Choice model based analysis of consumer preference for drone delivery service

Choice model based analysis of consumer preference for drone delivery service

Journal of Air Transport Management 84 (2020) 101785 Contents lists available at ScienceDirect Journal of Air Transport Management journal homepage:...

411KB Sizes 0 Downloads 26 Views

Journal of Air Transport Management 84 (2020) 101785

Contents lists available at ScienceDirect

Journal of Air Transport Management journal homepage: http://www.elsevier.com/locate/jairtraman

Choice model based analysis of consumer preference for drone delivery service Sang Hyun Kim 1 Korea Aerospace University, 76 Hanggongdaehak-ro, Deogyang-gu, Goyang-si, Gyeonggi-do, 10540, Republic of Korea

A R T I C L E I N F O

A B S T R A C T

Keywords: Drone Consumer preference Parcel delivery Discrete choice model Stated preference

It is anticipated that drones will soon be utilized for a range of applications, including delivery service. However, there has been a lack of research on consumer preference between drone delivery service and traditional delivery service. This is the first study to analyze the consumer preference for drone delivery based on a discrete choice model between the drone delivery service and traditional delivery services by truck or motorcycle. The discrete choice model is estimated using a stated preference survey, and potential consumers’ preference is analyzed for representative commodities with different price. The results show that the price and type of commodities in­ fluence consumer preference, which also depends on socio-demographic characteristics such as gender, age, and household income. Specifically, it was consistently observed in all cases that the younger the age, the higher the preference for drone delivery service. This study contributes to predicting the consumer preference for drone delivery service before real service offerings and to supporting the establishment of business strategies for companies who prepare for the new market of drone-based delivery.

1. Introduction Today, drones are being utilized not only for agriculture and photography but also for security, surveillance, and surveying (Teal Group Corporation, 2016). Of the various drone applications, delivery services for commercial goods or emergency medical supplies are currently being tested and implemented such as Amazon (Amazon and Amazon prime air, 2018), Alibaba (The Economic Times, 2017), and the DPD group in France (La Poste Group Press Department, 2016). They are expected to become a prominent drone application. Much attention has been given to drone delivery service, but there is still little officially launched service due to regulatory and technical is­ sues. Most countries have legal restrictions preventing drones from flying beyond visual line of sight or from flying automatically without a pilot, which makes it difficult for drone delivery service to begin in earnest, but their use will increase over the long term. As battery, pro­ pulsion, and communication technologies evolve, long-distance and long-term flight becomes possible and the marketability of drone de­ livery service is expected to improve as a result. As such, this study assumes that drones will be widely used in the near future, especially for the delivery of consumer goods. For example, Amazon Prime Air will deliver small packages to customers in 30 min or

1

less (Amazon and Amazon prime air, 2018). However, there has been a lack of research on consumer preference for the drone delivery service. Unlike traditional freight transport, of which the mode is determined by shippers such as logistics manager (Tavasszy and De Jong, 2013), re­ ceivers (i.e., consumers) can choose the mode of delivery for consumer goods. Online shopping malls in Korea, for instance, provide delivery options (e.g., by truck or motorcycle) to consumers, and consumers make a decision based on the speed and price of delivery. Indeed, some consumers are willing to pay more for the same-day shipping, then they choose the motorcycle delivery. As such, consumers will be aware of and choose the mode of delivery such as drone delivery when online retailers launch drone delivery service. Recall that some online retailers such as Amazon have already been advertising their upcoming new delivery option (i.e., drone) to potential customers. Hence, the choice behavior analysis in this new context is necessary. The market share of drone delivery service can be estimated with the knowledge on preference, so it is essential for the decision of whether or not to enter the new market enabled by drones. Therefore, this study contributes to analyze potential consumer preference for drone delivery service based on the analysis of mode choice behavior among traditional and new delivery services. Although the underlying choice model is well known and widely implemented, this is the first study to propose such an

E-mail address: [email protected]. Assistant Professor, School of Air Transport, Transportation, and Logistics.

https://doi.org/10.1016/j.jairtraman.2020.101785 Received 2 September 2019; Received in revised form 24 January 2020; Accepted 11 February 2020 Available online 19 February 2020 0969-6997/© 2020 Elsevier Ltd. All rights reserved.

S.H. Kim

Journal of Air Transport Management 84 (2020) 101785

analysis, and the results of this study will raise awareness of a new era in end-user delivery service. The proposed model considers not only traditional parcel delivery by trucks but also the delivery service by motorcycle, which is not widely used outside of Korea. The findings of this study will provide a reasonable estimation of the future market share of drone delivery services, so it will be useful for parcel delivery companies preparing to deploy drones in a new market.

and fleet have impacts only on some groups of passengers (Hess and Polak, 2006). Jou et al. studied passengers’ choice of access mode to Taiwan Taoyuan International Airport for improved airport access (e.g., high-speed rail) and found that both out-vehicle time and in-vehicle time need to be reduced (Jou et al., 2011). Unlike other studies analyzing customer preferences, Sha et al. modelled airlines’ decisions on city-pair route selection using a DCM (Sha et al., 2016). Their model can predict future evolution of the air transportation network based on the estimation of airlines’ preference. The DCM technique is also applied to freight transport mode choice, but the decision maker in these cases is usually freight shippers (Tavasszy and De Jong, 2013; McGinnis, 1979; Jeffs and Hills, 1990). Shinghal and Fowkes showed that the frequency of service is the most important factor for Indian firms to determine their freight transport mode, but service reliability is less important (Shinghal and Fowkes, 2002). On the other hand, Brooks et al. analyzed the mode choice behavior of Australian freight shippers and showed that the shippers place importance on the reliability of transport mode such as on-time delivery and short delays, as well as on the transit time and cost (Brooks et al., 2012). Recently, Chen et al. investigated consumers’ intention of using express cargo delivery service by motorcycle in urban areas of Taiwan, in which, similar to Korea, motorcycles are used for the delivery of business documents, small parcels, and meals (Chen et al., 2019). They showed that there is no significant difference in intention between gender but previous experience with the delivery service affects attitude toward using the service. To the best of our knowledge, there is no study on the consumer preference for drone delivery service, which is available in limited areas to date. However, it is expected that the drone delivery service will be a promising application of drones, especially for consumer goods. Most customers have not cared about what mode is used to deliver their or­ ders or products, but things can change if drone delivery service is widely used; consumers, not shippers, will make a decision on the de­ livery service for their business documents and small parcels. Thus, the analysis of (potential) consumer preference between drone delivery service and traditional delivery service is necessary, because previous studies do not deal with this new perspective of mode choice problems. Nevertheless, new technologies such as delivery drones may arise concerns on safety and reliability. In particular, people are unfamiliar with unmanned aircraft (e.g., drone), and they would have different perceptions of the technology. Molesworth and Koo (2016), and Lee et al. (2019) analyzed the choice behavior for unmanned (i.e., remotely-piloted) passenger aircraft and found that end users’ accep­ tance of such a technology is an important factor for their choices. In addition, the acceptance of newly introduced technology is affected by socio-demographic characteristics such as gender and age (Lee et al., 2019).

2. Literature review There have been many studies on the analysis of choice behavior such as transport mode choice. A Discrete Choice Model (DCM) de­ scribes individual behaviors to choose among discrete alternatives (McFadden, 1981; Ben-Akiva et al., 1985). One illustrative application of the DCM is transport mode choice, which is the third step of the Four-step model for travel demand (Hensher and Button, 2007). Transport mode choice is the process of modelling the behavior of passengers choosing a transport mode from origin to destination. Traditional travel demand analysis is based on an aggregate model that is inadequate when it comes to reflecting the variance of individual choices. To overcome this shortcoming, individual behavioral models were introduced in the early 1960s (Ben-Akiva et al., 1985). Among the individual behavioral models, logit models are more commonly used because of their relative ease of calculation and appli­ cation compared to others such as probit models (de Dios Ortuzar and Willumsen, 2011). Types of logit models are binary logit model, multi­ nomial logit model, nested logit model, and hybrid choice model that reflects latent variables (Ben-Akiva et al., 1985; de Dios Ortuzar and Willumsen, 2011; Greene and Hensher, 2003; Chorus and Kroesen, 2014). Binary and multinomial logit models are probabilistic models of an individual’s choice among two (binary) or three or more (multino­ mial) alternatives, and they assume an independent distribution among alternatives. The nested logit model is also a probabilistic choice model among multiple alternatives but uses generalized extreme value for correlated alternatives. The hybrid choice model can make more customized predictions through the introduction of latent variables for unobservable factors such as attitude and personal opinion, in addition to observable attributes of alternatives. Stated Preference (SP) is a technique to investigate individual pref­ erence by constructing hypothetical scenarios through statistical experiment design and providing the scenarios to individuals. It has been applied to marketing and transportation since the early 1970s (Kocur et al., 1981; Fowkes and Wardman, 1988). Davidson used the SP method for the estimation of air route demand (Davidson, 1973). Green and Srinivasan advanced the application of the SP method in trans­ portation by explaining the theory underlying conjoint analysis (Green and Srinivasan, 1978). Initially, the SP method was limited to ranking or rating analysis based on survey data, but Louvier and Hensher extended the method so that individuals could choose from various combinations, which enables the estimation of DCM (Louviere and Hensher, 1983). So, the SP method is widely adopted to the development of logit models. In the field of air transportation, individual behavioral models are applied to the airport selection problem, airline preference, mode choice of airport access, etc. Harvey showed that airport access time and flight frequency affect the choice of departure airport, using a multinomial logit model (Harvey, 1987). Furuichi and Koppelman, using a nested logit model, showed that airport access time and flight hour are important factors for both business passengers and travelers, with flight frequency found to be the most important (Furuichi and Koppelman, 1994). Proussaloglou and Koppelman built a conceptual framework for the analysis of air transport demand in competitive markets, using a DCM with airlines, flight schedules, and fare class (Proussaloglou and Koppelman, 1999). Hess and Polak proposed a nested logit model for the simultaneous choices of departure airport, airlines, and airport access mode, and they showed that flight frequency and airport access time have a significant effect on the choice, while other factors such as fare

3. Delivery-service choice model 3.1. Competitors of drone The purpose of this study is to analyze the choice behavior for de­ livery service, based on the assumption that a new delivery service by drone is introduced in addition to traditional delivery services by truck or motorcycle, and to propose discrete choice models for the drone de­ livery service. In the context of Korea, the drone delivery service is ex­ pected to compete with two ground-based delivery services: traditional parcel delivery by truck and “quick” delivery by motorcycle. The traditional parcel delivery is conducted by truck, so heavy or large items can be delivered. But the delivery takes usually one or two days, because it is managed through a hub-and-spoke system: every package is collected and sent to a central distribution center where it is categorized by destination. On the other hand, the quick delivery is carried out by motorcycle, so the size and weight of items are limited. But the delivery is done within a couple of hours, because the 2

S.H. Kim

Journal of Air Transport Management 84 (2020) 101785

motorcycle is relatively less affected by traffic and it is a point-to-point service. A delivery person picks up the package at the front door and delivers it directly to its destination. Due to its high price, the quick delivery service is usually used by companies for urgent documents. Although these two services aim at clearly different markets, drone delivery service may compete with both of them. Drones are better in delivery time than trucks and even motorcycles, and assumed to be a point-to-point service. Some customers may use the drone delivery service for small items if they want to get the items as soon as possible. In this case, the drone delivery service competes with truck delivery ser­ vice. On the other hand, companies will choose a faster delivery service for urgent documents regardless of cost, as long as the documents will be delivered safely. In this case, the drone delivery service competes with motorcycle delivery service. Note that this study assumes that drones will compete with either trucks (for small consumer goods) or motor­ cycles (for urgent business documents). Currently, online shopping malls in Korea utilize motorcycles for the same-day shipping, but drones will replace motorcycles and compete with trucks. On the other hand, drones will enter the monopoly market of urgent (business) document delivery, as a new competitor of motorcycles. In addition, drone is assumed to be safe and secure enough to be implemented for delivery services. However, (potential) consumers’ concerns on its safety and reliability can be implicitly reflected in the choice model, in the form of an alternative specific constant. Also, the drone delivery service is assumed to cost more than traditional delivery services, because the introduction of new technologies requires addi­ tional manpower, equipment, training, and facilities.

that customers are willing to use the faster but more expensive delivery service (i.e., drone) as the price of commodity increases. In addition, it was hypothesized that the kind of commodity also affects the choice behavior. As a result, effects of price and item on the choice behavior will be analyzed for the items given in Table. 1. Note that urgent doc­ uments are currently delivered by a motorcycle as stated before, and the exchange rate is about 1100 KRW per 1 USD (or 11 KRW per 1 cent), as of March 2019. 3.3. SP survey An SP survey was conducted to collect data on the preferences of potential consumers of drone delivery service. The sample size of the SP survey is closely related to the cost and period of the survey. There is no exact number known for the best sample size, but Bradely and Kores suggest 75–100 samples per segment (Bradley and Kroes, 1992). As the public is unfamiliar with drone technologies and associated applica­ tions, preferences for drone delivery service will depend on technology acceptance, which is assumed in this study to be affected by gender and age (Lee et al., 2019). Therefore, the number of samples for the SP survey is determined to be 400 people: 200 per each gender and 100 per each age group. The respondents are selected evenly considering gender and geographical areas among those who have experienced in online shopping. A summary of the respondents is given in Table 2. Note that income is monthly household income. The survey was conducted online for one month. The survey questionnaire asks respondents to choose the best alter­ native between two alternatives, which are truck versus drone for clothing and powdered formula or motorcycle versus drone for urgent document. Note that this study assumes only two alternatives are available for each item, because online shopping malls are likely to substitute drones for motorcycles. Two attributes, delivery time and cost, are given for each delivery service. Each attribute has two levels as given in Table. 3 and Table. 4. The delivery costs and times of truck and motorcycle delivery service are based on the average costs and times of the actual services. The delivery cost and time of drones will be calcu­ lated in practice based on the distance, speed, etc., but there is no business model for reference. Thus, they are set through discussions among field experts, considering the substitution of drones for motor­ cycles in the market of consumer goods delivery and the competition with motorcycles in the market of urgent document delivery. This is the reason why the delivery cost and time of drones in Table. 3 are different from those in Table. 4. For truck versus drone cases (i.e., clothing and powdered formula), the delivery cost of drones is assumed to be higher than that of trucks due to delivery speed difference. However, the drone delivery can be

3.2. Candidate items for drone delivery Due to the performance limitations of drones such as maximum takeoff weight, however, it is assumed that drone delivery service is avail­ able only for goods weighing less than 5 kg within a city area. In addi­ tion, it is also assumed that the drone delivery service will be implemented first by online retailers such as Amazon and Alibaba. So, based on the general categorization of the online retailers, a list of items suitable for the drone delivery service is made in five categories: clothing and beauty products; baby products; food and commodities; documents and office supplies; PCs and home appliances. Then, a pre­ liminary survey was conducted of 19 researchers with expertise in transportation at the Korea Transport Institute, in order to select the most suitable items for the drone delivery service. The 19 participants have been involved in several aviation and transport research projects, so they are assumed to be well-informed regarding the utility of drones. The selected items are given in Table 1. Note that clothing and powdered formula are consumer goods and consumers decide the delivery mode of goods, but the decision is made by senders for urgent documents. Clothing, powdered formula, and urgent documents were chosen by 47.4%, 47.4%, and 57.9% of respondents, respectively. Clothing and powdered formula are popular items of online shop­ ping: clothing is particularly one of the five most preferred items for online shopping in Korea (Statistics Korea, 2018). As well, they are not too large or heavy, which makes them suitable for drone delivery. Also, those items are usually delivered by trucks. A hypothesis was made that customers’ choice behavior depends on the price of delivered items when drone delivery service competes with truck delivery service, which is available at a low price. In other words, the hypothesis means

Table 2 Number of respondents according to socio-demographic characteristics. Socio-demographic characteristics Age

Education Income, USD

Table 1 List of items selected for drone delivery service. Item

Value (Item Price)

Competitor

Clothing

100,000 or 500,000 KRW (about 100 or 500 USD) 100,000 KRW N/A

Truck

Powdered formula Urgent document

Truck Motorcycle

Total

3

20–29 30–39 40–49 50 and above High school Undergrad Graduate Under 1 K 1–2 K 2–3 K 3–4 K 4–5 K 5–6 K 6–7 K Above 7 K

Number of respondents Male

Female

50 50 50 50 36 144 20 8 17 42 38 31 28 12 24 200

50 50 50 50 50 137 13 6 20 40 42 31 23 16 22 200

Total 100 100 100 100 86 281 33 14 37 82 80 62 51 28 46 400

S.H. Kim

Journal of Air Transport Management 84 (2020) 101785

3.4. Choice model formulation

Table 3 Attributes of truck and drone delivery service. Alternative

Attribute

Level 1 value

Level 2 value

Truck

Time Cost Time Cost

24 h 4000 KRW 1h 8000 KRW

48 h 6000 KRW 2h 12,000 KRW

Drone

The SP survey gives two alternatives (truck vs. drone or motorcycle vs. drone) because the main target markets of truck and motorcycle are different. Therefore, the DCM in this paper is a binary choice model. The survey questionnaire includes 32 choice questions (8 situations by 4 items) and additional questions on the socio-demographic characteris­ tics of respondents such as gender, age, education, and monthly household income. The characteristics are reflected in the DCM either as choice invariant variables or latent classes. A simple formulation of a binary logit model (BLM) (McFadden, 1981) with two attributes (xat : time in minutes and cost in KRW) and respondent characteristics (xch : gender, age, education, and monthly household income) is given in (1)–(4). Here cat denotes the coefficient vector for alternative attributes consisting of ctime and ccost , and cch de­ notes the coefficient vector for socio-demographic characteristics con­ sisting of cgen , cage , cedu , and cinc . The alternative specific constant (ASC) is applied only for traditional delivery service (by truck or motorcycle). Note that Gen is a dummy variable that is equal to 0 for male and 1 for female, and Edu is categorized into three groups: High school, Under­ graduate, and Graduate. Inc is also categorized into eight groups from below one million KRW to above seven million KRW per month. Time, Cost, and Age are continuous variables. U is the utility value of an alternative, and P is the probability of choice.

Table 4 Attributes of motorcycle and drone delivery service. Alternative

Attribute

Level 1 value

Level 2 value

Motorcycle

Time Cost Time Cost

1h 6000 KRW 20 min 7200 KRW

2h 16,000 KRW 40 min 19,200 KRW

Drone

conducted at a relatively low speed (e.g., delivery time of one or 2 h), which is still much faster than truck delivery. So, the delivery time of drones in these cases is assumed to be same as the motorcycle delivery time, although it can be shorter. It makes sense, considering the sub­ stitution of drones for motorcycles in the market of consumer goods delivery. On the other hand, the drone delivery service may use a different pricing strategy when competing with the motorcycle delivery service for urgent documents, because both are fast enough. So, the delivery cost of drones is similar to that of motorcycles. However, the drone delivery must be faster than the motorcycle delivery because the main customers in this market are companies who can pay more for faster delivery service. So, the delivery time of drones is set to one-third that of the motorcycle delivery. Each respondent is given 8 situations of delivery time and cost per item (e.g., 100 K KRW clothing, 500 K KRW clothing, 100 K KRW powdered formula, and urgent document). Hence, total 4 sets of 8 sit­ uations are presented in the survey questionnaire and the respondents are asked to make choices for 32 situations. The situations are generated using the Orthogonal Experimental Design Table in Kocur et al. (1981), and the corresponding Experimental Plan Code Number is 3a. Because there is no prior knowledge of the preference in drone delivery service, the orthogonal design enables to explore the search space properly. The designed situations are presented in Table 5. Note that Table 5 is applicable for both truck-drone and motorcycle-drone cases, because the number and level of attributes in both cases are identical. Also, it is notable that drone delivery is always faster than its competitor (i.e., truck or motorcycle), as given in Table 3 and Table 4. As a result, the drone delivery is superior in both the delivery time and cost to the motorcycle delivery in Situations 4 and 7, but these cases are intended to capture the acceptance of drone technologies, although it is assumed and noticed to the respondents that the drone delivery is safe and secure, avoiding potential selection bias due to concerns on a new technology.

ch ch Utrad ¼ cat ⋅xat trad þ c ⋅x þ ASC time age

Situation 1 Situation 2 Situation 3 Situation 4 Situation 5 Situation 6 Situation 7 Situation 8

Drone

Time

Cost

Time

Cost

1 1 1 1 2 2 2 2

1 1 2 2 1 1 2 2

1 2 1 2 1 2 1 2

1 2 2 1 2 1 1 2

(1)

​ Age þ cedu ​ Edu þ cinc ​ Inc þ ASC

þc

Udrone ¼ cat ⋅xat drone time

¼c

Timedrone þ ccost Costdrone

Ptrad ¼

expðUtrad Þ expðUtrad Þ þ expðUdrone Þ

Pdrone ¼

expðUdrone Þ expðUtrad Þ þ expðUdrone Þ

(2) (3) (4)

The characteristics of respondents can be used to form latent classes. Then, only the attributes of alternatives are explicitly included in the probability of selecting a delivery service (P); the characteristics are included in the membership function (M). The corresponding formula­ tion of a latent class logit model (LCLM) (Greene and Hensher, 2003) is given in (5)–(6). Here θch denotes the coefficient vector of membership function for socio-demographic characteristics and constl denotes a constant for latent class l. ASC is applied only for traditional delivery service using a dummy variable (dtrad ¼ 1 for traditional, 0 otherwise). A benefit of LCLM is that it can model utility coefficients differently for latent classes, as shown in (5). Note that the number of latent classes (L) is a design parameter determined by researchers but as Molesworth and Koo showed (Molesworth and Koo, 2016), the attitude towards un­ manned aircraft technology can be classified into two classes: trust in such a technology or not. Moreover, three or more classes failed to converge due to the singularity of variance matrix. Hence, the number of latent classes is set to two for this study. Both BLM and LCLM are esti­ mated with NLogit by Econometric Software. � L at X exp cat l ⋅xs þ ASCl dtrad � Ps ¼ Ml P (5) at at s exp cl ⋅xs þ ASCl dtrad l¼1

Table 5 Level of attributes of 8 situations. Traditional (Truck or Motorcycle)

​ Timetrad þ ccost ​ Costtrad þ cgen ​ Gen

¼c

where s is trad or drone � ch exp θch l ⋅x þ constl Ml ¼ P ch ch l expðθl ⋅x þ constl Þ

4

(6)

S.H. Kim

Journal of Air Transport Management 84 (2020) 101785

4. Results

Table 7 Estimated latent class logit model for low-price clothing.

4.1. DCM for low-price clothing: truck delivery as the competitor of drone delivery

Coefficients of utility function

Firstly, the results of BLM and LCLM for low-price clothing worth 100 K KRW are given in Table 6 and Table 7. The LCLM gives better statistics (e.g., log likelihood and adjusted R2 ) than the BLM, but the coefficients of membership function are not statistically significant, which means the estimation of membership function based on sociodemographic characteristics is not reliable. As indicated by the sign of ctime and ccost of both BLM and LCLM, the service attractiveness (i.e., utility value) decreases with delivery time and cost. The value of time (VOT), which is calculated by dividing ctime by ccost , of the BLM for low-price clothing is 1.00 KRW/min (about 0.0912 cent/min), whereas the VOTs of Classes 1 and 2 of the LCLM are 0.142 KRW/min (about 0.0129 cent/min) and 2.91 KRW/min (about 0.265 cent/min), respectively. Thus, Class 2 is willing to pay about 20 times more than Class 1, in order to save delivery time. Although Class 2 cares more of delivery time and cost (i.e., higher VOT) and the drone delivery service is much faster than the truck delivery service, Class 1 is more enthusiastic about the drone delivery regardless of time and cost as cost can be seen from the relatively low values of ctime 1 , c1 , and ASC1 . It means that Class 2 would think the drone delivery service is fast but too expensive. Based on the posterior estimate of the latent class probabil­ ities, Classes 1 and 2 account for 76.5% and 23.5% of the respondents, respectively. Coefficients cgen , cage , and cedu of the BLM (Table 6) indicate that fe­ male, old, or highly educated people prefer the truck delivery service. However, the results of cgen and cedu are not statistically significant. Meanwhile, high-income earners prefer the drone delivery service, which is fast but expensive. As a result, age and household income determine consumers preference for the drone delivery service of lowprice clothing.

Std.Err.

t-ratio

6.16E-05

6.14

8.21E-10

ccost

0.000377

2.17E-05

17.4

2.89E-15

c

cgen

0.141

0.0886

1.59

0.111

cage

0.00731

0.00355

2.06

0.0394

cedu

0.102

0.0861

1.19

0.234

cinc

0.0496

0.0234

2.12

0.0341

ASC

0.110

0.287

0.384

0.701

log likelihood ¼

ctime 1

9.91E-05

ccost 1

0.000696

ASC1

Std.Err.

0.603

ctime 2

0.00123

ccost 2

0.000422

ASC2

0.163

t-ratio

Two-tailed Pvalue

8.67E05 4.24E05 0.229

1.14

0.253

16.4

2.89E-15

2.64

0.00839

9.75E05 3.11E05 0.212

12.6

2.89E-15

13.6

2.89E-15

0.770

0.441

Latent class

Coefficient

Value

Std.Err.

t-ratio

Two-tailed Pvalue

Class 1

θgen 1

0.158

0.264

0.599

0.549

θ1

0.0163

0.0109

1.50

0.134

θinc 1

0.109

0.0725

1.50

0.133

θgen 2

0.394

0.710

0.555

0.579

0 (fixed parameter)

N/A

N/A

N/A

0.235

θedu 1

Class 2

const1

age

0.256

0.920

0.358

θ2

θedu 2 θinc 2

const2

Model statistics: log likelihood ¼ 0.40338.

1318.796, McFadden’s adjusted R2 ¼

Table 8 Estimated binary logit model for high-price clothing. Coefficient ctime ccost

Value

Std.Err.

0.000163

5.52E-05

0.000305

1.87E-05

t-ratio

Two-tailed P-value

2.96

0.0031

16.3

2.89E-15

cgen

0.0745

0.0798

0.934

0.350

cage

0.00294

0.00319

0.921

0.357

cedu

0.0882

0.0774

1.14

0.254

cinc

0.0698

0.0212

3.30

0.000981

ASC

0.0528

0.259

0.204

0.838

log likelihood ¼

1864.005, McFadden’s adjusted R2 ¼ 0.15779.

4.3. Low-price vs. high-price clothing: impact of commodity price on consumer preference The effect of commodity price is analyzed for clothing at different prices. Due to the unrealistic estimation of LCLM for high-price clothing, this analysis is based on the BLM given in Tables 6 and 8. The VOTs are 1.00 KRW/min (about 0.0912 cent/min) for low-price clothing and 0.534 KRW/min (about 0.0486 cent/min) for high-price clothing. In other words, customers are willing to pay more for fast delivery of lowprice clothing, as opposed to the hypothesis previously made. The re­ spondents seem to have no confidence yet in the reliability of drone delivery service, so they may not want to use the service for expensive items. Consequently, this analysis shows that consumer preference for the drone delivery service depends on the price of commodity, and it is inferred from the analysis that consumers still do not trust enough to use the new technology for expensive items.

Two-tailed P-value

0.000378

Class 1

age

Table 6 Estimated binary logit model for low-price clothing. Value

Value

Coefficients for class membership

The same analysis was conducted for high-price clothing worth 500 K KRW. The result of BLM is given in Table 8, but that of LCLM is omitted because the sign of ctime of a class is estimated to be positive, which is not realistic. It is reasonable to assume that customers prefer a shorter de­ livery time, but a positive ctime means the opposite. The VOT of the BLM for high-price clothing is 0.534 KRW/min (about 0.0486 cent/min). Among cch (i.e., cgen , cage , cedu , and cinc Þ, only cinc is statistically signifi­ cant, and its sign is negative. Hence, the utility value of truck delivery service decreases with household income. In addition, the inverse pro­ portion between truck’s utility and household income (i.e., negative cinc ) is consistent in low-price and high-price clothing.

time

Coefficient

Class 2

4.2. DCM for high-price clothing: truck delivery as the competitor of drone delivery

Coefficient

Latent class

1576.612, McFadden’s adjusted R ¼ 0.28764. 2

5

S.H. Kim

Journal of Air Transport Management 84 (2020) 101785

4.4. DCM for powdered formula: truck delivery as the competitor of drone delivery

Table 10 Estimated latent class logit model for powdered formula. Coefficient

Next, the results of BLM and LCLM for powdered formula worth 100 K KRW are given in Table 9 and Table 10. Note that the coefficients of gen age inc Class 2 membership function (i.e., θ2 ; θ2 ; θedu 2 ; θ2 , and const2 ) are omitted, because they are fixed at zero as given in Table 7. Similar to low-price clothing, the LCLM gives better statistics than the BLM, but the coefficients of membership function are not statistically significant, except for const1 . Thus, it is clear that there are different attitudes to­ ward drone delivery service (i.e., trust vs. distrust), but they are deter­ mined by factors other than socio-demographic characteristics. The VOTs are 0.678 KRW/min (about 0.0616 cent/min) for BLM, and 0.0592 KRW/min (about 0.00538 cent/min) and 4.82 KRW/min (about 0.438 cent/min) for Classes 1 and 2 of LCLM, respectively. Similar to the Class 2 for low-price clothing, Class 2 cares much more of delivery time and cost than Class 1 does. However, the positive sign of ASC2 indicates that Class 2 finds the truck delivery service to be satis­ factory for powdered formula and the drone delivery service to be too expensive. On the other hands, Class 1 is so enthusiastic about the drone delivery, similar to the Class 1 for low-price clothing. Classes 1 and 2 account for 83.2% and 16.8% of the respondents, respectively. Coefficient cinc of the BLM is still statistically significant for the powdered formula, and it is clear that high-income earners prefer the drone delivery service. Moreover, it is noticeable that cgen is also sta­ tistically highly significant and has the largest absolute value among the coefficients of the BLM. So, it is clear that gender is a dominant factor in the choice of delivery service for powdered formula and female re­ spondents prefer the truck delivery service to the drone delivery service. One possible explanation of this gender factor is that female respondents (i.e., the main customers of powdered formula) like to purchase powdered formula before it runs out, thus it rarely happens that powdered formula is urgently needed despite the high cost of the drone delivery service.

Value

Std.Err.

t-ratio

Two-tailed P-value

ctime 1 ccost 1

4.36E-05

8.46E-05

0.515

0.607

0.000737

4.34E-05

17.0

2.89E-15

0.892

0.226

3.95

7.85E-05

ctime 2

0.00119

1.29E-04

9.26

2.89E-15

0.000247

3.83E-05

6.44

1.19E-10

ASC1 ccost 2

ASC2

0.0850

0.280

0.304

0.761

0.0226

0.294

0.0770

0.939

age

0.00175

0.0117

0.150

0.881

θedu 1

0.186

0.282

0.660

0.509

0.000535

0.0776

0.00690

0.994

θgen 1 θ1

θinc 1

const1

log likelihood ¼

1.87

0.806

2.33

0.0200

1244.697, McFadden’s adjusted ρ2 ¼ 0.43690.

Table 11 Comparison of VOTs for clothing and powdered formula, both worth 100 K KRW. Latent Class

Clothing

Powdered formula

Class 1 Class 2

0.142 KRW/min 2.91 KRW/min

0.0592 KRW/min 4.82 KRW/min

e., negative cinc ) is consistent. Consequently, this analysis shows that consumer preference for the drone delivery service depends on the type of commodity, and the influence of socio-demographic characteristics differs from each other. 4.6. DCM for urgent document: motorcycle delivery as the competitor of drone delivery

Next, the choice behaviors for different items with the same price are compared. Overall, the VOT of powdered formula (of the BLM) is about 32% lower than that of (low-price) clothing. The comparison of VOTs by latent classes is given in Table 11. It is interesting that the VOT of powdered formula is higher than that of clothing for Class 2, whereas it is the opposite for Class 1. In other words, the difference in preference between two classes is more clear for powdered formula. Some people are more enthusiastic about the drone delivery for powdered formula than for clothing, whereas others are more sensitive to delivery time and cost for powdered formula. In addition, age is a statistically significant factor for clothing (i.e., cage of BLM), but gender plays the role for powdered formula (i.e., statistically significant cgen of BLM). Similar to the comparison between low-price and high-price clothing, however, the inverse proportion between truck’s utility and household income (i.

Previous sections analyzed DCMs for truck delivery service and drone delivery service. In this section, motorcycle delivery service is compared to the drone delivery service for urgent document. The result of BLM is given in Table 12, but that of LCLM is omitted because the sign of ctime of a class is estimated to be positive, which is not realistic. The VOT for urgent documents is 50 KRW/min (about 4.55 cent/min), which is much more higher than those for clothing and powdered formula. Considering the urgency of the delivery item and the delivery cost of the competitor (i.e., motorcycle delivery), this high value of time is reasonable. Coefficients cgen , cage , and cinc are statistically significant, and it is notable that the sign of cgen for urgent document is opposite to those for clothing and powdered formula. It means that the drone delivery service is preferred for urgent document by female respondents rather than male respondents. In other words, the male respondents are more con­ servative in accepting the new delivery service especially for important documents. In addition, the drone delivery service is preferred by young or high-income people, as indicated by positive cage and negative cinc .

Table 9 Estimated binary logit model for powdered formula.

Table 12 Estimated binary logit model for urgent document.

4.5. Clothing vs. powdered formula: impact of commodity type on consumer preference

Coefficient

Value

Std.Err.

t-ratio

Two-tailed P-value

Coefficient

time

0.000242

6.16655e-005

3.92

8.7599e-005

c

ccost

0.000357

2.16955e-005

16.5

2.88658e-015

c

cgen cage cedu cinc ASC

log likelihood ¼

0.303 0.00285 0.0565 0.0631 0.238

0.0893 0.00353

3.39 0.809

0.0860

0.657

0.0236

2.68

0.288

0.828

Value

Std.Err.

t-ratio

Two-tailed P-value

0.00600

0.00133

4.52

6.07881e-006

ccost

0.000120

5.64966e-006

21.2

2.88658e-015

0.000707

cgen

0.309

0.0795

3.88

0.000104

0.418

cage

0.00776

0.00315

0.511

cedu

0.0814

0.0768

1.06

0.289

0.00740

cinc

0.0525

0.0212

2.48

0.0132

0.408

ASC

0.0833

0.231

0.361

0.718

1564.185, McFadden’s adjusted R ¼ 0.29325.

time

log likelihood ¼

2

6

2.46

0.0139

1879.574, McFadden’s adjusted R ¼ 0.15075. 2

S.H. Kim

Journal of Air Transport Management 84 (2020) 101785

Note that the socio-demographic coefficients are applied only to the utility value of the motorcycle delivery service.

delivery. CRediT authorship contribution statement

4.7. Example analysis of consumer preferences

Sang Hyun Kim: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administra­ tion, Resources, Software, Supervision, Validation, Visualization, Writing - original draft, Writing - review & editing.

This section presents the estimated choice probabilities of an arbi­ trary group of potential customers. The socio-demographic character­ istics of the group are as follows. � � � �

Acknowledgments

Half males and half females. Age between 20 and 39. All holding a college degree (i.e., undergraduate). Monthly income between two million KRW (about 1820 USD) and five million KRW (about 4550 USD).

This work was supported by the Korea Transport Institute, and this paper is based on the result of Drone Traffic Safety Assessment Metrics Development (2018). References

The delivery time and cost of truck, drone, and motorcycle are as follows.

Amazon, Amazon Prime Air, 2018. www.amazon.com/Amazon-Prime-Air/b?ie¼UTF8 &node¼8037720011. (Accessed 6 November 2018). Ben-Akiva, M.E., Lerman, S.R., Lerman, S.R., 1985. Discrete Choice Analysis: Theory and Application to Travel Demand, 9. MIT press. Bradley, M., Kroes, E., 1992. Forecasting issues in stated preference survey research. In: Proceedings of the 3rd International Conference on Survey Methods in Transportation. Brooks, M.R., Puckett, S.M., Hensher, D.A., Sammons, A., 2012. Understanding mode choice decisions: a study of australian freight shippers. Marit. Econ. Logist. 14 (3), 274–299. https://doi.org/10.1057/mel.2012.8. Chen, H.-K., Chou, H.-W., Hung, S.-C., 2019. Interrelationships between behaviour intention and its influential factors for consumers of motorcycle express cargo delivery service. Transportmetrica: Transport. Sci. 15 (2), 526–555. https://doi.org/ 10.1080/23249935.2018.1509401. Chorus, C.G., Kroesen, M., 2014. On the (im-) possibility of deriving transport policy implications from hybrid choice models. Transport Pol. 36, 217–222. https://doi. org/10.1016/j.tranpol.2014.09.001. Davidson, J., 1973. Forecasting traffic on stol. J. Oper. Res. Soc. 24 (4), 561–569. https://doi.org/10.1057/jors.1973.105. de Dios Ortuzar, J., Willumsen, L.G., 2011. Modelling Transport. John Wiley & Sons. Fowkes, T., Wardman, M., 1988. The design of stated preference travel choice experiments: with special reference to interpersonal taste variations. J. Transport Econ. Pol. 27–44. Furuichi, M., Koppelman, F.S., 1994. An analysis of air travelers’ departure airport and destination choice behavior. Transport. Res. Pol. Pract. 28 (3), 187–195. https://doi. org/10.1016/0965-8564(94)90016-7. Green, P.E., Srinivasan, V., 1978. Conjoint analysis in consumer research: issues and outlook. J. Consum. Res. 5 (2), 103–123. https://doi.org/10.1086/208721. Greene, W.H., Hensher, D.A., 2003. A latent class model for discrete choice analysis: contrasts with mixed logit. Transp. Res. Part B Methodol. 37 (8), 681–698. https:// doi.org/10.1016/S0191-2615(02)00046-2. Harvey, G., 1987. Airport choice in a multiple airport region. Transport. Res. Gen. 21 (6), 439–449. https://doi.org/10.1016/0191-2607(87)90033-1. Hensher, D.A., Button, K.J., 2007. Handbook of Transport Modelling. Emerald Group Publishing Limited. Hess, S., Polak, J.W., 2006. Airport, airline and access mode choice in the san francisco bay area. Pap. Reg. Sci. 85 (4), 543–567. https://doi.org/10.1111/j.14355957.2006.00097.x. Jeffs, V.P., Hills, P.J., 1990. Determinants of modal choice in freight transport. Transportation 17 (1), 29–47. https://doi.org/10.1007/BF02125502. Jou, R.-C., Hensher, D.A., Hsu, T.-L., 2011. Airport ground access mode choice behavior after the introduction of a new mode: a case study of taoyuan international airport in taiwan. Transport. Res. E Logist. Transport. Rev. 47 (3), 371–381. https://doi.org/ 10.1016/j.tre.2010.11.008. Kocur, G., Adler, T., Hyman, W., Aunet, B., 1981. Guide to Forecasting Travel Demand with Direct Utility Assessment. Tech. Rep. UMITA-NH11-C001-82-1. La Poste Group Press Department, 2016. Dpdgroup Drone Delivers Parcels Using Regular Commercial Line. www.geopostgroup.com/en/news/dpdgroup-drone-delivers-par cels-using-regular-commercial-line. (Accessed 13 November 2018). Lee, J.-K., Kim, S.H., Sim, G.R., 2019. Mode choice behavior analysis of air transport on the introduction of remotely piloted passenger aircraft. J. Air Transport. Manag. 76, 48–55. https://doi.org/10.1016/j.jairtraman.2019.02.007. Louviere, J.J., Hensher, D.A., 1983. Using discrete choice models with experimental design data to forecast consumer demand for a unique cultural event. J. Consum. Res. 10 (3), 348–361. https://doi.org/10.1086/208974. McFadden, D., 1981. Econometric Models of Probabilistic Choice. McGinnis, M.A., 1979. Shipper attitudes toward freight transportation choice: a factor analytic study. Int. J. Phys. Distrib. Mater. Manag. 10 (1), 25–34. https://doi.org/ 10.1108/eb014464. Molesworth, B.R., Koo, T.T., 2016. The influence of attitude towards individuals’ choice for a remotely piloted commercial flight: a latent class logit approach. Transport. Res. C Emerg. Technol. 71, 51–62. https://doi.org/10.1016/j.trc.2016.06.017.

� Truck: 36 h and 5000 KRW (about 4.55 USD). � Drone: 30 min and 10,000 KRW (about 9.09 USD). � Motorcycle: one and half hour and 7000 KRW (about 6.36 USD). Then, the estimated choice probabilities of the drone delivery service for low-price clothing, powdered formula, and urgent document are 7.18%, 8.16%, and 48.7%, respectively. Note that the probabilities for clothing and powdered formula are estimated using LCLM, and that for urgent document is estimated using BLM. The result shows that the drone delivery service will be a serious challenge to the motorcycle delivery service but not to the truck delivery service, because the truck delivery service is so cheap and fast enough. 5. Conclusion This study analyzes potential consumers’ preference for drone de­ livery service when drones are an available option for the delivery of goods and urgent documents. The potential competitors of drone are truck and motorcycle, whose markets are distinguishable. Using a stated preference survey, several discrete choice models were estimated for clothing, powdered formula, and urgent documents. The drone delivery service competes with truck delivery service for clothing and powdered formula but competes with motorcycle delivery service for urgent documents. The results show that the preferences of potential customers for the drone delivery service depend on the price and type of commodities. The comparison between low-price and high-price clothing reveals that consumers are concerned about the reliability of the drone delivery service for expensive items. So, promotion activities for the drone de­ livery will be necessary. In addition, the VOT depends on commodities. The results also show that the socio-demographic characteristics of consumers influence the preference for the drone delivery service, but the importance of each characteristic varies from item to item. On the other hands, it is observed consistently in all models that young people are more likely to opt for the drone delivery service than old people are. However, there are some limitations in this study. First, the choice models are not calibrated with any revealed preference data, because the drone delivery service has not officially started. Then, the estimated models may not be reliable since it would be influenced by the noncalibrated model constants. Thus, the models need to be reviewed after the official launch of drone delivery service. Second, the survey respondents are unfamiliar with the drone delivery service, so their understanding may be very different from actual service. Future studies should deal with both issues together. Lastly, the results of LCLM show that the socio-demographic characteristics are not the main de­ terminants of the latent class, although they affect consumer preferences as shown in the results of BLM. Therefore, future studies need to take into account more factors regarding attitudes towards the drone 7

S.H. Kim

Journal of Air Transport Management 84 (2020) 101785

Proussaloglou, K., Koppelman, F.S., 1999. The choice of air carrier, flight, and fare class. J. Air Transport. Manag. 5 (4), 193–201. https://doi.org/10.1016/S0969-6997(99) 00013-7. Sha, Z., Moolchandani, K., Panchal, J.H., DeLaurentis, D.A., 2016. Modeling airlines’ decisions on city-pair route selection using discrete choice models. J. Air Transport. 63–73. https://doi.org/10.2514/1.D0015. Shinghal, N., Fowkes, T., 2002. Freight mode choice and adaptive stated preferences. Transport. Res. E Logist. Transport. Rev. 38 (5), 367–378. https://doi.org/10.1016/ S1366-5545(02)00012-1.

Statistics Korea, 2018. Monthly Online Shopping Survey in February 2018. http://kostat. go.kr/portal/korea/kor_nw/2/11/1/index.board?bmode¼read&bSeq¼&aSeq¼366 945&pageNo¼1&rowNum¼10&navCount¼10&currPg¼&sTarget¼title&sTxt¼. (Accessed 23 May 2018). Tavasszy, L., De Jong, G., 2013. Modelling Freight Transport. Elsevier. Teal Group Corporation, 2016. Teal Group World Civil UAS Market Profile and Forecast, 2016. The Economic Times, 2017. Alibaba’s Drones Deliver Packages to Islands. (Accessed 13 November 2018). https://economictimes.indiatimes.com/news/international/busin ess/alibabas-drones-deliver-packages-to-islands/articleshow/61545583.cms.

8