Park-and-Ride Choice Behavior in a Multimodal Network with Overlapping Routes

Abstract

This study estimates a discrete choice model from on-board survey data for park-and-ride (PNR) user’s station choice, and finds that users prefer a higher proportion of their trip time spent on transit than in a car. No evidence is found that users choose the PNR location that minimizes their overall travel time, but there is strong evidence that PNR locations closer to the direction of their destination are preferred. No significant relationship is found between PNR choice, user age, income, or gender. PNR facility amenities are found to be insignificant across all models tested, and the rail and bus rapid transit service provide a significant benefit to PNR user utility when compared to the express bus service. This study contributes to existing literature by accounting for route overlap among alternatives using a path size factor in the logit model estimation. A nested logit model for different transit types is estimated, but is surpassed in predictive ability by a multinomial and mixed logit model. Ultimately, the models are found to predict transit route choice more accurately than PNR location choice, improving on the PNR choice prediction by about 11%. Similarly, choice set alternatives that overlap with a user’s observed route are found to generally have higher choice probabilities than non-overlapping alternatives.

A literature review on park-and-rides

“…Key findings from the demand model studies are described here. Four studies were on mode choice models (Cornejo et al, 2014;Fan et al, 2016;Karamychev & van Reeven, 2011;Zhang, 2014), two studies were on station choice models (Pang & Khani, 2018;Webb & Khani, 2020), and one study was on preferred choice of scenarios (Cao & Duncan, 2019). The number of boardings was predicted by one study (Niles & Pogodzinski, 2016).…”

“…In the station choice models, travel time and trip-related information were used to predict PNR station choices by commuters. The studies used discrete choice models as their methods, such as binary logit (Cornejo et al, 2014), mixed logit (Pang & Khani, 2018), MNL (Cao & Duncan, 2019;Fan et al, 2016;Karamychev & van Reeven, 2011;Zhang, 2014), nested logit (NL) (Webb & Khani, 2020), and Poisson regression models (Niles & Pogodzinski, 2016).…”

“…But many other factors have the potential to affect mode choice. Future research can investigate some of the factors, which include street network data like speed limits, congestion, the monetary cost associated with other modes, multiple transit paths (Webb & Khani, 2020), payment information of the respondents, and transfer time between modes. Third, it was observed that machine learning algorithms have recently been introduced in the literature on travel mode choice; therefore, the application of machine learning techniques can be explored in predicting PNR demand.…”

“…Demand Model studies developed PNR demand models for their cities based on user surveys and travel time information. For future research, the association of mode choice with more street network data like speed limits, congestion, the monetary cost associated with other modes, multiple transit paths (Webb & Khani, 2020), payment information of the respondents, and transfer time can be explored. Alternative data collection methods like GPS traces, transit fare card data (Pang & Khani, 2018), and electronic surveys can be used to increase sample sizes.…”

Marginal Impacts of Park-and-Ride Facilities in the Twin Cities Region of the US

“…Some have applied spatial optimization methodologies to determine where P&R facilities are best located (9)(10)(11). Recent studies have examined P&R station choice (12) and willingness to walk additional distances between the parking facility and the transit stop (13). Others have examined the feasibility and impact of pricing at P&R facilities (14).…”