How bad is time variability for users in mobility services?

This paper establishes theoretical upper bounds on the ratio of the cost of time variability to the cost of time within an expected utility framework, demonstrating that for quadratic utility the ratio is at most half the squared coefficient of variation, thereby providing a data-light benchmark for assessing the economic significance of reliability improvements in mobility services.

The Gist

Imagine you are waiting for a bus. Sometimes it arrives exactly on time. Other times, it's 5 minutes late. Sometimes, due to a traffic accident, it's 30 minutes late.

This paper asks a very simple but profound question: How much does this "unpredictability" actually hurt us, and is there a limit to how bad it can get?

The authors, a team of researchers from Singapore and China, wanted to find a "worst-case scenario" rule of thumb. They wanted to know: If I am willing to pay $10 for a 10-minute ride, how much extra would I ever possibly pay just to make sure that ride is perfectly on time every single time?

Here is the breakdown of their findings, explained with everyday analogies.

1. The Problem: The "Anxiety Tax"

We all hate waiting. But we hate uncertainty even more.

• Scenario A: You know the bus takes exactly 10 minutes. You leave your house 10 minutes early. You are relaxed.
• Scenario B: The bus usually takes 10 minutes, but sometimes it takes 15, and rarely 30. To be safe, you leave 20 minutes early.

That extra 10 minutes you spend waiting at home is your "cost of time variability." You are paying with your time to buy peace of mind. The paper tries to figure out the maximum price tag on that peace of mind.

2. The "1.5x" Rule (The Big Discovery)

The researchers found a surprisingly simple "ceiling" or limit to how much time variability can cost you.

The Analogy: Think of a trip as a pizza.

• The base cost is the time it takes to get there if everything goes perfectly (the cheese and sauce).
• The variability cost is the extra toppings you add because you are worried the pizza might arrive cold or late.

The paper proves that the "anxiety toppings" can never cost you more than half the price of the pizza itself.

• If a perfect, on-time trip costs you $10 (in time and money), the worst-case scenario for a risky, unpredictable trip is that it costs you $15.
• It will never cost you $20 or $100. The chaos has a limit.

Why does this matter?
City planners and bus companies often argue, "We need to spend millions to make the bus 100% reliable!"
This paper says: "Hold on. Even if you fix the bus completely, the most you are saving the passengers is only 50% more than the cost of the trip itself." This helps governments decide if a project is worth the money. If the project costs more than that 50% savings, it might not be a good investment.

3. The "Dice Roll" vs. The "Coin Flip"

The paper looks at two different types of "bad luck":

• The Coin Flip (Poisson Process): This is like a bus that arrives randomly. Sometimes it's early, sometimes late, but the average is steady. In this common scenario, the "anxiety tax" is exactly 50% of the trip cost.
• The Loaded Dice (General Case): What if the bus is usually on time, but once a month it gets stuck in a massive, unpredictable traffic jam? The "badness" of the variability depends on how "spread out" the delays are.
  • If the delays are small and frequent, the cost is low.
  • If the delays are huge and rare (like a once-in-a-decade disaster), the cost goes up, but it is still capped by the formula: 0.5 × (How spread out the delays are)².

4. Who Are You? (The Risk Personality)

The paper also realizes that not everyone reacts to uncertainty the same way. They use three "personality types" to explain how people value reliability:

1. The "Time Saver" (Risk Neutral): This person just wants the trip to be short. They don't care if it's 10 minutes or 12 minutes, as long as the average is low. They are willing to take the gamble.
2. The "Worrywart" (Risk Averse): This person hates the spread. They hate that the bus might be 20 minutes late. They will pay extra to make the trip consistent, even if the average time doesn't change.
3. The "Disaster Prepper" (Prudent): This person is terrified of the "worst-case" scenario (the 30-minute delay). They are willing to pay a huge premium just to avoid that one rare, terrible event.

The paper provides a "benchmark" to help planners guess which type of person they are dealing with. If the users are mostly "Worrywarts," reliability is worth a lot. If they are "Time Savers," reliability is worth less.

5. The "Data-Light" Superpower

Usually, to figure out how much people value reliability, you have to survey thousands of people, asking complex questions like, "How much would you pay to reduce your delay variance by 10%?"

This paper says: You don't always need that much data.
Because they found a theoretical "ceiling" (the 1.5x rule), planners can make smart decisions early on.

• Example: "We are thinking of building a new dedicated bus lane. It will cost $1 million. Even if we assume the worst-case scenario where people hate uncertainty, the maximum value they get is only $500,000. Don't build it."

Summary

This paper is a safety net for decision-makers. It tells us that while waiting for an unpredictable bus is annoying and costs us time, it is not a bottomless pit of misery.

There is a mathematical limit to how much time variability can hurt us.

• The Rule: The cost of a chaotic trip is at most 1.5 times the cost of a perfect trip.
• The Takeaway: We can stop worrying about the "worst possible world" and start making practical decisions based on the fact that reliability has a price tag, and that price tag has a maximum limit.

It turns a complex economic nightmare into a simple rule of thumb: Uncertainty is expensive, but it's never that expensive.

Technical Summary

Here is a detailed technical summary of the paper "How bad is time variability for users in mobility services?" by Zang et al.

1. Problem Statement

Time variability (unpredictability in travel, waiting, or service times) is a pervasive feature of mobility services that causes significant welfare loss. While extensive literature exists on quantifying the Cost of Time Variability (COTV) (often referred to as the Value of Reliability), a critical theoretical gap remains: there is no established benchmark for the worst-case economic impact of time variability relative to the cost of time itself.

Without a theoretical upper bound, estimated COTVs lack a principled scale to determine if they are economically meaningful or negligible. This limitation hinders early-stage decision-making in transport appraisal, service design, and pricing, where detailed valuation data is often unavailable. The paper aims to answer: How bad can time variability be for users in the worst case?

2. Methodology

The authors develop a theoretical framework based on Expected Utility (EU) theory and extend it to Non-Expected Utility (Non-EU) models.

• Core Framework:

  • Utility Function: Models user utility $u(t)$ as a non-increasing function of service time $t$.
  • Risk Attitudes: Utilizes Relative Risk Aversion (RRA) (2nd-order preference) and Relative Prudence (RP) (3rd-order preference) to characterize user behavior.
  • Variability Premium ($\pi$): Defines the additional time a user is willing to "pay" (in time) to eliminate variability, satisfying $E[u(t)] = u(\mu + \pi)$.
  • Cost Definitions:
    • Cost of Time (COT): The monetary cost of the mean service time ($\mu$).
    • Cost of Time Variability (COTV): The additional monetary cost induced by variability.
  • Key Metric: The ratio $\rho = \text{COTV} / \text{COT}$, which normalizes the severity of variability against the baseline time cost.

• Analytical Approach:

  • Uses Taylor series expansions to approximate the variability premium and costs.
  • Derives theoretical upper bounds for the ratio $\rho$ under different utility assumptions (Quadratic vs. General Differentiable).
  • Extends the analysis to Dual Theory (DT) and Rank-Dependent Utility (RDU) to account for probability weighting and framing effects, introducing the concept of Dual Moments (measuring variability in the probability plane rather than the payoff plane).

3. Key Contributions

1. Theoretical Upper Bounds: Establishes the first theoretical "worst-case" benchmarks for the ratio of COTV to COT.
2. Quadratic Utility & Poisson Process: Demonstrates that under a quadratic utility function and a Poisson service process (common in queuing), the COTV is at most 50% of the COT. Consequently, the total cost under variability is bounded by 1.5 times the cost of a deterministic service.
3. General Bound via Coefficient of Variation (CV): For general service processes with quadratic utility, the ratio is bounded by $\rho \leq \frac{1}{2}CV^2$, where $CV$ is the coefficient of variation. This provides a data-light metric for early-stage screening.
4. Risk Preference Benchmarks: Identifies specific benchmark values for RRA and RP:
   • RRA = 1: The threshold where users value variance reduction equally to mean time reduction.
   • RP = 2: The threshold where users prefer reducing skewness (extreme delays) over reducing variance, even if it increases variance.
5. Robustness to Non-EU Models: Shows that the structural relationship between variability costs and risk preferences holds even in Dual Theory and Rank-Dependent Utility frameworks, though the specific measures of variability shift from variance (primal moments) to dual moments.

4. Key Results

A. Special Case: Quadratic Utility & Poisson Process

• Result: $\rho \leq 0.5$.
• Implication: The total cost of a mobility service with stochastic time is at most 1.5 times that of an identical deterministic service.
• Significance: This aligns with empirical observations of the "congestion multiplier" (often cited as 1.5) in transport economics, providing a theoretical justification for this heuristic.

B. General Case: Quadratic Utility (Arbitrary Process)

• Result: $\rho \leq \frac{1}{2}CV^2$.
• Implication: The severity of variability costs scales linearly with the square of the coefficient of variation. If variability is small relative to the mean, the cost is negligible; if variability is high, the cost grows quadratically.

C. General Case: Differentiable Utility (RRA & RP)

• Result: The ratio $\rho$ depends on three factors: $CV$, RRA, and RP.
  $$\rho \approx \frac{RRA \cdot CV^2}{2 + RRA \cdot RP \cdot CV^2}$$
• Marginal Benefits: For risk-averse and prudent users, the marginal benefit of reducing time variability is bounded by half the marginal benefit of reducing mean time ($\eta \leq 0.5$). This indicates diminishing returns on reliability improvements as risk preferences intensify.

D. Non-EU Extensions (Dual Theory & RDU)

• Dual Theory: The ratio depends on the second dual moment (related to the expected maximum of two draws) and the curvature of the probability weighting function.
• RDU: The ratio combines primal moments (variance) and dual moments, showing that both the utility function (risk aversion) and the probability weighting function (perception of risk) jointly determine the cost.

5. Significance and Practical Implications

• Early-Stage Decision Making: The derived bounds ($\rho \leq 0.5$ or $\rho \leq 0.5 CV^2$) serve as a data-light benchmark. Planners can screen and prioritize reliability improvement projects (e.g., adding lanes, dedicated bus lanes) without needing complex, user-specific survey data. If a project's estimated variability cost exceeds these theoretical bounds, it may be economically unjustifiable.
• Pricing and Design: The results provide a principled upper limit on users' willingness to pay (WTP) for reliability. This informs the pricing strategies for premium reliability-oriented services (e.g., high-speed rail vs. regular bus).
• User Segmentation: The benchmark values for RRA (1) and RP (2) allow policymakers to classify users based on their sensitivity to mean time, variance, and skewness (extreme delays), aiding in targeted service design.
• Theoretical Unification: The paper bridges the gap between empirical valuation studies and theoretical limits, clarifying that while variability costs are real, they are inherently bounded and predictable based on the degree of variability and user risk attitudes.

In summary, the paper transforms the understanding of time variability from a purely empirical estimation problem into a structured theoretical framework, offering clear, interpretable bounds that guide transport appraisal and service design under uncertainty.