1
00:00:04,580 --> 00:00:06,550
Using our regression
models, we would

2
00:00:06,550 --> 00:00:09,300
like to predict before
the season starts

3
00:00:09,300 --> 00:00:12,910
how many games the 2002
Oakland A's will win.

4
00:00:12,910 --> 00:00:15,450
To do this, we first
have to predict

5
00:00:15,450 --> 00:00:17,640
how many runs the
team will score

6
00:00:17,640 --> 00:00:20,070
and how many runs
they will allow.

7
00:00:20,070 --> 00:00:22,890
These models use
team statistics.

8
00:00:22,890 --> 00:00:26,100
However, when we are
predicting for the 2002 Oakland

9
00:00:26,100 --> 00:00:28,700
A's before the
season has occurred,

10
00:00:28,700 --> 00:00:31,980
the team is probably different
than it was the year before.

11
00:00:31,980 --> 00:00:34,670
So we don't know
the team statistics.

12
00:00:34,670 --> 00:00:38,180
But we can estimate these
statistics using past player

13
00:00:38,180 --> 00:00:39,520
performance.

14
00:00:39,520 --> 00:00:42,420
This approach assumes that
past performance correlates

15
00:00:42,420 --> 00:00:44,520
with future performance
and that there

16
00:00:44,520 --> 00:00:47,480
will be few injuries
during the season.

17
00:00:47,480 --> 00:00:50,850
Using this approach, we can
estimate the team statistics

18
00:00:50,850 --> 00:00:57,050
for 2002 by using the
2001 player statistics.

19
00:00:57,050 --> 00:01:00,360
Let's start by making a
prediction for runs scored.

20
00:01:00,360 --> 00:01:03,170
At the beginning
of the 2002 season,

21
00:01:03,170 --> 00:01:07,340
the Oakland A's had 24
batters on their roster.

22
00:01:07,340 --> 00:01:12,140
Using the 2001 regular season
statistics for these players,

23
00:01:12,140 --> 00:01:15,510
we can estimate that team
on-base percentage will be

24
00:01:15,510 --> 00:01:23,330
about 0.339 and team slugging
percentage will be about 0.430.

25
00:01:23,330 --> 00:01:27,350
We built the following linear
regression equation in R

26
00:01:27,350 --> 00:01:29,640
to predict runs scored.

27
00:01:29,640 --> 00:01:35,560
If we plug in 0.339 for
on-base percentage and 0.430

28
00:01:35,560 --> 00:01:40,060
for slugging percentage, we
predict that the 2002 Oakland

29
00:01:40,060 --> 00:01:45,060
A's will score about 805 runs.

30
00:01:45,060 --> 00:01:48,660
Similarly, we can make a
prediction for runs allowed.

31
00:01:48,660 --> 00:01:51,340
At the beginning
of the 2002 season,

32
00:01:51,340 --> 00:01:55,300
the Oakland A's had 17
pitchers on their roster.

33
00:01:55,300 --> 00:01:59,810
Using the 2001 regular season
statistics for these players,

34
00:01:59,810 --> 00:02:02,990
we can estimate that team
opponent on-base percentage

35
00:02:02,990 --> 00:02:07,960
will be about 0.307 and team
opponent slugging percentage

36
00:02:07,960 --> 00:02:12,800
will be about 0.373.

37
00:02:12,800 --> 00:02:15,590
Our regression equation
to predict runs allowed

38
00:02:15,590 --> 00:02:17,530
was as follows.

39
00:02:17,530 --> 00:02:20,370
By plugging in
0.307 for opponents

40
00:02:20,370 --> 00:02:25,990
on-base percentage and 0.373 for
opponents slugging percentage,

41
00:02:25,990 --> 00:02:33,070
we predict that the 2002
Oakland A's will allow 622 runs.

42
00:02:33,070 --> 00:02:37,460
We can now make a prediction for
how many games they will win.

43
00:02:37,460 --> 00:02:41,510
Our regression equation to
predict wins is as follows.

44
00:02:41,510 --> 00:02:48,370
We predicted 805 runs
scored and 622 runs allowed.

45
00:02:48,370 --> 00:02:50,800
We can plug in the difference
between runs scored

46
00:02:50,800 --> 00:02:53,620
and runs allowed to
predict that the A's

47
00:02:53,620 --> 00:02:59,520
will win 100 games in 2002.

48
00:02:59,520 --> 00:03:03,280
Paul DePodesta used a similar
approach to make predictions.

49
00:03:03,280 --> 00:03:05,790
It turns out that our
predictions and Paul's

50
00:03:05,790 --> 00:03:10,000
predictions closely
match actual performance.

51
00:03:10,000 --> 00:03:13,650
Our prediction for runs
scored was 805 runs.

52
00:03:13,650 --> 00:03:17,730
Paul predicted between
800 and 820 runs.

53
00:03:17,730 --> 00:03:21,400
And it turns out that the
2002 Oakland A's actually

54
00:03:21,400 --> 00:03:24,020
scored 800 runs.

55
00:03:24,020 --> 00:03:28,100
For runs allowed,
we predicted 622.

56
00:03:28,100 --> 00:03:33,079
Paul DePodesta predicted
between 650 and 670.

57
00:03:33,079 --> 00:03:39,050
It turns out that the Oakland
A's actually allowed 653 runs.

58
00:03:39,050 --> 00:03:43,250
For wins, we predicted that
they would win 100 games.

59
00:03:43,250 --> 00:03:47,670
Paul predicted that they would
win between 93 and 97 games.

60
00:03:47,670 --> 00:03:52,420
And they actually 103 games.

61
00:03:52,420 --> 00:03:55,070
These predictions
show us that by using

62
00:03:55,070 --> 00:03:58,480
publicly available data
and simple analytics,

63
00:03:58,480 --> 00:04:01,560
we can predict very close
to what actually happened

64
00:04:01,560 --> 00:04:05,340
before the season even started.

65
00:04:05,340 --> 00:04:08,870
It turns out that the A's
won a league record in 2002

66
00:04:08,870 --> 00:04:11,350
by winning 20 games in a row.

67
00:04:11,350 --> 00:04:13,940
And they won 1 more game
than the previous year

68
00:04:13,940 --> 00:04:17,160
and made it to the
playoffs once again.