Radiografía de las demandas en las comunidades mapuche

The first model represents the different search PC paths that a tourist can choose to follow in order to book a scenic train trip. Being based on a tree, the CEG framework has the necessary flexibility to represent directly the asymmetric un-

foldings that characterises this process. Observe that there are many events with probability zero in its event tree; see the dashed shape nodes in Figure 3.1. Omit- ting the corresponding edges allows us to further simplify the graphical topology of our model and also the computational complexity of the model search.

Figure 4.2: The best scored CEG for the PC sequence model. The stage partition is given by: u0 = {w0}, u1 = {w1}, u2 = {w2}, u3={w3, w9}, u4={w4, w10},

u5 ={w5, w6}, u6 = {w7}, u7 = {w8}, u8={w11, w12, w13, w14}, u9={w16},

u10={w15, w17}, u11 = {w18}, u12 = {w19}. No tourists went through position

w15.

Having a well-defined event tree a priori and assuming a hyper-stage for each set of situations that have the same geometric shape in Figure 3.1, I use the Algorithm 5 to look for the best stage structure configuration. Figure 4.2 depicts the MAP CEG model and Table 4.1 shows the posterior conditional probability mean of each stage with a 95% credible interval. The conditional probabilities of stages u3 and u8

are degenerated due to the domain conditions. Although the situationss3 and s23

are in the same stage u10 and so in the same position, I depict them using two

different positionsw15 and w17 to highlight the fact no tourist visited situation s3

in Figure 3.1 and so the positionw15 in Figure 4.2.

When a client starts examining his option to book a train ticket (position w0),

it is equally likely that he decides to visit a PC Ship or a PC Others. However this initial choice has a strong impact on the tourist’s choice of which PC to book

a train. To understand how we can read this from our CEG model, consider a client that initially visits a PC Others (position w2). Despite it being possible to

book a train at this point this is not likely to happen. This is because there is a clear predisposition (93%) to visit another PC. Note that stage u4 clearly favours

a PC Others (70%). It also plays a key role in this branch of the CEG model since it contains positions w4 and w10. Finally, observe that a tourist at position w6

(stage u5) has a small probability (25%) of booking a train. So we can conclude

that if a tourist goes first to a PC Others he will then probably book a train in a PC Others.

Stage State Space Mean (95% credible interval)(%)

u0 (Ship,Others) 52 (47,57) 48 (43,53) u1 (Booked,Searching) 0.4 (0,2) 99.6 (99,100) u2 (Booked,Searching) 7 (4,11) 93 (89,96) u3 (Ship,Others) 0 (0,0) 100 (100,100) u4 (Ship,Others) 30 (25,36) 70 (64,76) u5 (Booked,Searching) 25 (20,30) 75 (70,80) u6 (Booked,Searching) 56 (47,64) 44 (36,53) u7 (Ship,Others) 83 (71,92) 17 (8,29) u8 (Booked,Searching) 0 (0,0) 100 (100,100) u9 (1,2) 14 (8,23) 86 (77,92) u10 (1,2) 62 (38,83) 38 (17,62) u11 (1,2) 35 (28,42) 65 (58,72) u12 (1,2) 94 (88,97) 6 (3,12)

Table 4.1: Posterior mean and 95% credite intervals for the stages corresponding to the best scored PC sequence CEG depicted in Figure 4.2.

On the other hand, there is a good chance (62%≡0.996×0.75×0.83) that a tourist who initially visits a PC Ship (positionw1) actually books a scenic train in

position w1 there is a very tiny probability (0.4%) that he will book a train there.

So he probably proceeds to positionw5 and visits a PC Others (positionw5) where

he has a strong inclination (75%) to carry on searching (position w8) and so a

great tendency (83%) to return to the PCShip.

Recall that there are an extremely larger numbers of PCs gathered in the category Others compared with the category Ship. Therefore we can hypothesise that visiting a PC Others does not influence the subsequent tourist’s choice between going to a PC Others or a PC Ship, if it is the case. This allows us to justify the fact that the MAP CEG model gathers positionsw4 andw10at the same stage. So

apparently the variables P Ci, i= 2,3 associated with the decisions of which PC

to go given that the previous visited PC was a non-cruise PC (P Ci−1 =Others)

have indistinguishable conditional probability distributions.

Note that since positions w5 (P C1 = s, P C2 =o) and w6 (P C1 =o, P C2 =s)

are at the same stage the order of the two first visited PCs does not have any impact on the second decision with respect to keep searching or to book a train. Also observe that 90% of train bookings during the second visit happens in a PC Others.

Lastly, the MAP CEG model suggests that there are four different groups of tourists (positionsw16, . . . , w19) with respect to the option between public or cruise trains.

The great majority of clients (94%) who visit two non-cruise PCs successively before booking a train consecutively (position w19) prefer public trains. Having

visited a PC Others, a PC Ship and a PC Others consecutively (positionw17) the

chance that a client books a public train is reduced to 62%.

In contrast, the other two groups present a strong preference for cruise trains. Most tourists (65%) booking their trains in a PC Others without visiting two non- cruise PCs - sequences (o), (s,o) at positionw18- tend to book a cruise train. This

also happens with clients who arrive in a PC Ship after visiting two non-cruise PCs successively -sequence (0,0,s) at positionw18. Having not previously gone to two

have the highest probability (86%) to book a cruise train.