Residential Location , Mobility , and Travel Time : A Pilot Study in a Small-Size Italian Metropolitan Area

(is research concerns the topic of Land Use and Transport Interaction (LUTI) models. In particular, the patterns between residential households’ location and mobility choices are analyzed and simulated. (e attributes that influence household residential location choices belong to four categories: socioeconomic and mobility attributes of households and/or of their components; land use; real-estate market; transport system.(e paper presents the results of a pilot study on households’ location andmobility patterns in themetropolitan area of Reggio Calabria (Southern Italy).(e pilot study is divided into two stages. In the first stage, a survey allowed to collect information and identify existing patterns about residential and mobility choices of a sample of households. In the second stage, a residential location model is proposed and some preliminary calibrations are presented in a prototypal way. (e pilot study could be extended and improved in terms of spatial extension and sample dimension in order to allow a complete specification-calibration-validation process of the model. (e model development can support the land usetransport planning process in the Metropolitan City of Reggio Calabria.


Introduction
Land Use and Transport Interaction (LUTI) models simulate the two-way relationship between land use and transport systems in urban areas. e LU-to-T relationship influences transport, in the component of travel demand. On the other side, the T-to-LU relationship plays a prominent role in the spatial organization of the area, influencing activity location and land prices [1][2][3][4].
is paper investigates some elements concerning the T-to-LU relationship. In particular, it focuses on causal mechanism that links the transport network (supply) performances and costs, such as travel times [5,6], to the residential location and mobility choices (e.g., destination, mode).
As far as the modelling approach is concerned, the residential location models present in the literature can be grouped in two broad classes. e former is named gravitational-entropic, and it is based on gravitational theory and/ or on entropy maximization principle [12,13]. e latter is named discrete choice, and it has random utility as theoretical background. In the sphere of discrete choice models [14][15][16], the above described casual mechanism could be explained and simulated in terms of what extent travel cost (e.g., time), and other characteristics (attributes), maximizes the individual (dis)utility of a citizen (or other decisionmaker unit) associated with residential location choices. e paper presents the results of a pilot study on households' location and mobility patterns in the metropolitan area of Reggio Calabria (Southern Italy). e pilot study is finalized to address a number of issues [17] such as the target population, the structuring of the questionnaire and the survey execution, the mobility and residential location patterns, and the attributes that affect the residential location choice process. e pilot study is divided into two stages.
(i) e first stage concerns a survey, which has been designed and executed in order to collect information about residential and mobility choices of a sample of households living in the Italian Metropolitan City of Reggio Calabria (Italy). In 2014, a national law formally defined and introduced the metropolitan cities in Italy, which imposed the building of new institutional and functional assets to these new administrative entities [18,19]. (ii) e second stage concerns the specification and calibration of household residential choice models in a prototypal way, with the objective of investigating in which measure the residential location choices of households are affected by network travel times.
e results of the pilot study concern the following: (i) empirical evidences of households' residential location and mobility patterns; (ii) specifications and calibrations of a prototypal discrete choice model of residential location choices.
In the above context, the research concerns the identification of the factors/drivers that could influence household residential (re)locations inside a small-size metropolitan area, as the Metropolitan City of Reggio Calabria, in the medium-long term. ese elements could support the land use-transport planning process, in behalf of the institutional and functional building process of the Metropolitan City of Reggio Calabria. e development of the complete model is out of the scope of this research. e paper, after this introduction, is structured into four sections. Section 2 presents a state of the art of discrete choice models on residential location. Section 3 reports the theoretical background of the specified residential model and the main elements of the pilot study. Section 4 describes some preliminary results of a specification-calibration-validation process of a prototypal residential model. e research perspectives are reported in the last section.

Residential Location: A Literature Review on
Discrete Choice Models e scientific literature on residential location models can be grouped in two broad classes. e first class is composed of gravitational-entropic models, based on gravitational theory and/or on entropy maximization principle (e.g., [12,13]). To the second class belong the discrete choice models, which refer to random utility theory [14][15][16].
Without being exhaustive, the attention is focused on research papers relating on discrete choice models of residential location choice of households inside urban areas. ey may be classified according to the following criteria: (1) eoretical background; (2) Level of aggregation of decision-maker: aggregate/ disaggregate; (3) Typology of model attributes: 3(a) socioeconomic, 3(b) land use, 3(c) real estate market, and 3(d) transport.
As far as the theoretical background is concerned, it is worth noting that the location model is always based on the simulation of the agents' choices (behaviour). erefore, models can belong, or not, to the class of random utility theory with discrete choice model. e random utility theory with discrete choice model was initially proposed by McFadden [14] who proposed a discrete choice model to estimate the relative influence of location-specific attributes on residential location choices. Further studies were proposed by Anas and Duann [20] which were part of more complex system of models to simulate the whole LUTI process. Later, Frenkel et al. [21] developed a logit model that has been used to simulate the residential location choice of a category of inhabitants of a metropolitan area: the knowledge-workers. Some modelling approaches estimate effects of spatial distribution of urban freight facilities [22], also analyzing retail-store movements in relation to shopping trip behaviours [23]. e logit model is the most widespread one in literature. A logit model is specified in [24] to investigate the relationship between residential location choices, gender, and commuting trips in the Tel Aviv metropolitan area. Lee and Waddell [25] presented a nested logit model to investigate the patterns between residential mobility and location choice in Seattle (USA). Out of models belonging to the random utility theory, Manaugh et al. [26] presented a logistic regression analysis for estimating regional home-work location according to trip length in Montréal (Canada). In their study, sociodemographic variables at the individual and household level do not have enough explanation capacity of trip length variability. de Palma et al. [27] carried out a semi-hedonic regression of the selling prices of new and existing detached houses and apartments, based on a wide number of attributes regarding the real estate market of the Ile-de-France region.
According to the level of aggregation of decisionmaker, the difference is between considering the main agent who makes the decision (which is representative of the whole household) and whether the location choice is modelled as the result of an interaction among the members of the household (see [28]). Aggregate models use the whole household as in [29], where the residential location choices of households are estimated, by analyzing a sample of households living in Texas. Disaggregated models consider each single member, as in [30] that investigated the residential location choice process in a transitional housing market, as the one of Beijing (China). Marcucci et al. [31] worked on residential location preferences and relative decision power in three-member households of the city of Rome and of the Friuli-Venezia-Giulia region (Italy).
According to the typology of model attributes, the following four classes may be considered. Concerning socioeconomic attributes (3(a)), Prashker et al. [24] considered the age and sex as relevant attributes in their study, which reveals significant differences of behaviour between men and women. e work of Wu et al. [30] includes the level of education, income, and number of children as relevant variables in residential location choices. A common attribute to describe the land use characteristics (3(b)) is related to population density. Bhat and Guo [29] introduced it to capture the clustering effect of households. e presence of green areas [30] and the density of underground stations [32] are also considered as factors that affect the household location choices. e attributes of the real estate market (3(c)) are generally related to the floor-space stock and the selling price. Balbontin et al. [32] focused on the importance of dwelling and neighborhood attributes in residential location modelling in the center of Santiago de Chile. Ibeas et al. [33] studied interactions between transport and real-estate values in urban systems; they presented different kinds of hedonic models to estimate house price variations in metropolitan areas in relation to accessibility conditions. Some authors studied spatial effects related to interactions between trip generation, accessibility, and real estate values [34,35]. As far as transport attributes are concerned (3(d)), the distance from the CBD is frequently considered. Prashker et al. [24] considered travel distance and travel time by car and bus, while Wu et al. [30] considered average commuting time. Manaugh et al. [26] considered commuting distance from home to work as attribute that qualifies the transport system.
In relation to the literature review, the research contribution of the paper concerns the analysis of residential location process of households in the attempt to identify the attributes/factors that drive their choices, with a specific focus on the travel time attribute (3(d)), attribute related to transport system). e case study is a small-size metropolitan area, the Metropolitan City of Reggio Calabria (Southern Italy). e prototypal model presented considers the demand side of the residential location process, based on a behavioural approach by maximizing the utilities of households in choosing their residential location zone (random utility theory with discrete choice model). e input variables of the model are connected to travel time on the network from the residential origin to the job/school destinations, and the output variable is the probability of the household to choose a zone as residential location.

Residential Location: A Pilot Study on a
Small-Size Italian Metropolitan Area e pilot study aimed to address a number of issues, including the target population, the structuring of the questionnaire and the survey execution, the mobility and residential location patterns, and the attributes that affect the residential location choice process. e target population (households) was selected from a study area inside the Metropolitan City of Reggio Calabria ( Figure 1). e study area includes the Municipality of Reggio Calabria and a set of municipalities lying along the yrrenian and part of Ionian coast (depicted in light grey in Figure 1). e dimensions of the study area are about 560 km 2 with a population of about 275,000 inhabitants and 11 municipalities. e largest municipality is Reggio Calabria, which occupies an area of about 236 km 2 with about 182,551 of inhabitants. e study area represents the most populated portion of the Metropolitan City of Reggio Calabria, as it has 49% of the whole population, with an average population density of 491 inh./km 2 (against 108 inh./km 2 in the remaining part).
e study area was discretized into 14 zones. e Municipality of Reggio Calabria was subdivided into four zones called Center, North, South, and Hill.
As far as the transport infrastructures and services are concerned, the Tyrrhenian area has more infrastructural facilities (highway, double-line railway, and SS18 national road) and transport services than Ionian area (national road and single-line railway) in terms of quantity and quality. Transport infrastructures and services in the Tyrrhenian area have a European relevance and they represent principal connections to North Italy. e questionnaire is structured into two sections. e first section concerns information about the household unit, as a whole. e second section is dedicated to individual household component. e survey (first stage of the pilot study) was developed in two steps: (1) A preliminary version of a questionnaire is submitted to a small group of households, in order to validate it; (2) An advanced version of the questionnaire is submitted to a larger group.
e survey was proposed to respondents during two weeks, by interviewing them at home or via e-mail. e number of interviewed households was 51, for 182 components. Some papers in literature (see [17]) identify the sufficient sample size to execute a correct pilot study. Isaac and Michael [36] say that sample sizes "between 10 and 30 have some advantages, such as simplicity, easy calculation, Journal of Advanced Transportation and the ability to test hypotheses." Hertzog [37] makes different recommendations for sample size ranging from 10-15 in feasibility studies to 30 to 40 for pilot studies. In the opinion of the authors, the size of the sample is sufficient to execute a correct pilot study whose results are presented in the following. e next part of the section concerns the mobility and location patterns and the attributes that affect the residential location choice process.
It is structured as follows. Section 3.1 concerns the specification of a behavioural discrete choice model of residential location. Section 3.2 focuses on the characteristics of the households of the survey sample. Section 3.3 presents the analysis and discussion of the attributes obtained from the survey.

3.1.
eoretical Background. e proposed model is a behavioural discrete choice model of residential location, assuming the household as decision-maker. e residential location choice relates to the demand side of residential location process, which can be as the result of an "equilibrium" of the interaction between the demand of residence locations and the supply of land [3].
According to the random utility theory [14][15][16], a household i chooses a zone, among the set of available zones Q, o ∈ Q, to locate the residence by maximizing the associated perceived utility, U i o . As the utility, U i o , is a random variable, it is not possible to establish with certainty which zone will be chosen by the household for living/residing, while it is possible to express the probability that the zone o, belonging to Q, will be chosen from household as the probability that the perceived utility of zone o, U i o , is greater than the perceived utility of every other zone q, U i q : where Q is the choice set or the set of available alternatives (zones) for the household.
Assuming that the perceived utility, U i o , is independently and identically distributed (i.i.d.) according to a Gumbel across the alternatives, equation (1) is specified with a multinomial logit. Further model specifications proposed recently, based on different distributions for the utility (i.e., Fuzzy or Quantum Utility models), for a different choice level could be tested [35].  Journal of Advanced Transportation e probability of choosing a zone o as residential zone for household i is where o, zone; i, household; Q, total number of zones (available alternatives); and E[U i o ], expected value of perceived utility U i o . e expected value of perceived utility is a linear combination of attributes and weights: where x i o,k , value of the attribute k of the alternative (zone) o for the household i and β k , weight associated with attribute k to be calibrated. e attributes may belong to four categories: (a) socioeconomic attributes of household; (b) land use attributes; (c) real estate attributes of residential buildings; (d) transport system attributes.

Level of Aggregation of Decision-Maker.
e survey was carried out in order to obtain some elements about the structure of the sample of households. e analysis of the interviews shows that the sample is mainly composed of households with four components. e age groups of 25-34 and 55-65 years are the most represented (Figures 2(a) and 2(b)).

Relevant Attributes: Specification and Estimation.
For what concerns the case study, the specification of attributes for each category is reported as follows: (1) Socioeconomic and mobility attributes of household: professional/school status; origin-destination (OD) matrix of daily trips; and transport modes for homework and home-school trips; (2) Land-use: population, employees in industry sector, employees in retail sector, and employees in service sector; (3) Real-estate market: average sale price per zone for residential buildings; (4) Transport supply: density of bus stops per zone, density of rail stations per zone, and active accessibility of each zone.
e socioeconomic and mobility attributes were obtained by means of the survey, proposed to the sample of households living in the three defined territorial areas defined above (Figure 1). e land-use data concern the population and employment census data, available on the website of the Italian National Institute of Statistics [39]. e number of employees per zone of the study area was subdivided considering the following sectors: industry, retail, and services. e data of the real-estate market concern the average selling price per zone of residential buildings, which was obtained from the website of the Italian Real Estate Market Observatory (http://www.agenziaentrate.gov.it). e data of the transport system concern the supply of public transport, in particular the number of bus stops per zone and the number of railway stations per zone. e number of stops and stations is obtained from the information available on websites of public transport companies of the municipalities of the study area. e travel time has been estimated considering the private car as transport mode, on an average working day from 11 : 00 am to 12 : 00 am. Some information about the quality of the transport infrastructure [40] could be considered in the future.

Relevant Attributes: Analysis and Discussion
(1) Socioeconomic and Mobility Attributes of Household.
Respondents are categorized as employees in the public and private sectors, freelancers, housewives/husbands, or retired people. e number of the unemployed of the sample is small (Figure 3(a)). e respondents who are students are mainly undergraduates and high school students (Figure 3(b)). e home-work and home-school OD matrix of the sample of households for an average working day is presented in Table 1. e trips inside each territorial area (elements of the diagonal) present the highest percentages, in particular the trips inside the Municipality of Reggio Calabria (35.2%). A percentage of respondents (16%) travels to destinations located outside the study area. e number of trips between the different territorial areas is lower and they mainly have the Municipality of Reggio Calabria as destination.
e majority of the sample does not travel every day for home-work purpose, while the other part of the sample performs at least one daily trip (Figure 4(a)). Respondents who do not travel daily for home-school purpose are the majority (Figure 4(b)). e private car is the preferred mode to travel for the sample of households, followed by pedestrian trips and trips by bus ( Figure 5).
(2) Land-Use. Figures 6(a) and 6(b) show that the study area is characterized by the presence of a great pole, the central macro-zone of the Municipality of Reggio Calabria with more than 100,000 inhabitants (Figure 6(a)). e municipalities of Villa San Giovanni, Palmi, and Gioia Tauro (Tyrrhenian area) and the macro-zones of Reggio Calabria (North, South, and Hill) have more than 10,000 inhabitants. e central area of Reggio Calabria has more than 25,000 employees in the service sector, and the main part of     ere are three zones where the selling price of residential buildings is more than the average value ( Figure 7).
(4) Transport Supply. e presence of the pole inside the Metropolitan City of Reggio Calabria is in line with the high densities of bus stops and rail stations in that macro-zone. In the central area of Reggio Calabria, there are 5 bus stops/km 2 (Figure 8(a)) and 0.20 stations/km 2 (Figure 8(b)). One urban bus company operates inside the Municipality of Reggio Calabria providing transit services, while several bus companies offer suburban bus services in the Tyrrhenian and Ionian areas.
e active accessibility is estimated as the sum of travel times along the minimum path of the road transport network to reach all zones of the study area from a given zone. e value increases from central area to peripheral zones of the study area. e lowest values of accessibility (maximum values of travel times) are estimated for zones belonging to Ionian area (Figure 9). According to the calibration results, the whole set of calibrated models has been classified into three classes:

Residential Location: Model Calibration
(i) Models with incorrect sign and low statistical significance of parameters; (ii) Models with correct sign and low statistical significance of parameters; (iii) Models with correct sign and acceptable statistical significance.
e models belonging to classes (i) and (ii) are discarded and they are not further considered in the analysis. e models belonging to class (iii) present attributes related to the transport supply, in particular travel times on the road network (see Table 2). e specification of the utilities of models (iii) is reported in the following. e utility of model 1 is specified as the linear combination of parameters and attributes of travel time, t_HH_Comp, of the first four components of the household: (4) e utility of model 2 is calculated by means of equation (5) by considering a specific parameter for each of the main three territorial areas of the study area: Reggio Calabria, Tyrrhenian Area, and Ionian Area (see Figure 1): e utility of model 3 is specified by means of equation (6) introducing a dummy variable related to travel time threshold of 10 minutes from the city center: e calibrated parameters of models specified by means of equations (4)-(6) are reported in Table 3. Model of equation (4) takes into account each individual travel time for up-tofour-components household. e negative signs of the calibrated parameters are in line with the nature of the attribute: their values indicate that disutility increases if the household is formed by more than two components. Model of equation (5) considers travel times to reach the three macro-zones: Reggio Calabria, Tyrrhenian, and Ionian. e parameters indicate that the utility decreases more for the households living in macro-zone Tyrrhenian than the others. For what concerns model of equation (6), the parameter of travel time is negative (with a high t-student value), while the parameter of the dummy attribute is positive, indicating that the negative impact of travel time is mitigated for reduced values of travel time (less than 10 minutes). e low t-student value indicates low significance of the calibrated parameter. e calibrated models are validated by means of test on the likelihood ratio and on the goodness of fit of the model,  Journal of Advanced Transportation as statistic, ρ 2 , and adjusted statistic, ρ 2 _adj. e results of the tests are shown in Table 4. e null parameter of maximum likelihood function, L[0], has a constant value, while it assumes a decreasing value for the associated values of calibrated parameters, L[β(ML)]. Model 2 presents values of ρ 2 statistics equal to 0.47 and ρ 2 _adj equal to 0.40, which are slightly higher than the corresponding values associated with models 1 and 3.

Conclusions and Research Perspectives
e objective of the research was to estimate households' location and mobility choices inside a small-size Italian metropolitan area. e research was divided into two stages. e first stage concerned a survey in order to collect information and identify residential and mobility patterns of a sample of households. e second stage regarded the specification and calibration of some prototypal discrete choice models.
At this stage of the research, the models with values of calibrated parameters having correct sign and acceptable statistical significance are the ones with attributes connected with travel time on the road network. e presence of a relative great attraction pole of mobility in the metropolitan area, represented by the Municipality of Reggio Calabria, has relevant influence in the residential location choice of households, as the t-student values of the attributes t_center, t_tyrrhenian, t_ionian of model 2 show. Moreover, these first results show the role of travel times in residential location choices. For instance, the higher parameter's value estimated for the variable t_ionian with respect to the other two variables (model 2) confirms the low level of infrastructural facilities of the Ionian area.
e result of this pilot study can be useful in order to support the land use-transport planning process in the Metropolitan City of Reggio Calabria. In particular, a further development of this study in terms of spatial extension and sample dimension could allow for estimating the effects on households' residential choices caused by investments on new road transport infrastructures (e.g., a new highway that connects Tyrrenian and Ionian areas bypassing the zone "center" of Reggio Calabria). e following research perspectives are foreseen for the next future. e survey will be extended to households belonging to all municipalities of the metropolitan area of Reggio Calabria. e modelling framework will be extended and improved by means of a complete specification-calibration-validation process: new functional forms and utility specifications including socioeconomic, land use, and real estate market attributes in order to better capture the household location choice behaviour. e proposed model will be applied in order to support a strategic transport planning process [41]. e final aim is to support evaluation of effects of land use-transport scenarios on sustainability and equity [42,43].

Data Availability
e underlying data used to support the results of the study are not freely available due to respondents' privacy.

Conflicts of Interest
e authors declare that they have no conflicts of interest. Sum of travel times of home-work and home-school morning trips of all household members t_HH_Comp1. . . 4 Travel times of home-work and home-school morning trips for single household member (up to 4 members) t_center Travel times from center zone of Reggio Calabria to each zone (see Figure 1) t_thyrrhenian Travel times from Tyrrhenian zones to each zone (see Figure 1) t_ionian Travel times from Ionian zones to each zone (see Figure 1) t < 10 min Dummy attribute indicating if the zones are less than 10 minutes far from the city center