A user-centric multi-criteria evaluation of mobility solutions for people with psychiatric disabilities using data-driven persona

Abstract

13 This study aims at identifying which mobility solutions are most suitable for addressing the particular needs 14 of people with psychiatric disabilities. We developed a 3-step evaluation framework: (1) development of a 15 dataset (operator data and qualitative monitoring), (2) analysis of supply and demand of mobility solutions 16 for people with psychiatric disabilities, and (3) processing of a multi-criteria decision analysis. This 17 framework is used to analyze the ‘PAM’ service, an on-demand mobility solution specifically intended for 18 people with disabilities in Seine-Saint-Denis (Paris region, France). The results highlight the strong 19 performance of on-demand transport services in supporting routine mobility needs, while underscoring the 20 importance of robustness to preference and evaluation uncertainty in service assessment. From a policy 21 perspective, the framework provides local authorities with evidence-based recommendations to compare 22 mobility solutions and support the design or renewal of services better aligned with the specific needs of 23 people with psychiatric disabilities.

26 Keywords: People with disabilities; On-demand mobility; Multi-criteria decision analysis; Latent class 27 analysis; Persona; Travel behavior.

28 1. INTRODUCTION

29 Climate disruption and rapidly changing societal conditions accelerate transformations in mobility practices 30 and intensifying the need for equitable transport systems. Equity is understood as the fair distribution of 31 transportation benefits (e.g., access, speed, comfort) and costs (e.g., fares, time, pollution exposure) across 32 population groups (Bruzzone et al., 2023). Equity is a central objective in transport planning and is 33 particularly critical for people with psychiatric disabilities. In France, 8.7% of the population aged 15 or 34 older report at least one severe limitation in sensory, mental, cognitive, or psychological functions (DREES, 35 2023). Yet, despite growing interest in disability and transport, most studies focus on identifying needs and 36 barriers rather than evaluating how specific mobility solutions perform for this user group. 37 To compare mobility solutions, multi-criteria decision analysis (MCDA) is widely used for transport 38 infrastructure and services (Broniewicz and Ogrodnik, 2025). However, it has rarely been applied to 39 disability-related mobility. At the same time, user segmentation tools such as personas are increasingly 40 used in transport to represent heterogeneous travel behaviors and guide service design (Vallet et al., 2020). 41 Personas are often created through clustering methods (Salminen et al., 2020) such as Latent Class Analysis 42 (LCA) (Vermunt and Magidson, 2002). 43 This study combines LCA persona-based segmentation with MCDA to evaluate mobility services for 44 people with psychiatric disabilities using operational data. We address key research gaps: (1) Proposing a 45 user-centric multi-criteria decision analysis focusing on mobility solutions for people with psychiatric 46 disabilities (2) Enriching the MCDA with persona definition using both real operating data and insights 47 from academic literature.

48 2. METHODOLOGY

49 The general framework for evaluating mobility solutions for people with psychiatric disabilities is presented 50 in Figure 1.

1 Interviews with MetaScientific and PAM service the local observatory of Dataset grey literature operating data authority* mobility development INSEE data solutions

2 Latent class cluster analysis List of mobility solutions for Supply and people with psychiatric Demand disabilities* Analysis Creation of the persona*

Evaluation criteria* Multi-criteria Evaluation framework Decision Analysis Weighting scheme*

SMAA-PROMETHEE ranking of mobility solutions *with the local authority contributions

53 Figure 1 : a 3-step methodological framework

55 We adopt a 3-step methodological framework, First, the dataset is built (1), then we provide a supply and 56 demand analysis (2) to finally build the MCDA framework (3). The local authority contracts the operation 57 of the service to a dedicated operator. The service operator data are used to process a LCA in order to 58 identify trip categories (33197 trips for the year 2018). Methodological reviews highlight sensitivity to 59 variable selection and the need to complement quantitative outputs with qualitative insights when using 60 clustering methods (Salminen et al., 2020). The LCA is then completed with qualitative information about 61 people with psychiatric disabilities‘needs gathered from scientific and grey literature and interviews with 62 the local authority to build a user persona. Doing this offers a probabilistic and user-centric approach to 63 identify behavioral segments from travel patterns. In parallel, the meta-observatory of new mobility 64 solutions (Le Boennec et al., 2019) allowed us to select a first list of criteria and mobility solutions to be 65 discussed with the local authority. 66 By analyzing the persona, evaluation criteria from the meta-observatory were adapted based on user needs, 67 the operating data of the PAM service, and the objectives of the local authority (Fig. 1). Criteria weights 68 were defined using verbal statements method on a three-level scale (“not a priority”, “important”, “very 69 important”). Mobility solutions were then evaluated on a five-point scale, and weights and scores were 70 subsequently discussed and consolidated during a workshop. This evaluation frame is then used to process 71 the MCDA with the SMAA-PROMETHEE methods. While PROMETHEE–GAIA (Brans et al., 1986) 72 provides a deterministic representation of preferences based on fixed weights, Stochastic Multiobjective 73 Acceptability Analysis (SMAA) extends this framework by incorporating uncertainty to ensure the 74 robustness of the evaluation (Corrente et al., 2014). In SMAA, weights and, when relevant, evaluations are 75 treated as random elements.

76 This stochastic formulation thus systematically accounts for uncertainty in weights and/or evaluations and 77 yields a statistically grounded assessment of robustness and preference sensitivity for PROMETHEE II 78 rankings. We defined 6 alternatives scenarios of uncertainties (Table 1).

81 Table 1. Scenario description

Scenario Inputs Main purpose Weight Score noise2 name perturbed priority rule1 S1 Weight Baseline with the tightest weight constraints: Strict None only checks whether the winner holds when priorities are strictly non-overlapping

S2 Weights Baseline with a permissive weight space: tests Flexible None only sensitivity to imprecise priorities without score noise

S3 Weights + Assesses joint uncertainty when priorities are Strict ±1 (15% / 70% / scores strictly ordered, showing whether strict 15%) priorities buffer moderate score noise

S4 Weights + Stress test under strict priorities: persistence Strict ±1 (25% / 50% / scores of leaders here signals strong robustness to 25%) score noise

S5 Weights + Quantifies the combined impact of preference Flexible ±1 (15% / 70% / scores imprecision and realistic rating noise, 15%) indicating shortlist stability

S6 Weights + Maximal uncertainty (widest weight space + Flexible ±1 (25% / 50% / scores strongest score noise) 25%)

83 Weights priority: Strict: total weight of “Not a priority” < “Important” < “Very important”. Flexible: 84 each “Important” criterion has a higher weight than any “Not a priority” criterion, and each “Very 85 important” criterion has a higher weight than any “Important” criterion.

86 Score noise: Scores vary by -1 or +1. For scenario S3, 15% are lower by 1, 70% unchanged, 15% increase 87 by 1.

89 SMAA-PROMETHEE was performed using Rstudio software (R version: R- 4.3.3, 2025).

92 3. RESULTS

93 3.1 Mobility solutions to assess

94 In addition to conventional modes of transport (i.e. public transport, active modes), we used the meta95 observatory of new mobility solutions, completed with an international benchmark and discussions with 96 the local authority. The final set of solutions comprises public transport, walking, the PAM service, assisted 97 use of public transport, and pedicabs.

98 Within the PAM offer, a distinction is made between grouped and non-grouped trips, depending on whether 99 ridesharing is imposed on users. Variants also differ by powertrain (thermal or electric). In addition, the 100 evaluation includes an enhanced dynamic electric PAM configuration based on optimization algorithms, 101 although this option has not yet been implemented in practice.

102 3.2 LCA results and persona creation

103 A LCA was conducted and revealed four distinct mobility patterns. The resulting classes are presented in 104 Figure 2.:

108 Figure 2: LCA results – class identification

110 The results suggest that the dominant operational pattern in the dataset (Class 1) corresponds to high111 volume, schedulable sharing on short links for recurrent activities. The persona creation was then based on 112 this class. 113 Since no information was given about user profiles in the data, we completed this quantitative analysis of 114 trip segments by a qualitative analysis of the particular needs of people with psychiatric disabilities. 115 Three recurring mobility difficulties specific to people with psychiatric disabilities were identified through 116 interviews with the local authority and academic and grey literature (see Fig. 1). First, unplanned route 117 disruptions can generate confusion and disorientation, increasing the risk of unsuccessful trips. Second, 118 reading and interpreting travel information (such as signage, schedules, or maps) may hinder navigation 119 and create additional stress during journeys. Third, purchasing tickets from automated machines can be 120 challenging. Public transport may also involve uncertainty when boarding or alighting, waiting at stations, 121 or interacting with other passengers. 122 Based on these findings, a persona (Figure 3) was developed by combining (1) travel behaviors derived 123 from LCA results, (2) the mobility objectives and challenges identified, and (3) perceptions of the PAM 124 service.

Jeanne About Travel characteristics Jeanne is not aware of alternative Travel reasons: Activity (going to a day center) solutions to the PAM system and Travel distance from home: 7km has many prejudices about the Regular pattern: 4 working days per week, every use of public transportation, week. especially regarding the behavior Time: Departure at 8:30am, return at 6pm of other passengers towards her Grouping : Almost every time

Objectives and challenges How the PAM offer is perceived Jeanne has difficulty using digital tools, The PAM service is perceived as very useful for so she relies on the telephone making the trip due to the door-to-door service, Age : 71 years old reservation center. She is independent but the high rate of shared trips leads to Destination: Daycare center in making her trips as long as she significant delays Personality : No personality disorders. knows the route. The ability to plan Mobility challenges encountered Intellectual disabilities making it difficult to be her trip is also essential because she independent in making new routes. Fear of does not tolerate unexpected changes - Reading and understanding information in others. and will not use dynamic transport order to navigate Learning abilities: Can learn new routes with offers if they do not also offer planned - Purchasing a transport ticket from a vending guidance but is not willing to do so voluntarily routes machine - Knowing how to react to disruptions in the transport service used

127 Figure 3. Illustration of a persona created for people with psychiatric disabilities.

128 3.3 Evaluation criteria definition, weighting and scoring

129 Based on this persona and the meta-observatory framework, evaluation criteria were defined in line with 130 user needs, PAM operating data, and the objectives of the local authority (Fig. 1). Table 2 summarizes the 131 criteria, weights, and performance scores used in MCDA.