Travel Characteristics and Cost–Benefit Analysis of Bikeshare Service on University Campuses

Abstract

Bikeshare has emerged as a sustainable mobility solution not only for addressing the first- and last-kilometer problem but facilitating short- and medium-distance travel. While existing research predominantly focuses on city-level Bikeshare Programs (BSPs), there is a paucity of studies examining university campus BSPs, particularly in terms of quantitative analysis of trip frequency and system operation sustainability. This paper presents a systematical framework to investigate university campus BSPs from two complementary perspectives: users’ travel characteristics and operational sustainability. To achieve this, two successive self-reported questionnaire surveys were conducted on the campus of South China University of Technology in 2017 and 2020, respectively. Subsequently, a multinomial logistic regression model was developed to identify the key factors influencing users’ travel frequency. Finally, a cost–benefit analysis was developed to assess the operational sustainability of the system. The findings reveal two significant insights: (1) the system was profitable under the 2017 fare policy, with the potential to maximize profits by strategically increasing fares while enhancing service quality; and (2) in 2020, when the fare is adjusted closer to the predicted optimal value, there is an increase in the proportion of high-frequency users, accompanied by improved user experience, reduced difficulty in bike access/return, and slightly lower pricing satisfaction. This study provides a valuable method that can be extended to the restricted service communities for effective planning and evaluation of bikeshare systems.

1. Introduction

Contrary to private cars causing noticeable environmental problems, public bikeshare programs (BSPs) have been proven to be eco-friendly and even generate positive effects. BSPs occupy less road space and generate low- or zero-emission, especially greenhouse gas emissions features [1]. Since the first deployment (the White Bike project 1965, Amsterdam), BSPs have evolved through four generations characterized by dock configurations and fare policies: free bikes with dock stations to coin deposit bikes with dock stations, e-payment bikes with dock stations, and e-payment bikes dockless stations [2]. BSPs have been gaining worldwide popularity from travelers due to their user-friendliness, physical exercise, and environmental-friendly features, especially from the wide range of students with no or low income on university campuses.

Although BSPs can help improve mobility or transfer to metro stations as feeder modes [3,4], studies have reported not all BSPs are successful. Some features like poor service quality and high access costs may lead to failure [5]. For example, Transport for London reported the ratio of operating expenditure to customer income of its city BSP (Santander Cycle) for the whole year of 2017 is 21.35:11, and the operation requires financial support from local government, enterprises, and other organizations [6]. To well understand the BSP operation, many studies have been conducted from different perspectives, such as travel demand estimation [7,8], operation and pricing modes [1,9,10], station selection [11,12], system impact [13,14,15], and mobility equity comparison [16]. In terms of system usage costs, there are also significant variations across different countries and regions. Based on the operation of BSPs in nine cities (i.e., New York; Philadelphia; Boston; Montreal; Washington, D.C.; London; Paris; Changwon; and São Paulo), the daily membership fee ranges from KRW 1000 to USD 12, while the annual membership fees range from KRW 30,000 to USD 169 [10]. This price variation is strongly correlated with service targets, the number of users, and dispatch costs of the system. For example, JUMP bikeshare program in San Francisco with 1500 e-bikes operated at the restricted service area, one of the most popular dockless systems, provides an affordable membership plan with USD 5 annual membership (USD 5 per month in the second year) for disadvantaged populations on a limited income, and charges USD 0.15/min after the free initial 60 min (or USD 2.25 for each additional 15 min) [9]. Meanwhile, Qian et al. reported that the proportion of dockless JUMP system rebalancing activities (11%) is higher than dock-based Ford GoBike one (5.1%), but both are disproportionate to their service areas. Based on the average true withdrawal rate and the walking distance distribution from the 2010 to 2012 California Household Travel Survey data, Kou and Cai conducted a biking OD demand estimation to quantitatively compare it with station-based, dockless, and hybrid bikeshare systems, and illustrated that dockless users could save trip time by 10–15% with bike pickup/return [1]. By referring to the operating mode of the popular Citi Bike systems in North America, Yoon and Chow proposed a linear regression model to distribute daily trips to 1-Day and 3-Day Pass users, and designed a new pricing plan of (USD 18.50, USD 12) for (3-Day, 1-Day), which can increase monthly revenue at least 5.5% compared to a benchmark based on the public data [10].

Notably, university campuses have always been a hot spot for BSPs [17]. College students welcome bikeshare systems for the environmental benefits and high mobility, and service operators are interested in the low construction cost [18]. Researchers have studied campus BSPs regarding trip distribution [19], docking system layout [20], tracking system design [21], and environmental impact [22]. For example, the travel characteristics of campus cyclists were found to be different from that of general cyclists since the travel density of students is high and they welcome friendly alternative modes [23]. A study from a statistical model and real usage data from BIXI, Montreal, identified key factors to bikeshare flows, including weather conditions, time of day/week, the provision of cycle infrastructure, population density and job density, and the characteristics of the built environment around the stations (e.g., near to restaurants, commercial enterprises, and universities) [24]. Unlike the city-level travel behaviors, other researchers have reported the key factors influencing campus bikeshare usage, including students’ demographic attributes (e.g., age, grade, gender, monthly expenditure), private bike ownership, safety awareness, campus size, campus distance to transit stations and CBD area, etc. Some methods have been applied to identify the factors, such as spatial–temporal analysis [19], structural equation modeling [24], mixed-effects negative binomial modeling [23], descriptive analysis combined with multivariate regression and importance–performance analysis (IPA) [25], etc. Overall, many researchers have studied the influencing factors in campus BSP usage. Most importantly, deploying campus BSPs can free their users from purchasing private bikes or worrying about theft and maintenance, and can facilitate travel [26]. Existing studies mostly focus on the user side, on the attributes of users (students), users’ activities on campuses, and the riding environments. On the one hand, most existing studies focus on whether users will use bikeshare service rather than the frequency of usage, which is critical for understanding clustered behavior patterns and the long-term operational sustainability of BSPs. On the other hand, there is a lack of in-depth analysis regarding how service-related factors influence user behavior, such as the ease of bike check-in/check-out and the impact of the pricing policy. Very few studies have addressed the sustainability analysis of BSP systems regarding the perspective of users and operators [24], much less the gaming between the two sides of the BSP system, and in particular, the impact of the pricing strategy on the willingness to use, which is the determinant of the sustainable operation of a commercial program unless it is continuously subsidized by the government or the local administration. To address this issue, this paper developed a comprehensive cost–benefit analysis model by considering the quantitative relationship between price changes and users’ travel willingness. The case study was carried out on the Wushan campus of South China University of Technology (SCUT) based on before-and-after Revealed Preference (RP) questionnaire surveys for bikeshare travel in 2017 and 2020, respectively. Firstly, this study used the 2017 survey data to recognize the influencing factors on travel frequency, and then explained the relationship between the factors and the three kinds of frequency levels (high, medium, low) with a multinomial logistic regression model. Subsequently, the operational sustainability of the BSP was discussed using cost–benefit analysis. Finally, the supplementary data from the 2020 survey was used to validate the frequency choice and the cost–benefit analysis.

2. Data Collection and Analysis

This study focuses on campus BSPs and studies the bikeshare travel of college students. A case study was conducted on the Wushan Campus of SCUT. The targeted campus covers about an area of 1,826,000 square meters, with 27,800 students, and is surrounded by public transit hubs such as Wushan Subway Station and the campus bus terminal. Two sets of RP survey data were obtained in April–May 2017 and November–December 2020, respectively. A total of 391 valid questionnaires were collected from the 2017 survey, including 284 from bikeshare users; 131 valid questionnaires were collected from the supplementary 2020 survey, including 70 from bikeshare users. The 2017 survey was conducted offline by the students in the Department of Transportation through random sampling in university buildings; the 2010 survey was conducted online. The 2017 survey data are used to build travel frequency analysis and conduct cost–benefit analysis; the 2020 survey data are used to conduct the post-analysis that validates the proposed travel frequency prediction and the cost–benefit analysis. The same questionnaire in

Table 1. The list of questions for the questionnaire.

Questionnaire Contents
Part I: User characteristics (demographic)
What is your gender?
Which grade are you in?
Are you a bikeshare user?
Part II: Bikeshare trip characteristics
How often do you use a shared bike?
What is your motive to use a shared bike?
Do you use shared bikes for roundtrip or one way?
What is your usual travel purpose?
What is your usual travel distance?
What is your usual travel duration?
What is your usual bike-finding time?
What is your usual bike-finding distance?
Part III: Environment sensitivity
Will you choose the bikeshare service on rainy days?
Will you choose the bikeshare service on cold days?
Will you choose the bikeshare service on hot days?
Part IV: Bikeshare service access
Do you think it is difficult to find a bike?
What difficulties have you faced?
How do the difficulties influence your usage of a shared bike?
What is the maximum amount of time you accept to find a bike?
What is the maximum walking distance you accept to find a bike?
Are you willing to pay extra for the improvement of the bike-finding procedure?
Part V: Bikeshare fare policy
Are you satisfied with the current fare policy?
What other fare policy do you recommend?
What is the highest fare you are willing to pay for the bikeshare service?
Part VI: Non-bikeshare users
What are the reasons for not using shared bikes?
When the fare drops below how much would you consider using a shared bike?
Would you choose to use shared bikes when it is easier to find a bike?
Are you willing to pay extra for the improvement of the bike-finding procedure?

is used for both surveys, which consists of three main parts: respondents’ demographic attributes, reasons for not using shared bikes, and bikeshare users’ travel experience.

Figure 1 and Figure 2 illustrated the travel characteristics of bikeshare users in the two surveys, and one can obtain some interesting findings from the 2017 survey:

• Travel characteristics: 50.70% of the respondents used shared bikes with medium frequency (equal or more than 1 time but less than 3.5 times per week), and mainly for short-time riding (5–10 min, 62.32%).
• Bikes access: only 3.17% of the respondents find it easy to obtain an available bike, and spent an average of 5.5 min.
• Usage challenges: the inability to locate a bike (85.56%), bike damage (59.86%), and inaccurate bike positioning (28.17%). Additionally, weather-related factors such as rainfall (42.96%), low temperature (19.72%), and high temperature (10.92%) also play a significant role in adversely impacting the service.
• Non-users: The three main reasons for not using the bike-sharing service are owning a private bicycle (63.55%), difficulty in finding a shared bike (43.93%), or the complex registration process (24.3%). Additionally, relatively minor reasons include discomfort during rides (8.41%), high fares (7.48%), lack of bike-riding experience (7.48%), and poor road facilities (4.67%). In conclusion, inadequate bike deployment and untimely maintenance are the key factors contributing to low usage. Furthermore, respondents who own bicycles are primarily senior students (60.29%), and they do not use shared bicycles due to the short traveling time and insufficient vehicle availability.
• Comparative analysis: After four years of the BSP operation, the difficulty finding available bikes has decreased from 45.42% to 20%, and the proportion of respondents with high-frequency usage has increased from 22.18% to 30%. However, it is worth noting that the proportion of respondents who almost never use shared bikes has also risen from 13.03% to 21.43%. This may be attributed to students purchasing their own bikes after enrollment and the pricing model. In terms of price factors, the proportion of respondents satisfied with the pricing has decreased from 49.65% to 32.86%, while the proportion of dissatisfied respondents has increased from 2.11% to 7.14%. With the promotion of the BSP, the majority of travelers (40%) have increasingly favored a pricing model based on usage duration.

Furtherly, this study conducted in-depth analysis on respondents experiencing bike-finding difficulties, revealing significant disparities in walking distance and time expenditure compared to the overall sample (Figure 3). Specifically, 22.86% of the bike-finding difficulty group reported walking distances exceeding 500 m (vs. 15.02% in the overall sample), while 67.86% spent over 5 min searching (compared to 46% overall). Notably, 69.51% of this subgroup completed their bike trips within 10 min, indicating that their bike access time approximates actual riding time. These findings demonstrate that, unlike general urban BSPs [24], search efficiency constitutes a critical determinant of usage intention in campus-based short-distance travel scenarios.

3. Travel Frequency Analysis

3.1. Key Influencing Factors

Based on previous studies, the travel frequency of BSP users is strongly influenced by demographic, travel (purpose, distance, and duration), and environmental (natural weather and infrastructure) factors [27]. But the survey in this study shows travel frequency is dominated by the difficulty in finding a bike (e.g., searching time, walking distance) and the price. In this paper, the frequency is classified into three levels: high (3.5, higher), medium (1, 3.5), and low (0,1), based on the number of BSP trips per week.

Table 2. Distribution of bike-sharing travel frequency (in 2017).

Influencing Factors	Classification	Percentage of Travelers (%)
		High-Frequency	Medium-Frequency	Low-Frequency
Gender	Male	15.85%	27.82%	12.32%
	Female	6.34%	22.89%	14.78%
Attitude towards current pricing policy	Satisfaction	12.32%	25.00%	12.32%
	Okay	9.15%	25.00%	14.11%
	Dissatisfaction	0.70%	0.70%	0.70%
Desired pricing model	Based on ride time	5.63%	13.73%	4.57%
	Based on ride area	1.41%	5.28%	1.44%
	Based on ride distance	4.22%	13.73%	9.20%
	Multi-mode	10.91%	17.96%	11.92%
Difficulty finding a bike	Easy	0.70%	1.41%	1.06%
	Moderate	11.62%	27.11%	12.68%
	Difficult	9.86%	22.18%	13.38%
Time to find a bike	0–2 min	1.06%	5.99%	1.41%
	2–5 min	8.80%	26.76%	10.21%
	5–10 min	10.56%	14.44%	12.32%
	10 min or more	1.76%	3.52%	3.17%
Acceptable bike finding time	0–2 min	6.47%	13.92%	5.67%
	2–5 min	10.47%	25.19%	11.03%
	5–10 min	4.62%	7.95%	7.88%
	10 min or more	0.62%	3.65%	2.53%
Walking distance to find a bike	0–100 m	1.41%	4.23%	0.70%
	100–200 m	8.10%	19.37%	7.75%
	200–500 m	8.80%	22.18%	13.73%
	500 m or more	3.87%	4.93%	4.93%
Sensitivity regarding hot weather	Unaffected	12.32%	19.37%	9.51%
	Some influence	8.45%	26.76%	12.68%
	Seriously affected	1.41%	4.58%	4.92%
Sensitivity regarding rainy weather	Unaffected	2.82%	2.46%	3.17%
	Some influence	13.03%	26.41%	9.15%
	Seriously affected	6.34%	21.83%	14.79%
Sensitivity regarding cold weather	Unaffected	8.80%	13.73%	7.04%
	Some influence	7.39%	28.87%	14.44%
	Seriously affected	5.99%	8.10%	5.64%

shows the distribution of frequency. It shows that high-frequency travelers are mostly male (71.43%), satisfied with the pricing policy (half-hourly, 55.56%), not feeling difficulty finding a bike (56%), and not sensitive to hot (56%) or cold (40%) weather. Meanwhile, the low-frequency travelers are mostly female (54.55%), generally satisfied with the pricing policy (51.95%), in favor of multi-mode pricing policy (44.16%), feeling difficulty finding a bike (49.35%), with search duration of 5–10 min (45.45%) and walking distance of 200–500 m (50.65%), and more likely to be affected by weather conditions.

3.2. Travel Frequency Prediction Model

Travel frequency has a large impact on facilities design and planning, traffic impact analysis, system operation and sustainability analysis, as well as academic proposals and industrial applications [28]. In addition, compared with estimating travel frequency using the aggregated model, this study uses the disaggregated model to capture the characteristics of individual users.

In this paper, a multinomial logistic model is built to explain and estimate the probability of choosing the level of bikeshare travel frequency. Usually, the logistic models can be divided into ordered models and nominal ones. The former are often used to study variables with different levels, which assumes that all explanatory variables have the same estimation at different split points [29]. However, some researchers found that the ordered models might generate biased or inconsistent estimation when facing underreporting because they impose tight restrictions on the influence of the variables [30]. The latter have been applied or jointly applied to many classification problems in transportation, such as parking facility design [31] and accident severity [32]. This study presents a multinomial logistic regression model for travel frequency selection. The multi-class classification problem is modeled with two one-vs-one binary classification problems, and a standard logistic regression model is fitted for each subproblem. The two one-vs-one models are constructed as follows:

$$\ln ({\displaystyle \frac{p_2}{p_1}})={\alpha }_1+{\displaystyle \sum _i{\beta }_{1i}{x}_i}$$

(1)

$$\ln ({\displaystyle \frac{p_3}{p_1}})={\alpha }_2+{\displaystyle \sum _i{\beta }_{2i}{x}_i}$$

(2)

$$p_1+p_2+p_3=1$$

(3)

where p₁, p₂, and p₃ represent the probability of selecting low-, medium-, and high- frequency, respectively; x_i are the ith variable influencing the selection; α₁ and α₂ denote the intercepts, respectively; and β_1i and β_2i are the coefficients, respectively. In this study, low-frequency (p₁) is chosen as the reference level. The open-source Python (3.9) package “statsmodels” was used for parameter estimation.

The input variables are the ones that describe the users’ demographic or travel-related characteristics. A total of 284 records were used. The variables with no or weak statistical significance were eliminated by stepwise regression. The remaining 11 key factors are listed in

Table 3. Input variables of the trip frequency prediction model.

Influencing Factors	Factor Indicators	Variable Values
Personal Characteristics	Gender x1	1: Male, 2 *: Female
	Education Level x2	1: Undergraduate, 2 *: Graduate
Travel Characteristics	Trip purpose x3	1: Short-distance intra-district travel, 2: Medium- and long-distance inter-district travel, 3 *: Urgency/temporary use
	Duration x4	1: <5 min, 2: 5–10 min, 3 *: ≥10 min
	Difficulty finding a bike x5	1: Easy, 2: General, 3 *: Difficult
	Walking time to find a bike x6	1: <2 min, 2: 2–5 min, 3: 5–10 min, 4 *: ≥10 min
	Acceptable bike finding time x7	1: <2 min, 2: 2–5 min, 3: 5–10 min, 4 *: ≥10 min
	Walking distance to find a bike x8	1: <100 m, 2: 100–200 m, 3: 200–500 m, 4 *: ≥500 m
	Satisfaction with the current pricing policy x9	1: Satisfied, 2: General, 3 *: Unsatisfied
	Influence of rainy days x10	1: Unaffected, 2: Continue to use when it is not serious, 3 *: Give up using
	Influence of hot days x11	1: Unaffected, 2: Continue to use when it is not serious, 3 *: Give up using
	Influence of cold days x12	1: Unaffected, 2: Continue to use when it is not serious, 3 *: Give up using

* Indicates referent category.

.

After model estimation, the log-likelihood ratio test was conducted to validate the proposed model. The test statistic takes the following form:

$$D=-2LL=-2\times \ln ({\displaystyle \frac{\mathrm{likelihood} \mathrm{for} \mathrm{null} \mathrm{model}}{\mathrm{likelihood} \mathrm{for} \mathrm{proposed} \mathrm{model}}})$$

(4)

In the test, the −2LL value decreased from 600.96 to 511.01 with a p-value less than 0.001, which means that the proposed model is significantly better than the null model (random selection). Also, the significant value in the Pearson chi-square goodness of fit test was 0.172 > 0.05. The clarification results and the parameter estimation results are shown in

Table 4. Prediction results under the logistic regression model.

Observed Travel Frequency	Predicted Travel Frequency
	Low	Medium	High	Precision
Low	32	39	6	41.60%
Medium	20	112	15	76.20%
High	4	36	27	40.30%
Overall	19.20%	64.30%	16.50%	58.80%

and

Table 5. Estimated intercepts and coefficients of significant variables.

Frequency *	Variable **	Estimated B a	p-Value	Exp(B), OR b
medium	α1 intercept	0.573 (1.643)	0.727	-
	β1_3 trip characteristic, purpose: short-distance travel	0.793 (0.360)	0.028	2.211 (1.092, 4.478)
	β1_10 weather condition, rainfall: continue to use when not serious	−0.77 (0.35)	0.028	0.463 (0.233, 0.919)
	β1_12 weather condition, cold weather: continue to use when not serious	−1.128 (0.480)	0.019	0.324 (0.126, 0.829)
high	${\alpha }_2$ intercept	−3.234 (2.075)	0.119	-
	β2_1 user characteristic, gender: male	1.205 (0.41)	0.003	3.337 (1.495, 7.45)
	β2_3 trip characteristic, purpose: short-distance travel	1.052 (0.442)	0.017	2.863 (1.205, 6.803)
	β2_7 quality of service, acceptable accessing time: <2 min	2.153 (1.088)	0.048	8.612 (1.022, 72.599)
	β2_12 weather condition, cold weather: continue to use when not serious	−1.228 (0.53)	0.02	0.293 (0.104, 0.827)

* Low frequency is chosen as the reference category; ** redundant variables are set to 0; ^a standard error in parentheses; ^b 95% confidence interval in parentheses.

, respectively.

Table 5 lists the parameter estimation values, the p-values, and the odds ratio (OR) values. α and β correspond to the intercepts and the coefficients in the binary classification model expressed by Equations (1) and (2), with α₁ and β_{1_i} for the medium-/low-frequency selection, and α₂ and β_{2_i} for the high-/low-frequency selection. Due to the nonlinearity of the logistic regression model, the OR values (Exp(B)), instead of the direct parameter estimation values (B), are often used to reflect the change in the odds (p_i/p_j) under the effect of the coefficients. The values of coefficients (β_{1_i}, β_{2_i}) are determined by the values of the independent variables (x_i). With a specific study area (in aggregated analysis) or an individual user (in disaggregated analysis), the travel characteristics (the values of the independent variables) and the coefficients are determined, respectively. For example, if the students report a maximum acceptable bike searching time of less than 2 min (x₇ = 1), this significantly influences the odds of choosing high and low frequency (β_{2_7} = 2.153); however, it does not significantly influence the odds of choosing medium and low-frequency (β_{1_7} = 0). The model is intuitively correct, and the detailed explanation can be reached as follows:

• Trip purpose (x₃): Compared with urgent or temporary travelers (reference category), short-distance travelers have significantly higher odds of being medium- and high-frequency bikeshare travelers, with p₂/p₁ and p₃/p₁ being $e^{{\beta }_{1\_3}}$ = 2.211 and $e^{{\beta }_{2\_3}}$ = 2.863 times that of urgent travelers, respectively.
• Sensitivity to rainfall (x₁₀): Compared with the travelers who are more sensitive to rainfall (give up using, reference category), the insensitive travelers (continue when not serious) have significantly lower odds of being medium-frequency travelers, with p₂/p₁ being $e^{{\beta }_{1\_10}}=0.463$ times that of the sensitive travelers. However, no significant difference in the odds of being high-frequency travelers (p₃/p₁) is observed. With the equality constraint of Equation (3), this indicates that, compared with sensitive travelers, rainfall-insensitive travelers have a higher probability of being low- and high-frequency travelers (p₁, p₃) and a lower probability of being medium-frequency travelers (p₂).
• Sensitivity to cold weather (x₁₂): Compared with travelers who are more sensitive (give up using, reference category) to cold weather, the insensitive travelers (continue when not serious) have significantly lower odds of being medium- and high-frequency travelers, with p₂/p₁ and p₃/p₁ being $e^{{\beta }_{1\_12}}$ = 0.324 and $e^{{\beta }_{2\_12}}$ = 0.293 times that of urgent travelers, respectively.
• Gender (x₁): Compared with female travelers, male travelers have significantly higher odds of being high-frequency travelers, with p₃/p₁ being $e^{{\beta }_{2\_1}}$ = 3.337 times that of female travelers. However, no significant difference in the odds of being medium-frequency travelers (p₂/p₁) is observed. So, compared with female travelers, male travelers have a higher probability of being high-frequency travelers (p₃) and a lower probability of being low- and medium-frequency travelers (p₁, p₂).
• Maximum bike searching time (x₇): Compared with travelers who tolerate long bike searching time (accept bike searching time of 10 min and longer), travelers with low tolerance (accept when less than 2 min) have significantly higher odds of being high-frequency travelers, with p₃/p₁ being $e^{{\beta }_{2\_7}}$ = 8.612 times that of the high-tolerance group. However, no significant difference in the odds of being medium-frequency travelers (p₂/p₁) is observed. This change in odds indicates an increase in the probability of choosing high-frequency (p₃) and a decrease in the probability of choosing low- and medium-frequency (p₁, p₂) when bike searching time tolerance is low.

Compared with traditional studies, the above results show that, besides the conventional influencing factors (travel characteristics and travel environment), the bike finding difficulty and the sensitivity to weather conditions also have different effects on the odds of choosing different frequency levels.

4. Cost–Benefit Model

As a public service, BSPs influence citizens’ attitudes toward bike-sharing or even biking in general. If a BSP provides poor services or adopts unsustainable business practices (such as setting unreasonably low prices to expand ridership or over-pricing), travelers may switch to other modes, which will lead to lower utilization of non-motorized facilities and eventually hamper the future development of bike systems [33]. In this paper, a cost–benefit analysis model for campus BSPs is developed to understand the equilibrium of the market.

4.1. Cost Estimation

With the proposed travel frequency model, the proportion of each travel frequency group can be estimated, and the daily demand for bike-sharing trips in the area can be calculated using Equation (5).

$$N_D=\alpha (c)N_P{\displaystyle \sum _i{f}_i{p}_i}$$

(5)

where N_D is the total number of bikeshare trips per day; c denotes the fare (CNY/30 min); N_p represents the total population in the target area; α(c) is the proportion of bikeshare users under the fare c; f_i is the average number of trips per day per person for frequency group i; and p_i is the proportion of frequency group i, which is calculated from the survey data. To accurately estimate the mathematical relationship between fare changes and the proportion of shared bicycle usage, this study utilized the questionnaire question “What is the highest fare you are willing to pay for the bikeshare service?” (Table 1) to obtain the highest acceptable fare and the corresponding population proportion. Based on the respondents’ distribution range of CNY 0 to 5/30 min, a curve fitting was performed using the least squares method, resulting in the following relationship:

$$\alpha (c)=\left\{\begin{array}{ll}1\hfill & c=0\hfill \\ 1.17e^{-0.718c}\hfill & 5\ge c>0\hfill \\ 0\hfill & c>5\hfill \end{array}\right.$$

(6)

For a BSP, most of the operation cost is the bike purchasing and maintenance cost. This study assumes the operators aim to meet the demand, and the number of deployed bikes is no less than the number of users during peak periods. Therefore, the number of bikes should satisfy:

$$N_B={\displaystyle \frac{\beta N_D}{\gamma }}(1+\theta )$$

(7)

where N_B is the maximum number of bikes to be deployed; β denotes the peak-hour index (the ratio of peak-hour trips to total daily trips); γ is the average peak-hour bike turnover rate (trips/hour); and θ represents the proportion of backup bikes. The cost estimation model in this study considers bike purchasing, depreciation, and maintenance costs. The platform operation cost and the bike dispatching cost are not considered. According to Equation (7), the cost of the BSP can be calculated as follows:

$$C_V=N_B(1+k)c_0+N_B{c}_M$$

(8)

where C_V is the estimated cost of the BSP system; k denotes the annual bike depreciation rate; c₀ represents the purchase price per bike (CNY); and c_M means the annual maintenance cost per bike (CNY).

4.2. Benefit Estimation

This study assumes fare income as the BSP revenue since additional revenue, such as the deposit policy and advertisement revenue, vary greatly among operators. According to previous surveys, most campus trips last for less than 30 min, so the revenue is set to the product of the fare and the number of trips. Therefore, the annual revenue can be expressed as follows:

$$R_V=365\times N_{D}c$$

(9)

where R_V is the estimated annual revenue. The annual profit of the system can be calculated as follows:

$$P=R_V-C_V$$

(10)

where P is the estimated annual profit without considering the bike redistribution costs and platform operation ones.

5. Numerical Analysis

A case study is carried out in the Wushan Campus based on the survey data. There are about 27,800 students on campus, with a peak-hour index of 9.2% (β, from campus bus ridership investigation).

5.1. Benefit Analysis Based on 2017 Survey Data

Demand and annual revenue: The 2017 survey shows the proportion of low-frequency, medium-frequency, and high-frequency travel groups is 26.5%, 50.5%, and 23%, and the average number of trips per day in each group is 0.059, 0.356, and 1.333. The fare in 2017 was CNY 1/30 min. Based on Equations (5), (6), and (9), the total demand is about 7964, and the annual revenue is about CNY 2.986 million.

Operating cost: The peak-hour ratio is 9.2% from a campus bus ridership survey. Assume the backup rate is 50%, the peak-hour bike turnover rate is 1.2, the bike price is CNY 2000, the annual maintenance fee is CNY 500, and the bike depreciation rate is 25%. Based on Equations (7) and (8), the number of bikes is 916, and the annual cost is CNY 2.748 million.

Given the revenue and cost, the annual profit under the current fare is CNY 159 thousand. The profit under fluctuating fare is investigated to better understand the relationship between the bikeshare fare and system profit. Two types of operation modes (bike deployment policy) are considered:

• Stable fleet size: deploy bikes based on current demand (the demand when the fare is CNY 1/30 min), despite the fluctuation in demand under different fares, and
• Dynamic fleet size: deploy bikes in correspondence with the changing demand. The stable fleet size policy provides a constant annual cost, while the dynamic fleet size policy provides a fluctuating one.

Figure 4 shows the change in profit under the stable fleet size operation. When the fare is CNY 0.8–0.9/30 min, the cost and revenue are balanced. However, the maximum profit appears when the fare is about CNY 1.5/30 min, with a value of about CNY 0.3 million. Figure 5 shows the change in profit under the dynamic fleet size operation. The system also reaches a balance at CNY 0.8–0.9/30 min; but the maximum profit occurs at about CNY 2.5/30 min, with a value of over CNY 1.5 million.

According to the two figures, one can find the following results:

• If the BSPs operation agencies intend to maintain the current fleet size, they can achieve the maximum profit by slightly increasing the fare to about CNY 1.4/30 min.
• If the agencies, in the future, would like to deploy the dynamic fleet size practice, they can achieve an even higher maximum profit with the fare set to about CNY 2.4/30 min.
• The agencies can make an even higher profit by attracting more users. They can invest their profit in improving the service, such as increasing the number of bikes, expanding bike coverage, and redistributing bikes timely. The 2017 survey shows that a high percentage of people (43.93%) are non-BSP users due to the difficulties in finding a bike, but a very high percentage of them (86.92%) are willing to use it if more bikes are properly deployed and the bike searching time is reduced.

5.2. Before–After Analysis with 2020 Survey Data

In the 2020 survey, 131 valid questionnaires were collected, including 70 BSP users and 61 non-users. The distribution of the demographic characteristics and travel characteristics of the users are shown in Figure 2, and the distribution of the three travel frequencies within each variable is in

Table 6. Distribution of BSP travel frequency under different characteristics (2020).

Key Influencing Factors	Classification Values	Percentage of Travelers
		High-Frequency	Medium-Frequency	Low-Frequency
Gender	Male	18.57%	21.43%	24.29%
	Female	11.43%	7.14%	17.14%
Attitude towards current pricing policy	Satisfaction	8.57%	14.29%	10.00%
	Okay	20.00%	12.86%	27.14%
	Dissatisfaction	1.43%	1.43%	4.28%
Desired pricing model	Based on ride duration	12.86%	10.00%	17.14%
	Based on ride area	2.86%	0.00%	2.86%
	Based on ride distance	5.71%	5.71%	5.71%
	Multi-mode	8.57%	12.86%	15.72%
Difficult to find a bike Y/N	Easy	4.29%	7.14%	4.29%
	Moderate	20.00%	17.14%	27.14%
	Difficult	5.71%	4.29%	10.00%
Consumed time to find a bike	0–2 min	8.57%	8.57%	10.00%
	2–5 min	17.14%	15.71%	24.29%
	5–10 min	1.43%	1.43%	4.28%
	10 min or more	2.86%	2.86%	2.86%
Acceptable bike finding time	0–2 min	8.57%	10.00%	10.00%
	2–5 min	11.43%	11.43%	24.29%
	5–10 min	10.00%	5.71%	5.71%
	10 min or more	0.00%	1.43%	1.43%
Walking distance to find a bike	0–100 m	2.86%	8.57%	7.14%
	100–200 m	14.29%	8.57%	20.00%
	200–500 m	11.43%	7.14%	11.43%
	500 m or more	1.43%	4.28%	2.86%
Sensitivity regarding hot weather	Unaffected	11.43%	14.29%	12.86%
	Some influence	14.29%	12.86%	27.12%
	Seriously influence	4.29%	1.43%	1.43%
Sensitivity regarding rainy weather	Unaffected	0.00%	1.43%	7.14%
	Some influence	20.00%	18.57%	24.29%
	Seriously influence	10.00%	8.57%	10.00%
Sensitivity regarding cold weather	Unaffected	5.71%	7.14%	11.43%
	Some influence	21.43%	15.71%	27.15%
	Seriously influence	2.86%	5.71%	2.86%

.

Comparing the data from the two surveys in 2017 and 2020, one can find that:

• In 2020, the fare of bikeshares was raised to CNY 1.50/30 min, which is consistent with the findings from the benefit estimation in Figure 4. And, a bikeshare program with e-bikes (e-BSPs) began to emerge on campus, with a fare of CNY 1.5/15 min.
• From Figure 1 and Figure 2, one can observe the proportion of high-frequency users increases while that of medium-frequency users decreases. The ratio between high- and low-frequency users remains relatively stable, while the ratio between medium-frequency and low-frequency users decreases significantly.
• Changes in travel characteristics are also observed. The proportion of male users increases; the users are less sensitive to rainfall and cold weather; the bike-finding experience is improved, with decreased bike access difficulty and bike access time; and the acceptable bike access time remains stable. According to the proposed frequency estimation, a decrease in rainfall sensitivity (x₁₀) should lead to a decrease in the odds of being medium-frequency travelers (compared with low-frequency). The 2020 survey showed the consistent results.
• The frequency distribution in Table 1 and Table 3 is diversified. There is a significant decrease in the pricing satisfaction level for high-frequency users; more respondents find bike searching generally satisfying in each frequency group; and the influence of weather factors on low-frequency trips diminishes. The decline in pricing satisfaction among high-frequency travelers may be due to the introduction of the e-BSP. According to the 2020 survey, 74.27% of the users’ rides last less than 10 min, and the e-BSP service charges CNY 1.5/15 min. So, by choosing e-BSP, they do not need to pay more but can receive a faster and more comfortable riding experience.
• Regarding system operation, significantly more users prefer a duration-based pricing policy, with a proportion nearly matching that of those who prefer a combined pricing policy. The number of people who are generally satisfied with the current pricing policy increases. And the users are less willing to increase the fare to improve the service.
• There is also an increase in the proportion of non-BSP users (27.4% in 2017 to 46.6% in 2020). Overall, 55.7% said they bought private bikes or e-bikes due to a poor riding experience; 14.3% said they still had problems finding bikes; and 67.1% said they would join when bike finding is easier. The introduction of the campus e-BSP could also account for the significant decrease.

6. Conclusions

Based on two surveys carried out in 2017 and 2020, this paper studies the key influencing factors in campus bikeshare travel frequency and conducts a cost–benefit analysis. It finds out that:

• More than half of the respondents have used shared bicycles. In total, 60% of the users are high-frequency and medium-frequency travelers, and more people have been willing to use or increase the frequency of use if the number of bikes increases and the bikes are maintained timely.
• The travel frequency of bikeshare users is influenced by their demographic characteristics, travel characteristics, bike finding difficulty, and sensitivity to weather conditions. A multinomial logistic regression model was developed to describe the selection of the three bikeshare travel frequency levels (high, low, and medium).
• The cost–benefit analysis proves that, even with student users who have no income, the BSP agencies can still increase their profit by appropriately adjusting fares and upgrading service levels. Under the existing fleet size and assumptions, a reasonable price is suggested to be about CNY 1–1.4/30 min, and it should not be higher than CNY 2/30 min when the system is designed to be demand responsive in the future.
• For residential areas, industrial parks, and other restricted service communities, bikeshare programs have good prospects. The operation is profitable, the service can be of high quality, and it would be a good practice to promote sustainable transport.
• When operating the BSPs within a restricted service area, regional administrators (e.g., local governments, university management boards, and industrial park committees) will play critical roles alongside operators and users, particularly in preventing negative impacts caused by operators abruptly exiting the market after initial profitability. These impacts may include abandoned obsolete bicycles and unmanaged damaged vehicles. Preventive measures may include refundable deposit schemes for operating enterprises, competitive mechanisms involving two or more operators, and user feedback and penalty mechanisms.

This study includes an initial survey conducted in 2017 and a follow-up verification survey in 2020. These investigations provide data support for our research framework regarding travel demand prediction and system cost–benefit analysis. However, the relatively small number of questionnaires might have some impact on the validation of model parameters. Therefore, future research should address the following three issues: (a) validation of the reliability of the proposed BSPs cost–benefit framework under various university campuses and non-motorized travel demand patterns through expanding the sample size of the survey or field operational data; (b) improvements to travel frequency prediction models and the cost–benefit estimation model with consideration of bike rebalancing costs; and (c) competitive analysis when there are two or more bikeshare system agencies at the restricted service areas.