Library · The Fiddler: Solutions · Chapter 75

Can You Sweep the Series?

Revised Report an error

In a best-of-seven series the favoured team wins each game independently with probability pp. For which range a<p<ba<p<b is “winning in exactly five games” the single most likely of the eight possible series results?

The Fiddler, Zach Wissner-Gross, May 9, 2025(original post)

Solution

Writing q=1−pq=1-p, the favoured team wins in 4,5,6,74,5,6,7 games with probabilities p4,4p4q,10p4q2,20p4q3,p^4,\quad 4p^4q,\quad 10p^4q^2,\quad 20p^4q^3, and the losing results are smaller in this regime. Win-in-five beats win-in-four when 4q>14q>1 and beats win-in-six when 4>10q4>10q, i.e. q<14 and q>25 ⟹ 35<p<34.q<\tfrac14\ \text{and}\ q>\tfrac25\ \Longrightarrow\ \tfrac35<p<\tfrac34. So the range is 35<p<34\boxed{\tfrac35<p<\tfrac34} (that is, 0.6<p<0.750.6<p<0.75).

The computation

Tabulate all eight series outcomes (win or lose in 44 to 77 games) and, scanning pp, mark where “win in five” is the single largest. The marked band runs from 0.60.6 to 0.750.75.

import numpy as np
def best(p):
    q = 1 - p
    d = {'w4': p**4, 'w5': 4*p**4*q, 'w6': 10*p**4*q**2, 'w7': 20*p**4*q**3,
         'l4': q**4, 'l5': 4*q**4*p, 'l6': 10*q**4*p**2, 'l7': 20*q**4*p**3}
    return max(d, key=d.get)
ps = [p for p in np.linspace(.001, .999, 99999) if best(p) == 'w5']
print(round(min(ps), 3), round(max(ps), 3))     # 0.6  0.75

Extra Credit

Let p4p_4 be the probability that the Celtics (the favourites) sweep the Knicks in four games, and p7p_7 the probability that the series goes to seven games, with either team winning. Suppose pp is chosen uniformly at random from the interval (a,b)(a, b) found above, so that the likeliest result is the Knicks losing in five. How likely is it that p4p_4 is greater than p7p_7?

Solution

A Celtics sweep means winning the first four games: p4=p4p_4 = p^4. The series reaches a seventh game exactly when the first six games split three apiece, which happens in (63)=20\binom63 = 20 orders, so p7=20p3q3p_7 = 20p^3q^3, where q=1−pq = 1-p. Dividing by p3p^3, the sweep is likelier when p  >  20q3=20(1−p)3.p \;>\; 20q^3 = 20(1-p)^3 . As pp rises the left side increases and the right side decreases, so they cross exactly once. At p=35p = \tfrac35 the right side is 20×0.064=1.2820 \times 0.064 = 1.28, larger than 0.60.6. At p=34p = \tfrac34 it is 20/64=0.312520/64 = 0.3125, smaller than 0.750.75. So the crossing p∗p^* lies inside the interval, and the sweep is likelier for p∗<p<34p^* < p < \tfrac34. In terms of qq, the crossing solves 20q3+q−1=020q^3 + q - 1 = 0. That cubic has one real root, q∗≈0.323418q^* \approx 0.323418, so p∗≈0.676582p^* \approx 0.676582. The fraction of the interval above it is 0.75−0.6765820.75−0.6  ≈  0.4895,\frac{0.75 - 0.676582}{0.75 - 0.6} \;\approx\; \boxed{0.4895}, about 4949 percent. So for a slight majority of the plausible values of pp, a seven-game series is the likelier of the two, and Stephen A. Smith’s “closer to a sweep” is true only a little less than half the time.

The computation

The check plays out the series instead of using the formulas for p4p_4 and p7p_7. For a given pp it runs through all 272^7 win-loss sequences, stops each series once a team has four wins, and adds up the probability of a four-game Celtics sweep and of a seventh game. It then takes 30003000 evenly spread values of pp across the interval, standing in for the uniform choice, and counts how often the sweep is likelier. Finally it bisects for the crossing point.

from itertools import product
def outcomes(p):
    """Play out all 2^7 game sequences; series stops at four wins."""
    sweep = seven = 0.0
    for games in product((1, 0), repeat=7):        # 1 = Celtics win
        prob = 1.0
        for g in games: prob *= p if g else 1 - p
        c = k = 0
        for n, g in enumerate(games, 1):
            c += g; k += 1 - g
            if c == 4 or k == 4: break
        if n == 4 and c == 4: sweep += prob
        if n == 7: seven += prob
    return sweep, seven
a, b, M = 0.6, 0.75, 3000
grid = [a + (b - a)*(i + 0.5)/M for i in range(M)]   # uniform p, midpoints
share = sum(s > t for s, t in map(outcomes, grid)) / M
lo, hi = a, b                                      # where the two cross
for _ in range(50):
    mid = (lo + hi)/2
    s, t = outcomes(mid)
    if s > t: hi = mid
    else: lo = mid
print("share of p with sweep likelier: %.4f" % share)
print("crossover p = %.6f, share = %.5f" % (hi, (b - hi)/(b - a)))

It prints

share of p with sweep likelier: 0.4893
crossover p = 0.676582, share = 0.48945

The grid count agrees to within its spacing, and the bisection gives the boxed value.

Report an error on this page

Reports are stored by Netlify. See the privacy note.