Modeling the acceptability of crowdsourced goods deliveries: Role of context and experience effects

Transportation Research Part E 105 (2017) 18–38

Contents lists available at ScienceDirect

Transportation Research Part E journal homepage: www.elsevier.com/locate/tre

Modeling the acceptability of crowdsourced goods deliveries: Role of context and experience effects Aymeric Punel a, Amanda Stathopoulos b,⇑ a b

Civil and Environmental Engineering, Northwestern University, 2145 Sheridan Road, Evanston, IL 60208-3109, United States Civil and Environmental Engineering, Northwestern University, 2145 Sheridan Road, Tech #A312, Evanston, IL 60208-3109, United States

a r t i c l e

i n f o

Article history: Received 6 January 2017 Received in revised form 9 June 2017 Accepted 12 June 2017

Keywords: Crowdshipping Acceptance Peer-to-peer delivery Discrete choice Urban freight Stated choice experiment

a b s t r a c t Crowdshipping is a frontier in logistics systems designed to allow citizens to connect via online platforms and organize goods delivery along planned travel routes. The goal of this paper is to highlight the factors that influence the acceptability and preferences for crowdshipping. Through a survey using stated choice scenarios discrete choice models controlling for context and experience effects are specified. The results suggest that distinct preference patterns exist for distance classes of the shipment. In the local delivery setting, senders value transparency of driver performance monitoring along with speed, while longer shipments prioritize delivery conditions and driver training and experience. The model developed in this paper provides first key insights into the factors affecting preferences for goods delivery with occasional drivers. Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction The shipping industry is increasingly penetrated by digitization and technological change that gives shape to new delivery concepts. Anticipating new technologies is one of the critical components to improve performance and decrease negative impacts of goods deliveries (Macharis and Kin, 2017). The next wave of goods delivery innovation is likely to be impacted by emerging paradigms such as, delivery by ground drones, e-scooters, vehicle automation, pick-up boxes and crowd logistics (Savelsbergh and Van Woensel, 2016). For each of these innovations, a range of operational, business, legal and behavioral problems, need to be tackled. This paper provides an investigation of crowdsourced goods delivery, which has been defined as a ‘frontier’ (Wang et al., 2016) or ‘revolutionary change in city distribution’ (Macharis and Kin, 2017). Crowd-inspired logistics with delivery by occasional drivers has received growing attention in both industry and research communities. The study of crowdshipping is challenging due to the novelty, lack of operational uniformity and the lack of consolidated real-world systems that disseminate operational data. While there is no single formal definition of crowdshipping, in this paper we work with the following definition: a goods delivery service that is outsourced to occasional carriers drawn from the public of private travellers and is coordinated by a technical platform to achieve benefits for the involved stakeholders. The emergence of crowdsourced solutions in the shipping industry has the potential to radically alter the way shipments are organized, who will be the carriers of the future, and how the public expects package deliveries to be performed (Rougès and Montreuil, 2014). Service firms, which frequently originate as startups from outside the traditional logistics industry, manage online crowdshipping platforms where senders and drivers connect and negotiate the service. The platforms generally focus on shipments to individual households (B2C) or between consumers (B2B). The crowdship⇑ Corresponding author. E-mail addresses: [email protected] (A. Punel), [email protected] (A. Stathopoulos). http://dx.doi.org/10.1016/j.tre.2017.06.007 1366-5545/Ó 2017 Elsevier Ltd. All rights reserved.

A. Punel, A. Stathopoulos / Transportation Research Part E 105 (2017) 18–38

19

ping firm is responsible for recruitment, matching and pricing algorithms. Ensuring security of the shipment and the pickup/delivery is a crucial component. Firms variably deal with this by vetting and verification of participants, particularly drivers, making payment conditional on delivery, providing guarantees on delivery performance with basic and optional insurance and helping to organize safe public pick-ups/drop-offs. Technology and digitization allows detailed tracking, alerts and monitoring of performance that is leveraged in various ways by the platforms. Among the added value is the ability to track shipments in real time or to have detailed performance disclosure of other users. Payments are typically handled like ride-hailing firms, via credit card with the actual payment charged either at confirmed driver pickup or upon delivery. The main assets and disadvantages related to crowdshipping will take different forms for the various stakeholders. For senders, advantages relate to added value features such as tracking, transparency or flexibility in pick-up and delivery conditions, along with lower shipping costs. The potential carriers, on the other hand, are able to add an income to his/her commute in exchange for picking up a shipment (Miller et al., 2017). For service companies, the advantage of crowdshipping is the lower operating costs compared to traditional logistics operators. This is due to increased flexibility of assets, that is, storing facilities, supply stocks, vehicles and drivers (Rougès and Montreuil, 2014). From a broader societal perspective, the innovation has the potential to reduce traffic and energy-footprints of deliveries (McKinnon et al., 2015). At the same time, many challenges have been identified. They relate to trust and liability issues, maintaining a critical mass of couriers and customers in tandem, and fostering acceptability of new delivery models (Rougès and Montreuil, 2014). The overall re-organization of deliveries from traditional logistics to crowdshipping could even lead to negative impacts on the energy footprint of deliveries. The potential risk for rebound effects, such as increased travel and fuel consumption that might ultimately offset the benefits are discussed in Paloheimo et al. (2016). Crowdshipping is a growing contender in the shipping industry often promoted by non-traditional shipping companies, such as technology firms and retailers. Despite virtually no companies existing before 2012, at present there is a wave of interest in crowdshipping companies (Vuylsteke, 2016). For instance, companies such as US Deliv and Zipments (later acquired by Deliv) raised $7.85 million and $2.25 million respectively in 2013 (Office of Inspector General, 2014). Chinese crowdshipping pioneer Renren Kuaidi obtained $15 million in 2014 (German Industry and Commerce (HK), 2016). Crowdshipping concepts are launched by established firms, e.g. retailers like Walmart or logistics firm like DHL Myway trial in Stockholm) and as independent companies. While the majority of crowdshipping start-ups have emerged in the US (e.g. Postmates, Deliv, Roadie, Kaargo, UberRush), crowdshipping platforms are spread globally; with examples in Australia (e.g. PostRope, PPost), Colombia (Rappi), Nigeria (Max), China (Renren Kuaidi), and in Europe (e.g. Nimber in UK and Norway, Trunkrs in the Netherlands, PiggyBaggy in Finland) or across countries (Parcelio, Quincus). Despite the strong market interest, only a fraction of new crowdshipping companies succeed in establishing a lasting market by attracting and maintaining users within the system (Dablanc, 2016). The above highlights the critical role that behavior and acceptance plays in the viability, efficiency and maturing of crowdshipping delivery concepts. The goal of this paper is to study the determinants of crowdshipping acceptance among senders. The objectives are to (1) identify the representation and preference for the unique attributes related to crowdshipping (e.g. tracking of driver performance, increased control over delivery conditions), (2) use choice models to identify the factors that drive consumer preferences and acceptance of crowdshipping options, with focus on the role of; (i) shipping attributes, (ii) socio-demographic segments, (iii) user shipping experiences, (iv) different contexts, ranging from local to long-distance shipments. Finally, (3) the models will supply the first policy measures to assess sender behavior and early viability of the crowdshipping market in the US. Discrete choice models are estimated based on newly collected stated choice data to investigate the factors that influence the selection of crowdshipping alternatives. Competing crowdshipping delivery options are framed in terms of the utility each provides to senders and three different contexts are tested namely a local, medium and long-distance shipment. Various model structures accounting for absolute (scale) and relative preference for crowdshipping attributes, as well as correlated alternatives and random heterogeneity are investigated. The contribution of this paper is to provide a first investigation of crowdshipping acceptability. The innovativeness of the crowdshipping movement calls for an emergent research agenda on public reactions and acceptability. Understanding the acceptance of crowdshipping, both in the general public (senders/receivers) and among potential carriers is crucial for a number of reasons. It uncovers consumer preferences for an innovative shipping service, it allows us to forecast demand in the context of emerging logistics initiatives, company practices and policy variables related to crowdsourcing. Furthermore, the behavioral insights will contribute to improving operational models exploring crowd-sourcing and will help advance the underlying business models of crowdshipping logistics. This will improve recruitment, sender-driver-receiver experiences and help control external impacts and unintended effects. In a broader sense, it will contribute to building the ‘‘critical mass” necessary to establish a sustainable human-centered delivery system that ensures societal benefits (Rougès and Montreuil, 2014). The paper is organized as follows. Section 2 presents a literature summary to highlight the features that are likely to impact the use of crowdshipping. User characteristics, shipping attributes, experiences by the sender and the context of the delivery are analyzed. Section 3 describes the design of the survey, while the fourth section presents sample descriptive statistics. Section 5 presents the model structure to study senders’ preferences and acceptance toward using a crowdsourced delivery service. Section 6 summarizes and outlines future research directions.

20

A. Punel, A. Stathopoulos / Transportation Research Part E 105 (2017) 18–38

2. Literature on crowdsourced goods delivery: system functioning and behavioral influences The literature analysis examines the literature on shipping services relying on crowd resources. Notably, there is a lack of behavioral research in this area, needed to supplement and contextualize the operational concepts. Therefore the literature review includes strands from broader on-demand and sharing mobility systems with a similar organization to crowdshipping to define the expected impact of shipping attributes, personal factors, experiences and context. 2.1. Shipping innovation and behavior A number of models and analysis frameworks have been proposed to evaluate behavior surrounding goods delivery innovations. These include; the potential for collaborative delivery (Chowdhury, 2016); acceptance of off-hour and delivery timing (de Jong et al., 2016; Holguín-Veras et al., 2007; Marcucci and Gatta, 2017), receiver choice between off-hour or distribution center delivery (dell’Olio et al., 2016); consumer choice between, and mode of travel for, e-shopping vs. physical commerce (Barone et al., 2013; Hsiao, 2009) and last-mile customer parcel pickup behavior (Collins, 2006). However, there are virtually no academic studies of behavior related to collaborative or shared resource delivery. Given this we turn to an investigation of logistics system operations using the crowd, to later address the behavioral insights from a broader mobility literature. 2.2. Shipping systems based on crowd-resources The literature on crowdshipping is concentrated in the area of operations. Various researchers have examined the impact of relying on different types of crowd-resources to complete deliveries, generally with a local delivery focus. Chen and Pan (2016) build a methodological approach applied to a crowdshipping system composed of a taxi fleet. The two-step decision tool seeks the optimal path based on both offline taxi trajectory mining and online parcel routing. While the model is not tested on real data, it defines a framework to study the challenges of matching shipments with the taxi-routes without decreasing service. Kafle et al. (2017) build a framework where crowdsourced cyclists and pedestrians are employed for first/last legs of trucking trips. By optimizing the relay points and matching crowd resources with truck routes and schedules the total truck mileages and costs can be reduced. Wang et al. (2016) model a large-scale assignment optimization system of citizen-workers performing the last-mile delivery from pick-up stations. Realistic motion patterns for people and parcels are extracted from empirical data. Results suggest that companies can reduce costs and delivery failure risks by relying on citizens and pick-up stations. Arslan et al. (2016) study a peer-to-peer platform that matches delivery tasks with ad-hoc private drivers as a dynamic pick-up and delivery problem. Archetti et al. (2016) study vehicle routing strategies with mixed drivers (professional and occasional) seeking to fulfill deliveries while minimizing cost. Advantages of employing occasional drivers are found to relate to cost savings. Paloheimo et al. (2016) examine the application of PiggyBaggy crowdsourced delivery model for the pickup and delivery of library books. A central concern when shifting to crowd-delivery is the risk of increased mileages to satisfy deliveries. The case-study suggests that crowdshipping has the potential to reduce environmental impacts, in part due to small detours and use of bikes for the deliveries. Erickson and Trauth (2013) evaluate the potential of a crowdsourced delivery service from a management perspective. They find that the main challenges that organizations will face relate to ensuring quality and timeliness of work tasks. A general finding in the operational crowdshipping literature is that crowdshipping promises benefits in terms of reducing cost, failed delivery risk, or overall mileages. However, these results hinge critically on assumptions about demand and acceptance factors. The number of willing crowdparticipants on both the sender and receiver side, along with their movement patterns and mode choices underpin the design and impact of these systems. The willingness of occasional drivers or pedestrian/bikers to detour from existing plans are also fundamental acceptance problems that similarly need to be included in business model and operational designs. Turning the focus instead to behavior and preference analysis, we note a lack of research that directly addresses the acceptance of crowdshipping. Hence, the following literature analysis will take a broader perspective by examining related literature strands. Drawing on papers from the logistics and parcel delivery setting, along with sharing economy mobility systems (ridesourcing and carpooling), the role of service attributes, personal features, and the delivery context is outlined. 2.3. Service attributes Public acceptance towards a new service will vary as a function of its specific attribute performance. Crowdshipping acceptance can be analyzed as a service at the intersection of package delivery and personal transportation ridesourcing. Indeed, for both ridesourcing and crowdshipping, the user requests, through a mobile/Internet application, a door-to-door transportation service performed by a (non-professional) driver. In the ridesourcing literature, users are found to place the highest value on travel time, flexibility, convenience, reliability, and perception of security (Agatz et al., 2012; Furuhata et al., 2013) when requesting a ride. The cost and time savings along with the convenience appear to be the main assets of these systems (Shaheen, 2016). Comparing traditional shipping and crowdshipping, the main difference relates to the role of the carriers, who in the latter case are typically occasional and/or non-professional drivers. While arrangements vary across service operators, this sta-

A. Punel, A. Stathopoulos / Transportation Research Part E 105 (2017) 18–38

21

tus can lead to issues of liability, trust and reputation. To compensate for customer concerns, crowdshipping companies often have two-way rating systems for each transaction similar to other platforms, such as ridesourcing (Uber or Lyft) or short-term lodging (Airbnb, Turnkey). Esper et al. (2003) find that transparency and being able to select among carriers lead to an increased willingness to purchase in an e-retail setting. This can be assumed to extend to delivery services preferences. Ratings can ensure transparency of service performance and maintaining service quality even with non-trained shippers (Hall and Krueger, 2015) and appear to be central for the end users (Cabral and Hortacsu, 2010; Panda et al., 2015). 2.4. Socio-demographic features Identifying the relevant personal characteristics is important to begin predicting both the willingness to try as well as the transition to persistent crowdshipping use. Firstly, as an app-based service relying on connected mobile platforms to communicate and manage work tasks, crowdshipping is more likely to attract younger segments. Indeed, users of app-based ondemand ride services, such as Uber, tend to be younger than the general population (Rayle et al., 2016; Shaheen et al., 2016). This is also true more broadly for collaborative economic systems (Panda et al., 2015), heavily reliant on the use of the Internet (Belk, 2014). While the millennial generation is more likely to adopt sharing systems, it also appears that people with higher levels of education are more likely users. This is evidenced in the ridesourcing and car sharing cases (Rayle et al., 2016). Instead, income has a less clear connection with sharing system adoption. Efthymiou et al. (2013) find that Greeks in low to medium income classes are more likely to join car-sharing. In the case of ridesourcing Rayle et al. (2016) find that users’ incomes almost match that of the total population with some tendency for the low-income category to be underrepresented. Dias et al. (2016) find that lower income individuals in the US have a lower propensity to use both ridesourcing and car sharing services, likely due to cost considerations. The role of gender in crowdshipping decisions is similarly hard to infer from the literature. Whereas the authors could find no relevant literature on gender effects related to shipping behavior, gender differences in the collaborative economy do exist, albeit with contradictory findings. For sharing and ridesourcing industries some studies find that proportions of men and women in sharing systems match the general population (Shaheen et al., 2016). Other works suggest conflicting findings: either that women are more likely users (Caulfield, 2009; Delhomme and Gheorghiu, 2016) versus women being underrepresented in sharing communities (Anderson, 2014). Miller et al. (2017) find that female crowdshipping drivers are equally likely to accept a delivery but less so for evening deliveries. A possible explanation for the gender effects relates to safety issues (Panda et al., 2015; The Heights Board, 2016). Concerning underlying motivations, crowdshipping is likely motivated by a range of factors. The public participates in sharing initiatives for reasons, ranging from pecuniary to opportunities for making social connections (Bellotti et al., 2015). Crowdshipping has a potential to reduce vehicle movements and environmental impacts (McKinnon et al., 2015). In line with this, users may be motivated by environmental concerns (Hamari et al., 2016; Piscicelli et al., 2015). Alternatively, personal considerations might drive acceptance. Paloheimo et al. (2016) find ‘trying something new’ to be a leading motivation. Other works suggest acceptance can depend on the potential to save money (Bellotti et al., 2015; Hamari et al., 2016) or opportunities to make social connections (Bellotti et al., 2015). 2.5. Experience and Contextual features While personal transportation is an event that many users experience on a daily basis, shipping is likely to be more infrequent, even with the increase in e-retailing. Personal experiences of shipping are likely to impact the preferences for shipping attributes and innovation. Several works have proposed reference-dependent preferences in passenger (Hess et al., 2008) and goods transport settings (Masiero and Hensher, 2011; Masiero and Rose, 2013). These models account for the experience by allowing separate treatment of improvement and deterioration from a reference level. Most studies find evidence for asymmetry in the preferences, where losses give rise to larger disutility than gains generate utility (Stathopoulos and Hess, 2012). Experience may also impact preferences in other ways. Transportation research has looked at the impact of experience versus traffic state information on repeated route decisions. There is evidence of less switching when experience outperforms noisy information (Srinivasan and Mahmassani, 2003). Similarly, before extensive experience is gained, drivers are more sensitive to information than their own experience (Ben-Elia et al., 2008). Hensher and Ho (2016) proposed a mode choice model where preferences for mode-specific attributes were scaled according to a past experience measure. Overall, there is evidence for statistically significant preferences for the existing service (status quo) when other factors remain constant. Salkeld et al. (2000) termed this effect a ‘veil of experience’. Finally, the context of the shipment, regarding the package features and the shipment distance, may impact the crowdshipping decision. Identifying the acceptance related to shipment type and context is relevant to develop services with tailored business models. Rougès and Montreuil (2014) examine 18 existing crowdshipping companies and identify 5 principal types of crowdshipping services, according to the shipment distance and stakeholder goals (clients and carriers). Crowdshipping services range from the local scale to long-distance shipments. Local last-mile delivery, which is similar to ride-hailing, requires a dynamic organization to (near) instantly fulfill users’ requests, generally of small and time-sensitive packages. On the other hand, longer distance deliveries typically have more heterogeneous shipment sizes and their organization is more similar to the concept of carpooling (on-the-go transportation). This implies less urgency and more time for stakeholders to interact for the organization.

22

A. Punel, A. Stathopoulos / Transportation Research Part E 105 (2017) 18–38

The literature summary has drawn insights from research on ridesharing, the sharing economy, on-demand transport applications, and shipping innovation to delineate the factors that are most likely to contribute to acceptance of crowdshipping and that will guide our model specifications. However, it is important to point out that crowdshipping has several unique features with a bearing for acceptance analysis. Most notably the shift from personal transportation to goods delivery is likely to modify the impact of individual features and attributes. Consequently, it is necessary to investigate crowdshipping, bearing in mind both the similarities and differences to personal mobility systems. The literature on the market structure and service factors has guided the development of the choice experiment. The insights drawn from the behavioral literature from broader settings has instead guided the modeling specifications to account for context, references and experience of users.

3. Data collection The data for this study was collected from a survey administered in the United States in June 2016. The survey was composed of five main parts: – – – –

Preliminary questions to map previous experience with both traditional parcel and crowdsourced shipping. Stated choice scenarios of hypothetical shipments using crowdshipping drivers. Package delivery and shipping service attribute rating and opinion questions. Attitudinal statements designed to measure motivations for using crowdshipping ranging from sharing and sense of community to tangible goals such as saving money (7-point Likert scales). – Demographic information on respondent education, income, gender and age. 3.1. Stated choice scenarios Stated choice methods present decision makers with choices among hypothetical alternatives, designed to reveal the relative value associated with different attributes (Louviere et al., 2000). The stated choice tasks were a key component in the survey as they revealed systematic influences on the acceptance of crowdshipping options. The scenarios were designed to be similar to existing crowdshipping Internet platforms where senders post delivery jobs and available occasional drivers bid on them. Respondents were given a specific framing for a shipment, in terms of a package dimension and distance. They then chose among three alternative crowdshipping drivers offering to deliver the shipment. A pilot study was carried out in the spring of 2016 to identify relevant service attributes and levels for crowdshipping senders. Additionally, a literature analysis (see Section 2) guided the selection and presentation of the attributes and levels. Three main groups of attributes were selected for the choice experiment (an example is shown in Fig. 1); – Traditional shipping attributes represented by delivery cost and shipment duration; – Control over delivery conditions. Compared to traditional carrier shipments, crowdshipping offers opportunities for tracking and exchanges to alter delivery conditions. The delivery convenience was conceptualized as three attributes in the survey. Respondents were given the opportunity to define (a) pickup day, (b) pickup timing, (c) delivery timing, contrasted with having to adapt to the driver’s schedule; – Driver training and experience. A third set of features of the novel system centers on the acceptance of non-professional drivers delivering packages using their current vehicles. This was captured as three separate attributes, namely driver’s expertise (professional or occasional carrier) along with monitoring of driver performance via both recorded experience (number of packages he/she has already carried) and rating from previous deliveries (represented by a star rating). Respondents were presented with a sequence of scenarios, representing three different shipping contexts. The contexts varied by shipment distance and package size relating to different reference points for cost and time performances (see Table 1). – Context 1: A short-distance trip (5 miles), assimilated to a last-mile local delivery, for a large package (size of a television); – Context 2: A medium-distance trip (100 miles) for a medium package (size of a backpack); – Context 3: A long-distance trip (1400 miles) for an extra-large package (size of a mattress). In line with literature findings about the role of experience for processing choices, the design of alternatives and attributes sought to anchor the experiment on actual experiences. It has been established that to minimize biases related to hypothetical choices, experiments should ideally be related to real experiences (Hensher, 2010). Given that only a minority has experienced these systems, the design was based on actual experiences from the existing crowdshipping market to date. The combination of distance and package size was defined according to the most frequently shipped items drawn from a

A. Punel, A. Stathopoulos / Transportation Research Part E 105 (2017) 18–38

23

Fig. 1. Example choice scenario: medium distance.

Table 1 Choice experiment design table. Attribute

Description and levels

Cost

Levels of price by context (in dollar $):
Short distance, 5 levels: 15/18/21/24/27
Medium distance, 5 levels: 23/28/33/38/43
Long distance, 5 levels: 190/230/271/312/352

Time

Levels of time by context (in hours):
Short distance context: 1/4/8
Medium distance context: 24/48
Long distance context: 48/72/96

Professional driver (dummy) Driver experience (dummy) Driver rating: 4.5 stars (dummy) Driver’s rating: 5 stars (dummy) Decide pick-up: day (dummy) Decide pick-up: time (dummy) Decide delivery: time (dummy)

1 = driver is professional; 0 = driver is occasional 1 = driver has fulfilled at least one crowdshipping request; 0 = driver has never fulfilled a crowdshipping request 1 = driver has experience with a 4.5-star rating; 0 = driver has experience with a 4-star rating 1 = driver has experience with a 5-star rating; 0 = driver has experience with a 4-star rating 1 = respondent decides day of pick-up; 0 = driver decides day of pick-up 1 = respondent decides time of the day for pick-up; 0 = driver decides time of the day for pick-up 1 = respondent decides time of the day for delivery; 0 = driver decides time of the day for the delivery

crowdshipping operator’s database shared with the research team. Around 20,000 US observations were analyzed to define realistic reference scenarios and ranges. To obtain more realistic WTP and sensitivity analysis measures for a new and non-mandatory service like crowdshipping, the experiment included an opt-out rather than being designed as a forced choice. In addition, the use of an opt-out or reference alternative is recommended in choice experiment design guidelines to enhance realism and credibility of experiments (Bateman et al., 2002). Following each crowd-shipment scenario with three driver options, a set of additional options was displayed. A fourth option corresponded to a traditional delivery service with price, delivery time, and pickup/delivery conditions drawn from an online request form of a traditional shipper (DHL) for the size-distance combination. Respondent were then asked to either confirm their crowdshipping choice, switch to a traditional delivery method, not send

24

A. Punel, A. Stathopoulos / Transportation Research Part E 105 (2017) 18–38

the package at all (option 5) or defer the decision if they were undecided (option 6, don’t know). This two-step presentation has the advantage of encouraging respondents to first consider the crowdshipping choice carefully, and learn which delivery attributes they care about, before being exposed to the wider set of alternatives that was more realistic for the setting. This approach was replicated during modeling to account for whether people seriously considered the crowdshipping option. In total, each respondent was presented with 9 different scenarios, three for each context, characterized by eight attributes. For the online presentation, scenarios were randomly assigned from a full factorial design of attribute-alternative combinations. The target scenarios included only combinations of three alternatives with no occurrence of binary dominance or unrealistic scenarios. 4. Sample description and summary statistics The survey was conducted online using Qualtrics survey software (www.qualtrics.com) with respondents recruited through an online respondent panel. The survey aimed to gain insights on potential shippers from the general public and their evaluation of crowdsourced delivery options. Table 2 presents a summary of the socio-demographic characteristics of the respondents. The sample is in majority composed of women (60%), 9% points higher than the national average (US CENSUS, 2012a). The sample respondents are younger than the national average of Internet users (US CENSUS, 2012b), with lower representation of respondents above the age of 65. The population in the sample is more highly educated than the American average as more than half of respondents (54%) have a Bachelor’s degree or higher compared to the national average of the Internet user population of 37%. Concerning work status, half of the sample is full time employed, while 8% are students. 5% state to be unemployed which is in line with the July 2016 unemployment statistics of 4.9% (BLS, 2016). The median household income falls within the $40,000–$59,999 category, therefore close to the national average of $53657 (DeNavas-Walt and Proctor, 2015). Correlation across demographic features was also examined by comparing the proportions for the 5 demographic factors. Evidence for variation in employment status was found to be linked to the remaining demographics. In line with expectations, the proportion of high income respondents in the sample increased for higher education attainment. The insight about some degree of correlation between characteristics was used to define the sociodemographics to include in the choice models (Section 6). 4.1. Web based survey response dynamics Relying on an Internet-based survey has potential implications for representativeness and obtaining response rates from different groups (Diana, 2012). Given the increasing penetration of Internet and the advantages in accessing large samples in limited time-frames, it is likely that Internet-based surveys will continue to grow. Until now the findings concerning the impact of online stated choice surveys on representativeness or response behavior is mixed. Web surveys might produce samples with a bias of age (young or middle ranges), male respondents or towards higher income and educational attainment (Kwak and Radler, 2002; Lindhjem and Navrud, 2011; Marta-Pedroso et al., 2007; Olsen, 2009). Importantly, studies vary in the impact of using the internet mode. Looking at response behavior, several studies suggest that there are no significant differences in WTP estimates for respondents answering via web versus face-to-face and mail (Marta-Pedroso et al., 2007; Olsen, 2009; Windle and Rolfe, 2011). There is some indication that internet can provide higher response quality and less errors than a telephone interview in particular for more demanding choice tasks (Börjesson and Algers, 2011). In the context of the present study, it was valuable to obtain responses from people familiar with the use of the Internet given the reliance of crowdshipping platforms on the web and device use. Web surveys indeed have higher rates of more technologically sophisticated respondents (Kwak and Radler, 2002). Moreover, to reduce heterogeneity compared to a national US panel, respondents were drawn from four US states with relatively large populations and developed economies. The survey gathered responses from a total of 587 respondents over the age of 18. After carefully screening for excessively fast completion time, response to an attention filter question, significant share of missing personal information, or strongly repetitive response patterns for blocks of questions, 56 respondents were removed (9.5%). The final sample contained responses from 531 respondents used for the modeling. Despite the lack of representativeness of the sample for some of the reference demographics, this is unlikely to impact the derivation of behavioral estimates that is the core objective of this work. The more serious risk relates to the realism of the experiment and the translation to behavioral reality. The development of the choice experiment as well as the modeling has sought to strike a balance between representativeness and realism of the scenarios. Lastly, given that there is no established reference population of crowdshippers, our analysis will focus on analyzing parameter ratios (willingness to pay) and policy scenarios to derive insights on the sensitivities and reactions to crowdshipping features rather than studying market-shares which are always biased in stated choice data. 4.2. Sample descriptives Table 2 also presents a comparison between the sample fraction that tried crowdshipping versus the rest. With the caveat of the small sample size of real-life users, some trends emerge for the sample demographics. Females are under-represented among the actual users of crowdshipping, who also tend to be younger and have higher income.

25

A. Punel, A. Stathopoulos / Transportation Research Part E 105 (2017) 18–38 Table 2 Key sample demographic characteristics. Survey U.S. population (Count = 531) statistics (2012)

Crowdshipping user in survey (Count = 35)

Never used crowdshipping in survey (Count = 496)

Gender

Female Male

60.08% 39.92%

49.13%a 50.87%a

40.00% 60.00%

61.49% 38.51%

Age (intervals)

Less than 34 35–44 45–64 65 and older

53.86% 22.41% 20.90% 2.83%

46.44%a 14.19%a 28.21%a 11.16%a

60.00% 25.71% 14.29% 0.00%

53.33% 22.22% 21.41% 3.03%

Household annual income (intervals)

Less than $39,999 $40,000–$79,999 More than $80,000

37.67% 40.68% 21.65%

– – –

28.57% 45.72% 25.71%

38.31% 40.32% 21.37%

9.98% 35.97%

34.15%a 28.36%a

8.57% 40.00%

10.10% 35.56%

54.05%

37.49%a

51.43%

54.35%

75.71% 4.71% 19.58%

63.09%b 4.56%b 32.35%b

88.57% 0.00% 11.43%

74.80% 5.04% 20.16%

Education (>25 years old) High school or less Some college or associate’s degree Bachelor’s degree or higher Employment status (>15 years old)

Employed unemployed Not in labor force

Notes: a U.S. Census Bureau. b BLS. Pearson’s Chi-squared two-proportions test was applied to examine correlation between the 5 demographics proportions. Employment status proportions were found to differ significantly for the remaining categories. The proportion of high income was higher for higher education attainment.

It is interesting to note that nearly 7% of the sample have already used a crowdsourced delivery service. A large fraction of respondents (43.3%) has never heard of crowdshipping before taking the survey, while 50% were familiar with the concept (28.1% under a different name). This might appear surprising given that the rate of using more established ride-sharing applications in a similar range of 7.5% for at least weekly use and 12% for at least monthly use (among 2173 registered US voters (Morning Consult, 2015)). As can be observed in Fig. 2, results differ across age categories. Millennial respondents (15–34) are more familiar with crowdshipping, while 25–44 year-old respondents are more likely to have tried it. The majority of respondents over 45 are unfamiliar with the overall concept. It is reasonable to assume that there is a correlation between general shipping propensity and crowdshipping acceptance. This leads us to compare traditional parcel sending and receiving to the use of crowdshipping. Notably, all three behaviors tend to be of the same magnitude across age-classes (see Fig. 3). The exception is for the 25–34 group who sends and receives more shipments overall, but is also much more likely to use crowdshipping. The opposite occurs for the least active shipment group of respondents over 55. The survey gathered opinions about the ranking of goods delivery services. Fig. 4 shows the ranking of shipping features from first to third most important. Cost is by far the leading feature when selecting a shipping option, which is consistent with a 2015 supply chain report suggesting that customers prefer affordable over fast delivery options (Gibson et al., 2015). It is however interesting to note that while speed is the most often cited second priority, the second most popular attribute relates to delivery integrity. Whereas the delivery guarantee is certainly an important feature, it was not directly included in the current experiment given that the setup focused on the competition among drivers. The insurance is typically covered by the shipping company and does not vary by carrier. The delivery handling is indirectly taken into account by the expertise and rating attributes in the experiment. In fact, the most common respondent-suggested criterion was related to the carrier’s reputation. Respondents’ preference for package integrity is strongly related to the trust in well-known traditional carriers. For the crowdshipping companies, who cannot offer similar guarantees, the corresponding assurance is given through a reputation or rating systems related to past delivery performance. A somewhat contradictory finding was that while people seem to care about the integrity of their package, the availability of insurance is not among their main criteria when looking for a shipping service. The survey also explored the attitudes towards crowdsourced deliveries. Respondents were asked to express their level of agreement on 6 positive and 6 negative statements about crowdshipping. Results are presented in Fig. 5. Globally, levels of agreement were aligned across different respondent groups. For positive statements, the consensus was higher overall, with cost being the most valued benefit. The environmental benefit revealed the most difference in opinions, with experienced users and the youngest respondents being in favour, while older respondent did not agree equally on this advantage of crowdshipping. The more pragmatic motivation for using crowdshipping related to a better use of vehicle space was a leading factor mainly for older respondents (55 plus) and experienced users. A breakdown table including the full question format and shares of agreement is included in the appendix. When considering the potential drawbacks of crowdshipping, opinions differed both across statements and respondent groups. The claims that crowdshipping is less efficient or complicated to use

26

A. Punel, A. Stathopoulos / Transportation Research Part E 105 (2017) 18–38

55 and over

2%

45 to 54

6%

15% 14%

35 to 44

8%

25 to 34

8%

15 to 24

56%

26% 22%

54% 28%

26%

5% 0%

28%

29%

25% 10%

20%

43% 37%

29% 30%

40%

41% 50%

60%

70%

80%

90%

100%

I have already used crowdshippping I have already heard about crowdshipping, but I have never used it I have already heard about crowdshipping, but I didn't know it was called like this I have never heard about crowdshipping Fig. 2. Crowdshipping familiarity by age category (n = 531).

49%

50% 45% 40%

37%37%

35% 30%

26% 24%25%

25% 16% 13% 11%

15% 13% 11%

11%11%

20% 15% 10%

3%

5% 0%

55 and over

45 to 54

35 to 44

25 to 34

15 to 24

Respondents having already used crowdshipping Respondents sending more than one package a month Respondents receiving more than one package a month Fig. 3. Shipping experience divided by age category; Crowdshipping (n = 35; 7%), sending 2+ packages (n = 175; 33%) and receiving 2+ packages (n = 372; 70%) per month.

were not supported by the majority of respondents. However, interestingly, the higher scores for these two items were given by the group of people who have already experienced a crowdsourced delivery service. This suggests that reality might differ from the expectations regarding the ease of use of such a platform. Across the board, higher income respondents were more concerned about crowdshipping disadvantages, from trust to driver credentials. On the other hand, those who have tried crowdshipping were less worried about the absence of professional drivers and the reliability of the delivery.

5. Modeling The objective of the models is to determine which attributes are most relevant, what role socio-demographic features play and to examine the impact of the shipping context in relation to crowdshipping acceptance. The models were defined as Multinomial Logit (MNL) and Error Component Random Parameter (or Mixed) Logit models (RPL). The decision-makers n (senders) are assumed to derive utility Unj from the set of shipping options. The discrete choice models presented in the remainder of this section were estimated with Biogeme (Bierlaire, 2003).

27

A. Punel, A. Stathopoulos / Transportation Research Part E 105 (2017) 18–38

Delivery Cost

370

84

28 66

Package Received in its Integrity

105 120

33

Speed

142

88 16

Convenience

15

Respect of Scheduled Delivery Time

14

Reliability in Pick Up Timing

10

Insurance's Availability 0

60 45

113 76

32 37 36 38 50

First Concern

100

150

Second Concern

200

250

300

350

400

Third Concern

Fig. 4. Ranking of shipping attributes [top three, sorted on first concern] (n = 531).

Positive features

Negative features

Fig. 5. Spider chart of agreement with positive and negative crowdshipping statements (n = 531).

5.1. Model specification In this paper we employ a model including error components and random parameters (RPL). The utility a decision maker n receives from choosing alternative i among shipping options j is decomposed in a systematic component V ni and a random error component eni . The systematic component is defined as a linear additive function of observed attributes K such that (Eq. (1))

V ni ¼

K X a0k xki

ð1Þ

k¼1

where a is interpreted as the marginal utility associated with attribute k. If the error term is assumed to be independent and identically distributed (iid) with extreme value type 1 distribution the probability that a decision is chosen is given by the V conditional logit formula Pni ¼ Pe niV nj . j

e

The limitations inherent in assuming that preferences are fixed for all attributes and constant across alternatives can be overcome by including random parameters (Train, 2009). Eq. (2) shows the RPL utility function for the crowdshipping alternatives including random taste parameters and error components.

28

A. Punel, A. Stathopoulos / Transportation Research Part E 105 (2017) 18–38

U ni ¼ a0 xni þ l0n yni þ b0n zni þ eni

ð2Þ

The terms xnj , ynj , znj represent vectors of observed variables related to the decision makers and the alternatives. a is a vector of fixed coefficients representing the taste of the sample and eni represents the random term, assumed to be Independent and Identically Distributed (IID) extreme value. To account for the fact that respondents may view hypothetical scenarios systematically different (e.g. Scarpa et al., 2005) the utility is further partitioned into ln, a vector of random terms with a mean of zero. The vector of bn coefficients represent the random taste of decision maker n. The vectors that vary over decision makers (ln, bn) do so with densities f ðlÞ and f ðbÞ that need to be defined by the researcher. The density f ðlÞ is specified as normally distributed with Nð0; r2 Þ capturing the common variance of the three crowdshipping alternatives. Different assumptions for the density f ðbÞ is tested in modeling allowing random taste variation across people in the sample. Commonly used distributions include normal, lognormal and triangular (Hensher and Greene, 2003). The RPL model also accounts for the pseudo-panel nature of the data, i.e. that each respondent faces a sequence of crowdshipping scenarios. The repeated choices introduce correlation in the unobserved factors over time (Train, 2009). The likelihood is shown in Eq. (3) and the probability of making a certain sequence of choices t(n) is defined in Eq. (4). To derive estimates of the random parameters, it is necessary to integrate over the densities f ðlÞ, f ðbÞ and enj . The choice probability for alternative i is thus conditional on the values of on b and l and derives the product over the repeated choices of each respondent and can be written:

ea xni þln yni þbn zni eV ni Lni ðbn ; ln Þ ¼ P a0 x þl0 y þb0 z ¼ P V nj nj nj nj n n je je 0

0

0

ð3Þ

The integration is performed with simulation deriving the choice probability over all possible parameter values (Eq. (4)).

ZZ PtðniÞ ¼

P

eV ni P V f ðbjXÞuðlj0; r2 Þdbdl nj t¼1...n je

ð4Þ

where f(
) is the joint density function of random attribute parameter vectors and u(
) the normal density function for the error component. MHLS draws (Hess et al., 2006) are used to boost the simulation efficiency, testing different numbers of draws until estimates stabilize. Based on model fit, analysis of the coefficient of variation and the reasonableness of the random parameter estimates, a lognormal distribution was selected for the cost parameter, for each of the three shipment contexts. This implies lnðbÞ  NðhLN ; rLN Þ where hLN is the mean and rLN is the standard deviation parameter to be estimated. For interpretation, the mean, standard deviation and median of the lognormal distribution are computed using the respective formulas in Hess et al. (2005) and Train (2009).1 To verify the presence of asymmetry, cost and time were modeled as separate parameters for gains and losses (dell’Olio et al., 2016). The reference level was the average price and time used to design the experiment. To analyze the potential role of experience on crowdshipping preferences the scale of the utility was allowed to vary according to different experience-features (Swait and Louviere, 1993). To accomplish this the observed utility in Eq. (3) is multiplied for each individual by the scaling function s ¼ ðexpðk01 CS:USER þ k02 SENDER þ   ÞÞ. The candidate experience factors tested in the model are: real experience using crowdshipping (CS.USER), various levels of discrete parcel shipper (SENDER) and receiver experience, and possibility of socio-demographic factors impacting the estimated experience parameters k. The functional form of the experience measure follows Hensher and Ho (2016). Our model tests several forms of the experience scaling, and finds evidence that experience measures vary by the delivery context.

6. Results The MNL and RPL model parameters and associated robust t-ratios are reported by context (local, medium, long) in Table 5. The presented results are derived from extensive specification testing. The systematic utilities for crowdshipping (CS), traditional (TRAD), no shipment (NO) and don’t know (DK) are expressed as follows.

V CS1...3;n ¼ ðexpðk1  CS:USER þ k2  SENDERÞÞ  ascCS þ aþ  xnþ þ a  xn þ ln  yn þ bn  zn V TRAD;n ¼ ascTRAD þ a  xn V NO;n ¼ ascNO V DK;n ¼ ascDK

1

hLN ¼ ehLN þ

r2

LN 2

ðmeanÞ;

ð5Þ

pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2

rLN ¼ hLN : erLN  1 ðst devÞ; medLN ¼ ehLN ðmedianÞ

A. Punel, A. Stathopoulos / Transportation Research Part E 105 (2017) 18–38

29

6.1. Overall model evaluation: MNL and experience conditioned RPL The specification of the three CS options V CS1...3;n includes gains and losses specification ðaþ ; a Þ and experience conditioning. To facilitate comparison, the error components and random parameters are layered on the basic MNL for each context with no further changes than removing insignificant parameters. The RPL models outperform their MNL counterpart for each context (see Table 5). Interestingly, the error components become insignificant once experience conditioning is included, with exception of the medium context. Random parameters were tested for cost, time and all experience coefficients. No evidence of random effects was found for travel-time, while cost, and some of the experience parameters were found to have significant random heterogeneity. The RPL models stabilized at 500 draws using MHLS. The choice of distributional assumption for the random cost parameters was based on the theoretical advantages of the log-normal distribution that maintains the expected (negative) sign for each person. The lognormal assumption also had a better goodnessof-fit and smaller coefficient of variation than other distributional assumptions tested (normal and uniform). The experience measures were fitted with a normal distribution to allow for a mix of negative and positive effects. The coefficient of variation is calculated as cov ¼ st:dev rpl =meanrpl . This provides a measure of the remaining preference heterogeneity in the sample. The higher coefficient of variation in the local context suggests there is more unexplained variability in these decisions. Both specifications are presented to illustrate the different insights derived from each. The focus of the model specification is to identify individual heterogeneity (systematic and random) in attribute sensitivity, acceptance of the crowdshipping alternatives and impact of the shipping context. The most significant effects overall are due to the context, with evidence for three clearly delineated markets, differing more in terms of attribute sensitivity than according to respondent’s sociodemographic status. Given this, the following results are discussed by attribute type and highlight the different sensitivities according to shipment distance. 6.2. Experience effects A number of potential experience effects, relating to crowdshipping and traditional shipping and receiving packages were tested as experience conditioning effects (i.e. effects that rescale the crowdshipping attributes). For local and long distances the effect of having sent at least two parcels in the last month (dummy) was significant, while having tried crowdshipping (dummy) was significant in all contexts. A surprising finding is that the different experiences have opposite effects on the heteroscedasticity of the crowdshipping attributes. When deriving the exponential of the CS_User parameters (0.605 for local, 0.536 for medium) this implies that utility is rescaled downwards for the innovative options for users that tried crowdshipping in the past. On the contrary, having experience with shipping multiple packages in the past month with traditional means, which 1/3 of respondents had done, leads to the opposite effect. For local deliveries, the effect (scale = 1.242) implies that the perceived value of crowdshipping increases, likely due to the increased convenience. In the long distance shipments, the effect is the opposite. Crowdshipping experience leads to higher scale of those attributes, while sending  2 packages leads to lower value of innovative attributes. Further information is revealed when considering the normal random parameters fitted to the experience factors. Looking at the sending  2 packages effect, the normally distributed estimates suggest that both positive and negative preferences exist in the sample. Specifically, in the local setting, 28% of the distribution is below zero. For long distance, 69% of the distribution is negative. These findings imply that for both settings about 1/3 of respondents have an experience scaling effect that is contrary to the sensitivity implied by the mean sign (illustrated in Fig. 7). This highlights the nuances inherent in modeling experience effects, and the significant heterogeneity related to how experiences translate into absolute preferences. 6.3. Analysis of preferences and referencing The time and cost parameters remain significant with the expected negative sign throughout. An important exception occurs for delivery time in the long-distance context. Attempts to transform the long distance time variable did not result in a significant effect. Evidence for gain-loss asymmetry was tested for time and cost in all contexts. Significant asymmetry was found only for cost in the short context. Curiously, the asymmetry gives more weight to gains than to losses, thereby going against most established literature (e.g. Masiero and Rose, 2013). By studying the willingnessto-pay estimates in Table 3 we observe a trend towards marginally declining willingness to pay for decreasing delivery time. For a local delivery, a day of delivery time saved is worth paying 41 $ for, while a day lost requires 13 $ in compensation. The unexpected finding reflects that our sample yields more utility from improving cost performance, than they suffer from cost increases. This suggests that in the local delivery market users are less sensitive to cost increases, which might suggest that the presented costs fell in an acceptable range, and users were inclined to search for ‘a good deal’ albeit not shunning higher costs. The medium distance, inter-city market, is willing to pay around 2 $ for a day of delivery time saved. The second group of attributes deals with the control over delivery conditions, in the form of being able to schedule the hour or day of the pick-up and delivery. Notably, getting to decide the day of the week is more valuable than picking the hour

30

A. Punel, A. Stathopoulos / Transportation Research Part E 105 (2017) 18–38

Fig. 6. Elasticity of competition between crowdshipping and other shipping options.

for the local and long distance senders. Incidentally, this might be related to the larger size of the shipment in these two cases. The finding suggests that daily schedules and activities are a more important driver of preferred shipment pick-up. Turning instead to consider the drop-off condition, the effect is only present in the long-distance crowdshipment. It appears that local shipments are more concerned with the pick-up scheduling related to the respondent’s own schedule, while longer shipments shift the attention to the final delivery. This latter finding may relate to increasing need for control of a delivery that spans more distance and time. This is in line with Esper et al. (2003) noting that carrier transparency led to higher propensity to choose that option. The third set of attributes, uniquely related to crowdshipping, seizes on the driver’s training and performance in the platform. Longer distance emphasizes the driver training and experience in the platform, while shorter distance decisions are more guided by driver ratings. The star-rating parameters can be understood as a ‘bonus’ over the 4-star rating which is the minimum tested in the experiment. 6.4. Socio demographics It is important to note that while several socio-demographic factors are found significant in the basic MNL model, most of these effects become insignificant in the RPL specifications. The general findings point towards higher income individuals being more likely to try crowdshipping. Respondents with higher level of employment and lower education were also more likely to abandon the traditional shipment, potentially due to the opportunity to save time and money by doing so. These effects did not hold up when the correlation patterns and random heterogeneity was controlled for. The sole sociodemographic impact that stays significant is gender effect for long distance (men prefer traditional shipments) and the effect of low income for long distance shipments (higher income favors trying crowdshipping). This finding is aligned with some of the results from passenger mobility ride-hailing (Dias et al., 2016) where higher income classes are more likely users. This implies that crowdshipping is more likely to compete by focusing on unique service quality attributes, rather than competing just on cost and time. The findings related to the socio-demographics also highlight the emerging nature of the crowdshipping market where it is difficult to identify established acceptance patterns. To summarize, the findings suggest that there are distinct preference patterns in the three markets. In the local delivery market, customers are highly time-sensitive and oriented towards monitoring intra-system experience and ratings compared to traditional shipper training. The medium distance context implies similar emphasis on crowdshipping experience but overall lower willingness to pay for improving performance, in particular a significantly lower sensitivity for delivery

31

A. Punel, A. Stathopoulos / Transportation Research Part E 105 (2017) 18–38

0.4 0.3 0.2 0.1 0.0

kernel distribution

Density of Crowdshipping user experience normal dist. (Long Dist. Mean=0.422)

-4

-2

0

2

4

kernel distribution

0.0 0.2 0.4 0.6 0.8 1.0

Density of Sender>2 experience normal dist. (Long Dist. Mean=-0.204)

-4

-2

0

2

4

0.3 0.2 0.1 0.0

kernel distribution

Density of Sender>2 experience normal dist. (Local Dist. Mean=0.598)

-4

-2

0

2

4

Fig. 7. Kernel density of RPL parameters describing crowdshipping and traditional sender experience (Normal distributions).

Table 3 Willingness to Pay (WTP) for attributes by context. Local (5 miles) Willingness to pay for:

MNL (cost dec.)

MNL (cost inc.)

RPL (cost dec.)

RPL (cost inc.)

D RPL/MNL (Local)

D RPL/MNL (Local)

Time savings/losses per hour Time savings/losses (day) Professional driver status Driver CS experience Driver 4.5 star rating Driver 5 star rating Driver 4.5/5 star rating Schedule delivery time Schedule pick-up day Schedule pick-up time

0.56/h 13.4 1.53 4.15 2.95 4.38 – – 4.59 2.54

1.70/h 40.8 4.6 12.47 8.88 13.17 – – 13.8 7.65

0.42/h 10.1 1.26 3.61 1.69 2.19 – – 3.06 2.17

1.03/h 24.7 3.10 8.85 4.15 5.37 – – 7.51 5.32

0.75 0.75 0.82 0.87 0.57 0.50 – – 0.54 0.70

0.61 0.61 0.67 0.71 0.47 0.41 – – 0.54 0.70

Medium (100 miles)

Long (1400 miles)

Willingness to pay for:

MNL

RPL

D RPL/MNL (Med.)

MNL

RPL

Time saving/losses per hour Time saving/losses (day) Professional driver status Driver CS experience Driver 4.5 star rating Driver 5 star rating Driver 4.5/5 star rating Schedule delivery time Schedule pick-up day Schedule pick-up time

0.11/h 2.64 – 10.08 – 3.78 – – 2.86 3.49

0.08/h 1.92 – 5.95 – 2.06 – – 2.59 2.42

0.73 0.73 – 0.59 – 0.54 – – 0.91 0.69

– – 38.44 83.02 – – 48.58 18.63 43.49 23.07

– – 31.35 77.39 – – 46.52 17.00 39.57 20.21

D RPL/MNL (Long)

0.82 0.93

0.96 0.91 0.91 0.88

32

A. Punel, A. Stathopoulos / Transportation Research Part E 105 (2017) 18–38

Table 4 Market changes in response to crowdshipping performance variation (local and long distance deliveries).

All 5-star drivers All drivers experienced All drivers professional Always select pickup timing Always select pick-up day Always select delivery timing

Local MNL

Local RPL

D RPL/MNL

Long MNL

Long RPL

D RPL/MNL

7.71% 2.34% 2.11% 2.47% 4.87% na

6.04% 2.59% 2.24% 2.71% 4.27% na

0.784 1.106 1.060 1.100 0.877 na

12.34% 7.90% 5.74% 4.58% 5.18% 1.46%

12.40% 7.57% 4.96% 4.16% 5.17% 1.39%

1.005 0.959 0.864 0.910 0.997 0.954

time. When evaluating a long-distance shipment, traditional professional training of the driver and delivery control emerge as decisive factors while delivery duration is not significant.

6.5. Willingness to pay and sensitivity analysis Market shares derived from this model should be treated with caution given the stated choice nature of the data. Instead, to draw useful insights from this model the focus is on marginal trade-offs and the relative changes in shares of the alternatives as a function of changes in the qualitative variables. Table 3 shows the WTP for each attribute. For the RPL model these are derived from parameter ratios using the mean of the lognormal cost-distribution. Comparing the models, the specifications give similar results. With few exceptions, the RPL reveals lower willingness to pay estimates (shown in Table 3 as D < 1). This suggests that the initial model with only systematic preference heterogeneity overestimates how much potential senders would be willing to pay for the new delivery solution. The WTP parameters of the long distance shipment may appear high but need to be analyzed in the context of the overall costs in this distance class. While the local and medium contexts had similar price ranges (20–30 $), the average long distance shipment price was $271. The WTP and market share sensitivity appear lower than traditional freight service estimates. It is, however, difficult to compare our results to traditional shipping literature given that logistics costs reported by carriers also include labor and vehicle costs, which are not relevant for the sender decision analyzed here (Tavasszy and De Jong, 2014). Further insights can be drawn from analyzing market share changes that determine the % change in choice probabilities, in response to a % change in a given attribute performance. Given the declared choice nature of our data, and a lack of a reference market to re-calibrate the constants, caution must be applied in interpreting sensitivity results. Rather than presenting point elasticities that are impacted by the point of departure of the calculation, Fig. 6 presents the % change in demand for a range of sample-enumerated attribute changes. Given that % changes are not feasible for discrete features, pseudoelasticities are derived by comparing the base-line market-shares to scenarios where, case-by-case, each service attribute, such as a 5-star driver status, is imposed for all the crowdshipping options, while all else is left unaltered. The calculation takes the ratio of the change/the mid-point of the before-after change scenario, computed for each observation then averaged. Table 4 shows the pseudo-elasticity sensitivity analysis for the discrete features for the local and long distance contexts. The estimates suggest a strong consistency for the two model forms, while the response sensitivity is consistently higher for the long distance setting. The presented percentages can be interpreted as the effect of ensuring all drivers have the maximum performance, e.g. all carriers are professionally trained, on the overall market share gain of crowdshipping. The sensitivity estimates are in the range of 2–12%. This can be interpreted as a sensitivity analysis measuring the maximum market share crowdshipping would gain from other options if the highest attribute performance could be ensured. Many of these changes can be regarded as rather modest. Overall, the experience attribute group yields a stronger impact, compared to the delivery control attributes. More insights can be derived from observing the competition between the crowdshipping versus the remaining options, namely the traditional and opt-out alternatives. Fig. 6 shows sample-enumerated scenarios of market-share changes as the status quo is compared to the ‘best’ performance of crowdshipping features. While caution should be exercised given the hypothetical nature of the experiment, the models have sought to make the reactions of the respondents as realistic as possible by using experienceconditioned models. The plot shows that the market-share competition occurs between crowdshipping options and both traditional and the no-shipping option. The amount of market-share that is gained by crowdshipping when reaching top performance for each driver feature ranges up to a 12% gain. The most pronounced performance rewards relate to improving driver ratings (for either distance), and to driver experience (long-distance) and selecting the day of pickup (local shipment). There is evidence for more direct competition between innovative and traditional shipment options in the local context. Instead, for longer distances, the majority of the market-share that crowdshipping attracts is drawn from the ‘no-shipment’ option. This points towards one of the distinctive features of crowdshipping. That is, it is not yet clear from the research if it competes directly with traditional shipping options, or rather opens the door to shipping objects in settings that are not viable, or prohibitive, with traditional shipping solutions (Rougès and Montreuil, 2014).

Table 5 Model results.

Number of parameters Number of observations Final log-likelihood Rho-square LR ratio test (v2 critical value) Variable name

Experience features

Socio-demographic

RPL - Local

MNL - Medium

RPL - Medium

MNL - Long

MNL - Long

18 1593 Obs. 2142.15 0.249 154.39 (v2 = 3.84)

19 1539 Obs. 531 Ind. 2064.955 0.277

19 1593 Obs. 1325.981 0.535 147.526 (v2 = 9.49)

15 1539 Obs. 531 Ind. 1252.218 0.561

16 1593 Obs. 1804.55 0.368 33.75 (v2 = 3.84)

17 1593 Obs. 531 Ind. 1787.67 0.374

Est.

Rob. t-test

Est.

Rob. t-test

Est.

Rob. t-test

Est.

Rob. t-test

Est.

Rob. t-test

Est.

Rob. t-test

– – – 0.0635 0.191 – –

– – – 3.79 7.10 – –

– 0.108 0.292 0.792 0.564 – 0.836 – 0.876 0.486 –

– 5.07 3.56 5.63 5.11 – 7.54 – 10.62 5.74 –

– – – – – 3.34 1.59 2.05 1.32 0.129 0.389 1.110 0.520 – 0.673 – 0.942 0.668 –

– – – – – 7.04 5.3 8.82 7.09 5.39 3.87 5.9 3.7 – 5.86 – 9.24 6.49 –

0.131 – – – – – – – – 0.0146 – 1.32 – – 0.495 – 0.375 0.457 –

11.61 – – – – – – – – 2.43 – 5.11 – – 3.61 – 2.24 3.40 –

– 1.48 0.368 – – – – – – 0.0186 – 1.45 – – 0.501 – 0.631 0.590 1.51

– 10.62 4.01 – – – – – – 2.55 – 5.27 – – 3.12 – 3.42 3.66 0.01

0.0212 – – – – – – – – – 0.815 1.76 – 1.03 – 0.395 0.922 0.489 –

21.66 – – – – – – – – – 8.92 11.49 – 10.64 – 3.87 10.04 4.97 –

0.023 – – – – – – – – – 0.721 1.78 – 1.07 – 0.391 0.91 0.465 –

19.59 – – – – – – – – – 8.27 12.69 – 10.31 – 3.5 9.84 4.81 –

Crowdshipping user Crowdshipping user [random normal - mean] Crowdshipping user [random normal - std dev] Send  2 package/m Send  2 package/m [random normal - mean] Send  2 package/m [random normal - std dev]

0.502 – – 0.217 – –

3.40 – – 3.41 – –

0.624 – – – 0.598 1.040

3.04 – – – 4.27 4.96

0.391 – – – –

3.05 – – – –

0.292 – – – – –

1.92 – – – – –

0.355 – – 0.13 – –

3.58 – – 2.45 – –

– 0.422 1.05 – 0.204 0.418

– 1.95 3.15 – 3.17 3.52

Male 15–24 year old 25–34 year old  Male 45–54 year old  Male 55 year old and over Low income Medium income High income Low education Employed - full & part time Employed - full time

– 0.702 – – – – – 0.465 – – –

– 1.95 – – – – – 1.99 – – –

– – – – – – – – – – –

– – – – – – – – – – –

– – 0.469 0.965 0.574 0.724 0.615 – 0.426 0.327 –

– – 3.03 2.41 3.15 3.97 3.46 – 2.28 2.20 –

– – – – – 0.434 – – – – –

– – – – – 1.78 – – – – –

0.598 – – – – – – – – – 0.437

2.85 – – – – – – – – – 2.03

0.701 – – – – – – – – – –

2.35 – – – – – – – – – –

Cost Cost [random lognorm. dist. - mean] Cost inc. [random lognorm. dist. - std. dev] Cost increase Cost decrease Cost inc. [random lognorm. dist. - mean] Cost inc. [random lognorm. dist. - std. dev] Cost dec [random lognorm. dist. - mean] Cost dec [random lognorm. dist. - std. dev] Time Professional driver Experienced driver Driver 4.5 star rating Driver 4.5&5 star rating Driver 5 star rating Decide delivery: time Decide pick-up: day Decide pick-up: time Crowdshipping st dev

A. Punel, A. Stathopoulos / Transportation Research Part E 105 (2017) 18–38

Shipping attributes

MNL - Local

(continued on next page)

33

34

Table 5 (continued) MNL - Local

ASC

Crowdshipper 1 Crowdshipper 2 Crowdshipper 3 Traditional shipment No shipment No opinion

RPL - Local

MNL - Medium

RPL - Medium

MNL - Long

MNL - Long

0 0.00138 0.0741 0.523 0.451 0.00385

na 0.02 0.91 2.55 2.31 0.02

0 0.0458 0.0442 0.917 0.815 0.360

na 0.48 0.49 3.86 3.24 1.37

0 0.0883 0.0949 0.643 4.28 3.65

na 0.56 0.59 1.39 9.01 8.08

0 0.267 0.145 1.78 5.9 5.27

na 1.38 0.71 2.3 7.06 6.47

0 0.0869 0.243 4.12 3.37 4.6

na 0.90 2.49 13.44 13.53 16.2

0 0.15 0.233 4.52 3.96 5.20

na 1.52 2.28 12.96 12.36 15.34

– – – – – – – – –

– – – – – – – – –

– – – 0.125 0.269 0.035 0.308 0.446 0.129

– – – – – – – – –

– – – – – – – – –

– – – – – – – – –

0.244 0.158 0.228 – – – – – –

– – – – – – – – –

– – – – – – – – –

– – – – – – – – –

– – – – – – – – –

– – – – – – – – –

– – – – –

– – – – –

– 2.147 1.449 – 1.739

– – – – –

– – – – –

– – – – –

0.649 – – – –

– – – – –

– – – – –

– – – – –

– – – 2.488 2.049

– – – – –

Derived parameters

COV COV COV Coefficients of variation COV COV

cost [lognormal] cost inc. [lognormal] cost dec. [lognormal] crowdshipping user [normal] Send > 2 Package/m [normal]

Note: 500 MHLS used for all RPL models.

A. Punel, A. Stathopoulos / Transportation Research Part E 105 (2017) 18–38

Lognormal parameters

Mean - cost. Std. dev. - cost. Median - cost. Mean - cost inc. Std. dev. - cost inc. Median - cost inc. Mean - cost dec. Std. dev. - cost dec. Median - cost dec.

A. Punel, A. Stathopoulos / Transportation Research Part E 105 (2017) 18–38

35

7. Summary and future directions Traditional delivery services, such as United States Postal Service or Fedex now co-exist with a myriad of e-retailers and start-ups offering alternative service models, using crowd resources. There is thus growing interest in the role of freight delivery innovation using crowdsourced vehicles and drivers. Crowdshipping systems make use of existing resources and have the potential to increase system efficiency for operators and at the same time provide a radically revised experience for consumers. Given the advancements by shipping start-ups and retail trials, there is an urgent need for a research agenda to examine the feasibility, the potential as well as the pitfalls, of crowdsourced deliveries. There is currently a gap in the literature related to acceptability and relative preference for crowdshipping attributes that this paper seeks to address. This paper provides a first exploration of how the public evaluates hypothetical shipping scenarios where different types of deliveries are fulfilled by non-professional shippers and compared to traditional shipping options. The experiment is designed as a shipping platform where crowdshipping drivers are compared. Several innovative features are included to characterize the bidding situation including driver expertise and rating as well as increased consumer control over the pick-up and delivery planning than is typically observed in traditional shipping. The choice models developed include traditional attributes and socio-demographic effects. In addition, to account for referencepoint impacts, asymmetric preference parameters were included. Finally, to deal with the variable rate of experience related to the novel shipping option, models with experience-conditioning were developed for each context. The results confirm the importance of shipment cost in line with existing literature on freight delivery preferences. Most shipping attributes, from shipment duration to driver training had variable impact in the different shipping distance scenarios. The RPL model findings point to fairly distinct preference patterns in the local, inter-city, and long distance markets. Differences are more pronounced for driver performance attributes than for socio-demographics. The experience scaling showed that both direct crowdshipping and traditional sender experience significantly impacted the value of other attributes. Strong differences were revealed with opposite effects for direct and indirect (traditional sender) experience, as well as changing effects by distance and within the groups. A notable finding was that users with direct experience of innovative shipping appeared to have a more critical stance towards crowdshipping attributes in the local setting. Our findings suggest that past experiences with the new option does not generate a strictly positive utility as expected from previous work. Instead a more complex image emerged in this research. This finding might relate to also experiencing the potential disadvantages of the service as suggested by the lower ratings among experienced users when rating positive and negative crowdshipping statements (Fig. 5 & appendix). The contribution of the research findings presented here are twofold. First, they add to the body of evidence on the acceptability of new crowd-sourced systems. Parallel work is seeking to study acceptance of public agencies, crowdshipping firms, receivers and the crowd-carriers that make the new system possible. Secondly, the paper offers specific insights on how to frame a crowdshipping investigation and provides first findings from a US sample related to attribute sensitivities, acceptance drivers, and derived WTP and market sensitivity measures. The model findings suggest insights for operational research about the ranking of attributes for different distance settings. The research also gives insights for company decisions where the findings can guide business model design. While a more dynamic and impatient on-demand process rules local deliveries, where time is the priority, longer distances lower these requirements and senders pay more attention to features of the delivery related to managing the shipping process. A limitation of the study lies in the fact that only a limited number of delivery situation combinations are presented in the experiment. For future work, as the crowdshipping market matures, it would be ideal to anchor the shipment on respondents experience. Moreover more factors can be accounted for, including combinations of shipment object, value, purpose/urgency, conditions of traffic-congestion, alongside the distance class. Given that the current results can provide priors for subsequent work, a follow-up experiment can be designed to control more of the variation in shipment motivations. Lastly, it is important to note that several stakeholders are relevant to the functioning of crowd logistics, from receivers, to drivers, senders, service company operators and policy makers. In this paper we focus on the potential customers, or shippers. This emphasis is justified from the fact that the early literature suggests that shippers and drivers are the core stakeholders to build critical mass. This interaction is similar to ride-hailing systems where demand depends on having enough drivers, and attractiveness for drivers to sign up hinges on the level of demand. Importantly, the decision process of shippers is bound to be fundamentally different from that of the driver carrying out the delivery. A first investigation of occasional drivers comparing a pure commute to a commute including a delivery (detour) and profit is examined in Miller et al. (2017). Future efforts will also address the role of the receiver, who is to date less of a driving force in the decision around the shipment. Notwithstanding, this is progressively changing, as receivers are increasingly involved in the shipment quality, and with e-retailing, many times the sender and receiver are the same person. Acknowledgements This research is based upon work supported by the National Science Foundation Partnerships for Innovation: Building Innovation Capacity (PFI:BIC) Grant No. 1534138 Smart CROwdsourced Urban Delivery (CROUD) System.

36

Appendix A. Reasons to try and reservations about crowdshipping globally and by respondent category

CSP1 CSP2 CSP3

CSP5

CSP6

CSN1 CSN2

CSN3 CSN4 CSN5 CSN6

% of Agree – Global

% of Agree – Age: 15–24

% of Agree – Age: 25–34

% of Agree – Age: 35–44

% of Agree – Age: 45–54

% of Agree – Age: 55 and over

% of Agree – Shipping experience

% of Agree – No shipping experience

% of Agree – Crowdshipping experience

% of Agree – No crowdshipping experience

. . . it allows me to save money . . . it is eco-friendly . . . I like the idea of taking advantage of the empty space in one’s car . . . it is more convenient than traditional delivery services . . . I am able to communicate with the drivers for the whole delivery time . . . the whole delivery (payment, request, monitoring) is done through my smartphone

90.21 59.70 61.58

88.75 73.75 57.50

93.20 62.62 63.59

87.39 51.26 56.30

86.15 47.69 56.92

91.80 60.66 75.41

87.43 65.71 64.57

91.57 56.74 60.11

91.43 74.29 74.29

90.12 58.67 60.69

73.45

71.25

76.70

66.39

75.38

77.05

74.29

73.03

80.00

72.98

72.88

72.50

76.70

66.39

67.69

78.69

74.86

71.91

80.00

72.38

75.14

72.50

76.70

74.79

81.54

67.21

77.14

74.16

77.14

75.00

I have reservations about trying crowdshipping because ...

% of Agree – Global

% of Agree – Age: 15–24

% of Agree – Age: 25–34

% of Agree – Age: 35–44

% of Agree – Age: 45–54

% of Agree – Age: 55 and over

% of Agree – Shipping experience

% of Agree – No shipping experience

% of Agree – Crowdshipping experience

% of Agree – No crowdshipping experience

. . . I wouldn’t trust the drivers . . . I don’t want to communicate my information to a stranger . . . drivers are not professional carriers . . . I am worried about the delivery conditions . . . the system seems complicated to use . . . it does not seem as efficient as traditional delivery services

61.39 56.50

70.00 61.25

64.56 57.28

61.34 57.98

50.77 50.77

50.82 50.82

52.57 52.57

65.73 58.43

48.57 54.29

62.30 56.65

52.54

51.25

59.22

48.74

44.62

47.54

46.86

55.34

40.00

53.43

64.78

66.25

65.05

67.23

64.62

57.38

62.29

66.01

48.57

65.93

15.63

13.75

19.42

13.45

12.31

13.11

20.57

13.20

22.86

15.12

22.98

23.75

25.73

23.53

20.00

14.75

25.71

21.63

31.43

22.38

Note: ‘‘% of Agree” in table is re-coded from 7-point scale by grouping the top three levels, from somewhat agree to strongly agree.

A. Punel, A. Stathopoulos / Transportation Research Part E 105 (2017) 18–38

CSP4

I would like to try crowdshipping because . . .

A. Punel, A. Stathopoulos / Transportation Research Part E 105 (2017) 18–38

37

References Agatz, N., Erera, A., Savelsbergh, M., Wang, X., 2012. Optimization for dynamic ride-sharing: a review. Eur. J. Oper. Res. 223, 295–303. http://dx.doi.org/ 10.1016/j.ejor.2012.05.028. Anderson, D.N., 2014. Not just a taxi”? For-profit ridesharing, driver strategies, and VMT. Transportation (Amst) 41, 1099–1117. http://dx.doi.org/10.1007/ s11116-014-9531-8. Archetti, C., Savelsbergh, M., Speranza, M.G., 2016. The vehicle routing problem with occasional drivers. Eur. J. Oper. Res. 254, 472–480. http://dx.doi.org/ 10.1016/j.ejor.2016.03.049. Arslan, A., Agatz, N., Kroon, L.G., Zuidwijk, R.A., 2016. Crowdsourced Delivery: A Dynamic Pickup and Delivery Problem with Ad-hoc Drivers. ERIM Report Series Reference. Barone, V., Crocco, F., Mongelli, D.W.E.E., 2013. Models of choice between shopping and E-shopping. Appl. Mech. Mater. 442, 607–616. http://dx.doi.org/ 10.4028/www.scientific.net/AMM.442.607. Bateman, I.J., Carson, R.T., Hanemann, W.M., Hanley, N., Lee, M.J., Loomes, G., Mourato, S., Pearce, D.W., Elgar, E., Day, B., Hanemann, M., Hanley, N., Hanemann, W.M., Hanley, N., Lee, M.J., Loomes, G., Mourato, S., Pearce, D.W., Elgar, E., 2002. Economic Valuation with Stated Preference Techniques: A Manual. Edward Elgar. Belk, R., 2014. You are what you can access: sharing and collaborative consumption online. J. Bus. Res. 67, 1595–1600. http://dx.doi.org/10.1016/j. jbusres.2013.10.001. Bellotti, V., Ambard, A., Turner, D., Gossmann, C., Demkova, K., Carroll, J.M., 2015. A muddle of models of motivation for using peer-to-peer economy systems. In: Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems – CHI ’15. ACM Press, New York, USA, pp. 1085– 1094. http://dx.doi.org/10.1145/2702123.2702272. Ben-Elia, E., Erev, I., Shiftan, Y., 2008. The combined effect of information and experience on drivers’ route-choice behavior. Transportation (Amst) 35, 165– 177. http://dx.doi.org/10.1007/s11116-007-9143-7. Bierlaire, M., 2003. BIOGEME: a free package for the estimation of discrete choice models. In: Swiss Transp. Res. Conf. Börjesson, M., Algers, S., 2011. Properties of internet and telephone data collection methods in a stated choice value of time study context. J. Choice Model. 4, 1–19. Bureau of Labor statistics, 2016. Employment Situation https://www.bls.gov/news.release/archives/empsit_08052016.htm Cabral, L., Hortacsu, A., 2010. The dynamics of seller reputation: evidence from eBay. J. Ind. Econ. 58, 54–78. http://dx.doi.org/10.1111/j.14676451.2010.00405.x. Caulfield, B., 2009. Estimating the environmental benefits of ride-sharing: a case study of Dublin. Transp. Res. Part D Transp. Environ. 14, 527–531. http://dx. doi.org/10.1016/j.trd.2009.07.008. Chen, C., Pan, S., 2016. Using the Crowd of Taxis to Last Mile Delivery in E-Commerce: A Methodological Research. Springer International Publishing, pp. 61– 70. http://dx.doi.org/10.1007/978-3-319-30337-6_6. Chowdhury, M.S., 2016. Managing the freight deliveries in manhattan: opportunities for collaborative city logistics measures. In: International Conference on Transportation and Development 2016. American Society of Civil Engineers, Reston, VA, pp. 162–170. http://dx.doi.org/10.1061/9780784479926. 015. Collins, A.T., 2006. Travel Behaviour in the Context of Parcel Pickups (No. ITLS-WP-15-14). The Australian Key Centre in Transport and Logistics Management The University of Sydney. Dablanc, L., 2016. Digital market places for urban freight: is digital city logistics a disruption to the urban freight routine? Presentation in The VREF Conference on Urban Freight 2016: Plan for the future sharing urban space 17–19 October 2016, Gothenburg. de Jong, G., Kouwenhoven, M., Ruijs, K., van Houwe, P., Borremans, D., 2016. A time-period choice model for road freight transport in Flanders based on stated preference data. Transp. Res. Part E Logist. Transp. Rev. 86, 20–31. http://dx.doi.org/10.1016/j.tre.2015.12.004. Delhomme, P., Gheorghiu, A., 2016. Comparing French carpoolers and non-carpoolers: which factors contribute the most to carpooling? Transp. Res. Part D Transp. Environ. 42, 1–15. http://dx.doi.org/10.1016/j.trd.2015.10.014. dell’Olio, L., Moura, J.L., Ibeas, A., Cordera, R., Holguin-Veras, J., 2016. Receivers’ Willingness-to-Adopt Novel Urban Goods Distribution Practices. Transp Res. Part A Policy Pract. http://dx.doi.org/10.1016/j.tra.2016.10.026. DeNavas-Walt, C., Proctor, B.D., 2015. Income and Poverty in the United States: 2014. Diana, M., 2012. Internet mobility survey sampling biases in measuring frequency of use of transport modes. Transp. Res. Rec. J Transp. Res. Board 66–73. http://dx.doi.org/10.3141/2285-08. Dias, F.F.F., Lavieri, P., Garikapati, V.M., Astroza, S., Pendyala, R.M., Bhat, C.R., 2016. A Behavioral Choice Model of the Use of Car-Sharing and Ride-Sourcing Services. Technical paper, Department of Civil, Architectural and Environmental Engineering, The University of Texas at Austin. Efthymiou, D., Antoniou, C., Waddell, P., 2013. Factors affecting the adoption of vehicle sharing systems by young drivers. Transp. Policy 29, 64–73. http:// dx.doi.org/10.1016/j.tranpol.2013.04.009. Erickson, L.B., Trauth, E.M., 2013. Getting work done: evaluating the potential of crowdsourcing as a model for business process outsourcing service delivery. In: Annual Computers and People Research Conference. ACM Press, New York, USA, pp. 135–140. http://dx.doi.org/10.1145/2487294.2487323. Esper, T.L., Jensen, T.D., Turnipseed, F.L., Burton, S., 2003. The last mile: an examination of effects of online retail delivery strategies on consumers. J. Bus. Logist. 24, 177–203. http://dx.doi.org/10.1002/j.2158-1592.2003.tb00051.x. Furuhata, M., Dessouky, M., Ordóñez, F., Brunet, M.-E., Wang, X., Koenig, S., 2013. Ridesharing: the state-of-the-art and future directions. Transp. Res. Part B 57, 28–46. German Industry and Commerce (HK), 2016. Crowd-shipping: Is Crowd-sourced the Secret Recipe for Delivery in the Future?. Gibson, B.J., Clifford, D.C., Ishfaq, R., 2015. The State of the Retail Supply Chain – Essential Findings of the Fifth Annual Report. Hall, J.V, Krueger, A.B., 2015. An Analysis of the Labor Market for Uber’s Driver-Partners in the United States. Hamari, J., Sjöklint, M., Ukkonen, A., 2016. The sharing economy: why people participate in collaborative consumption. J. Assoc. Inf. Sci. Technol. 67, 2047– 2059. http://dx.doi.org/10.1002/asi.23552. Hensher, D.A., 2010. Hypothetical bias, choice experiments and willingness to pay. Transp. Res. Part B Methodol. 44, 735–752. http://dx.doi.org/10.1016/j. trb.2009.12.012. Hensher, D.A., Greene, W.H., 2003. The mixed logit model: the state of practice. Transportation (Amst) 30, 133–176. http://dx.doi.org/10.1023/ A:1022558715350. Hensher, D.A., Ho, C.Q., 2016. Experience conditioning in commuter modal choice modelling – Does it make a difference? Transp. Res. Part E Logist. Transp. Rev. 95, 164–176. http://dx.doi.org/10.1016/j.tre.2016.09.010. Hess, S., Bierlaire, M., Polak, J.W., 2005. Estimation of value of travel-time savings using mixed logit models. Transp. Res. Part A Policy Pract. 39, 221–236. http://dx.doi.org/10.1016/j.tra.2004.09.007. Hess, S., Rose, J.M., Hensher, D.A., 2008. Asymmetric preference formation in willingness to pay estimates in discrete choice models. Transp. Res. Part E Logist. Transp. Rev. 44, 847–863. http://dx.doi.org/10.1016/j.tre.2007.06.002. Hess, S., Train, K.E., Polak, J.W., Polak, J.W., 2006. On the use of a Modified Latin Hypercube Sampling (MLHS) method in the estimation of a Mixed Logit Model for vehicle choice. Transp. Res. Part B 40, 147–163. http://dx.doi.org/10.1016/j.trb.2004.10.005. Holguín-Veras, J., Silas, M., Polimeni, J., Cruz, B., 2007. An investigation on the effectiveness of joint receiver-carrier policies to increase truck traffic in the off-peak hours: Part I: the behavior of receivers. Networks Spat. Econ. 7, 277. http://dx.doi.org/10.1007/s11067-006-9002-7. Hsiao, M.-H., 2009. Shopping mode choice: physical store shopping versus e-shopping. Transp. Res. Part E Logist. Transp. Rev. 45, 86–95.

38

A. Punel, A. Stathopoulos / Transportation Research Part E 105 (2017) 18–38

Kafle, N., Zou, B., Lin, J., 2017. Design and modeling of a crowdsource-enabled system for urban parcel relay and delivery. Transp. Res. Part B Methodol. 99, 62–82. http://dx.doi.org/10.1016/j.trb.2016.12.022. Kwak, N., Radler, B., 2002. A comparison between mail and web surveys: response pattern, respondent profile, and data quality. J. Off. Stat. 18, 257–273. Lindhjem, H., Navrud, S., 2011. Are internet surveys an alternative to face-to-face interviews in contingent valuation? Ecol. Econ. 70, 1628–1637. http://dx. doi.org/10.1016/j.ecolecon.2011.04.002. Louviere, J.J., Hensher, D.A., Swait, J.D., 2000. Stated Choice Methods: Analysis and Application. Cambridge Univ Press. Macharis, C., Kin, B., 2017. The 4 A’s of sustainable city distribution: innovative solutions and challenges ahead. Int. J. Sustain. Transp. 11, 59–71. http://dx. doi.org/10.1080/15568318.2016.1196404. Marcucci, E., Gatta, V., 2017. Investigating the potential for off-hour deliveries in the city of Rome: retailers’ perceptions and stated reactions. Transp Res. Part A Policy Pract. http://dx.doi.org/10.1016/j.tra.2017.02.001. Marta-Pedroso, C., Freitas, H., Domingos, T., 2007. Testing for the survey mode effect on contingent valuation data quality: a case study of web based versus in-person interviews. Ecol. Econ. 62, 388–398. http://dx.doi.org/10.1016/j.ecolecon.2007.02.005. Masiero, L., Hensher, D.A., 2011. Shift of reference point and implications on behavioral reaction to gains and losses. Transportation (Amst) 38, 249–271. http://dx.doi.org/10.1007/s11116-010-9302-0. Masiero, L., Rose, J.M., 2013. The role of the reference alternative in the specification of asymmetric discrete choice models. Transp. Res. Part E Logist. Transp. Rev. 53, 83–92. http://dx.doi.org/10.1016/j.tre.2013.03.001. McKinnon, A., Browne, M., Whiteing, A., 2015. Green Logistics: Improving the Environmental Sustainability of Logistics. Kogan Page Publishers. Miller, J., Nie, M., Stathopoulos, A., 2017. Crowdsourced urban package delivery: modeling traveler willingness to work as crowdshippers. Accept. Publ. Transp. Res. Rec. Morning Consult, 2015. National Tracking Poll [WWW Document]. (accessed 11.1.16). Office of Inspector General, 2014. Using the ‘‘Crowd” to Deliver Packages. United States Postal Service. Olsen, S.B., 2009. Choosing between internet and mail survey modes for choice experiment surveys considering non-market goods. Environ. Resour. Econ. 44, 591–610. http://dx.doi.org/10.1007/s10640-009-9303-7. Paloheimo, H., Lettenmeier, M., Waris, H., 2016. Transport reduction by crowdsourced deliveries – a library case in Finland. J. Clean. Prod. 132, 240–251. http://dx.doi.org/10.1016/j.jclepro.2015.04.103. Panda, R., Verma, S., Mehta, B., 2015. Emergence and acceptance of sharing economy in India. Int. J. Online Mark. 5, 1–17. http://dx.doi.org/10.4018/ IJOM.2015070101. Piscicelli, L., Cooper, T., Fisher, T., 2015. The role of values in collaborative consumption: insights from a product-service system for lending and borrowing in the UK. J. Clean. Prod. 97, 21–29. http://dx.doi.org/10.1016/j.jclepro.2014.07.032. Rayle, L., Dai, D., Chan, N., Cervero, R., Shaheen, S., 2016. Just a better taxi? A survey-based comparison of taxis, transit, and ridesourcing services in San Francisco. Transp. Policy 45, 168–178. http://dx.doi.org/10.1016/j.tranpol.2015.10.004. Rougès, J.-F., Montreuil, B., 2014. Crowdsourcing delivery: new interconnected business models to reinvent delivery. In: 1st International Physical Internet Conference, pp. 1–19. Salkeld, G., Ryan, M., Short, L., 2000. The veil of experience: do consumers prefer what they know best? Health Econ. 9, 267–270. http://dx.doi.org/10.1002/ (SICI)1099-1050(200004)9:3 267::AID-HEC511 3.0.CO;2-H. Savelsbergh, M., Van Woensel, T., 2016. 50th anniversary invited article—city logistics: challenges and opportunities. Transp. Sci. 50, 579–590. http://dx.doi. org/10.1287/trsc.2016.0675. Scarpa, R., Ferrini, S., Willis, K., 2005. Performance of error component models for status-quo effects in choice experiments. In: Applications of Simulation Methods in Environmental and Resource Economics. Springer-Verlag, Berlin/Heidelberg, pp. 247–273. http://dx.doi.org/10.1007/1-4020-3684-1_13. Shaheen, S.A., 2016. Mobility and the sharing economy. Transp. Policy 51, 141–142. http://dx.doi.org/10.1016/j.tranpol.2016.01.008. Shaheen, S.A., Chan, N.D., Gaynor, T., 2016. Casual carpooling in the San Francisco Bay Area: understanding user characteristics, behaviors, and motivations. Transp. Policy 51, 165–173. http://dx.doi.org/10.1016/j.tranpol.2016.01.003. Srinivasan, K.K., Mahmassani, H.S., 2003. Analyzing heterogeneity and unobserved structural effects in route-switching behavior under ATIS: a dynamic kernel logit formulation. Transp. Res. Part B Methodol. 37, 793–814. http://dx.doi.org/10.1016/S0191-2615(02)00063-2. Stathopoulos, A., Hess, S., 2012. Revisiting reference point formation, gains-losses asymmetry and non-linear sensitivities with an emphasis on attribute specific treatment. Transp. Res. Part A Policy Pract. 46. http://dx.doi.org/10.1016/j.tra.2012.08.005. Swait, J., Louviere, J., 1993. The role of the scale parameter in the estimation and comparison of multinomial logit models. J. Mark. Res. 30, 305. http://dx.doi. org/10.2307/3172883. Tavasszy, L.A., De Jong, G., 2014. Modelling Freight Transport. Elsevier. The Heights Board, 2016. All-Female Rideshare Service Only Temporary Solution [WWW Document]. (accessed 4.20.16). Train, K., 2009. Discrete Choice Methods with Simulation. Cambridge University Press. US CENSUS, 2012a. Population by Age and Sex: 2012. Current Population Survey, Annual Social and Economic Supplement. US CENSUS, 2012b. Reported Internet Usage for Individuals 3 Years and Older, by Selected Characteristics: 2012. Vuylsteke, D., 2016. How Much You Should Ask for to Transport an Item. PiggyBlog [WWW Document]. (accessed 7.31.16). Wang, Y., Zhang, D., Liu, Q., Shen, F., Lee, L.H., 2016. Towards enhancing the last-mile delivery: an effective crowd-tasking model with scalable solutions. Transp. Res. Part E Logist. Transp. Rev. 93, 279–293. http://dx.doi.org/10.1016/j.tre.2016.06.002. Windle, J., Rolfe, J., 2011. Comparing responses from internet and paper-based collection methods in more complex stated preference environmental valuation surveys. Econ. Anal. Policy 41, 83–97. http://dx.doi.org/10.1016/S0313-5926(11)50006-2.