Global urban visual perception varies across demographics and personalities

TL;DR

The paper addresses how demographics and personality shape urban visual perception across diverse cities using SPECS, a global, demographically balanced dataset of 1,000 participants rating 400 street-view images across 10 indicators. It introduces four new perceptual dimensions (live nearby, walk, cycle, green) alongside the traditional six, and analyzes demographic and location effects with robust statistics, including Welch's ANOVA and post-hoc tests. A key finding is that demographic and personality groups exhibit location-dependent differences, and that machine-learning predictions trained on global datasets exhibit magnitude bias when applied to diverse urban contexts. The work emphasizes the need for locally tuned, human-centric perception models and demonstrates how the new indicators relate to traditional ones, offering actionable insights for context-aware urban planning and design. It also provides open access to SPECS and code, enabling further exploration of demographic and personality factors in urban perception.

Abstract

Understanding people's preferences is crucial for urban planning, yet current approaches often combine responses from multi-cultural populations, obscuring demographic differences and risking amplifying biases. We conducted a largescale urban visual perception survey of streetscapes worldwide using street view imagery, examining how demographics -- including gender, age, income, education, race and ethnicity, and personality traits -- shape perceptions among 1,000 participants with balanced demographics from five countries and 45 nationalities. This dataset, Street Perception Evaluation Considering Socioeconomics (SPECS), reveals demographic- and personality-based differences across six traditional indicators -- safe, lively, wealthy, beautiful, boring, depressing -- and four new ones -- live nearby, walk, cycle, green. Location-based sentiments further shape these preferences. Machine learning models trained on existing global datasets tend to overestimate positive indicators and underestimate negative ones compared to human responses, underscoring the need for local context. Our study aspires to rectify the myopic treatment of street perception, which rarely considers demographics or personality traits.

Figures (5)

• Figure 1: Research gaps and key questions in urban visual perception studies, followed by our methodology to bridge them. The profile survey included demographics (gender, age group, education level, annual household income, and race and ethnicity) and personality traits (extraversion, agreeableness, conscientiousness, neuroticism, openness). The human perception ratings included the traditional six indicators (safe, wealthy, lively, beautiful, depressing, and boring) and new proposed four indicators (live nearby, walk, cycle, and green) for pairwise comparisons. Source of imagery: Mapillary and KartaView contributors. We acknowledge the use of icons from The Noun Project, created by various authors and licensed under CC BY 3.0.
• Figure 2: Statistical difference of perception scores by demographics. Perception Q scores are calculated from ratings by participants in each group for all (without location grouping) and each location. Welch's ANOVA was used for demographic attributes with only two groups and for demographic attributes with more than two groups, we performed the Games-Howell post-hoc test. Minimum sample size $n$ (rated images with at least four pairwise comparison by participants per group) is shown for each demographic profile. Locations with no significant differences in any indicator and demographic groups with fewer than $n$ samples are not shown. Significance thresholds at $p<0.05$.
• Figure 3: Statistical difference of perception scores by personality traits. Perception Q scores are calculated from ratings by participants in each group for all (without location grouping) and each location. We performed the Games-Howell post-hoc test and the minimum sample size $n$ (rated images with at least four pairwise comparison by participants per group) is shown for each personality trait $\leq$Q1 and $\geq$Q3 quartile. Locations with no significant differences in any indicator and personality groups with fewer than $n$ samples are not shown. Significance thresholds at $p<0.05$.
• Figure 4: Street-level images perception Q scores statistical comparisons based on where people are from (top), where SVIs were taken (middle), and rated by people from their own versus other cities (bottom). Welch's ANOVA was used to compared the combined scores by location (All) against the ML model predictions (top) and to compare scores by each location against the ML model predictions (middle). Minimum sample size $n$ (rated images with at least four (top) or 22 (middle) pairwise comparison by participants per location) is shown. Significance threshold at *$p<0.05$.
• Figure 5: Correlation between our four new indicators (rows) and the six predominantly used indicators (columns) across all locations. Pearson correlation R-values where all correlations are significant ($p<0.05$) except for the correlations of the green indicator and the positive indicator wealthy and negative indicators boring and depressing. Minimum sample size $n$ (rated images with at least 22 pairwise comparison by participants per indicator) is shown.