42 1. Streaks

0 20 40 60 80 100 120 140

Game number

0.2

0.4

0.6

0.8

1

Winning

percentage

Figure 23. Windowed winning percentage for 1984 Baltimore Orioles

parameter Black is easy to describe in terms of this figure; it is the

area of the black region.

It is clear that if a team is consistent, then Black will be very

small, while if a team is streaky, Black will probably be large. Thus,

this parameter might be able to distinguish between the block-Bernoulli

and the Bernoulli models. To decide which model fits a given team

better, Albert and Bennett compute the winning percentage of the

team. Then they obtain, by simulation, the distribution of the pa-

rameter Black under both models. The actual value of Black for the

team is computed and compared with the two distributions. For each

distribution, a p-value is reported. If the first p-value of the observa-

tion is larger than the second p-value, then they claim that the first

model fits the team’s performance better than the second model.

In fact, they go further and use the ratio of the p-values as the

odds that the team is consistent (or streaky). For example, if the p-

values for a given team, under the consistent and streaky models, are

.08 and .30, respectively, then they say that the probability that the