Undoubtedly, estimating a hybrid model contributes to the understanding of determinants of individual **route** **choice** **behavior** in urban networks. Findings suggest that individuals generally prefer shorter, faster and less congested routes, but also that their characteristics, their spatial abilities and their behavioral patterns significantly influence their preferences and could even bring them to ignore better alternatives because they are comfortable with their current choices. Further research could concentrate on the simplification of the model specification with a lower number of **variables**, on the consideration of the concept of landmark similarity in the **route** **choice** model and not only in the **choice** set generation, and on the analysis of the effect of **latent** **variables** on the **choice** set formation process when a joint model of **choice** set generation and **route** **choice** is proposed and estimated.

Show more
34 Read more

A comparison between observable level-of-service attributes and **latent** attitudinal **variables** finds the latter to have a greater influence on aggregate travel mode shares. For example, our model predicts that a 10 % increase in attitudes towards comfort and con- venience across the sample population will reduce the market share for car by 8.4 %, where a 10 % increase could be taken heuristically to represent a modest improvement in attitudes. 3 In comparison, a 10 % increase in car travel times across the sample population is expected to reduce the market share for car by a meager 0.9 %. Individual attitudes may change through situational forces such as those mentioned in the previous paragraph in the short-term and through changes in values brought on about by family learning and socialization in the long-term. While the mode **choice** sub-model provides useful infor- mation on how better to market public transit, the structural component of the attitudes sub-model offers guidance on how each of these different policies might best be imple- mented. All three attitudinal constructs are significantly and positively explained by the three value constructs, suggesting that the indirect influence of each of the value constructs on the relative appeal of public transit can be decomposed into these three constituent competing effects. Interestingly, the cumulative indirect effect of values toward security on the comparative attractiveness of public transit is negligible. In fact, values toward hedonism have the strongest negative indirect effect on transit ridership levels, followed by

Show more
16 Read more

The relationship between space syntax measurements and bicycle volumes was first investigated. Table 4.2 shows the coefficients and statistics for the regression models with global integration and local integration as the sole explanatory **variables**. Global integration is not statistically significant and can hardly explain the actual bicycle volumes. Local integration, however, is statistically significant and exhibits a good R-squared value. Moreover, the regression results indicate that local integration is positively related to bicycle volumes, which is as expected. This finding suggests that local integration provides stronger explanatory power than global integration in modeling bicycle movement. The bicycle is extremely convenient for short-range trips, however, it is not suitable for a long-distance travel because it is human-powered. Therefore, local integration, which only considers the accessibility of a road segment within a limited travel distance, is more appropriate in modeling bicycle traffic.

Show more
31 Read more

Previous **route** **choice** studies have mostly focused on the effects of observable factors, such as **route** attributes and socio-economic characteristics (Dalumpines & Scott, 2017a; Jan et al., 2000). According to (Jan et al., 2000), factors affecting **route** **choice** decisions can be classified into four categories including travellers’ attributes (such as age, gender, education, income, etc.), **route** attributes (such as traffic conditions, speed limits, number of turns, pavement quality, etc.), trip attributes (such as trip purpose, travel time, etc.), and circumstances (such as weather conditions, time of day, traffic information, etc.). However, **route** **choice** decisions might not be exclusively dependent on these observable **variables**, but also on **latent** **variables**, which cannot be directly observed, and measured, such as attitudes, perceptions, and lifestyle preferences (Gärling et al., 1998; Hurtubia et al., 2014; McFadden, 1986, 1999). Since every decision maker may have a different perception of these **variables**, they are considered to be intrinsically subjective (Raveau et al., 2010). The explicit incorporation of these **latent** constructs in the **choice** process improves the explanatory power of these models (Ben-Akiva et al., 2002; Prato et al., 2012; Walker, 2001). Accordingly, different segments of the population, characterized by some of these **latent** constructs, might also have different **choice** behaviours (Hurtubia et al., 2014). However, the **latent** behavioural heterogeneity among the population has mostly been ignored by assuming that all the individuals in the sample population have similar levels of driving experience, spatial knowledge, familiarity with the road network, ability to process information, motivation to compare all the considered alternatives, etc. Ignoring these sources of heterogeneity could reduce the explanatory power of the model and introduce errors in model’s forecasts (Ben-Akiva et al., 1993).

Show more
209 Read more

En s’appuyant sur le cadre défini pour l’identification de chemin dans le cas unimodal, une méthode « map-matching » multimodale est développée pour résoudre le problème plus général où les trajets peuvent être effectués à l’aide de différents modes et que ce dernier est inconnu. Nous déduisons à la fois le chemin et le mode simultanément à partir de données variées. L’algorithme de génération de chemins candidats est étendu pour gérer les réseaux multimodaux, et pour générer des chemins multimodaux, où un mode de transport est associé à chaque tronçon de **route**. Le statut **latent** contient le mode et le lieu, and la corrélation entre les deux est utilisées. Par exemple, si le mode de transport est le bus, the chemin doit suivre les lignes de bus existantes. En plus du signal GPS, l’accélération et le signal Bluetooth participent aussi à la collecte d’information sur la mobilité, et ils sont donc intégrés dans le modèle probabiliste de mesure à l’aide d’un modèle de mesure correspondant à chaque capteur. Le modèle pour l’accélération fournit un indicateur de déplacement qui peut être utilisé pour déterminer le mode de transport. Les données du signal Bluetooth fournissent le nombre d’appareils Bluetooth dans les environs, ce qui peut-être exploité par exemple pour identifier l’utilisation des transports publics dans le cas d’un grand nombre d’appareils identifiés. Cette approche est flexible pour deux raisons. D’abord, d’autres types de données de capteurs peuvent être intégrés, pour autant qu’un modèle de mesure correspondant au type de capteur est fourni. Ensuite, tout réseau de transports publics peut être ajouté ou supprimé en fonction des besoins et de la disponibilité. Les données enregistrées lors d’un trajet ne nécessitent pas de subir un prétraitement sous forme de segment de mobilité unimodal ; ainsi, le risque de fausse segmentation est atténué. Des expériences numériques présentent des visualisations sur carte de certains exemples de trajets, et une analyse de la performance de la détection du mode de transport.

Show more
130 Read more

We were limited in this analysis to categorical and often dichotomously rated pain indicators. Future models could be built from more normally distributed pain rat- ings, ideally from validated pain assessment scales. The resulting **latent** **variables** would be more normally dis- tributed, which would then allow us to associate them with potential biomarkers, using powerful parametric statistical methods. The necessary observed **variables** could be obtained by clinicians in the field, or over the phone. Pd and Ps scores could then be generated in the field by web or phone-based applications running multi- variate regression classification models. Future studies might then use these phenotypes as outcome measures in clinical pain management trials. If we could demon- strate specific treatment effects of analgesic, antidepres- sant or cognitive enhancing therapies on Ps, Pd (and potentially Pc) respectively, then treatment decisions could be individualized on the basis of an individual’s Ps, Pd and Pc scores.

Show more
10 Read more

Eq. (1) is denoted as “X i = Disc(ξ i , α (i) )”, as function of ξ i and α (i) . Early in the twen- tieth century Pearson (1900), Pearson and Pearson (1922), see also the bibliography in Goodman (1981), have proposed the polychoric correlation, i.e. the Pearson’s correlation among the corresponding **latent** **variables** for measuring association among ordinal vari- ables. In the eighties, the practitioners of covariance structure models, using packages such as LISREL or EQS, widened the scope of these models, originally conceived for continuous **variables**, by using, for ordinal **variables**, polychoric correlations the same way they used Pearson’s correlations for continuous **variables** (see e.g. Muthén (1983, 1984), Jöreskog et al. (2002) among others). A relevant results is due to Olsson (1979), who describes for the first time a maximum likelihood algorithm for computing the polychoric correlations. Jöreskog (2002) discusses a way for testing the normality based on the chi square distance or on the log-likelihood ratio, but does not make a thorough discussion of the meaning of the underlying normality hypothesis.

Show more
23 Read more

When you are traveling in a novel city, you probably need to check maps and plan a **route** from the nearest subway station to your hotel. Most often, there is more than one **route** **choice**. For example, there may be several stations nearby, so you need to decide which one to use and choose the best **route**. However, the definition of the “best **route**” is ambiguous. It might refer to the shortest one, the simplest one, or the one easiest to find. According to previous findings on **route** **choice**, people do not always choose the shortest **route** (Bailenson, Shum, & Uttal, 1998, 2000). Instead, their **route**-finding decisions depend on the use of general heuristics (Bailenson et al., 1998; Brunyé et al., 2012; Brunyé, Mahoney, Gardony, & Taylor, 2010; Yang & Schwaninger, 2011). For instance, people tend to select the street most in line with the target (Hochmair & Karlsson, 2005), or prefer the extreme routes over the middle ones although all the routes were the same length and required the same number of turns (Christenfeld, 1995). Furthermore, the routes with straight initial segments are preferred even though the routes may not be the shortest in distance (Bailenson et al., 1998, 2000).

Show more
tasks but not necessarily effective in text model- ing. At the beginning of training, the **latent** vari- able z contains a small amount of information about the input sentence. Many **latent** units of z are pulled towards the prior early to optimize an objective function before they capture useful in- formation (Hoffman et al., 2013; Sønderby et al., 2016). Without a cost annealing strategy or a con- straint on the decoder (Bowman et al., 2016; Chen et al., 2017; Yang et al., 2017), z would be en- tirely ignored for the remaining training steps. In our work, we aim at developing a simple variant of the LSTM-VAE model to address this common training issue. We observe that pairing the input sentence with multiple **latent** **variables** improves **latent** variable usage. In addition, we present a method that leverages multiple **latent** **variables** to further boost the performance of the baseline LSTM-VAE model.

Show more
A number of comments may be made about this **choice** process. In proposing it, Mahmassani and colleagues have recommended that the decision to switch **route** from that previously used should only be made subject to the additional constraint that the absolute saving in travel cost is greater than some fixed minimum value (any smaller savings being imperceptible to the user). Furthermore, the question arises as to how this **choice** process should be implemented for the individual's first trip between this origin- destination pair; this is an important issue, as it has been shown that the final state of a network under such decision rules is very likely to be dependent on the initial conditions. The most obvious starting condition of choosing the minimum perceived cost **route** may therefore be inadvisable (It may prove difficult to move away from an unrealistic initial pattern). An alternative would be to apply a straight cost minimising rule for a number of days, as a start-up period, before introducing boundedly rational **choice**. When the aim of the study is a before-and-after assessment of some measure, then a sensible starting point for the simulation of the `after' situation are the choices which prevailed from the `before' scenario.

Show more
39 Read more

The first group of cycle planners are ones which are covering travelling around the city. An example which could be used is the Vancouver cycle planner (Cycling **Route** Planner, 2007). The user has a possibility to set his preferences, such as priority usage of cyclist facilities instead of major roads, shortest path, least elevation gain path, least traffic pollution or mostly vegetated path. Then the planner finds ideal path which is based on these preferences and display the path on the map with overall information about it along with directions. Another even more detail approach is used in the OPT for health – **route** planner for San Francisco (OPT for Health, 2010). Cyclists are not the only target group but possibility to adjust the searching for ideal **route** is here even more complex. The user is allowed to set values for each attribute which is considered (e.g. most bicycle friendly, major road, traffic access restricted…). This setting is supposed to lead to improvement of searched path. On the other hand there are other types of cycle planners which are trying to be still complex even with limited amount of adjustable preferences such as San Francisco Bicycle trip planner (2009). In this planner the user can choose only from three types of paths – shortest, balanced or biker friendly and setting of maximum grade.

Show more
16 Read more

LATHAT VARIABLN MODELS FOR MIXED MANIFEST VARIABLES Irini Moustaki London School of Economics and Political Science University of London Submitted in Fulfilment of the Requirement for the Degree of Do[.]

187 Read more

However, there are a few notable exceptions. One is the performance improvements demon- strated in two different papers using a method for redistributing the weight of principal components (PCs) in factorized DSMs (Caron, 2001; Bulli- naria and Levy, 2012). In the latter of these pa- pers, the factorization of **latent** **variables** in DSMs is used to reach a perfect score of 100% correct answers on the TOEFL synonym test. This re- sult is somewhat surprising, since the factorization method is the inverse of what is normally used.

The **route** **choice** problem is presented as discrete **choice** problem. Five **route** **choice** models mentioned in the literature (MNL, CNL, PCL, PSL and C-Logit) have been theoretically described. A large scale network is used to generated routesets for a sample of origins and destinations (26 x 26 zones). From the literature a probit simulation technique is adopted and implemented in Matlab. **Route** **choice** probabilities are estimated for all routes (2148 in total, of which 2010 significant), based on an arbitrary spread parameter. The Logit-based **route** **choice** models are then calibrated against a random sample of routesets and validated against all routesets (538 relevant sets in total). The validation process included an analysis of the model performance to characteristic of the routeset. From this analysis the PCL model is chosen as applicable for model use.

Show more
84 Read more

The two-stage model described by Equation 1 can accommodate a wide range of possible behaviors. Indeed, if each **choice** set C(A) is single-valued, then we will see that any **choice** data can result from the postulated structure. More generally, the only restriction imposed by the framework itself has to do with the alternatives contained in certain “behavioral indi¤erence classes.” (See Section 2.2 for further discussion of these points.) And it follows that a manipulator with access to all three psychological **variables** has nearly complete control over the decision maker’s choices.

Show more
17 Read more

The main purpose of this article is to guide practitioners in their **choice** of a methodology for the estimation of a binary **choice** model with potentially endogenous explanatory **variables**. We are primarily interested in assessing the performance and robustness of an estimator that has been proposed recently (Lewbel et al., 2012; Dong and Lewbel, 2015). This estimator, which was originally described in Lewbel (2000), has been of greater interest lately after Dong and Lewbel (2015) provided a quite simple (step‐by‐step) estimation procedure. The availability of an estimation package in one of the most popular statistical software further increased its popularity. 1 Lewbel’s estimator relies on a continuously distributed, strictly exogenous “special regressor” that has to satisfy a large support condition. When such a special regressor is available and satisfies all required assumptions, Lewbel’s estimator may be a relevant alternative to more common control function and Maximum Likelihood (ML) estimators. 2

Show more
32 Read more

Under these assumptions which is the most probable **choice** Alice will make in each scenario? Answering this question requires formulating a **choice** model that explains her choices. However, estimating the impact of information within the framework of a **choice** model is hardly a simple task. Since information is assumed to change the level of uncertainty in the **choice** situation, the main challenge is to model travelers' response to information. Thus, the assumption of perfect information generally embedded in random utility-based **route**-**choice** models, i.e. without information effects, is discarded. These models assume drivers’ cognition is governed by rational **choice**. They do not explicitly abstract **behavior** under uncertainty; rather they apply different sets of assumptions on the distributions of the unobservable factors (Watling & van Vuren, 1993). In this context, flexible error terms as applied in mixed discrete **choice** models such as Mixed Logit (Srinivasan & Mahmassani, 1999) and Mixed Probit (Mahmassani & Liu, 1999) have proved of added value. Using dynamic specifications is another approach to reduce uncertainty by updating the level of utility over time (Horowitz, 1984). It is also possible to assume that travelers' cognition reflects bounded rationality (Simon, 1982) whereby choices are made based on simple decision rules or heuristics in the form of thresholds of accepting possible outcomes. Srinivasan & Mahamassani, (2003) applied this approach by using a Mixed Logit

Show more
33 Read more

Suppose initially that much the greater proportion of the OD flow is initially on **route** 2. In this case, delays at the junction are small since congestion at the junction is small for any reasonable signal control policy. Thus in this case travel times are dominated by the uncongested travel times and so the travel time on **route** 2 will exceed the travel time along **route** 1. Thus, in the dynamical system illustrated in figure 1, flow will naturally swap from **route** 2 to the quicker **route** 1 as time passes. This will happen under any reasonable signal control policy including P 0 . Under reasonable assumptions, eventually

Show more
14 Read more

The Dutch Design Week (DDW) 2017 attracted around 335,000 visitors from both the Netherlands and other countries during a week with more than around 610 exhibitions in 110 locations spread over the city Eindhoven. Due to specific travel purposes, visitors normally have specific destinations targeting the DDW exhibitions. Meanwhile, they might also visit the city for taking a break, a meal or a touring in between exhibition activities. The mixed environment makes modeling **behavior** of DDW visitors more complex than shoppers and tourisms only. Will visitors’ **behavior** be influenced by city built environment, except by exhibitions’ location? In this perception we research how the exhibitions and city built environment together influence visitors’ **route** **choice** becomes the interest. Except the OD (original and destination place) and trip character, the shops, transportation facilities and other built environment besides the **route** may also attract them to change their visiting **route** and destination. In this paper, we will pay special attention to the **route** **choice** between different exhibitions. Knowing how the built environment influences on pedestrian **behavior** can help the event organizer distribute the exhibitions and facilities more efficiently. Thus, visitors can visit more exhibitions and have a better experience.

Show more
BikeAnjo: A group of biking fanatics founded in 2010 that helps new people to ride a bike, as well as organising cycling trips around the city of S˜ao Paulo. With the use of BikeAnjo a large variety of cyclists can be reached, since both experienced and inexperienced cyclists will be part of this community. It is expected that deviations of the mean speeds will be high, but the average speed tends to be a good representation since there will be cyclists from both sides of the spectrum. However, as stated by Stinson & Bhat (2004), there is a differ- ence in **route** **choice** factors between different levels of experience. Inexperienced cyclists tend to give more value to factor related to separation from cars, while experienced cyclists are more sensitive to factors related to travel time.

Show more
51 Read more