Analysis of Travel Choices and Scenarios for Sharing Rides
Final Report

Chapter 3. Encouraging Carpooling Using App-Based Incentive Tools

With the proliferation of shared mobility options, it is important to understand traveler behavior and decision choices (among vehicle/mode, vehicle occupancy, service types, and times) and how various travel cost and travel performance factors can affect those choices. Research and experience show that carpooling represents an important opportunity to enhance the efficiency of urban transportation systems and reduce congestion. Carpooling and vanpooling have been part of metropolitan planning organizations and departments of transportation travel demand management (TDM) programs for decades. Various agencies are further supporting these programs by offering incentives for travelers to carpool and rideshare.

Some transportation agencies are working with emerging app-based carpooling and navigation services, like Scoop™, Waze™, Metropia™, Agile Mile™, and Hytch™, to pilot incentives for ridesharing in private vehicles. For example, Contra Costa Transportation Authority piloted a partnership with Scoop™ to test cash-based carpool incentives through the Scoop™ app from January 2017 to June 2018.¹ The agency tested a $2 incentive per passenger-ride (deducted from the mileage fee the rider was to pay the driver, minus $1 going to Scoop) coupled with a $2 per trip driver incentive and found that the two sides of the incentive were equally important; when the driver incentive was removed, shared trips decreased by 20 percent as carpool matching became unreliable. Similarly, King County Metro Transit recently partnered with Scoop™ and the University of Washington in a 5-month pilot that tested a $2 incentive for riders and drivers for each carpool trip via Scoop's™ application (which also facilitates distance-based payment from the passenger to the driver). Shen, Wang, and Gifford summarized the results of the pilot, finding that the Scoop™ incentive reduced single-occupancy vehicle (SOV) commute trips, with an estimated decrease of nearly 1,000,000 vehicle miles traveled (VMT), effectively resulting in a VMT reduction at a cost of $0.40 per mile.² These pilots, and others described below, have shown that incentives may be effective in converting SOV trips to carpool trips and point the way toward a precise study of the effect of incentive levels on sharing in the app-based carpooling context.

Beyond incentive size, many studies have been conducted to identify factors that limit carpooling adoption. Generally, these factors may be divided into four main categories: (1) convenience, (2) trust issues, (3) lack of awareness, and (4) lack of incentive, with the first two recognized as the most critical. Emphasizing trust issues as an example in their paper covering different ways of carpooling, Kurth and Hood noted that if someone is interested in carpooling using a work connection, it would be advantageous not only because the poolers have similar work schedules, but also because the work organization establishes a basic trust level between participants.³ Gardener and Abraham mention that drivers dislike delegating control and value their solitary personal space.⁴ Therefore, they conclude that privacy issues and the fear of riding with strangers are limitations to carpooling success. In another study, the availability of potential carpooling partners and their relationships were dominant elements in affecting this fear.⁵ This was significant enough that Correia and Viegas mark the psychological barrier of riding with strangers as one of two main problems limiting the future of carpooling.⁶

Overcoming psychological barriers is a critical step in adopting carpooling. Drive-alone users may, if motivated, potentially become carpool users and carpool users may alternate between driver and passenger roles based on their travel needs, daily activities, car availability, the perception of transportation costs, conditions, etc.

The central question is: what influences a user's choice between riding (or driving) alone versus carpooling, and what specific financial levers can be pulled to influence this choice? To provide an answer, it is helpful to better understand: (1) the user socio-demographics, temporal and spatial activity characteristics, and the transportation conditions influencing the mobility option decision, (2) what triggers and motivates individuals to transition from one mobility option to another, and (3) the factors that could sustain the behavior change over the long-term.

Making mobility options available, while essential to behavior change, does not mean that travelers will change their behavior immediately or at all. What is missing is actively presenting the mobility options to commuters and engaging them through appropriate means (e.g., gamification and incentives) in order to trigger the desired behavior changes. As such, engagement becomes the second framework behavior tenet with the goal to better understand how it positively affects traveler choices.

While many app-based carpooling solutions are emerging, data on incentives and their impact on sharing rates are hard to obtain. For the purpose of this study, data previously gathered by two companies (Metropia™ and Hytch™) were analyzed by these companies to answer questions FHWA subsequently posed to them. These reported results should be considered examples, as findings from both models are not representative of the other app-based carpooling systems in the marketplace due to significant variations in how each app-based carpooling solution interfaces with its users. More importantly, the data gathered from both apps preceded this study, so the ability to conduct deliberate tests in line with research questions for this study was limited, but the research team considered looking at retrospective data to be meaningful. One important limitation was that travel choice data prior to app use was in general not gathered, thereby making it hard to attribute behavior change to specific app use.

The two app-based systems used in the study, Metropia™ and Hytch™, are described in the following pages.

Metropia's™ app includes a complete suite of available mode options and services include driver navigation, dynamic "social carpool" pairing, transit, ride hailing, micro-transit, biking, and walking. For the purpose of the study, only the choice making surrounding carpooling was included. Within the app, Metropia's™ Driving Up Occupancy (DUO) is the module that enables carpooling rewards. DUO is a dynamic tool that tracks drivers and passengers, who are both DUO users, once they are inside a vehicle. DUO leverages the existing types of social relationships in the user's daily life (e.g., working for the same company, being a member of the same recreational/sports team, living in the same household, etc.) to alleviate the trust concerns and psychological barriers of riding with strangers. The impact of incentives on Metropia's™ users was analyzed using app data combined with Metropia's™ micro-survey tool. In addition to identifying variables affecting the magnitude and trend of carpool use, a time-series analysis is presented to better understand how incentive changes over time affect the carpool use trend (i.e., can carpool behavior be sustained if incentives are reduced over time?). This is directly related to the sustainability of monetary resources offered by the incentive provider (typically a public agency or employer).

The Hytch™ platform enables users to simultaneously take advantage of multiple sources of reward funding from different entities. These rewards may be affiliate-based, open to all travelers in a geo-fenced area, or available to travelers attending a large event, depending on the goals of each funding entity. Typical rewarding entities are major employers and local governments seeking to reduce employee parking costs or curtail congestion. The app verifies, measures, and rewards the user for travel decisions that are pre-specified as reward eligible (e.g., nonmotorized travel, carpooling, vanpooling, or taking transit). Like the Metropia™ app, reward sponsors could, if they desire, vary reward amounts to test impacts on travel choices and continued engagement with the Hytch™ app.

Impact of Rewards and Incentives using Data from Metropia Driving Up Occupancy Platform

Metropia's™ DUO 1.0 version, data from which Metropia analyzed for this study, allows a drive-alone user to earn points based on the selected departure time, while for each successfully paired carpool (DUO) trip the driver earns his/her driving points plus a bonus of half of the points each passenger earns (while not taking the points away from the passenger).⁷ Beyond passively collecting data, Metropia's™ platform includes a micro-survey tool which allows questions to be asked directly to the users, using real-time applications, to understand behavior and travel patterns better and to obtain pertinent user characteristics, producing response rates significantly higher than traditional surveys. Metropia™ users included in the study were able to select one of three mobility options (drive alone, carpool driver, and carpool passenger) and earn points for each trip. Roughly 60 points are awarded for each trip between the driver and the passenger. These points can be translated into dollars on a conversion rate of 1,200 points to $5.

Metropia's™ DUO service was initiated in Austin and El Paso, Texas, and Tucson, Arizona, in January 2016, and the micro-survey feature became available July 1, 2017. Since then, the micro-survey tool has been used on various occasions to acquire user information to support the platform's behavior engine and to validate backend predictive algorithms. From July 1, 2017, through September 30, 2017, Metropia™ conducted its first wave of micro-surveys to better understand user socio-demographic characteristics.

According to Metropia™, it sought to create a rewards platform to encourage and sustain desired behaviors that reduce congestion, including shifting driving out of the heaviest peak periods and encouraging carpooling. Metropia's™ effort preceded the most recent inquiry of the Federal Highway Administration (FHWA) into app-based carpooling incentives, and, as a result, did not test many questions related to the impacts of incentive changes on behavior modification.

Metropia's™ analysis relied on data collected responses from the micro-survey along with the participants' mobility option choices (drive alone or carpool) from January 2016 through December 2017 coupled with passively tracked trip data. Figure 8 illustrates the pertinent timeframes, while table 11 summarizes the number of micro-survey participants by the market.

Figure 8. Graph. Driving Up Occupancy timeframes and user data definition.
(Source: Metropia™)

Out of the 1,260 users, 644 had a completeness rate of all the survey questions of greater than 40 percent. Metropia™ imputed missing data, where possible, in conducting analysis for this study.⁸ The basic concept behind imputing data is that the number of users participating in the survey and the number of questions administered in the survey form a matrix that is generally sparse (i.e., not all users respond to all questions). The Multivariate Imputation by Chained Equations (MICE) process⁹ was utilized by Metropia™ in this study to complete the sparse matrix, and the micro-survey data were used as explanatory variables. After Metropia™ completed the imputation process, data pertaining to the travel behavior (mobility option and reward points), trip characteristics (travel time and distance), and day of the week (weekday versus weekend) were attached by Metropia™ to its 644 users, resulting in a dataset of 16,897 observations (9,224 for weekday and 7,673 for weekend) based on an average of 14.3 and 11.9 months of activity for each user on weekday and weekend, respectively.

Based on the mobility option selected, Metropia™ mapped each user to one of the following four roles for each trip and computed the proportion of each role for over an observation period:

1. Drive alone: always a driver; never in a carpool.
2. Carpool passenger only: always a passenger; never been a driver.
3. Carpool driver only: always a carpool driver.
4. Carpool both: has been a carpool driver and passenger.

Metropia's™ data support both cross-sectional and longitudinal analysis and thus a three-tiered framework was developed, as illustrated in figure 9.

Figure 9. Diagram. Tiered analysis framework.
(Source: Metropia™)

Metropia's™ dataset allowed a breakdown of results by a host of demographic and trip characteristics. As the focus of the analysis here is on the broader impacts of the type of strategy that Metropia™ deployed, some detail available from Metropia™ was less relevant to the findings focused on in this document.¹⁰ One area of interest, though, was whether those with growing usage over time were demographically different and/or exhibited other behaviors different from those showing declining participation. This was of interest as those with growing usage are more likely going to be long-term participants in app-based carpooling than those with declining usage.

For the period between January 2016 and December 2017, over 85 percent of the users had used Metropia's™ platform in the last 6 to 15 months. The time-series data associated with the trend analysis reflected monthly carpool utilization for those users who had used the platform for at least 6 months. Carpool use in the Metropia™ app varies by time of day (peak versus non-peak), day of the week (weekday versus weekend), activity type, and user socio-demographic characteristics. Furthermore, segmenting by commute purpose (i.e., travel primarily associated with work or school); the user's familiarity with a specific market; age; education level; and years of driving appeared to have significant explanatory value in attempting to understand the differences (behavior) in carpool-passenger and drive-alone modes.

Metropia™ stated that its primary goal was to create and deploy a behavioral engine that establishes and maintains a new mobility habit for users, while being financially sustainable for partnering agencies. Table 12 summarizes the distribution of the selected users by three markets.

For these markets, Metropia™ divided carpool users into groups that exhibited an increasing trend, decreasing trend, or no trend in utilization, based on the Mann-Kendall test.^11,12 The Mann-Kendall trend test is a non-parametric test to detect significant trends in time series and requires that trend to be monotonic, meaning that for an increasing trend observation, observation y needs to be higher than observation y – 1. Based on the Mann-Kendall test of the 453 users, 308 users or 68 percent had no trend, 93 users or 21 percent had an increasing trend, and 52 users or 11 percent had a decreasing trend. Of the 145 users who had a trend, 24 percent were from Austin, 44 percent from Tucson, and 32 percent from El Paso. Figure 10 illustrates the temporal carpool use for these users, providing the following observations:

• Increased carpool use over time among more than 50 percent of the users for whom a trend has been identified.
• Carpool use is higher for the decreasing trend initially, indicating that some users potentially are interested in the reward points only and jump on the opportunity, but later their interest diminishes as the reward points decrease.
• Carpooling has a slower start among users with an increasing trend, potentially indicating that it takes some time for the users to fully appreciate carpool as a mobility option. As they get more comfortable, carpool use increases.
• Growth of the increasing trend becomes flat after 9 months, potentially indicating that carpool has reached an equilibrium state for users.

Figure 10. Graph. Temporal carpool use.
(Source: Metropia™)

Metropia™ undertook a time-series analysis to better understand how incentive changes over time affect the carpool use trend (i.e., can carpool behavior be sustained if incentives are reduced over time?). This is an important aspect of the overall policy framework, since it is directly related to the ability of the public agency to support a desired change in travel behavior in a cost-effective manner. It is also tied to the underlying principle of using incentives as a mechanism to break an old habit and support a new habit that can be maintained over time with an affordable cost structure.

Figure 11 illustrates how DUO reward points changed over time for both carpool drivers and passengers, indicating that reward points were reduced about 14.5 percent for drivers and 3.3 percent for passengers during the analysis period for this study. User carpooling levels tended to stay the same or increase, rather than decrease, despite declining rewards, suggesting that Metropia™ has created an award structure that is associated with users sustaining desired behaviors.

Figure 11. Graph. Carpool driver and passenger incentive point trend.
(Source: Metropia™)

Impact of Rewards and Incentives using Examples from Hytch™¹³

Hytch™ provided a range of examples about partner reward structures. Since such partnership arrangements provide the financing that may lead to behavior change, understanding these arrangements and the motivations of partners to enter into them offers insights on the potential to expand app adoption and usage. First, the City of South Bend, Indiana, is using Hytch™ to pay 50 cents per mile to the general community to take qualified people to specific work locations by carpool. Second, an anonymous Fortune 100 partner company is deploying Hytch™ to pay for sharing rides by carpool or vanpool to or from a specific pilot site in San Diego. Third, the City of Spring Hill, Tennessee, contracted to deploy Hytch's™ "corridor rule" to reward citizens when they take longer, but less congested routes within a targeted corridor. Finally, the teledentistry company SmileDirectClub™ contracted with Hytch™ to pay employees and their co-travelers when they carpool to and from specific parking locations with limited parking supply. Their rewards also apply to employees using public transit and other modes.

One feature of Hytch Rewards™ is that the company has arranged to offset carbon emissions from all recorded trips even when monetary rewards to users are not provided. Hytch™ records carpooling trips but does not match carpoolers. Instead, carpools are arranged by individuals, reward partner organizations, or another firm that provides such matching.

In 2018 and 2019, Hytch™ provided rewards to 10,889 individual users, for their over 12 million miles of travel rewarded. During this period, the total average user reward was $23.62, amounting to about 2 cents per rewarded mile.

Over time, Hytch™ reward levels have tended to decrease gradually, largely since reward partners typically seek to find the most cost-effective means to achieve their goals. As can be seen in figure 12, reductions in the number of rewarded miles (which would be contrary to the goals of the reward partners) were not seen until per-mile reward amounts declined to below 2 cents.

Figure 12. Diagram. Total miles per user per month compared to average rewards per mile.
(Source: Middle Tennessee State University Data Science Institute)

According to Hytch™, it is concerned about attrition of app users and interested in finding the lowest cost rewards that sustain desired behavior. Minimizing attrition is important to realizing long-term benefits of applications. Normal attrition rates have hovered at or below 1 percent per month. An attrition rate of 1 percent per month means that half of the participants would have dropped out in 50 months, or that the average participant is, under normal conditions, expected to stay in the system for about four years. But when user rewards have, in some instances, been eliminated entirely for many trips (excluding carbon emission offsets, which have continued), participation drop off is substantial. As can be seen in table 13, and also in figure 13 use fell immediately following a large reduction in reward levels in June 2019. The share of users taking their last trip using the app grew each month after the reduction in rewards, reaching 4.4 percent in November 2019. A high positive correlation of 0.73 was found with the elimination of rewards and permanent disengagement from the app. (This finding was made through Middle Tennessee State University Data Science Institute running a correlation matrix, which indicates the mutual strength of the relationship between two variables, from -1 to +1, known as the coefficient of correlation.)

Figure 13. Percentage of trips with no rewards and last trips per month.
(Source: Middle Tennessee State University Data Science Institute)

Responses to varying reward amounts may be considered a "natural experiment," where the differing conditions faced by participants were due to factors outside of the control of the platform company. Random assignment would be preferable for this research need, since markets, programs, and reward amounts in this natural experiment could vary somewhat, but random assignment would require sufficient planning and budgeting. Despite this note of caution, changes in behavior from the natural experiment, especially within group behavior changes when reward levels have been varied, may be the result of the changes in reward amounts rather than other factors. When examining "within user" data, or changing behavior of individual users over time, the lack of random assignment becomes less of a concern since users are not being compared to others. While, in theory, it is possible that different users might respond differently to changing reward levels, there is no inherent reason that users found to be in one of these three categories (increasing, stable, and decreasing reward levels) would respond differently to changing incentives than would users ending up in other categories.

Data from Hytch™ allows researchers to find correlations between per-mile reward values and user miles that are rewarded. Table 14 is quite instructive in this regard and shows that a reward level of 2 cents per mile appears to yield indistinguishable results in monthly Trips per User from higher reward levels, but substantially better results than for lower reward levels. Monthly average awards of $7.54 for participants receiving 2 cents per mile, as shown below, appear to be a very affordable cost as compared to other transportation investment options (which are not explored as part of this research).

Table 14 shows that higher average per-mile reward levels correspond to average trip lengths that are shorter than with lower reward levels. Hytch™ explained that some reward sponsors cap reward amounts on a per-trip basis, suggesting that those receiving the highest per-mile rewards were taking shorter trips, where no cap is imposed. Where such a cap exists, the expected manifestation of behavior change resulting from a higher per-mile reward structure would be more trips taken by those inclined to take shorter trips, as such trips would be handsomely rewarded. The data here do not show that result.

Another area of inquiry with the Hytch Rewards™ data requested by FHWA was for Hytch™ to report on individual distance per user and trips per user with rewarded miles where reward levels have, over time, declined noticeably, stayed the same, and increased noticeably. Such an inquiry may help formulate the most cost-effective strategies for allocating reward funds and could answer the following FHWA questions:

1. For users offered reward levels that remain constant, how, if at all, do distance per user and trips per user change over time (and can participation duration effects be separated from seasonality effects)?
2. Does starting with higher reward levels lead to higher initial distance per user and trips per user than starting with lower reward levels? If so, is the higher average maintained even when reward levels are cut?
3. For users where reward levels rise, do distance per user and trips per user rise in a corresponding fashion?
4. Does the manner and timing of reward redemption affect behavior? For example, if Hytch™ participants were required to be eligible for a minimum of $10 in rewards prior to redemption, would users become more responsive to incentives as they neared redemption eligibility?

To begin to answer FHWA's questions, Hytch™ reviewed its data and shared a number of products the company generated from its data. The data and related products showed expected seasonal effects (e.g., travel picks up after the winter months). The reduction in reward miles when per-mile rewards dipped below 2 cents was also very apparent. Overall, though, clear patterns were difficult to discern. Future controlled experiments may be able to answer these questions.

Hytch's™ existing data could help guide such controlled experiments. To answer the third question above (i.e., "For users where reward levels rise, do distance per user and trips per user rise in a corresponding fashion?"), Hytch™ has only a very small number of users to whom this condition applied. Nonetheless, the data, shown in the table below, suggest that there might be something worthy of further exploration.

For this sample, table 15 shows that increasing reward levels appear to correlate with increased average miles per user. This is particularly apparent beginning in week 12 of 2019, or two weeks after average reward per mile jumped (and continued edging up in subsequent weeks). The rewards increased in weeks where seasonal driving might have increased, which might account for a bit of the shift, but results from this small sample were far higher than would likely be due to seasonality. Nevertheless, caution in interpreting the data is advised. The number of week-over-week users varied, sometimes substantially. After some time, reward miles declined despite per-mile reward levels remaining high, so it is an open question as to whether behavior changes are sustained. These data suggest that providing increasing reward levels to a greater number of participants (randomly assigned) in a future study could make for a worthy test even though, as noted earlier, higher per-mile reward levels were not correlated with higher numbers of reward trips across the entire study population.