WEBVTT

00:00:02.040 --> 00:00:05.440
We will now consider an example
that illustrates the

00:00:05.440 --> 00:00:09.030
difference between the notion
of independence of a

00:00:09.030 --> 00:00:12.990
collection of events and
the notion of pairwise

00:00:12.990 --> 00:00:16.460
independence within
that collection.

00:00:16.460 --> 00:00:18.290
The model is simple.

00:00:18.290 --> 00:00:22.700
We have a fair coin which
we flip twice.

00:00:22.700 --> 00:00:25.460
So at each flip, there
is probability 1/2

00:00:25.460 --> 00:00:28.950
of obtaining heads.

00:00:28.950 --> 00:00:32.340
Furthermore, we assume that the
two flips are independent

00:00:32.340 --> 00:00:33.770
of each other.

00:00:33.770 --> 00:00:38.760
Let H1 be the event that the
first coin toss resulted in

00:00:38.760 --> 00:00:43.655
heads, which corresponds to this
event in this diagram.

00:00:46.370 --> 00:00:51.700
Let H2 be the event that the
second toss resulted in heads,

00:00:51.700 --> 00:00:56.570
which is this event
in the diagram--

00:00:56.570 --> 00:01:02.245
the two ways that we can have
the second toss being heads.

00:01:04.800 --> 00:01:08.590
Now, we're assuming that the
tosses are independent.

00:01:08.590 --> 00:01:13.500
So the event heads-heads has a
probability which is equal to

00:01:13.500 --> 00:01:16.539
the probability that the first
toss resulted in heads--

00:01:16.539 --> 00:01:17.390
that's 1/2--

00:01:17.390 --> 00:01:20.200
times the probability that the
second toss resulted in heads,

00:01:20.200 --> 00:01:21.460
which is 1/2.

00:01:21.460 --> 00:01:23.730
So the product is 1/4.

00:01:23.730 --> 00:01:26.880
We have probability 1/4
for this outcome.

00:01:26.880 --> 00:01:31.430
Now, the total probability of
event H1 is 1/2, which means

00:01:31.430 --> 00:01:36.560
that the probability of what
remains should be 1/4, so that

00:01:36.560 --> 00:01:39.320
the sum of these two
numbers is 1/2.

00:01:39.320 --> 00:01:42.270
By the same argument, the
probability of this outcome,

00:01:42.270 --> 00:01:45.880
tails-heads , should be 1/4.

00:01:45.880 --> 00:01:47.940
We have a total of 3/4.

00:01:47.940 --> 00:01:50.070
So what's left is 1/4.

00:01:50.070 --> 00:01:53.080
And that's going to be the
probability of the outcome

00:01:53.080 --> 00:01:55.500
tails-tails .

00:01:55.500 --> 00:02:01.470
Let us now introduce a new
event, namely the event that

00:02:01.470 --> 00:02:05.040
the two tosses had
the same result.

00:02:05.040 --> 00:02:10.728
So this is the event that we
obtain either heads heads or

00:02:10.728 --> 00:02:11.978
tails-tails.

00:02:14.210 --> 00:02:18.560
Schematically, event C
corresponds to this blue

00:02:18.560 --> 00:02:21.290
region in the diagram.

00:02:21.290 --> 00:02:26.760
Is this event C independent
from the events H1 and H2?

00:02:26.760 --> 00:02:30.520
Let us first look for pairwise
independence.

00:02:30.520 --> 00:02:34.440
Let's look at the probability
that H1 occurs

00:02:34.440 --> 00:02:37.150
and C occurs as well.

00:02:37.150 --> 00:02:39.820
So the first toss resulted
in heads.

00:02:39.820 --> 00:02:43.060
And the two tosses had
the same result.

00:02:43.060 --> 00:02:47.200
So this is the same as the
probability of obtaining heads

00:02:47.200 --> 00:02:48.620
followed by heads.

00:02:51.340 --> 00:02:54.780
And this corresponds to just
this outcome that has

00:02:54.780 --> 00:02:56.090
probability 1/4.

00:02:59.930 --> 00:03:07.250
How about the product of the
probabilities of H1 and of C?

00:03:07.250 --> 00:03:09.242
Is it the same?

00:03:09.242 --> 00:03:12.290
Well, the probability
of H1 is 1/2.

00:03:15.240 --> 00:03:16.900
And the probability of C--

00:03:16.900 --> 00:03:17.850
what is it?

00:03:17.850 --> 00:03:21.450
Event C consists of
two outcomes.

00:03:21.450 --> 00:03:24.030
Each one of these outcomes
has probability 1/4.

00:03:24.030 --> 00:03:27.550
So the total is, again, 1/2.

00:03:27.550 --> 00:03:32.430
And therefore, the product of
these probabilities is 1/4.

00:03:32.430 --> 00:03:34.829
So we notice that the
probability of the two events

00:03:34.829 --> 00:03:38.020
happening is the same as the
product of their individual

00:03:38.020 --> 00:03:42.370
probabilities, and therefore,
H1 and C

00:03:42.370 --> 00:03:44.065
are independent events.

00:03:46.610 --> 00:03:52.500
By the same argument, H2 and C
are going to be independent.

00:03:52.500 --> 00:03:55.690
It's a symmetrical situation.

00:03:55.690 --> 00:03:59.940
H1 and H2 are also independent
from each other.

00:03:59.940 --> 00:04:02.740
So we have all of the conditions
for pairwise

00:04:02.740 --> 00:04:04.470
independence.

00:04:04.470 --> 00:04:08.710
Let us now check whether
we have independence.

00:04:08.710 --> 00:04:12.730
To check for independence, we
need to also look into the

00:04:12.730 --> 00:04:16.329
probability of all three events
happening and see

00:04:16.329 --> 00:04:19.070
whether it is equal to the
product of the individual

00:04:19.070 --> 00:04:20.420
probabilities.

00:04:20.420 --> 00:04:23.770
So the probability of all
three events happening--

00:04:23.770 --> 00:04:27.450
this is the probability
that H1 occurs and H2

00:04:27.450 --> 00:04:30.750
occurs and C occurs.

00:04:30.750 --> 00:04:34.260
What is this event?

00:04:34.260 --> 00:04:37.460
Heads in the first toss, heads
in the second toss, and the

00:04:37.460 --> 00:04:39.770
two tosses are the same--

00:04:39.770 --> 00:04:44.280
this happens if and only
if the outcome is

00:04:44.280 --> 00:04:46.780
heads followed by heads.

00:04:46.780 --> 00:04:51.060
And this has probability 1/4.

00:04:51.060 --> 00:04:54.480
On the other hand, if we
calculate the probability of

00:04:54.480 --> 00:05:01.850
H1 times the probability of H2
times the probability of C, we

00:05:01.850 --> 00:05:07.490
get 1/2 times 1/2 times
1/2, which is 1/8.

00:05:07.490 --> 00:05:10.900
These two numbers
are different.

00:05:10.900 --> 00:05:15.020
And therefore, one of the
conditions that we had for

00:05:15.020 --> 00:05:18.340
independence is violated.

00:05:18.340 --> 00:05:23.800
So in this example, H1,
H2, and C are pairwise

00:05:23.800 --> 00:05:29.040
independent, but they're not
independent in the sense of an

00:05:29.040 --> 00:05:32.330
independent collection
of events.

00:05:32.330 --> 00:05:34.345
How are we to understand
this intuitively?

00:05:37.800 --> 00:05:46.520
If I tell you that event H1
occurred and I ask you for the

00:05:46.520 --> 00:05:49.940
conditional probability
of C given that H1

00:05:49.940 --> 00:05:53.050
[occurred], what is this?

00:05:53.050 --> 00:05:57.360
This is the probability that
we obtain, given that the

00:05:57.360 --> 00:06:01.790
first event is heads, the first
result is heads, the

00:06:01.790 --> 00:06:05.030
only way that you can have the
two tosses having the same

00:06:05.030 --> 00:06:08.970
result is going to be in
the second toss also

00:06:08.970 --> 00:06:12.020
resulting in heads.

00:06:12.020 --> 00:06:15.680
And since H2 and H1 are
independent, this is just the

00:06:15.680 --> 00:06:19.340
probability that we have heads
in the second toss.

00:06:19.340 --> 00:06:22.110
And this number is 1/2.

00:06:22.110 --> 00:06:26.720
And 1/2 is also the same as the
probability of C. That's

00:06:26.720 --> 00:06:31.380
another way of understanding the
independence of H1 and C.

00:06:31.380 --> 00:06:36.510
Given that the first toss
resulted in heads, this does

00:06:36.510 --> 00:06:41.580
not help you in any way in
guessing whether the two

00:06:41.580 --> 00:06:45.750
tosses will have the
same result or not.

00:06:45.750 --> 00:06:49.290
The first one was heads, but the
second one could be either

00:06:49.290 --> 00:06:52.870
heads or tails with
equal probability.

00:06:52.870 --> 00:06:57.670
So event H1 does not carry any
useful information about the

00:06:57.670 --> 00:07:04.210
occurrence or non-occurrence of
event C. On the other hand,

00:07:04.210 --> 00:07:12.210
if I were to tell you that
both events, H1 and H2,

00:07:12.210 --> 00:07:14.560
happened, what would
the conditional

00:07:14.560 --> 00:07:17.560
probability of C be?

00:07:17.560 --> 00:07:21.250
If both H1 and H2 occurred, then
the results of the two

00:07:21.250 --> 00:07:27.400
coin tosses were identical, so
you know that C also occurred.

00:07:27.400 --> 00:07:31.540
So this probability
is equal to 1.

00:07:31.540 --> 00:07:36.000
And this number, 1, is
different from the

00:07:36.000 --> 00:07:41.040
unconditional probability
of C, which is 1/2.

00:07:41.040 --> 00:07:46.190
So we have here a situation
where knowledge of H1 having

00:07:46.190 --> 00:07:51.409
occurred does not help you in
making a better guess on

00:07:51.409 --> 00:07:53.980
whether C is going to occur.

00:07:53.980 --> 00:07:57.490
H1 by itself does not carry
any useful information.

00:07:57.490 --> 00:08:02.320
But the two events together,
H1 and H2, do carry useful

00:08:02.320 --> 00:08:03.630
information about C.

00:08:03.630 --> 00:08:06.360
Once you know that H1 and
H2 occurred, then C

00:08:06.360 --> 00:08:08.040
is certain to occur.

00:08:08.040 --> 00:08:12.430
So your original probability for
C, which was 1/2, now gets

00:08:12.430 --> 00:08:14.610
revised to a value of 1.

00:08:14.610 --> 00:08:19.720
So H1 and H2 carry information
relevant to C. And therefore,

00:08:19.720 --> 00:08:25.290
C is not independent from these
two events collectively.

00:08:25.290 --> 00:08:28.810
And we say that events H1.

00:08:28.810 --> 00:08:31.420
H2, and C are not independent.