Would you like to inspect the original subtitles? These are the user uploaded subtitles that are being translated:
1
00:00:00,004 --> 00:00:03,003
- If you haven't worked with Bayesian analysis before,
2
00:00:03,003 --> 00:00:05,006
dealing with prior probabilities and base rates,
3
00:00:05,006 --> 00:00:08,005
then the concept can be a little unclear.
4
00:00:08,005 --> 00:00:10,003
I'd like to work through a classic example
5
00:00:10,003 --> 00:00:14,008
from Tversky and Kahneman called at The Taxicab Problem
6
00:00:14,008 --> 00:00:17,001
that will demonstrate how Bayesian analysis works
7
00:00:17,001 --> 00:00:19,000
and give you better intuition.
8
00:00:19,000 --> 00:00:23,005
Let's start with the facts as proposed within the problem.
9
00:00:23,005 --> 00:00:27,006
A cab was involved in a hit-and-run accident at night.
10
00:00:27,006 --> 00:00:30,008
There are two cab companies working in the city,
11
00:00:30,008 --> 00:00:33,001
Green and Blue.
12
00:00:33,001 --> 00:00:35,008
Furthermore, we know that a witness
13
00:00:35,008 --> 00:00:39,003
identified the cab as blue.
14
00:00:39,003 --> 00:00:42,006
However, 85% of the cabs in the city are green
15
00:00:42,006 --> 00:00:46,001
and only 15% are blue.
16
00:00:46,001 --> 00:00:49,005
Next, we need to know about witness reliability.
17
00:00:49,005 --> 00:00:52,005
The court tested the reliability of the witness
18
00:00:52,005 --> 00:00:53,009
under the circumstances
19
00:00:53,009 --> 00:00:56,007
that existed on the night of the accident.
20
00:00:56,007 --> 00:00:59,009
And concluded that the witness correctly identified
21
00:00:59,009 --> 00:01:03,000
each one of the two colors 80% of the time
22
00:01:03,000 --> 00:01:06,002
and failed 20% of the time.
23
00:01:06,002 --> 00:01:10,006
If there are any attorneys watching this video, don't worry.
24
00:01:10,006 --> 00:01:14,000
This is merely an example and it's not meant to be taken
25
00:01:14,000 --> 00:01:20,001
as a reflection of witness reliability in the real world.
26
00:01:20,001 --> 00:01:23,008
So your question is, what is the probability
27
00:01:23,008 --> 00:01:25,008
that the cab involved in the accident
28
00:01:25,008 --> 00:01:29,004
was actually blue rather than green?
29
00:01:29,004 --> 00:01:34,002
So we're looking at the probabilities for blue versus green.
30
00:01:34,002 --> 00:01:36,001
The answer might surprise you.
31
00:01:36,001 --> 00:01:38,009
And Neel Ocean, in an internet posting,
32
00:01:38,009 --> 00:01:41,001
provides an intuitive explanation
33
00:01:41,001 --> 00:01:43,002
of how to arrive at the correct answer,
34
00:01:43,002 --> 00:01:46,005
which is 41%.
35
00:01:46,005 --> 00:01:52,002
And you can find the answer at the website listed below.
36
00:01:52,002 --> 00:01:54,009
Let's walk through Neel Ocean's explanation
37
00:01:54,009 --> 00:01:58,004
of why the answer is 41%.
38
00:01:58,004 --> 00:02:02,002
We need to visualize the answer as a decision tree.
39
00:02:02,002 --> 00:02:07,005
At the top, we know that our base rate of cabs in the city
40
00:02:07,005 --> 00:02:11,008
are 85% green and 15% blue.
41
00:02:11,008 --> 00:02:13,009
The witness is 80% accurate.
42
00:02:13,009 --> 00:02:15,004
So if a cab is green,
43
00:02:15,004 --> 00:02:18,007
then the witness will say green 80% of the time
44
00:02:18,007 --> 00:02:22,000
and be correct and will be wrong 20% of the time,
45
00:02:22,000 --> 00:02:24,008
meaning that the cab is actually blue.
46
00:02:24,008 --> 00:02:26,008
And we have the same for the blue side.
47
00:02:26,008 --> 00:02:28,005
If the cab is actually blue,
48
00:02:28,005 --> 00:02:31,004
then it will be guessed 80% of the time.
49
00:02:31,004 --> 00:02:32,003
If it's green,
50
00:02:32,003 --> 00:02:35,009
then 20% of the time the witness will be wrong.
51
00:02:35,009 --> 00:02:39,009
Now we take compound probabilities to see how likely it is
52
00:02:39,009 --> 00:02:43,007
we end up in each branch of this decision tree.
53
00:02:43,007 --> 00:02:46,006
So 85% of the time the cab is green
54
00:02:46,006 --> 00:02:50,007
and it is guessed as being green 80% of the time.
55
00:02:50,007 --> 00:02:54,003
.85 times .80, is .68.
56
00:02:54,003 --> 00:02:58,003
Same thing for if the cab is actually blue,
57
00:02:58,003 --> 00:03:00,000
but it's guessed green.
58
00:03:00,000 --> 00:03:05,003
We have .85 times .20, which leads to .17.
59
00:03:05,003 --> 00:03:08,005
And then we have similar calculations for blue
60
00:03:08,005 --> 00:03:14,002
leading to probabilities of .12 and .03.
61
00:03:14,002 --> 00:03:17,007
Now remember, that the witness said the cab is blue.
62
00:03:17,007 --> 00:03:21,000
So we need to calculate the probability
63
00:03:21,000 --> 00:03:24,001
that the witness guessed blue
64
00:03:24,001 --> 00:03:26,008
when the cab was actually blue.
65
00:03:26,008 --> 00:03:28,007
So we used the two values in the middle
66
00:03:28,007 --> 00:03:31,002
where the cab was actually blue.
67
00:03:31,002 --> 00:03:34,001
And we get this calculation,
68
00:03:34,001 --> 00:03:37,005
.12 when the cab is actually blue,
69
00:03:37,005 --> 00:03:41,001
divided by .12 plus .17,
70
00:03:41,001 --> 00:03:44,009
which is when the cab is blue and was guessed blue.
71
00:03:44,009 --> 00:03:48,005
But also when the cab was guessed green,
72
00:03:48,005 --> 00:03:50,006
but was actually blue.
73
00:03:50,006 --> 00:03:55,002
So we have .12, divided by .12, plus .17.
74
00:03:55,002 --> 00:03:59,006
And that leads to a result of .41.
75
00:03:59,006 --> 00:04:01,003
In the next movie, I will show you
76
00:04:01,003 --> 00:04:04,005
how to create what is called a classification matrix
77
00:04:04,005 --> 00:04:05,009
so we can get the values that we need
78
00:04:05,009 --> 00:04:09,000
to perform this calculation in Excel.
6113
Can't find what you're looking for?
Get subtitles in any language from opensubtitles.com, and translate them here.