Parici, le déplacement heureux et citoyen

Overview

For the past decade, car occupancy rate[1] in industrial societies has continuously dropped, despite the rise of environmental concern. On top of the environmental challenge, this multiplication of private cars can negatively impact transport efficiency in urban area: In Paris, the road network approach saturation at peak hours.

Carpooling, the sharing of car journeys amongst multiple passengers, could provide an efficient solution to this problem. Recent years of the digital era have seen the rise of ridesharing network to connect drivers with passengers[2]. However, as these social networks don’t process user data in real time, they cannot provide a solution for flexible, one-time rides on a daily basis. As a consequence, these services have been limited to long haul car-pooling.

An alternative technology, taking advantage of GPS navigation devices supported by smartphones, is real-time ridesharing. Despite being advertised as ridesharing, all services created on this principle so far are actually ridesourcing[3] : Uber, the most famous of these companies, outsources ride requests to taxi companies.

Genuine ridesharing on the contrary, requires the driver and the passengers to share a common destination or at least itinerary. Several reasons can explain why no successful ridesharing network has yet been worked out[4]. Two inhibitory factors stand out to us: Comfort, i.e., the unpredictable pleasantness of sharing one’s ride with a stranger, and marginal economy, i.e., the lack of monetary incentive.

I propose to design a car-sharing app to promote on the fly car sharing by subverting these two obstacles. The program is called Parici.

 

Method: modeling person-to-person affinity with generosity

Parici will link car drivers going to their destination and willing to offer a hitch, with hitchhikers who are going in the same direction. The difference with existing carpooling model is that the key incentive will be social experience instead of money.

To avoid freeriding exploitation of the program, the hitcher will nevertheless agree on paying a participation fee beforehand, undisclosed to the driver. But we propose that this money subserve the social experience: after the interaction, the driver will choose to charge the fee or not. This binary decision will be used to build a model of affinity specific to each driver, by training a simple binary classifier to predict the monetary outcome of the ride based on the personality profile of the hitcher. The underlying idea is that the happier the driver, the more likely he will be on waiving the fee.

When joining the program, each person will take a scientific personality test, based for example on the Big Five personality traits model. This will provide an objective measure of their personality in a multidimensional space of personality traits, upon which a supervised learning algorithm will be able to predict outcome of ridesharing.

 

Preliminary analysis

To explore the feasibility of using personality trait scale for this project, we downloaded the result of an online survey [5] using Cattell’s 16 Personality Factors Test with answer. The data comprises answers of 49159 participants to 163 items related to 10 different psychological traits. Data are ordinal with a scale from 1 (I disagree) to 5 (I completely agree).

 

Fig1

 

 

Figure 1 Plot A: Cross-correlation analysis of psychological features

This personality test provides at most 163 features for a model. This is a relatively small set of features. To ensure that this scale contains enough information, we needed to check first that items were not correlated. We thus pooled all participants together and computed the cross-correlation matrix between every pair of items of the questionnaire. We first noticed each trait contained two inversely correlated axes of variation: for most traits, half of the items were positively correlated, and the other half was negatively correlated. (look at the10x10 white-bordered square box diagonal). This is because items assess several correlated lower level factors beneath each global factor.

A more disturbing observation is the existence of correlation between items of different traits (10×10 white-bordered boxes outside of the diagonal), although traits are supposed to have little or no correlation among them, by definition. This suggests the information provided by the items of this particular scale may be redundant, potentially hindering its ability to separate personalities using these items as features.

 

Figure 2 Plot B: Cross-correlation analysis of individual psychological profile

This would especially be the case if people responding to surveys were likely to obtain the same score at psychological tests. We therefore checked the similarity in psychological profile of a sample of individuals, by computing the cross-correlation matrix between 100 randomly sampled subjects. The mean correlation of these 100 individuals in trait space was 0.1658 ± 0.1754 of standard deviation. This estimate thus suggests that variation in psychological traits between individuals is not correlated. This is not so trivial, as the individual determinants underlying these psychological variables are likely to influence social network use, or even self-reporting, potentially biasing the population sampled. This is a difficulty we could be faced with when deploying Parici. On the other hand, this correlation matrix clearly highlights the existence of similarity in psychological trait between some individuals (look for yellow squares on the side of the diagonal).

 

Overall, despite high between-trait correlations, we predict this dataset could be used to cluster psychological profiles in a multidimensional trait space. This sketchy analysis also indicates that the scale of the Cattell’s 16 Personality Factors is sensitive enough, despite the overlap between factors, to measure high (plot B, yellow squares), and low (plot B, blue squares) between subjects. However, the project would benefit from more thorough exploratory analysis of other scales of personality measures to identify the ones with best discriminatory power. It is important to note that we don’t make any assumption regarding compatibility between subjects (data, outside of dating agency field, is non-existent), and thus we are not able to infer a model of compatibility from analysis of the data alone. We will need the results of social interactions between Parici users to build such a model.

 

Additional caveats for the project are:

– The lack of research modeling person-to-person affinity on a psychological scale. Our model will thus have to be initiated to an arbitrary value and be built and updated online, which may hinder its take-off. However, we expect the perspective of participating in a great social experience to be a powerful drive for users to engage in the program.

– Difficulty to obtain reliable personality data from users. Self-reported test results should always be taken with precaution. Interestingly, recent research[6] has shown that the pattern of social network usage can be mined to predict people’s personality. These data suggest that it will be possible to infer user’s personalities transparently, and possibly with better features (i.e., more decorrelated)

– A too weak incentive for people to participate in the program and the critical number of users enrolled in the program for it to compete with alternate mode of transportation. We count on a favorable evolution of the socioeconomic context to solve these two difficulties: Municipalities are setting up increasingly strict regulations for personal car usage, which will foster adoption of the program.

Moreover, it will prepare the urban population to transition to a new habit of motorized transport usage, which in the future will likely not rely on personal ownership of a vehicle, but rather, on sharing a fleet of vehicle.

 

Conclusion

In summary, we will subvert current obstacles faced by carpooling, by introducing a network relying on an exquisite mix of digital facilitation, social engineering and momentum: unpredictability will be part of the experience. On top of being eco-friendly, this network will also promote a type of open sociality that has been lost in our western societies.

 

[1] https://www.eea.europa.eu/data-and-maps/indicators/occupancy-rates-of-passenger-vehicles/occupancy-rates-of-passenger-vehicles

[2] http://tech.eu/features/481/ride-sharing-europe-carpooling-blablacar/

[3] https://www.its.dot.gov/itspac/Dec2014/RidesourcingWhitePaper_Nov2014.pdf

[4] http://dynamicridesharing.org/inhibitors.php

[5] https://openpsychometrics.org/_rawdata/

[6] http://www.sciencedirect.com/science/article/pii/S002200001300072X

Advertisements