inde inite period with freedom to determine the bus fare. In the other situation, the markets 'N' (Andradas/São João da Boa Vista), 'O' (Jacutinga/Itapira), 'P' (Extrema/Bragança Paulista) and 'Q' (Selviria/Ilha Solteira) could be classi ied as 'semi-urban'.
Using this methodology, the public authority has grounds on which to initiate a bidding process, aiming at a future delegation of interstate bus transport service with urban characteristics. The bidding notice must ensure the economic viability of the market, considering 15 years operation with controlled bus fare.
The markets 'L' and 'M' could be classi ied as road if the absolute threshold value is considered as a reference (63 points). However, these markets have overall score values closer to the threshold value and should not be de initively classi ied (they are displaced in the indeterminate zone). The markets 'L' and 'M' require further studies to improve the accuracy of the model, taking into account some speci icities of each city.  In the end, this result makes it possible to af irm that the developed method was able to rank the markets. Through this technique, the classi ication of the markets occurred more objectively, according to the pre-established criteria of FPV, EPV and Sub-EPV.

DISCUSSION
The MCDA-C proved to be appropriate for the proposed goal. However, this technique requires careful de inition of criteria and choice of specialists. Even in situations where there is a restriction on the number of specialists, it is possible to take advantage of the existing literature to de ine the FPV, EPV and Sub-EPV.
In the case of this paper, de ining urban characteristics as being an agglomeration of information on social, political-institutional and technical-operational aspects proved effective. Subsequently, the weighting of the criteria and their respective allocation in the Macbeth software are also sensitive activities and must be preceded by justi ications. Once all the steps have been completed, the method becomes robust and able to substantiate the decision of the public authority.

CONCLUSION
This paper shows how scienti ic theory, based on a Multi-Criteria Decision Analysis-Aid -Constructivist (MCDA-C) technique, can be useful when handling policy decision-making. As a constructivist approach, the method opens space for discussion and brainstorming, enabling the creation of duly substantiated regulations.
The study presented a method to classify the interstate bus transport services in Brazil as 'semi-urban' or 'road'. This method complements the one presented by Brandão Filho et al. (2016) and becomes useful for public authority decision-making. The 'semi-urban', delegated by Union concession, requires bidding process and must be provided by urban buses. The 'road', authorized by the public authority, needs to be provided by coaches. The model developed measures the degree of 'urban characteristics' using the MCDA-C technique, supported by the MACBETH software.
The study was important as a subsidy for the decision-making of National Land Transport Agency (ANTT) managers. As a result, at least four markets should be indicated to conduct bidding processes.
The proposed technique can help solve local problems in other countries in the same way that it has been successful inr Brazil. In order to adapt the method, some points of view can be inserted or excluded. In addition, where appropriate, the indicator weights can be changed as desired.
As a suggestion for new research, it would be possible to test new criteria to de ine urban characteristics and compare them to this study, in ways such as: using Big Data through smartphones to characterize demand and low of people between municipalities; using the Social Vulnerability Index (IPEA, 2015) as an element of social and urban pro ile integration; and hierarchizing and weighting the municipalities according to their size.
As a lesson learned, it has been proven that transport policies should not be implemented with incomplete regulation; correcting their failures may require much more effort in the future.