A survey using Google Forms was conducted, and selected respondents were chosen from Surabaya city and Malang city in Indonesia. The selection of Surabaya and Malang as research locations was based on their unique characteristics. Surabaya is known as the capital of the Province of East Java and is a trading and industrial city. Malang is known as a city of education and recreation that requires safe and comfortable public transportation. Although the cities provide public transportation, its role has been decreasing. Instead, many people are switching to motorcycles or private cars.

To achieve the objectives, a literature study was conducted on the online taxi mode choice. Using stated preference technique, a questionnaire was designed to acquire data on travel preference. Then, the binary logit model was proposed to describe the preferences of passenger’s personalized travel behavior and Importance Performance Analysis was conducted to understand the performance of online taxis based on passenger perceptions. A survey using Google Forms was conducted for five weeks and 200 responses were collected.

The questionnaire included (a) personal profile information (age, gender, income, education, and occupation),

(b) Preference for choosing personalized travel mode (online taxi or conventional taxi), (c) the evaluation of action principle in choosing personalized travel mode (travel cost and pick up time). The respondents were selected randomly from Surabaya city and Malang city. The respondents must have had no more than two months of experience in using an online-taxi, be 15-65 years old, and have the online taxi mobile applications as well as experience in using the apps. The definition of variables on the scale is presented in Table 1.

In this study, a conventional taxi refers to a road-based taxi that utilizes a taximeter (conventional taxi) and an online taxi refers to a taxi ordered by using electronic applications. Modal choice selection is one of the basic and important issues in transportation; the choice of traveling by either conventional taxi or online taxi is also associated with personal characteristics and travel characteristics.

The Stated Preference approach has been widely used in transportation, as well as in modal choice selection. This approach examines individual responses to a series of experimentally designed choice alternatives, which are typically described in terms of combinations of attributes with several predefined levels. Hensher used Stated Preference (SP) in his study to determine the value of time [7]. The Stated Preference approach offers the modal choices at once to respondents, then the respondents sort their choices to show the level of preferences of their choices.

To understand the modal choice, several formulas can be utilized. One of them is the logit model, which is used to forecast the probability of an event by fitting data. The decision of mode selection is influenced by individual and transit characteristics and it can be disclosed by using the binary logistic model [8]. Furthermore, a study regarding the adoption of on-demand ride services (Uber) was conducted in California. The study aimed to estimate binary logit models of the adoption of Uber/Lyft with and without the inclusion of attitudinal variables [9].

In this paper, the binary logit model is proposed to describe the preferences of passenger’s personalized travel behavior, whether travel fare and pick up time influencing the passenger’s decision to use online taxi or not, wherein passengers shift from online taxi to conventional taxi. The binary logit model was used because, in this paper, the dependent variable consists of only two categories, which are whether the passengers use the online taxi or any conventional taxi.

The result of personalized travel mode prediction is travel by either online taxi or conventional taxi. Below, xⁱ refers to the i^th passenger. Total prediction is the probability of the event (online taxi or conventional taxi). Assuming that n variables exist from x¹, x², … xⁿ, the logit value of the unknown binomial probability is defined as the following linear function [2]:

(1)

The unknown parameter ^β_k is estimated through maximum likelihood estimation, which characterizes the contribution that the k^th variable makes to the logit probability value (k= 1, 2, … n), and ^β₀ refers to an intercept. From Equation 2, Equation 3 is obtained as follows:

(2)

Assuming that the utility function could be written as:

(3)

Eq 2 could be formulated as logistic regression:

(4)

Where f(z) Є (0,1) refers to the probability of an event.

Importance-Performance Analysis (IPA) has been widely adopted in a variety of business sectors for understanding customer satisfaction, identifying areas for improvement, and prioritizing resource allocation [10]. Importance-Performance Analysis (IPA) measures perceived performance based on the importance of service with a two-dimensional plot [11]. A study to measure-out of the customer satisfaction factors in the online taxi mobile app service was performed by using Importance-Performance Analysis (IPA) [12]. Another study about the motorcycle taxi was conducted in Indonesia, Importance-Performance Analysis was applied and the results showed that better provision of trip rates has become a significant indicator for a business partner during the partnership process [13]. The data are presented in a matrix, with the X-axis and the Y-axis depicting attribute importance and attribute satisfaction or performance, respectively, with four quadrants based on their rankings. Attributes located in quadrant I have high importance but low performance, which warrants managers or stakeholders to concentrate their effort and resources on; quadrant II has attributes with both high importance and performance, which managers or stakeholders need to maintain good work on. Attributes in quadrant III have low importance and performance rankings and are of low priority for resource allocations; finally, attributes in quadrant IV have low importance but high performance and are thus possibly “overkill”, for which managers or stakeholders might direct their resources elsewhere [14]. Fig. (1) shows the Importance-Performance Analysis (IPA) matrix [11].

Based on the Ministry of Transportation Regulation No. 46 of the year 2014 on the minimum service standards for road transportation with non-fixed routes, the attributes for the study were developed. The list of research attributes is shown in Table 2.

Fig. (1). The Importance-Performance Analysis (IPA) Matrix.

The detail of respondents’ characteristics for this study is shown in Table 3.

After the data from the questionnaire were collected, the online taxi performance analysis was carried out with Importance-Performance Analysis, and the results are shown in Table 7.

To evaluate the influence of travel expense and pickup time, in this study, selection ratios of the conventional taxi and the online taxi were evaluated with the binary logit model. The results for travel cost analysis are shown in Fig. (2).

Fig. (2) indicates that with the increase of online taxi travel costs, the selection of the ratio of online taxi decreases, while that of the conventional taxi increases. When the travel cost of an online taxi is cheaper (Rp.15.000) than a conventional taxi, the probability to use an online taxi is 0.902 (90.2%) but the point at which the travel cost of an online taxi is exactly the same as the conventional taxi, the probability to use the online taxi decreases to 0.704 (70.4%). A midpoint travel cost between online taxis and conventional taxis is Rp. 9.606,20. It is shown that the maximum price gap between an online taxi and conventional taxi is Rp.9.606,20. If the travel cost of online taxis is more expensive by Rp. 9.606,20 than conventional taxis, passengers would prefer using conventional taxis and the probability to use conventional taxis significantly increases. The result shows that online taxi fare is a sensitive variable that determines the decision of the passengers whether to use an online taxi or conventional taxi. The travel fare is a sensitive variable for the passengers, and the traveler’s willingness to pay the maximum price gap is Rp.9.606,20. This is because the majority of respondents are the group of people, mainly students, with a median income that they receive from their parents of less than Rp. 1.000.000,00/month (one million rupiahs per month) including the high-school students (32.5%) and university students (26.5%), therefore the budget for the travel expense is limited.

Modal choice probability based on pick up time is an analysis same as the method of modal choice probability based on travel cost. The results of the analysis for modal choice based on pick-up time are shown in Fig. (3).

Fig. (2). Graph of modal choice probability based on travel cost (Rupiah).

Fig. (3). The graph of modal choice probability based on pick up Time (minutes).

As shown in Fig. (3), when passengers care more about pick-up time, the selection ratio of online taxi increases. The midpoint pick-up time of online taxis and conventional taxis is not visible in Fig. (3), because the analysis using the binary logit model showed that the midpoint for pick-up time is 13.43 minutes, but the offered choices for pick-up time to respondents were only from 0-10 minutes. If the pick-up times of online taxis are 13.43 minutes longer than conventional taxis, the probability to choose conventional taxis increases, and passengers would prefer to use conventional taxis than online taxis. A more positive difference in pick up times leads to shorter pick-up times and a greater probability to choose online taxis; consequently, the probability to choose conventional taxis significantly decreases. The research result shows that for passenger willing to wait for the online taxi little longer, more than 10 minutes offered as maximum pick-up time in the questionnaire sheet, it is because the main purpose of traveling by online taxi is for study purposes like going to campus or study group, and they have enough spare time to wait little longer but not more than 13.43 minutes.

The performance of online taxi was analyzed by Importance-Performance Analysis, and the attributes were identified as being in quadrant I (needs improvement), quadrant II (keep up the good work), quadrant III (low priority), or quadrant IV (possible overkill). The results showed that all attributes met passengers’ expectations of online taxi performance and the passengers were satisfied. This is proven with no attributes found in quadrant I that need improvements. Attributes in quadrant III are of low priority because these do not significantly influence the respondents’ perception of online taxi performance. Furthermore, attributes in quadrant IV are possible overkill attributes, as these attributes do not significantly influence the respondents' perception about online taxi performance but the online taxi still provided these attributes. To maintain this performance, stakeholders need to pay attention to improve attributes in quadrant II, which contains attributes that need to be focused to have continued good work. They must improve the quality of attributes in quadrant II and reduce the effort for attributes in quadrant III and quadrant IV.