WEBVTT

00:00:00.290 --> 00:00:03.030
An important reason why
conditional probabilities are

00:00:03.030 --> 00:00:06.540
very useful is that they allow
us to divide and conquer.

00:00:06.540 --> 00:00:09.550
They allow us to split
complicated probability modes

00:00:09.550 --> 00:00:12.370
into simpler submodels
that we can then

00:00:12.370 --> 00:00:14.410
analyze one at a time.

00:00:14.410 --> 00:00:18.200
Let me remind you of the Total
Probability Theorem that has

00:00:18.200 --> 00:00:20.690
his particular flavor.

00:00:20.690 --> 00:00:25.060
We divide our sample space into
three disjoint events--

00:00:25.060 --> 00:00:27.020
A1, A2, and A3.

00:00:27.020 --> 00:00:29.830
And these events form a
partition of the sample space,

00:00:29.830 --> 00:00:32.299
that is, they exhaust
all possibilities.

00:00:32.299 --> 00:00:34.960
They correspond to three
alternative scenarios, one of

00:00:34.960 --> 00:00:36.550
which is going to occur.

00:00:36.550 --> 00:00:39.610
And then we may be interested
in a certain event B. That

00:00:39.610 --> 00:00:43.180
event B may occur under
either scenario.

00:00:43.180 --> 00:00:46.060
And the Total Probability
Theorem tells us that we can

00:00:46.060 --> 00:00:49.760
calculate the probability of
event B by considering the

00:00:49.760 --> 00:00:54.700
probability that it occurs under
any given scenario and

00:00:54.700 --> 00:00:58.080
weigh those probabilities
according to the probabilities

00:00:58.080 --> 00:01:00.950
of the different scenarios.

00:01:00.950 --> 00:01:04.129
Now, let us bring random
variables into the picture.

00:01:04.129 --> 00:01:06.750
Let us fix a particular
value--

00:01:06.750 --> 00:01:08.360
little x--

00:01:08.360 --> 00:01:14.110
and let the event B be the event
that the random variable

00:01:14.110 --> 00:01:17.880
takes on this particular
value.

00:01:17.880 --> 00:01:20.490
Let us now translate the
Total Probability

00:01:20.490 --> 00:01:23.100
Theorem to this situation.

00:01:23.100 --> 00:01:26.650
First, the picture will look
slightly different.

00:01:26.650 --> 00:01:29.560
Our event B has been replaced
by the particular event that

00:01:29.560 --> 00:01:32.229
we're now considering.

00:01:32.229 --> 00:01:35.210
Now, what is this probability?

00:01:35.210 --> 00:01:39.229
The probability that event B
occurs, having fixed the

00:01:39.229 --> 00:01:43.770
particular choice of little x,
is the value of PMF at that

00:01:43.770 --> 00:01:45.670
particular x.

00:01:45.670 --> 00:01:48.100
How about this probability
here?

00:01:48.100 --> 00:01:51.220
This is the probability that the
random variable, capital

00:01:51.220 --> 00:01:53.870
X, takes on the value
little x--

00:01:53.870 --> 00:01:56.100
that's what a PMF is--

00:01:56.100 --> 00:01:58.090
but in the conditional
universe.

00:01:58.090 --> 00:02:00.980
So we're dealing with
a conditional PMF.

00:02:00.980 --> 00:02:03.700
And so on with the
other terms.

00:02:03.700 --> 00:02:08.389
So this equation here is just
the usual Total Probability

00:02:08.389 --> 00:02:13.600
Theorem but translated
into PMF notation.

00:02:13.600 --> 00:02:17.220
Now this version of the Total
Probability Theorem, of

00:02:17.220 --> 00:02:23.520
course, is true for all
values of little x.

00:02:23.520 --> 00:02:27.450
This means that we can now
multiply both sides of this

00:02:27.450 --> 00:02:35.530
equation by x and them
sum over all

00:02:35.530 --> 00:02:39.200
possibles choices of x.

00:02:39.200 --> 00:02:42.680
We recognize that here we have
the expected value of the

00:02:42.680 --> 00:02:44.770
random variable X.

00:02:44.770 --> 00:02:48.620
Now, we do the same thing
to the right hand side.

00:02:48.620 --> 00:02:50.610
We multiply by x.

00:02:50.610 --> 00:02:56.820
And then we sum over all
possible values of x.

00:02:56.820 --> 00:02:58.950
This is going to be
the first term.

00:02:58.950 --> 00:03:02.020
And then we will have
similar terms.

00:03:02.020 --> 00:03:03.600
Now, what do we have here?

00:03:03.600 --> 00:03:08.500
This expression is just the
conditional expectation of the

00:03:08.500 --> 00:03:12.330
random variable X under
the scenario that

00:03:12.330 --> 00:03:14.820
event A1 has occurred.

00:03:14.820 --> 00:03:20.920
So what we have established is
this particular formula, which

00:03:20.920 --> 00:03:24.230
is called the Total Expectation
Theorem.

00:03:24.230 --> 00:03:28.190
It tells us that the expected
value of a random variable can

00:03:28.190 --> 00:03:31.810
be calculated by considering
different scenarios.

00:03:31.810 --> 00:03:35.130
Finding the expected value under
each of the possible

00:03:35.130 --> 00:03:37.660
scenarios and weigh them.

00:03:37.660 --> 00:03:40.400
Weigh the scenarios according
to their respective

00:03:40.400 --> 00:03:41.610
probabilities.

00:03:41.610 --> 00:03:43.350
The picture is like this.

00:03:43.350 --> 00:03:46.200
Under each scenario, the
random variable X has a

00:03:46.200 --> 00:03:48.640
certain conditional
expectation.

00:03:48.640 --> 00:03:50.880
We take all these
into account.

00:03:50.880 --> 00:03:53.329
We weigh them according
to their corresponding

00:03:53.329 --> 00:03:54.300
probabilities.

00:03:54.300 --> 00:03:58.290
And we add them up to find
the expected value of X.

00:03:58.290 --> 00:04:00.750
So we can divide and conquer.

00:04:00.750 --> 00:04:04.950
We can replace a possibly
complicated calculation of an

00:04:04.950 --> 00:04:09.720
expected value by hopefully
simpler calculations under

00:04:09.720 --> 00:04:12.670
each one of possible
scenarios.

00:04:12.670 --> 00:04:18.810
Let me illustrate the idea
by a simple example.

00:04:18.810 --> 00:04:22.170
Let us consider this PMF, and
let us try to calculate the

00:04:22.170 --> 00:04:25.320
expected value of the associated
random variable.

00:04:25.320 --> 00:04:30.950
One way to divide and conquer
is to define an event, A1,

00:04:30.950 --> 00:04:33.850
which is that our random
variable takes values in this

00:04:33.850 --> 00:04:37.900
set, and another event, A2,
which is that the random

00:04:37.900 --> 00:04:41.070
variable takes values
in that set.

00:04:41.070 --> 00:04:43.540
Let us now apply the Total
Expectations Theorem.

00:04:43.540 --> 00:04:46.330
Let us calculate all the terms
that are required.

00:04:46.330 --> 00:04:48.070
First, we find the
probabilities of

00:04:48.070 --> 00:04:49.550
the different scenarios.

00:04:49.550 --> 00:04:53.600
The probability of event
A1 is 1/9 plus 1/9 plus

00:04:53.600 --> 00:04:55.950
1/9 which is 1/3.

00:04:55.950 --> 00:05:02.450
And the probability of event
A2 is 2/9 plus 2/9 plus 2/9

00:05:02.450 --> 00:05:05.580
which adds up to 2/3.

00:05:05.580 --> 00:05:09.320
How about conditional
expectations?

00:05:09.320 --> 00:05:13.090
In a universe where event A1 one
has occurred, only these

00:05:13.090 --> 00:05:14.760
three values are possible.

00:05:14.760 --> 00:05:19.030
They had equal probabilities,
so in the conditional model,

00:05:19.030 --> 00:05:20.910
they will also have equal
probabilities.

00:05:20.910 --> 00:05:23.170
So we will have a uniform
distribution over

00:05:23.170 --> 00:05:25.320
the set {0, 1, 2}.

00:05:25.320 --> 00:05:27.660
By symmetry, the expected
value is going

00:05:27.660 --> 00:05:29.110
to be in the middle.

00:05:29.110 --> 00:05:31.880
So this expected value
is equal to 1.

00:05:31.880 --> 00:05:36.030
And by a similar argument, the
expected value of X under the

00:05:36.030 --> 00:05:40.090
second scenario is going to be
the midpoint of this range,

00:05:40.090 --> 00:05:41.940
which is equal to 7.

00:05:41.940 --> 00:05:47.440
And now we can apply the Total
Probability Theorem and write

00:05:47.440 --> 00:05:51.620
that the expected value of X is
equal to the probability of

00:05:51.620 --> 00:05:55.520
the first scenario times the
expected value under the first

00:05:55.520 --> 00:06:00.440
scenario plus the probability
of the second scenario times

00:06:00.440 --> 00:06:05.040
the expected value under
the second scenario.

00:06:05.040 --> 00:06:08.050
In this case, by breaking down
the problem in these two

00:06:08.050 --> 00:06:11.560
subcases, the calculations
that were required were

00:06:11.560 --> 00:06:16.420
somewhat simpler than if you
were to proceed directly.

00:06:16.420 --> 00:06:18.990
Of course, this is a rather
simple example.

00:06:18.990 --> 00:06:22.330
But as we go on with this
course, we will apply the

00:06:22.330 --> 00:06:25.440
Total Probability Theorem in
much more interesting and

00:06:25.440 --> 00:06:26.690
complicated situations.