Probability · April 29, 2026

Ride-sharing: What the Wait Time Tells You

Advanced Probability
Time: 00:00

An 8-minute wait sounds like evidence of surge pricing. The math says otherwise: it barely shifts your estimate from the prior. Using exponential wait-time distributions, Bayes' theorem, and simple cost-benefit analysis, this puzzle explores why observed data can be surprisingly uninformative.

It’s 8:47 am. You open your ride-share app and see an estimated wait of 8 minutes. No surge badge is visible — but you know the app sometimes withholds that information until after you confirm. Meanwhile, the subway entrance is half a block away.

You have exactly the kind of data a careful rider collects. Over hundreds of trips during your morning commute window, you’ve established the following:

  • Surge probability. Surge pricing is active on 30% of trips during this window.
  • Wait time without surge. Driver arrivals follow an exponential distribution with mean 5 minutes (λ = 1/5 per minute).
  • Wait time with surge. Fewer drivers respond to non-surge-adjusted demand; arrivals follow an exponential with mean 12 minutes (λ = 1/12 per minute).
  • Fares. A normal ride costs $9; a surge ride costs $18.
  • Your alternative. The subway costs $3 and gets you there in 20 minutes. You value your time at $0.50 per minute.

The challenge

The app quotes an 8-minute wait. No additional information is available. Using only that observed wait time and your historical data:

  1. What is the probability that surge is currently active?
  2. What is the expected total cost of booking the ride (fare + time cost)?
  3. Should you book the ride or take the subway?
  4. What is the critical wait time — the threshold below which the ride is the better choice and above which the subway wins?

Express your answer to Part 1 as a probability to three significant figures. For Part 4, solve for the threshold algebraically or numerically — both approaches are valid.

Interactive Supplement
The Surge Decision Explorer

Explore this puzzle visually with an interactive diagram — drag sliders, watch the geometry update in real time, and build intuition before you solve.

Open interactive →
💡 Hint

The key is recognizing that observing a wait time gives you evidence about surge status — but evidence, not certainty. Start by writing down what each hypothesis predicts about seeing an 8-minute wait.

Under the no-surge hypothesis, wait times are Exp(1/5). What is the probability density of observing w  = 8 minutes? Do the same for the surge hypothesis using Exp(1/12). These two density values are your likelihoods.

Once you have both likelihoods, combine them with the prior probabilities (30% surge, 70% no-surge) to get the posterior probability of surge. The ratio of weighted likelihoods is the quantity you need.

For the decision, compute the expected total cost of the ride (probability-weighted fare plus time cost) and compare it to the fixed cost of the subway alternative.

For the threshold in Part 4: write the break-even condition as an equation in w. You will find that the posterior itself depends on w, making this an implicit equation — solve it numerically by evaluating both sides at a few candidate values.


Solution

Step 1 — Set up the likelihoods

Let S denote the event “surge is active.” Wait times follow an exponential distribution in both cases:

  • No surge: f(w | ¬S) = λ 0e−λ0w, with λ0 = 1/5
  • Surge active: f(w | S) = λ 1e−λ1w, with λ1 = 1/12

Evaluating at w = 8:

f(8 | S)  =  (1/12) e−8/12  ≈  0.04279
f(8 | ¬S)  =  (1/5) e−8/5  ≈  0.04038

Step 2 — Apply Bayes’ theorem

With prior P(S) = 0.30:

P(S | w) = f(w | S) · P(S)
f(w | S) · P(S) + f(w | ¬S) · P(¬S)
= 0.04279 × 0.30
0.04279 × 0.30  +  0.04038 × 0.70
P(S | w = 8) ≈ 0.312  (31.2%)
Part 1 answer

Closed-form expression

For exponential wait times the posterior has a clean closed form. Dividing numerator and denominator by the numerator and simplifying:

P(S | w) = 1
1  +  (1−q) · λ0 · e−(λ0−λ1)w
q    λ1

With our parameters: constant prefactor = (0.7/0.3) × (12/5) = 56/15 ≈ 5.600, decay rate = λ0−λ1 = 1/5−1/12 = 7/60 per minute. Longer waits increase the posterior toward 1; shorter waits pull it toward the prior.

Step 3 — Expected cost of booking

E[fare] = 0.312 × $18  +  0.688 × $9 = $5.62 + $6.19 = $11.81
E[time cost] = 8 × $0.50 = $4.00
E[total] = $15.81

Step 4 — Compare to subway

Subway total = $3.00 + 20 × $0.50 = $13.00
Part 3 answer
Take the subway — it saves $2.81 in expected cost.
$15.81 ride vs $13.00 subway

Step 5 — The decision threshold

The break-even wait time w* satisfies:

P(S | w*) · $18  +  [1−P(S | w*)] · $9  +  0.50w*  =  $13.00

Because P(S | w*) itself depends on w*, this is implicit. Evaluating numerically:

w (min) P(S|w) E[fare] Time cost Total vs Subway
2 min 18.4% $10.66 $1.00 $11.66 −$1.34 ✓
4 min ≈ w* 22.2% $11.00 $2.00 $13.00 $0.00 ≈
6 min 26.4% $11.38 $3.00 $14.38 +$1.38 ✗
8 min 31.2% $11.81 $4.00 $15.81 +$2.81 ✗
Decision rule (Part 4)
Book the ride if the quoted wait is less than 4 minutes.
Take the subway if the quoted wait is 4 minutes or more.
At the threshold w* ≈ 4 min, P(surge | w*) ≈ 22.2% — surge probability is actually lower than the prior because short waits are more consistent with normal driver availability.

The deeper lesson

Notice how little the 8-minute wait actually shifts the posterior — from a 30% prior to only 31.2%. The two distributions overlap heavily in this range: an 8-minute wait is plausible under both surge and no-surge conditions. Long waits are informative about surge; short waits are informative about no-surge; but the middle range is genuinely ambiguous.

The decision to take the subway is not primarily driven by updated beliefs about surge — it is driven by the time cost of waiting. Even at a normal fare, an 8-minute wait plus whatever the remaining ride takes makes the subway competitive. This is why the threshold is as low as 4 minutes: the time value of waiting dominates the fare uncertainty.

Further Reading
Bertsekas, D. P., &amp; Tsitsiklis, J. N. (2008). <em>Introduction to Probability</em> (2nd ed.). Athena Scientific. [Chapter 3: General Random Variables; exponential distribution and memoryless property.]
Gelman, A., Carlin, J. B., Stern, H. S., &amp; Rubin, D. B. (2013). <em>Bayesian Data Analysis</em> (3rd ed.). CRC Press. [Chapter 2: Single-parameter models; conjugate priors.]
DeGroot, M. H., &amp; Schervish, M. J. (2012). <em>Probability and Statistics</em> (4th ed.). Addison-Wesley. [Chapter 6: Large-Sample Methods; likelihood-based inference.]
Hall, J. D. (2019). Is Uber a substitute or complement for public transit? <em>Journal of Urban Economics</em>, 108, 1&ndash;14. https://doi.org/10.1016/j.jue.2018.09.003
Interactive diagram → Browse more puzzles