5. Data from Various Sports 41

0 25 50 75 100 125 150

Game number

0.45

0.5

0.55

0.6

0.65

Winning

percentage

Figure 22. Block winning percentages for a streaky team

of the nine 18-game blocks. Figure 22 shows an example of the block

winning percentages for a team whose season percentage is .55.

A team whose win-loss sequence arises from the above block-

Bernoulli process will be said to be streaky, while one whose win-loss

sequence arises from a Bernoulli process will be called consistent. Of

course, we do not expect all teams to behave in one of these two ways;

we need a parameter that can be estimated and that does a good job

of distinguishing between these two models. As we said above, our

windowed difference statistic does not distinguish very well between

the two models we posited.

Albert and Bennett define their parameter, called Black, as fol-

lows. Given a team’s win-loss sequence for a season, they compute

the windowed winning percentages for all blocks of 12 games. (So in

a 162-game season, there are 151 such blocks, starting at games 1 to

151.) They plot these percentages, along with a horizontal line whose

y-value is the season winning percentage. An example of this plot

is shown in Figure 23; the team is the 1984 Baltimore Orioles. The