WEBVTT

00:00:00.650 --> 00:00:04.210
In this segment we revisit the
concept of conditional

00:00:04.210 --> 00:00:08.170
expectation and view it
as an abstract object

00:00:08.170 --> 00:00:09.930
of a special kind.

00:00:09.930 --> 00:00:12.650
To get going, let us start with
something simple, the

00:00:12.650 --> 00:00:13.840
concept of a function.

00:00:13.840 --> 00:00:16.379
Let's say a function
h that maps real

00:00:16.379 --> 00:00:18.130
numbers to real numbers.

00:00:18.130 --> 00:00:20.520
As a concrete instance,
consider the quadratic

00:00:20.520 --> 00:00:25.180
function that maps a number
x to its square.

00:00:25.180 --> 00:00:29.370
Consider now a random variable,
capital X. What do

00:00:29.370 --> 00:00:33.710
we mean when we write h of X?

00:00:33.710 --> 00:00:35.860
For h defined--

00:00:35.860 --> 00:00:39.990
for example in this particular
way as a quadratic function--

00:00:39.990 --> 00:00:44.770
h of X is defined to be
a random variable.

00:00:44.770 --> 00:00:46.500
Which random variable?

00:00:46.500 --> 00:00:49.250
It is the random variable that
takes the value little x

00:00:49.250 --> 00:00:53.700
squared whenever capital X, the
random variable, happens

00:00:53.700 --> 00:00:56.690
to take the value little x.

00:00:56.690 --> 00:00:59.700
And this is the random variable
that we usually

00:00:59.700 --> 00:01:04.080
denote as the random
variable X squared.

00:01:04.080 --> 00:01:06.810
Now let this come to conditional
expectations.

00:01:06.810 --> 00:01:09.780
The conditional expectation of
a discrete random variable is

00:01:09.780 --> 00:01:11.480
defined by this formula.

00:01:11.480 --> 00:01:15.490
It is like the ordinary
expectation except that we now

00:01:15.490 --> 00:01:19.060
live in a conditional universe
in which the random variable

00:01:19.060 --> 00:01:23.650
capital Y is known to have
taken a value little y.

00:01:23.650 --> 00:01:27.490
And therefore, instead of using
the ordinary formula for

00:01:27.490 --> 00:01:31.990
expectations that involve the
PMF of X, we now use that

00:01:31.990 --> 00:01:36.400
formula but with the conditional
PMF of X, which is

00:01:36.400 --> 00:01:38.860
the appropriate PMF
that applies to

00:01:38.860 --> 00:01:41.450
this conditional universe.

00:01:41.450 --> 00:01:44.250
And if it happens that the
random variable capital X is

00:01:44.250 --> 00:01:47.289
continuous, we would have an
alternative formula but of the

00:01:47.289 --> 00:01:51.120
same kind, where the summation
is replaced by an integral and

00:01:51.120 --> 00:01:54.789
the PMF is replaced by a PDF.

00:01:54.789 --> 00:01:57.910
Now let us look at this
quantity here.

00:01:57.910 --> 00:02:00.650
We have fixed some particular
little y.

00:02:00.650 --> 00:02:02.020
Calculate this quantity.

00:02:02.020 --> 00:02:05.390
And what we get is a number.

00:02:05.390 --> 00:02:09.949
It is a number, but the value of
that number depends on the

00:02:09.949 --> 00:02:11.840
choice of little y.

00:02:11.840 --> 00:02:15.080
If I give you a different little
y then you will get

00:02:15.080 --> 00:02:18.260
another number for this
conditional expectation.

00:02:18.260 --> 00:02:21.170
This means that this quantity
here is really a

00:02:21.170 --> 00:02:23.290
function of little y.

00:02:23.290 --> 00:02:25.579
And let us give a name
to this function.

00:02:25.579 --> 00:02:29.800
Let us call this function g.

00:02:29.800 --> 00:02:36.670
Now that we have defined g we
can ask, what is this object?

00:02:36.670 --> 00:02:39.500
It's a function of
capital Y. It's a

00:02:39.500 --> 00:02:40.860
function of a random variable.

00:02:40.860 --> 00:02:44.030
So it should be a random
variable by itself.

00:02:44.030 --> 00:02:48.590
By analogy, with the earlier
concrete example, it is the

00:02:48.590 --> 00:02:55.579
random variable that takes the
numerical value g of little y

00:02:55.579 --> 00:02:59.300
whenever capital Y happens to
take the value little y.

00:02:59.300 --> 00:03:03.080
But g of little y has been
defined to be the same as this

00:03:03.080 --> 00:03:05.150
conditional expectation.

00:03:05.150 --> 00:03:08.230
So it's the random variable
whose value is this

00:03:08.230 --> 00:03:11.940
conditional expectation, which
is a particular number, if

00:03:11.940 --> 00:03:16.150
capital y happens to take
the value little y.

00:03:16.150 --> 00:03:19.200
This particular random variable
that we have defined

00:03:19.200 --> 00:03:23.590
here, g of capital Y, we call
it the abstract conditional

00:03:23.590 --> 00:03:27.820
expectation of the random
variable X, given the random

00:03:27.820 --> 00:03:29.860
variable Y.

00:03:29.860 --> 00:03:33.960
To summarize, this notation
here stands

00:03:33.960 --> 00:03:35.665
for a random variable.

00:03:35.665 --> 00:03:39.590
It is the random variable whose
numerical value turns

00:03:39.590 --> 00:03:43.510
out to be this one if the value
of the random variable

00:03:43.510 --> 00:03:48.550
capital Y happens
to be little y.

00:03:48.550 --> 00:03:52.640
It is a function of capital Y.
Once we know the value of

00:03:52.640 --> 00:03:57.850
capital Y, then the value of the
conditional expectation is

00:03:57.850 --> 00:03:58.950
well defined.

00:03:58.950 --> 00:04:00.100
It is known.

00:04:00.100 --> 00:04:03.600
And it's equal to this
particular number.

00:04:03.600 --> 00:04:05.690
It is of course a
random variable.

00:04:05.690 --> 00:04:08.720
And as a random variable, it
has all the attributes that

00:04:08.720 --> 00:04:10.320
random variables have.

00:04:10.320 --> 00:04:12.440
For example, it has a
distribution, that

00:04:12.440 --> 00:04:14.250
is, a PMF or a PDF.

00:04:14.250 --> 00:04:16.399
It has a mean of its own.

00:04:16.399 --> 00:04:18.829
And it has a variance
of its own.

00:04:18.829 --> 00:04:22.680
So what will be next in our
agenda is to talk about these

00:04:22.680 --> 00:04:27.820
attributes of this special
random variable, and also to

00:04:27.820 --> 00:04:29.150
use it in several examples.