WEBVTT
00:00:00.840 --> 00:00:03.890
In the previous example, we had
a model where the result
00:00:03.890 --> 00:00:07.620
of the first coin toss did not
affect the probabilities of
00:00:07.620 --> 00:00:10.270
what might happen in
the second toss.
00:00:10.270 --> 00:00:13.580
This is a phenomenon that we
call independence and which we
00:00:13.580 --> 00:00:15.660
now proceed to define.
00:00:15.660 --> 00:00:18.230
Let us start with a first
attempt at the definition.
00:00:18.230 --> 00:00:20.280
We have an event, B,
that has a certain
00:00:20.280 --> 00:00:22.160
probability of occurring.
00:00:22.160 --> 00:00:26.050
We are then told that event A
occurred, but suppose that
00:00:26.050 --> 00:00:31.230
this knowledge does not affect
our beliefs about B in the
00:00:31.230 --> 00:00:34.800
sense that the conditional
probability remains the same
00:00:34.800 --> 00:00:38.290
as the original unconditional
probability.
00:00:38.290 --> 00:00:41.800
Thus, the occurrence of A
provides no new information
00:00:41.800 --> 00:00:47.230
about B. In such a case, we
may say that event B is
00:00:47.230 --> 00:00:50.000
independent from event A.
00:00:50.000 --> 00:00:53.420
If this is indeed the case,
notice that the probability
00:00:53.420 --> 00:01:02.120
that both A and B occur, which
is always equal by the
00:01:02.120 --> 00:01:05.960
multiplication rule to the
probability of A times the
00:01:05.960 --> 00:01:11.630
conditional probability of B
given A. So this is a relation
00:01:11.630 --> 00:01:12.890
that's always true.
00:01:12.890 --> 00:01:17.100
But if we also have this
additional condition, then
00:01:17.100 --> 00:01:20.810
this simplifies to the
probability of A times the
00:01:20.810 --> 00:01:26.280
probability of B.
00:01:26.280 --> 00:01:30.780
So we can find the probability
of both events happening by
00:01:30.780 --> 00:01:35.240
just multiplying their
individual probabilities.
00:01:35.240 --> 00:01:39.720
It turns out that this relation
is a cleaner way of
00:01:39.720 --> 00:01:43.420
the defining formally the
notion of independence.
00:01:43.420 --> 00:01:47.060
So we will say that two
events, A and B, are
00:01:47.060 --> 00:01:51.610
independent if this
relation holds.
00:01:51.610 --> 00:01:56.120
Why do we use this definition
rather than the original one?
00:01:56.120 --> 00:01:59.950
This formal definition has
several advantages.
00:01:59.950 --> 00:02:04.300
First, it is consistent with
the earlier definition.
00:02:04.300 --> 00:02:09.960
If this equality is true, then
the conditional probability of
00:02:09.960 --> 00:02:16.400
event B given A, which is the
ratio of this divided by that,
00:02:16.400 --> 00:02:20.510
will be equal to the probability
of B. So if this
00:02:20.510 --> 00:02:25.250
relation holds, then this
relation will also hold, and
00:02:25.250 --> 00:02:30.160
so this more formal definition
is consistent with our earlier
00:02:30.160 --> 00:02:32.710
intuitive definition.
00:02:32.710 --> 00:02:36.370
A more important reason is that
this formal definition is
00:02:36.370 --> 00:02:40.780
symmetric with respect to the
roles of A and B. So instead
00:02:40.780 --> 00:02:45.210
of saying that B is independent
from A, based on
00:02:45.210 --> 00:02:49.640
this definition we can now say
that events A and B are
00:02:49.640 --> 00:02:52.420
independent of each other.
00:02:52.420 --> 00:02:56.400
And in addition, since this
definition is symmetric and
00:02:56.400 --> 00:03:01.440
since it implies this condition,
it must also imply
00:03:01.440 --> 00:03:03.010
the symmetrical relation.
00:03:03.010 --> 00:03:06.230
Namely, that the conditional
probability of A given B is
00:03:06.230 --> 00:03:09.770
the same as the unconditional
probability of A.
00:03:09.770 --> 00:03:13.320
Finally, on the technical
side, conditional
00:03:13.320 --> 00:03:17.000
probabilities are only defined
when the conditioning event
00:03:17.000 --> 00:03:19.150
has non-zero probability.
00:03:19.150 --> 00:03:23.060
So this original definition
would only make sense in those
00:03:23.060 --> 00:03:28.490
cases where the probability of
the event A would be non-zero.
00:03:28.490 --> 00:03:32.100
In contrast, this new definition
makes sense even
00:03:32.100 --> 00:03:35.800
when we're dealing with zero
probability events.
00:03:35.800 --> 00:03:39.690
So this definition is indeed
more general, and this also
00:03:39.690 --> 00:03:42.490
makes it more elegant.
00:03:42.490 --> 00:03:44.740
Let us now build some
understanding of what
00:03:44.740 --> 00:03:46.970
independence really is.
00:03:46.970 --> 00:03:50.870
Suppose that we have two events,
A and B, both of which
00:03:50.870 --> 00:03:54.380
have positive probability.
00:03:54.380 --> 00:03:57.640
And furthermore, these two
events are disjoint.
00:03:57.640 --> 00:03:59.860
They do not have any
common elements.
00:03:59.860 --> 00:04:02.380
Are these two events
independent?
00:04:02.380 --> 00:04:04.330
Let us check the definition.
00:04:04.330 --> 00:04:11.720
The probability that both A and
B occur is zero because
00:04:11.720 --> 00:04:13.900
the two events are disjoint.
00:04:13.900 --> 00:04:15.760
They cannot happen together.
00:04:15.760 --> 00:04:19.390
On the other hand, the
probability of A times the
00:04:19.390 --> 00:04:23.350
probability of B is positive,
since each one of the two
00:04:23.350 --> 00:04:25.890
terms is positive.
00:04:25.890 --> 00:04:29.280
And therefore, these two
expressions are different from
00:04:29.280 --> 00:04:33.210
each other, and therefore this
equality that's required by
00:04:33.210 --> 00:04:36.290
the definition of independence
does not hold.
00:04:36.290 --> 00:04:39.180
The conclusion is that these
two events are not
00:04:39.180 --> 00:04:40.290
independent.
00:04:40.290 --> 00:04:43.930
In fact, intuitively, these two
events are as dependent as
00:04:43.930 --> 00:04:45.510
Siamese twins.
00:04:45.510 --> 00:04:48.710
If you know that A occurred,
then you are sure
00:04:48.710 --> 00:04:50.530
that B did not occur.
00:04:50.530 --> 00:04:55.140
So the occurrence of A tells you
a lot about the occurrence
00:04:55.140 --> 00:04:57.700
or non-occurrence of B.
00:04:57.700 --> 00:05:00.280
So we see that being independent
is something
00:05:00.280 --> 00:05:03.680
completely different from
being disjoint.
00:05:03.680 --> 00:05:07.730
Independence is a relation
about information.
00:05:07.730 --> 00:05:11.270
It is important to always keep
in mind the intuitive meaning
00:05:11.270 --> 00:05:12.450
of independence.
00:05:12.450 --> 00:05:16.750
Two events are independent if
the occurrence of one event
00:05:16.750 --> 00:05:20.660
does not change our beliefs
about the other.
00:05:20.660 --> 00:05:23.560
It does not affect the
probability that the other
00:05:23.560 --> 00:05:26.600
event also occurs.
00:05:26.600 --> 00:05:29.640
When do we have independence
in the real world?
00:05:29.640 --> 00:05:33.650
The typical case is when the
occurrence or non-occurrence
00:05:33.650 --> 00:05:37.590
of each of the two events A
and B is determined by two
00:05:37.590 --> 00:05:41.720
physically distinct and
non-interacting processes.
00:05:41.720 --> 00:05:45.230
For example, whether my coin
results in heads and whether
00:05:45.230 --> 00:05:48.530
it will be snowing on New Year's
Day are two events that
00:05:48.530 --> 00:05:51.250
should be modeled
as independent.
00:05:51.250 --> 00:05:53.610
But I should also say that there
are some cases where
00:05:53.610 --> 00:05:56.940
independence is less obvious and
where it happens through a
00:05:56.940 --> 00:05:59.610
numerical accident.
00:05:59.610 --> 00:06:02.730
You can now move on to answer
some simple questions where
00:06:02.730 --> 00:06:05.700
you will have to check for
independence using either the
00:06:05.700 --> 00:06:07.880
mathematical or intuitive
definition.