Scholarly article on topic 'The Simulation of Shopping Trips at Urban Scale: Attraction Macro-Model'

The Simulation of Shopping Trips at Urban Scale: Attraction Macro-Model Academic research paper on "Economics and business"

CC BY-NC-ND
0
0
Share paper
OECD Field of science
Keywords
{"End-consumer choices" / "urban freight demand" / "shopping trip"}

Abstract of research paper on Economics and business, author of scientific article — Francesco Russo, Antonio Comi

Abstract This paper presents advancement on the calibration of a models system allowing us to estimate the goods attracted within urban and metropolitan areas. It is a component of a general modelling framework proposed by authors developed in order to support ex-ante assessment of city logistics measures. The primary characteristic of this general framework is the representation of interacting behaviour of commodity consumers and commodity suppliers/shippers/retailers. In fact, this framework allows us to simulate goods movements at an urban scale combining urban passenger and commodity flows (commodity flows are generated in order to support a given need), and to investigate the existing relationships among end-consumers and other involved decision-makers (e.g. producers, wholesalers, retailers).

Academic research paper on topic "The Simulation of Shopping Trips at Urban Scale: Attraction Macro-Model"

Available online at www.sciencedirect.com

SciVerse ScienceDirect

Procedia - Social and Behavioral Sciences 39 (2012) 387 - 399

The Seventh International Conference on City Logistics

The simulation of shopping trips at urban scale: Attraction

macro-model

Francesco Russoa, Antonio Comib*

aDepartment of Computer Science, Mathematics, Electronics and Transportation, Mediterranea University of Reggio Calabria, Feo

di Vito, 89060 Reggio Calabria, Italy bDepartment of Enterprise Engineering, Tor Vergata University of Rome, Via del Politecnico 1, 00133 Rome, Italy

Abstract

This paper presents advancement on the calibration of a models system allowing us to estimate the goods attracted within urban and metropolitan areas. It is a component of a general modelling framework proposed by authors developed in order to support ex-ante assessment of city logistics measures. The primary characteristic of this general framework is the representation of interacting behaviour of commodity consumers and commodity suppliers/shippers/retailers. In fact, this framework allows us to simulate goods movements at an urban scale combining urban passenger and commodity flows (commodity flows are generated in order to support a given need), and to investigate the existing relationships among end-consumers and other involved decision-makers (e.g. producers, wholesalers, retailers).

© 2012 Published by Elsevier Ltd. Selection and/or peer-review uuder r esponsibility o f 7th International Conference on City Logistics

Keywords: End-consumer choices; urban freight demand; shopping trip

1. Introduction

Today, most cities have to deal with the large number of trucks and vans delivering goods in the urban area, balancing the economic viability of businesses located in the city against environmental sustainability. In this context, there is worldwide interest in setting up a sustainable development strategy to identify and define measures to achieve a continuous long-term improvement in quality of life by

* Corresponding author. Tel.: +39-06-72597059; fax: +39-06-72597053. E-mail address: comi@ing.uniroma2.it

1877-0428 © 2012 Published by Elsevier Ltd. Selection and/or peer-review under responsibility of 7th International Conference on City Logistics doi:10.1016/j.sbspro.2012.03.116

creating sustainable communities able to manage and use resources efficiently and effectively, use the ecological and social innovation potential of the economy and, at the end, ensure prosperity, environmental protection and social cohesion. Thus, several measures have been implemented by cities in order to make urban mobility more sustainable and reduce the environmental impact of freight transport. Consequently, for ex-ante assessment, identification and classification of implementable measures and models useful for the simulation of different scenarios are required (Russo and Comi, 2010a). At this aim, several methods and models have been proposed in order to evaluate ex-ante the results to be pursued by their implementation (Ambrosini et al., 2008; Chow et al., 2010). But, we find few studies treating the overall problem of urban freight transport simulation; existing models mainly simulate some aspects of the restocking process and, in particular, consider inter-urban freight movements (Ogden, 1992; Bowyer et al., 2008). They focus on the movements between firms (producers) and distribution centres on a wide scale. Few of them considers the possibility of combining freight and passenger flows (Oppenheim, 1994), hence of representing the interacting behaviour of commodity consumers and commodity suppliers/shippers/retailers. Thus, there are difficulties forecasting the impacts and simulating the effects of transportation measures on an urban scale. The end-consumer movements are mainly studied within the passenger mobility and few studies have considered them belonging to urban freight distribution as the final part of supply chain (Russo and Comi, 2002 and 2003; Gonzalez-Feliu et al., 2010). In fact, as several surveys have confirmed (Nuzzolo et al., 2008; Comi et al., 2011) the freight arrives to urban area mainly for satisfying the end-consumers' demand. Hence, the urban goods movements related to end-consumers' needs are mainly pull movements (the products arrive on the market when required, i.e. consumers with their demand influence production and determine freight movements). The push movements are, in general, triggered by the first ring of the freight movement chain (the producer seeks to anticipate consumer demand, and the freight arrives on the market before it is required) and refer to wider scale (e.g. regional scale).

In this background, the authors have proposed a general framework for simulating goods movements at urban scale. The proposed models allow us to simulate the end-consumer mobility in order to obtain the freight quantities required in the study area and disaggregated for freight types (Russo and Comi, 2010b). The commodity flows are estimated considering that they are moved in order to support a given end-consumer need. This type of movement is defined attraction. It regards the connection between zones in which the goods are bought by the end-consumer and zones where the goods are consumed (end-consumer trips). On the other hand, in the urban area we have also acquisition movements. This type of movements regards the connection between zones where the retailers take the goods and zones where they sell them. Thus, attraction analyses freight trips made by end-consumers; acquisition analyses restocking trips (e.g. goods movements from distributor to retailer).

The general modelling system has been developed on two levels and concerns a medium-size city, and a disaggregated approach for each decisional level is applied. It could be considered an open architecture; in fact, it is possible, in the considered architecture, to introduce specific models from the current literature allowing us to take into account the impacts due to the city logistics implementation on end-consumer goods demand and restocking process of urban retail businesses: • commodity level,

O attraction macro-model; referring to end-consumer quantities, it has general socio-economic data (residents, number of employees, etc.) as input and gives as output the goods quantity required by them (demand in goods quantity for each od pair);

O acquisition macro-model; it concerns logistics trips from the retailer's standpoint; in the literature several types of models have been developed and different decision-makers are considered for each choice level; this model receives as input the goods quantity required in each traffic zone by the

retailer and, in analysing the restocking process, it gives the quantity that is acquired inside or outside the study area (demand in goods quantity for each macro-area);

• vehicle level:

O service macro-model; it receives as input the demand in quantity for a macro-area and gives as output quantity for each consignment, zone and vehicles needed for restocking;

O path macro-model; it receives as input the demand in vehicles and gives as output the departure/arrival time and path used; it can be both static and dynamic.

In the following, the attraction macro-model will be discussed and each model will be detailed (section 2). The section 3 presents the calibration results on the basis of 200 end-consumer interviews, carried out in a suburb of Rome. These results are the advancement on the calibrations developed by the authors in the course of multi-year research. In particular, Russo and Comi (2010b), starting from statistic-descriptive models recalled in Cascetta (2009) for shopping purpose trips (both for durable and nondurable goods), have developed a prototypical model for converting these trips into quantities. These calibrations have been mainly developed to test the goodness of the proposed modelling framework. Comi and Conte (2011) have studied the end-consumer shopping behavioural in a suburb of Rome and have proposed some statistic-descriptive models to simulate the shopping purpose trips considering more than two freight types. Recalling the dataset used by Comi and Conte (2011), the section 3 proposes a joint disaggregate behavioural models that allow us to highlight that the choice processes is a hierarchical choice process and each choice dimension can be influenced by each other (e.g. the choice of purchase dimension could be influenced by destination zone but, on the other hand, the choice of destination zone can be influenced by the purchase dimension to be done). Finally, in section 4 some conclusions are given.

2. The macro-model specification

At urban scale, two different types of goods movements can be identified: end-consumer and logistic movements. End-consumer movements refer to movements made by end-consumers (customers) travelling from their residence/consumption zone to others where they make their purchases. The logistics movements are those connected with restocking process and allow shops and warehouses to be restocked. The attraction macro-model focuses on end-consumer movements.

The attraction macro-model refers to end-consumer quantities, and it has general socio-economic data (e.g. residents, number of employees) as input and gives as output the goods quantity required by them (goods quantity flows by end-consumer). This macro-model allows us to analyse the freight quantity required by end-consumers. We assume that the attracted freight flow is generated by the consumption of goods, as part of the conduct of a given, generic urban activity undertaken by consumers.

The attraction macro-model consists of a set of elementary models that allow us to calculate, as final output, the goods quantities (disaggregated by freight type) that are consumed in a zone and purchased (thus required) in another. In this approach, defined as trip-based, the first two models allow the O/D (Origin/Destination) matrices in trips to be calculated, whether round trip or trip chain.

Referring to goods quantity flows by end-consumers, given that goods flows are estimated to support a given end-consumers' need, the total quantity of freight type s attracted from zone d, Qs,tot,d, can be calculated as:

QsMd= Qsd + QEsd = X QsM + QEs,d = EE ™PJdim)-dim+ QEsd =

o dim o

= YXTRIPo ' P [d/os\ p [dim/dos]-dim + QEsi = ^^n(o)x • p [x/os ]• p [d/os ]• p [dim/dos]• dim + QEsd

dim o dim o x

where,

Qs,d: is the goods quantity bought/sold in d given by the demand of end-consumers

living/working in a zone o within the study area;

QEs,d: is the goods quantity bought/sold in d given by the demand of end-consumers

living/working in a zone z external to the study area;

Qs,od : is the goods quantity bought in zone d by end-consumers living/working in zone o (sold

by retailers of zone d);

TRIPso: is the number of trips for purchase of freight type s with origin in the inner zone o;

TRIPsM(dim): is the number of trips for purchases of freight of type s, from o to d, concluding with a purchase of dimension dim;

Dim: is the dimension of purchase;

n(o): is the number of end-consumers (e.g. families) of zone o;

p[x/os]: is the probability for end-consumer E conditional upon having o as zone of residence

and purchasing freight of type s, of undertaking x trips in a set time with x equal to 0, 1, ..., n; it is estimated by a generation model;

p[d/os]: is the probability of trips being undertaken by end-consumer E going to destination d

conditional upon leaving from o for purchases of type s; it is estimated by a distribution model;

p[dim/dos]: is the probability to conclude a trip with a purchase of dimension dim (0, dim1, dim2, ...., dimn) conditional upon undertaking a trip from zone o to zone d for a purchase of goods type s; it is estimated by a dimension choice model.

The generation model gives the probability p[x/os], for end-consumer E conditional upon having o as

zone of residence and purchasing freight of type s, of undertaking x trips in a set time with x equal to 0, 1,

..., n. It can be determined as:

p [x/os] =prob

Ux > Ux-

yx' ^ x,x'e.\0,1,...,n}

Where, Ux is the perceived utility of undertaking x trips by end-consumer E.

The distribution model gives the probability p[d/os] of trips being undertaken by end-consumer E going to destination d conditional upon leaving from o for purchases of type s. If DoE is the set of alternatives for the decision-maker E that begins her/his trip from zone o, this probability can be expressed as:

p[d/os] =prob[ud>ud,] yd' * d, d' e D^ (2)

where Ud is the utility perceived by end-consumer E of going from o to d. It should be noted that typically the destination chosen for carrying out an activity is not a traffic zone but one (or more) elementary destination (such as a shop, an office, etc.) within it. The traffic zone d is therefore a compound

alternative consisting of the aggregation of Md elementary alternatives. However, it is possible to hypothesize a relationship between the size of an alternative and a set of observable size variables. For example, the size variables could include an attraction measure such as shops.

These two elementary models allow us to obtain the O/D matrices in terms of trips made by users to buy some goods (products). In this approach, the generation and distribution models are those that traditionally allow trips to be calculated (Hunt and Stefan, 2007). In the literature there are several models available, such as descriptive or behavioural both for round trip and trip chain. While all are usable on their own, the difficulties in employing them in a more general architecture could lie in the complexity of the downstream model to convert trips into quantities and to simulate the choice of shop type (retail shop, store, cafes, bars, restaurants, private services, etc.).

To convert the trips into quantity, a dimension choice model is proposed. This model allows us to give a quantity dimension (dim) to each shopping trip. It gives the probability p[dim/dos] that a trip concludes with a purchase of dimension dim (0, dim1, dim2, ...., dimn) conditional upon undertaking a trip from zone o to zone d for a purchase of goods type s. The probability p[dim/dos] can be expressed as:

p[dim/dos] =prob Udim > Udim- Vdim' ^ dim, dim' e\0,...,dimn} (3)

where Udim is the perceived utility that a trip concludes with purchases of dimension dim.

The next section presents the calibration of previous described models. Both descriptive and behavioural models have been calibrated in relation to different aggregation level of freight (freight types, e.g. durable goods and not durable goods, and class of products, e.g. foodstuffs, home accessories, clothing) and category of user (e.g. workers). Respect to the models present in the literature, we propose joint disaggregate models in order to highlight that the choice processes is a hierarchical choice process and each choice dimension can be influenced by each other.

3. The macro-model calibration

3.1. The dataset

The models have been developed using the results of some surveys carried out in a suburb of the city of Rome where more than 200 households have been interviewed. The attention is on shopping journey, considering both home based trips and non-home based trips (e.g. home-work-shopping-work-home). The survey has allowed us to investigate the purchasers' behaviours. In particular, the interviews have been structured in two sections: the former related to infer the personal characteristics of interviewed (e.g. job, age, family composition, income), the latter related to collect data on journey and purchases (e.g. freight types, frequency of purchase trip, origin and destination of trip, transportation mode, dimension of purchases).

The 62% are female and the 38% are male. Referring to income, about the 65% have declared to have an income less than 40,000 €/year. The 52% are employed (Comi and Conte, 2011).

This survey also allows us to characterize trips in terms of frequency and purchased freight types. In particular, the analysis of characteristics of purchases and transportation behaviours has pursued us to use two different freight classifications. The former that provides two classes: durable and non-durable goods. It gave the best results, in Italy, in simulating end-consumer trips at an urban scale (Russo e Comi, 2010b). The latter consists of six classes: foodstuffs, home accessories, stationery, clothing, household and personal hygiene and other. It has been used by Filippi et al. (2010) for the assessment of city

logistics scenarios in the inner area of Rome. From surveys, it is emerged that each family weekly buys and consumes about 52.3 kg and undertakes about 2.4 trips for shopping. These results are similar to those revealed in other Italian cities and towns.

3.2. The generation model

The generation or trip frequency model estimates the mean number of relevant trips TRlPs,o undertaken for buying the types s by the generic user with origin in zone o.

The emission models can be classified in two main categories: behavioural and descriptive models. The mean number of trips undertaken by the end-consumer, departing from o, for the purchase of freight type s, trip[os] can be calculated as:

where n(o) is the number of end-consumers (e.g. families) of zone o.

The Binomial and Multinomial Logit models are the random utility models most frequently used to simulate p[x/os]. From our dataset, it is emerged that the probability of undertaking more than two relevant trips in a week is negligible (less than 10%); a Multinomial Logit with three alternatives - not undertaking or undertaking one or more than one trip in a week - has been calibrated and validated. The systematic utility has been specified as linear function of the following attributes:

• age, it is a dummy variable equal to 1 if end-consumer's age is less than 40, 0 otherwise;

• fam, it is a dummy variable equal to 1 if the end-consumer's family has less than 4 components, 0 otherwise;

• inc, it is a dummy variable equal to 1 if end-consumer's income is less than 40,000 €/year, 0 otherwise;

• gend, it is a dummy variable equal to 1 if end-consumer's gender is female, 0 otherwise;

• wk, it is a dummy variable equal to 1 if end-consumer is employed, 0 otherwise;

• lAAo, it is the active accessibility index of end-consumer's zone o;

• ASA, it is the Alternative Specific Attribute.

The active accessibility index lAAo has been calculated as:

trip [os]=^ x ■ p [x/os] .

The flow of trips from zone o can then be expressed as follows:

TRlPso = n(o)x ■ p [x/os ]

lAAo = ln AAo - min (AAz) j max (AAz) - min (AAz)J

where AAx is the active accessibility of zone x estimated as:

EPj: the number of employees at retail establishments related to the considered freight type

in zone j,

distj the distance between zone x and j, calculated on the road network according the path of

minimum generalised travel cost, a1 and a2: the parameters.

The Tables 1 and 2 report the results of calibration for non-durable and durable goods. The models have been calibrated using the Maximum Likelihood (LM) estimator within the classic theory of statistical inference. The presented models are the result of several specifications and calibrations based on different combinations of possible attributes. The models which performed the best statistical significances are given. We can see that all parameters have the expected signs and most of them are statistically significant. Parameter analysis shows the important role of active accessibility for nondurable goods; in fact, we can see that increasing the accessibility, the number of trips increase. It confirms that end-consumers living in high accessible zones prefer to undertake more than one trip per week for purchasing non-durable goods (i.e. daily consumption products). Referring to age, results confirm that young people undertake more trips. Referring to family components, it is emerged that small families prefer to make few trips. The sign of income parameters confirm that low income people make fewer trips for non-durable goods (e.g. foodstuffs), but many trips for durable goods (i.e. they use to visit many shops and undertake many trips before to purchase). The negative sign of worker parameters confirm that the probability to have many trips for week decreases for employees; in fact, in general, they have little time for shopping. Finally, as expected, the signs of parameters relative to gender confirm that women undertake more trips than men.

Models of this type should be considered as a tool for quantitative analysis of the determinants of urban mobility rather than an operational tool. Its application for the simulation of travel demand for purchases in an entire city area could require a considerable amount of information. However, this is not necessarily true for all behavioural models and operative trip frequency models are often used for the simulation of large scale systems as proposed by Comi and Conte (2011). These models allow us to estimate the average number of shopping trips by the mean values of socio-economic variables such as income, gender, etc., and level-of-service attributes such as accessibility.

Table 1. Behavioural generation model for non-durable purchase weekly trips

Attribute Parameter Alternatives

No trip One trip per week More than one trip per week

Age (age) ßage 0.054 (1.1)

Family components (fam) ßfarn -1.630 (-3.5)

End-consumer's income (inc) ßinc -0.477 (-1.2)

End-consumer's gender (gend) ßgend 1.092 (2.6)

End-consumer's job (wk) ßwk -1.142 (-3.0)

Active Accessibility (IAAo) ßlAA -0.317 (-2.1)

Alternative Specific Attribute (ASAx) ßx -6.111 (-3.0) -1.629 (-2.7)

Accessibility ai = 0.373 (1.7) «2 = -3.355 (-2.2)

(f 0.38

Table 2. Behavioural generation model for durable purchase weekly trips

Attribute Parameter Alternatives

No trip One trip per week More than one trip per week

Age (age) ßage 0.425 (1.9)

Family components (fam) ßfam 0.975 (3.6)

End-consumer's income (inc) ßinc 0.128 (-1.3)

End-consumer's gender (gend) ßgend 1.364 (5.2)

End-consumer's job (wk) ßwk -1.261 (-2.8)

Alternative Specific Attribute (ASAx) ßx 0.712 (1.4) 0.732 (1.4)

P2 0.27

3.3. The distribution model

Distribution model expresses the percentage (probability) p[d/os] of trips undertaken by end-consumers going to destination d, given departure zone o for purchases of freight type s. The calibrated distribution models have a Multinomial Logit structure:

p[d / os] = exp(Vd y £ exp(Vd.)

The systematic utility Vd has been expressed linear combination of the attributes of possible destinations in relation to the zone of origin o:

Vd = £ PjXjd = Pm •Ep: + Pdist • distod + Pod • OD

where,

EPd : is the number of employees at retailer employment related to freight type s in zone d; it

is expressed in thousands;

distod: the distance between zone o and d, calculated on the road network according the path of

minimum generalised travel cost; it is expressed in kilometers; OD: is a dummy variable equal to 1 for intrazonal trip, 0 otherwise.

The Table 3 reports the calibrated models for all identified freight types. The models have been calibrated using the Maximum Likelihood (LM) estimator within the classic theory of statistical inference. The presented models are the result of several specifications and calibrations based on different combinations of possible attributes. According to the behavioural interpretation, the distribution model simulates the choice of a destination among possible alternatives. It should be noted that typically the destination chosen for carrying out an activity (i.e. to buy something) is not a traffic zone but one (or more) elementary destination (such as a shop) within it. The traffic zone d is therefore a compound alternative composed of the aggregation of elementary alternatives. Further analysis has been developing in order to calibrate others models considering the size function as proposed by Ben-Akiva and Lerman (1985) and Cascetta (2009). Further developments also regard the choice set modelling. In fact, the distribution model has to consider a huge number of elemental alternative destinations, which is not

always the tested realty. The end-consumer could choose the zone where to buy within a pre-defined and well-known choice set according to some specific attributes (e.g. brands, price).

Table 3. Distribution models for purchase weekly trips

Attribute Freight types

home accessories household and

non durable durable foodstuffs stationery clothing personal hygiene other

Employees ßetrm 0.680 0.212 0.521 0.303 0.269 0.171 0.229 0.157

(EPd;) (4.7) (3.6) (3.5) (3.1) (3.0) (1.8) (1.7) (1.8)

Distance (distad) ßdist -0.121 (-1.4) -0.029 (-1.7) -0.090 (-1.3) -0.024 (-1.4) -0.011 (-1.3) -0.133 (-1.3) -0.529 (-1.4)

Intra-zonal ßod 4.753 3.726 4.739 4.323 4.564 3.826 5.353 4.048

trip (OD) (2.6) (2.9) (2.7) (2.1) (2.6) (2.0) (2.6) (2.5)

0.45 0.18 0.44 0.32 0.37 0.20 0.56 0.24

3.4. The dimension choice model

It should be noted that when the end-consumer arrives in a zone, she/he could or not purchase something. Thus, an intermediate model has been included in the general framework in order to estimate the probability to buy or not (Fig. 1). The probability to purchase has been estimated by a Binomial Logit model:

p[Purchase / dO] = exp(vpurchase exp(vpurchase ) + eXp(VmpurchaSe )]

where the systematic utility to purchase, Vpurchase, has been expressed as linear function of the following attributes related to socio-economic characteristics of end-consumer and passive accessibility index of zone d.

The passive accessibility index, IAP' k , has been calculated as:

APd - min (APz) I max (APz)- min (APz) where APx is the passive accessibility of zone x estimated as:

APP = Y(PoPi T • exp[«4 • distix]

Popi: the number of residents of zone i,

distix: the distance between zone i and x, calculated on the road network according the path of

minimum generalised travel cost; a3 and a4: parameters.

The Table 4 reports the calibrated parameters. The obtained results confirm that end-consumers having chosen to travel through low accessibility zones prefer to buy, and high income end-consumers prefer to visit many shops before purchasing. The same happens for old and non-employed people. The member of large family tends to do not buy. It happens because that other member could do it.

Fig. 1. Modelling of purchase process

Table 4. Purchase model for all freight types

Passive Accessibility (IAPd) Age (age) Family components (fam) End- consumer's job (wk) End-consumer's income (inc) End-consumer's gender (gend) ASA no purchase

Alternative purchase purchase purchase purchase purchase purchase no purch.

Value -6.2 (-2.1) 0.69 (1.5) 0.85 (1.2) -0.48 (-1.2) 0.30 (1.6) -0.24 (-1.5) -2.48 (-2.9)

Accessibility a3 = 1.22(1.9) a4 = -6.22 (2.9)

P2 0.27

Then, a dimension should be assigned to each purchase. The dimension choice model allows us to give a quantity dimension (dim) to each shopping trip. It gives the probability p[dim/dos] that a trip concludes with a purchase of dimension dim (dimh dim2, ...., dimn) conditional upon undertaking a trip from zone o to zone d for a purchase of goods type s. The calibrated dimension choice model has a Multinomial Logit structure:

p[ dim / dos] = exp(VSsm)l £ exp(Vdim,)

/ dim'

The systematic utility Vdim has been expressed linear combination of the attributes of end-consumer (EC,) and purchase (e.g. freight type, PCj) and journey (e.g. passive accessibility of purchase zone, JOk):

Vdim = £ ßi • ECi +X ßj ■PCj ++X ßk JOk

Three dimensional alternatives have been considered. Different dimension classes have been defined for the two main freight types as reported in Table 5. The Tables 5 also reports the results of dimension choice model calibration. In particular, it should be noted that, as revealed by surveys, the passive accessibility of zone plays a key-role in the choice of purchase dimension. In fact, the survey has evidenced that trip to non-easy accessible zone pushes to make larger purchases. Other analysis is in progress in order to identify the exiting relationships between the choice of low-accessible zone and the prices of products.

Table 5. Dimension choice model for non-durable and durable purchases

Attribute Parameter Non-durable Durable

alternatives alternatives

dim[ dim2 dim3 dim1 dim2 dim3

< 4 kg (4, 7 kg] > 7 kg < 0.3 kg (0.3, 2 kg] > 2 kg

Age (age) ßage -1.152 (-1.9) -2.667 (-2.2)

Family components (fam) ßfarn 0.798 (1.7) -0.280 (-1.9)

End-consumer's income (inc) ßinc 0.165 (2.3) 0.286 (1.9)

End-consumer's gender (gend) ßgend 2.138 (2.7) -1.854 (-1.9)

End-consumer's job (wk) ßwk -0.469 (-[.[) 0.009 (1.5)

Passive Accessibility (lAPd) ßlAP -2.068 (-2.1) -1.016 (-1.7)

Alternative Specific Attribute (ASAX) ßx 1.618 (2.6) -1.861 (-1.3) 2.713 (2.2) 2.002 (1.3)

P2 0.39 0.52

4. Conclusions

This paper presents a modeling system for estimating the goods attracted by urban area. It consists of three main models: generation, distribution and dimension choice models. The generation and distribution models allow us to estimate the O/D matrices for purchases. Hence, the dimension choice model allows us to give a dimension to each trip undertaken by end-consumers.

The freight modeling system has been specified and calibrated on the basis of real test cases (suburb of Rome), considering different freight types. Models have been specified in order to create integration with passenger mobility. In fact, this interaction is especially relevant in the urban context where congestion is an effect shared and generated by both markets and presumably the decision-makers take this effect into consideration before making a transport decision.

This modeling system can be successfully used for the estimation of urban freight flows by road in the initial assessment of future scenarios, as well as to calculate the impacts due to new city logistics measure implementation. It can be considered a tool developed to support ex-ante assessment and to capture the effects due to urban freight transport measures on end-consumer behavior. In fact, the choice, if or not to undertake a trip and where to go for purchasing, can be influenced by socio-economic attributes of user

(e.g. income, number of family members, etc.) or transportation attributes (e.g. generalized travel cost or accessibility with respect to the possible destination for the trip purpose). City logistics measures impact on this latter class of attributes.

The dimension choice model aims at converting trips into quantities. The quantity of goods purchased by end-consumer depends on freight types and is mainly influenced by socio-economic attributes of customers (e.g. income, age, gender) or characteristics of sold freight (e.g. trademark) as well as of accessibility of purchasing zone.

Hence, the attraction macro-model allows us to investigate how urban policies modifying the transportation attributes for passenger or the sale network can modify the end-consumer demand and thus, the attracted goods quantity.

Further analyses are required in order to improve these results and to investigate how the trade characteristics of retailer location can influence the end-consumer's choices.

Acknowledgements

Authors would like to thank Emanuele Conte for his support in carrying out the surveys and the analysis of the dataset.

References

[1] Ambrosini C, Meimbresse B, Routhier J, Sonntag H. Urban freight policy-oriented modelling in Europe. In: Taniguchi E,

Thompson RG, editors. Innovations in city logistics, Nova Science Publishers, Hauppauge NY, U.S.A.; 2008, p. 197-211.

[2] Ben-Akiva M, Lerman SR. Discrete choice analysis: Theory and application to travel demand. The MIT Press, Cambridge,

Massachusetts, U.S.A.; 1985.

[3] Bowyer D, Thompson RG, Spiridonos F. Melbourne freight movement model. In: Taniguchi E, Thompson RG, editors.

Innovations in city logistics, Nova Science Publishers, Hauppauge NY, U.S.A.; 2008, p. 213-229.

[4] Cascetta E. Transportation systems engineering: Model and application. Springer; 2009.

[5] Chow JYJ, Yang CH, Regan A. State-of-the art of freight forecast modeling: lessons learned and the road ahead. In:

Transportation 2010; 37(6): 1011-1030.

[6] Comi A, Conte E. A modelling system for estimating freight quantities attracted by cities. In: Pratelli A, Brebbia CA, editors.

Urban transport XVII - Urban transport and the environment in the 21st Century, WITpress, Southampton, U. K.; 2011, p. 423-434.

[7] Comi A, Delle Site P, Filippi F, Nuzzolo A. Ex-post assessment of city logistics measures: The case of Rome. In: Mussone L,

Crisalli U, editors. Transport management and land-use effects in presence of unusual demand, Franco Angeli, Milan, Italy; 2011; p. 235-252.

[8] Filippi F, Nuzzolo A, Comi A, Delle Site P. Ex-ante assessment of urban freight transport policies. Procedia - Social and

Behavioral Sciences 2010; 2(3): 6332-6342.

[9] Gonzalez-Feliu J, Toilier F, Routhier JL. End-consumer goods movement generation in French medium urban areas. Procedia -

Social and Behavioral Sciences 2010; 2(3): 6189-6204.

[10] Hunt JD, Stefan KJ. Tour-based microsimulation of urban commercial movements. Transportation Research Part B 2007; 41(9): 981-1013.

[11] Nuzzolo A, Crisalli U, Comi A. Metropolitan freight distribution by railways. In: Taniguchi E, Thompson RG, editors. Innovations in city logistics, Nova Science Publishers, Hauppauge NY, U.S.A.; 2008, p. 351-367.

[12] Ogden KW. Urban goods movement. Ashgate, Hants, England;1992.

[13] Oppenheim N. Urban travel demand modeling. John Wiley & Son, New York; 1994.

[14] Russo F, Comi A. Urban freight movement: A quantity attraction model. In: Sucharov LJ, Brebbia CA, Benittez F, editors. Urban transport VIII: Urban transport and the environment in the 21st Century, WITpress, Southampton, U. K.; 2002, p. 831840.

[15] Russo F, Comi A. Urban freight movements: quantity attraction and distribution models. In: Beriatos E, Brebbia CA, Coccossis H, Hungolos A, editors. Sustainable planning & development, WITpress, Southampton, U. K.; 2003, p. 711-720.

[16] Russo F, Comi A. A classification of city logistics measures and connected impacts. Procedia - Social and Behavioral Sciences 2010; 2(3): 6355-6365.

[17] Russo F, Comi A. A modelling system to simulate goods movements at an urban scale. In: Transportation 2010; 37(6): 9871009.