🔬 physics

Scaling law of individual urban tour behavior

By analyzing Foursquare check-in data, this study reveals that urban tour lengths follow a truncated power-law distribution and proposes a tour terminate-continue model that successfully reproduces this pattern along with other fundamental scaling laws of human mobility.

Original authors: Xu-Jie Lin, Yitao Yang, Wei-Peng Nie, Xiao-Yong Yan

Published 2026-03-09

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Xu-Jie Lin, Yitao Yang, Wei-Peng Nie, Xiao-Yong Yan

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine your daily life as a series of little loops. You wake up at home (your "base"), grab coffee, drop off the kids, go to work, maybe stop at the gym, and then head back home. In the world of data science, this entire loop is called a "tour."

For a long time, scientists studying how people move around cities were like cartographers who only drew the straight lines between points. They knew how far people traveled and how often they visited specific spots, but they completely ignored the shape of the loops themselves. They didn't ask: How many stops do people usually make before heading home? Is it usually one stop, or do they go on a wild 10-stop adventure?

This paper, "Scaling law of individual urban tour behavior," is like a detective story that finally solves the mystery of these loops.

The Big Discovery: The "Truncated Power-Law"

The researchers looked at millions of check-in records from Foursquare (like a digital diary of where people go) in New York and Los Angeles. They found a surprising pattern:

Most people's tours are short and sweet. You go to work, maybe grab lunch, and go home. But occasionally, someone goes on a massive tour: work, gym, grocery store, dry cleaner, dinner, and then home.

The distribution of these tour lengths follows a "Truncated Power-Law."

The Analogy: Imagine a staircase where the first few steps are huge, and the steps get smaller and smaller as you go up. Most people are on the bottom steps (short tours). A few people are on the middle steps. Almost no one reaches the very top (super long tours).
The "Truncated" part: The staircase doesn't go on forever. It gets cut off. Why? Because you can't stay out for 20 stops; you get tired, the city closes, or you just have to go home.

The Problem with Old Models

Before this study, scientists used models like the EPR (Exploration and Preferential Return) model.

The Old Model Analogy: Imagine a dog on a leash. The dog runs around, sometimes sniffing a new bush (exploring), sometimes running back to a spot it liked before (preferential return).
The Flaw: These old models were great at predicting where the dog went, but they were terrible at predicting when the dog would stop and go home. They didn't have a "stop button." They just kept the dog running until the simulation ended, which didn't match real human behavior.

The New Solution: The "Tour Terminate-Continue" (TTC) Model

The authors built a new model called TTC. Think of this model as a decision-making robot that mimics a human's internal monologue at every stop.

At every single stop you make, the robot asks two questions:

"Should I go home now?" (The Terminate Decision)
- The model discovered a secret rule: The longer your tour gets, the less likely you are to stop.
- Analogy: If you've only been out for 10 minutes, you might think, "I'm bored, let's go home." But if you've already visited three places, you think, "I'm already this far out, I might as well grab a coffee too!"
- The model uses a math formula to say: "As the tour gets longer, the chance of going home drops." This explains why short tours are common, but long ones happen too.
"Where do I go next?" (The Continue Decision)
- If the robot decides not to go home, it has to pick a new spot.
- It has a "curiosity meter" (parameter $\theta$ ).
- Analogy: Sometimes you want to try a brand new restaurant (exploring). Sometimes you just want to go to your favorite taco place you've been to a hundred times (revisiting). The model balances these two desires perfectly.

Why This Matters (The "So What?")

This isn't just about math; it's about understanding how our cities work.

For City Planners: If we know people tend to stop after 2-3 errands, we can design neighborhoods where schools, shops, and parks are close together. This encourages "efficient tours" where people can do everything in one loop without driving across the whole city.
For Delivery Trucks: Trucks don't just drive randomly; they do tours too. If we understand the "stop button" logic, logistics companies can plan better routes, using fewer trucks and saving fuel.
For Animal Behavior: Interestingly, this same logic applies to animals! A squirrel foraging for nuts also has a "tour" (leave the tree, find nuts, return). This model might help biologists understand how animals decide when they've gathered enough food to go home.

The Bottom Line

This paper is a breakthrough because it stopped looking at human movement as a random walk and started looking at it as a series of intentional loops.

By realizing that people get "locked in" to a tour the longer they are out, the researchers created a model that can predict not just where we go, but how long our adventures last. It's like finally understanding the rhythm of the city's heartbeat, rather than just counting the beats.

1. Problem Statement

While individual human mobility has been extensively studied, existing models often overlook the tour behavior mechanism. A "tour" is defined as a chain of trips starting and ending at a fixed base location (e.g., home) with intermediate stops.

The Gap: Previous models (e.g., EPR, SHMU) successfully reproduce statistical laws like Zipf's law (location frequency) and Heaps' law (number of visited locations) but fail to accurately reproduce the distribution of tour length (the number of intermediate stops before returning home).
The Question: What is the statistical distribution of urban tour lengths, and what underlying mechanism governs the decision to terminate a tour and return to the base?

2. Methodology

A. Empirical Data Analysis

Dataset: The authors analyzed Foursquare check-in data from New York and Los Angeles, comprising 255,204 trips from 15,499 users.
Processing:
- Identified the most frequently visited location for each user as the "base location."
- Extracted tours from trajectories (sequences of trips between base and intermediate stops).
- Analyzed the distribution of tour length ( $l$ ).
Finding: The tour length distribution follows a truncated power-law:
$P(l) \sim l^{-\alpha}e^{-l/\beta}$
where $\alpha$ is the scaling exponent and $\beta$ is the cutoff value. This indicates that while short tours are common, the probability of very long tours drops exponentially due to city size or scheduling constraints.

B. The Proposed Model: Tour Terminate-Continue (TTC)

To reproduce this distribution, the authors proposed the TTC model, which integrates a tour termination mechanism with location selection rules.

Tour Termination Mechanism:
- At each step of a tour, an individual decides whether to terminate (return to base) or continue.
- The probability of termination ( $P_{term}$ ) depends on the current tour length ( $l$ ) following a power-law decay:
  $P_{term} = \rho l^{-\gamma}$
  $P_{cont} = 1 - \rho l^{-\gamma}$
- Here, $\rho$ is the initial termination probability, and $\gamma$ characterizes how the likelihood of stopping decreases as the tour gets longer.
Location Selection Mechanism:
- If the tour continues, the individual chooses between a new location or a previously visited location.
- This is governed by a competition parameter $\theta$ :
  $P_{new} = \frac{\theta}{\theta + \sum m_k}$
  $P_{old, k} = \frac{m_k}{\theta + \sum m_k}$
- $m_k$ represents the visit frequency of location $k$ . This captures the "preferential return" behavior.
Spatial Extension (d-TTC Model):
- To address spatial constraints (distance), the model was extended to d-TTC by incorporating a gravity-based mechanism for selecting new locations:
  $P^{(j)}_{new} \sim \frac{O_i O_j}{d_{ij}^\sigma}$
- Where $O$ is population (attractiveness), $d$ is distance, and $\sigma$ is the distance decay exponent.

3. Key Contributions

Discovery of Universal Scaling Law: The paper establishes that urban tour lengths universally follow a truncated power-law distribution across different cities, a pattern previously neglected in mobility modeling.
Novel Mechanism (TTC Model): It introduces a simple, parameter-efficient model that explicitly simulates the tour generation process (termination vs. continuation), which traditional random walk models (EPR, SHMU) lack.
Unified Framework: The model successfully reproduces four distinct statistical laws simultaneously:
- Tour length distribution (Truncated Power Law).
- Termination probability scaling ( $P_{term} \sim l^{-\gamma}$ ).
- Heaps' Law (relationship between visited locations and trips).
- Zipf's Law (frequency of location visits).
- Radius of Gyration distribution (spatial spread).
Analytical Derivation: The authors derived analytical solutions for the tour length distribution and the growth of visited locations, validating the simulation results mathematically.

4. Results

Model Performance:
- The TTC model achieved a Kolmogorov-Smirnov (KS) distance of 0.022 (NY) and 0.027 (LA) for tour length distribution, significantly outperforming the EPR model (KS $\approx$ 0.15–0.20) and SHMU model (KS $\approx$ 0.35).
- The EPR and SHMU models failed to capture the high frequency of short tours and the rapid decay of long tours because they lack a specific termination mechanism.
Parameter Sensitivity:
- $\rho$ (Initial Termination): Increasing $\rho$ leads to shorter tours, higher return frequency to the base, and a smaller radius of gyration.
- $\gamma$ (Scaling Exponent): Increasing $\gamma$ makes termination less sensitive to tour length, slightly increasing exploration and the number of visited locations.
- $\theta$ (Exploration): Increasing $\theta$ increases the tendency to visit new locations but does not affect the tour length distribution itself.
- $\sigma$ (Distance Decay): Increasing $\sigma$ (stronger distance penalty) reduces the radius of gyration but has negligible effect on tour length or visit frequency patterns.
Generalizability: The model was also tested on heavy truck GPS data in Beijing and Shanghai, where it successfully reproduced tour length distributions and scaling laws, confirming its applicability beyond human mobility.

5. Significance and Implications

Theoretical Advancement: The study bridges the gap between statistical physics models of mobility and transportation science's tour-based frameworks. It proves that the "decision to stop" is a critical, distinct mechanism from the "decision to move."
Urban Planning & Policy:
- Understanding the parameters ( $\rho, \gamma$ ) allows policymakers to simulate how changes in urban layout or service accessibility might encourage longer tours (e.g., combining shopping and commuting), potentially reducing total travel costs and emissions.
- It suggests that improving the accessibility of intermediate stops (schools, shops) can increase the efficiency of daily tours.
Logistics Optimization: The model's application to freight trucks offers a framework for optimizing delivery routes and dynamic matching of trucks to orders, reducing the number of vehicles needed and mitigating environmental impacts.
Broader Applications: The tour-based framework is applicable to animal foraging behavior, suggesting a universal mechanism for "home-return" mobility across biological and social systems.

In conclusion, this paper provides a fundamental shift in modeling human mobility by prioritizing the tour structure over simple trip chains, offering a robust, unified model that explains both the spatial and temporal statistical properties of individual movement.