WEBVTT

00:00:00.270 --> 00:00:03.330
When we have independence,
does anything interesting

00:00:03.330 --> 00:00:05.640
happen to expectations?

00:00:05.640 --> 00:00:08.400
We know that, in general, the
expected value of a function

00:00:08.400 --> 00:00:12.180
of random variables is not the
same as applying the function

00:00:12.180 --> 00:00:14.430
to the expected values.

00:00:14.430 --> 00:00:18.820
And we also know that there are
some exceptions where we

00:00:18.820 --> 00:00:20.770
do get equality.

00:00:20.770 --> 00:00:23.820
This is the case where we are
dealing with linear functions

00:00:23.820 --> 00:00:28.040
of one or more random
variables.

00:00:28.040 --> 00:00:34.980
Note that this last property
is always true and does not

00:00:34.980 --> 00:00:39.160
require any independence
assumptions.

00:00:39.160 --> 00:00:42.070
When we have independence,
there is one additional

00:00:42.070 --> 00:00:45.490
property that turns
out to be true.

00:00:45.490 --> 00:00:49.650
The expected value of the
product of two independent

00:00:49.650 --> 00:00:52.340
random variables is
the product of

00:00:52.340 --> 00:00:55.360
their expected values.

00:00:55.360 --> 00:00:58.580
Let us verify this relation.

00:00:58.580 --> 00:01:02.160
We are dealing here with the
expected value of a function

00:01:02.160 --> 00:01:09.289
of random variables, where the
function is defined to be the

00:01:09.289 --> 00:01:10.539
product function.

00:01:13.300 --> 00:01:18.660
So to calculate this expected
value, you can use the

00:01:18.660 --> 00:01:22.210
expected value rule.

00:01:22.210 --> 00:01:27.580
And we are going to get the sum
over all x, the sum over

00:01:27.580 --> 00:01:34.930
all y, of g of xy, but in this
case, g of xy is x times y.

00:01:34.930 --> 00:01:39.740
And then we weigh all those
values according to the

00:01:39.740 --> 00:01:44.039
probabilities as given
by the joint PMF.

00:01:44.039 --> 00:01:53.930
Now, using independence, this
sum can be changed into the

00:01:53.930 --> 00:01:56.440
following form--

00:01:56.440 --> 00:02:00.520
the joint PMF is the product
of the marginal PMFs.

00:02:03.120 --> 00:02:06.250
And now when we look at the
inner sum over all values of

00:02:06.250 --> 00:02:12.010
y, we can take outside the
summation those terms that do

00:02:12.010 --> 00:02:19.260
not depend on y, and so this
term and that term.

00:02:19.260 --> 00:02:28.680
And this is going to yield a
summation over x of x times

00:02:28.680 --> 00:02:34.650
the marginal PMF of X, and then
the summation over all y

00:02:34.650 --> 00:02:39.630
of y times the marginal PMF of
Y. But now we recognize that

00:02:39.630 --> 00:02:46.230
here we have just the expected
value of Y. And then we will

00:02:46.230 --> 00:02:48.880
be left with another expression,
which is the

00:02:48.880 --> 00:02:59.780
expected value of X. And this
completes the argument.

00:02:59.780 --> 00:03:08.520
Now, consider a function of X
and another function of Y. X

00:03:08.520 --> 00:03:10.480
and Y are independent.

00:03:10.480 --> 00:03:14.560
Intuitively, the value of X
does not give you any new

00:03:14.560 --> 00:03:18.740
information about Y, so the
value of g of X does not to

00:03:18.740 --> 00:03:23.740
give you any new information
about h of Y. So on the basis

00:03:23.740 --> 00:03:28.230
of this intuitive argument, the
functions g of X and h of

00:03:28.230 --> 00:03:31.660
Y are also independent
of each other.

00:03:31.660 --> 00:03:34.670
Therefore, we can apply the
fact that we have already

00:03:34.670 --> 00:03:40.050
proved, but with g of X in the
place of X and h of Y in the

00:03:40.050 --> 00:03:45.030
place of Y. And this gives us
this more general fact that

00:03:45.030 --> 00:03:48.120
the expected value of the
product of two functions of

00:03:48.120 --> 00:03:52.270
independent random variables is
equal to the product of the

00:03:52.270 --> 00:03:55.520
expectations of these
functions.

00:03:55.520 --> 00:03:59.180
We could also prove this
property directly without

00:03:59.180 --> 00:04:01.690
relying on the intuitive
argument.

00:04:01.690 --> 00:04:05.970
We could just follow the same
steps as in this derivation.

00:04:05.970 --> 00:04:09.840
Wherever there is an X, we
would write g of X, and

00:04:09.840 --> 00:04:14.130
wherever there is a Y, we would
write h of Y. And the

00:04:14.130 --> 00:04:17.100
same algebra would go through,
and we would end up with the

00:04:17.100 --> 00:04:22.000
expected value of g of X times
the expected value of h of Y.