WEBVTT

00:00:00.640 --> 00:00:03.900
We have already introduced the
concept of the conditional PMF

00:00:03.900 --> 00:00:08.720
of a random variable, X, given
an event A. We will now

00:00:08.720 --> 00:00:12.150
consider the case where we
condition on the value of

00:00:12.150 --> 00:00:18.690
another random variable Y. That
is, we let A be the event

00:00:18.690 --> 00:00:22.500
that some other random variable,
Y, takes on a

00:00:22.500 --> 00:00:26.310
specific value, little y.

00:00:26.310 --> 00:00:28.760
In this case, we're talking
about a conditional

00:00:28.760 --> 00:00:33.240
probability of the
form shown here.

00:00:33.240 --> 00:00:34.420
The conditional probability--

00:00:34.420 --> 00:00:38.110
that X takes on a specific
value, given that the random

00:00:38.110 --> 00:00:42.160
variable Y takes on another
specific value.

00:00:42.160 --> 00:00:46.300
And we use this notation to
indicate those conditional

00:00:46.300 --> 00:00:48.200
probabilities.

00:00:48.200 --> 00:00:52.750
As usual, the subscripts
indicate the situation that

00:00:52.750 --> 00:00:54.070
we're dealing with.

00:00:54.070 --> 00:00:56.280
That is, we're dealing with the
distribution of the random

00:00:56.280 --> 00:00:59.310
variable X and we're
conditioning on values of the

00:00:59.310 --> 00:01:02.260
other random variable, Y.

00:01:02.260 --> 00:01:05.990
Using the definition now of
conditional probabilities this

00:01:05.990 --> 00:01:10.020
can be written as the
probability that both events

00:01:10.020 --> 00:01:15.450
happen divided by the
probability of the

00:01:15.450 --> 00:01:16.700
conditioning event.

00:01:21.840 --> 00:01:25.490
We can turn this expression
into PMF notation.

00:01:25.490 --> 00:01:28.330
And this leads us to
this definition

00:01:28.330 --> 00:01:30.390
of conditional PMFs.

00:01:30.390 --> 00:01:34.289
The conditional PMF is defined
to be the ratio

00:01:34.289 --> 00:01:35.900
of the joint PMF--

00:01:35.900 --> 00:01:38.440
this is the probability
that we have here--

00:01:38.440 --> 00:01:41.330
by the corresponding
marginal PMF.

00:01:41.330 --> 00:01:44.200
And this is the probability
that we have here.

00:01:44.200 --> 00:01:47.190
Now, remember that conditional
probabilities are only defined

00:01:47.190 --> 00:01:51.210
when the conditioning event has
a positive probability,

00:01:51.210 --> 00:01:53.770
when this denominator
is positive.

00:01:53.770 --> 00:01:58.200
Similarly, the conditional PMF
will only be defined for those

00:01:58.200 --> 00:02:03.990
little y that have positive
probability of occurring.

00:02:03.990 --> 00:02:08.680
Now, the conditional PMF is a
function of two arguments,

00:02:08.680 --> 00:02:10.570
little x and little y.

00:02:10.570 --> 00:02:14.500
But the best way of thinking
about the conditional PMF is

00:02:14.500 --> 00:02:21.270
that we fix the value, little
y, and then view this

00:02:21.270 --> 00:02:25.100
expression here as
a function of x.

00:02:25.100 --> 00:02:28.390
As a function of x, it gives
us the probabilities of the

00:02:28.390 --> 00:02:32.690
different x's that may occur in
the conditional universe.

00:02:32.690 --> 00:02:37.480
And these probabilities must,
of course, sum to 1.

00:02:37.480 --> 00:02:39.579
Again, we're keeping y fixed.

00:02:39.579 --> 00:02:42.510
We live in a conditional
universe where y takes on a

00:02:42.510 --> 00:02:43.870
specific value.

00:02:43.870 --> 00:02:46.110
And here we have the
probabilities of the different

00:02:46.110 --> 00:02:47.840
x's in that universe.

00:02:47.840 --> 00:02:49.385
And these sum to 1.

00:02:52.130 --> 00:02:56.370
Note that if we change the value
of little y, we will, of

00:02:56.370 --> 00:02:59.740
course, get a different
conditional PMF for the random

00:02:59.740 --> 00:03:04.090
variable X. So what we're really
dealing with in this

00:03:04.090 --> 00:03:08.910
instance is that we have a
family of conditional PMFs,

00:03:08.910 --> 00:03:13.500
one conditional PMF for every
possible value of little y.

00:03:13.500 --> 00:03:16.170
And for every possible value
of little y, we have a

00:03:16.170 --> 00:03:20.380
legitimate PMF who's
values add to 1.

00:03:20.380 --> 00:03:22.320
Let's look at an example.

00:03:22.320 --> 00:03:25.630
Consider the joint PMF
given in this table.

00:03:25.630 --> 00:03:31.690
Let us condition on the event
that Y is equal to 2, which

00:03:31.690 --> 00:03:34.480
corresponds to this row
in the diagram.

00:03:39.520 --> 00:03:44.829
We need to know the value of the
marginal at this point, so

00:03:44.829 --> 00:03:50.800
we start by calculating the
probability of Y at value 2.

00:03:50.800 --> 00:03:53.560
And this is found by adding
the entries in

00:03:53.560 --> 00:03:55.400
this row of the table.

00:03:55.400 --> 00:03:58.920
And we find that this
is 5 over 20.

00:03:58.920 --> 00:04:01.700
Then we can start calculating
entries of

00:04:01.700 --> 00:04:03.730
the conditional PMF.

00:04:03.730 --> 00:04:09.230
So for example, the probability
that X takes on

00:04:09.230 --> 00:04:14.580
the value of 1 given that Y
takes the value of 2, it is

00:04:14.580 --> 00:04:20.320
going to be this entry, which
is 0, divided by 5/20, which

00:04:20.320 --> 00:04:22.360
gives us 0.

00:04:22.360 --> 00:04:27.710
We can find the next entry, the
probability of X taking

00:04:27.710 --> 00:04:33.180
the value of 2, given that Y
takes the value of 2 will be

00:04:33.180 --> 00:04:38.400
this entry, 1/20 divided
by 5/20.

00:04:38.400 --> 00:04:39.730
So it's going to be 1/5.

00:04:42.990 --> 00:04:46.180
And we can continue with
the other two entries.

00:04:46.180 --> 00:04:53.490
And we can actually even plot
the result once we're done.

00:04:53.490 --> 00:04:59.560
And what we have is that at 1,
we have a probability of 0.

00:04:59.560 --> 00:05:02.860
At 2, we have a probability
of 1/5.

00:05:07.588 --> 00:05:13.200
At 3, we have a probability
of 3/20 divided

00:05:13.200 --> 00:05:19.070
5/20, which is 3/5.

00:05:19.070 --> 00:05:23.665
And at 4, we have, again,
a probability of 1/5.

00:05:26.520 --> 00:05:31.040
So what we have plotted here
is the conditional PMF.

00:05:31.040 --> 00:05:36.060
It's a PMF in the variable x,
where x ranges over the

00:05:36.060 --> 00:05:40.710
possible values, but where we
have fixed the value of y to

00:05:40.710 --> 00:05:43.120
be equal to 2.

00:05:43.120 --> 00:05:46.560
Now, we could have found this
conditional PMF even faster

00:05:46.560 --> 00:05:50.600
without doing any divisions
by following the intuitive

00:05:50.600 --> 00:05:53.200
argument that we have
used before.

00:05:53.200 --> 00:05:55.740
We live in this conditional
universe.

00:05:55.740 --> 00:05:59.150
We have conditioned on
Y being equal to 2.

00:05:59.150 --> 00:06:03.740
The conditional probabilities
will have the same proportions

00:06:03.740 --> 00:06:06.660
as the original probabilities,
except that they needed to be

00:06:06.660 --> 00:06:09.050
scaled so that they add to 1.

00:06:09.050 --> 00:06:13.160
So they should be in the
proportions of 0, 1, 3, 1.

00:06:13.160 --> 00:06:17.670
And for these to add to 1, we
need to put everywhere a

00:06:17.670 --> 00:06:19.640
denominator of 5.

00:06:19.640 --> 00:06:24.470
So the proportions are indeed
0, 1, 3, and 1.

00:06:24.470 --> 00:06:30.730
Pictorially, the conditional PMF
has the same form as the

00:06:30.730 --> 00:06:36.640
corresponding slice of the joint
PMF, except, again, that

00:06:36.640 --> 00:06:40.159
the entries of that slice are
renormalized so that the

00:06:40.159 --> 00:06:42.810
entries add to 1.

00:06:42.810 --> 00:06:45.480
And finally, an observation--

00:06:45.480 --> 00:06:49.280
we can take the definition of
the conditional PMF and turn

00:06:49.280 --> 00:06:53.300
it around by moving the
denominator to the other side

00:06:53.300 --> 00:06:57.040
and obtain a formula, which
is a version of the

00:06:57.040 --> 00:06:59.450
multiplication rule.

00:06:59.450 --> 00:07:03.850
The probability that X takes a
value little x and Y takes a

00:07:03.850 --> 00:07:07.880
value little y is the product
or the probability that Y

00:07:07.880 --> 00:07:11.080
takes this particular value
times the conditional

00:07:11.080 --> 00:07:15.420
probability that X takes on the
particular value little x,

00:07:15.420 --> 00:07:19.970
given that Y takes on the
particular value little y.

00:07:19.970 --> 00:07:23.960
We also have a symmetrical
relationship if we interchange

00:07:23.960 --> 00:07:29.430
the roles of X and Y. As we
discussed earlier in this

00:07:29.430 --> 00:07:32.520
course, the multiplication rule
can be used to specify

00:07:32.520 --> 00:07:34.210
probability models.

00:07:34.210 --> 00:07:36.990
One way of modeling two
random variables is by

00:07:36.990 --> 00:07:39.580
specifying the joint PMF.

00:07:39.580 --> 00:07:43.060
But we now have an alternative,
indirect, way

00:07:43.060 --> 00:07:45.260
using the multiplication rule.

00:07:45.260 --> 00:07:49.580
We can first specify the
distribution of Y and then

00:07:49.580 --> 00:07:54.200
specify the conditional
PMF of X for any given

00:07:54.200 --> 00:07:56.510
value of little y.

00:07:56.510 --> 00:08:00.540
And this completely determines
the joint PMF, and so we have

00:08:00.540 --> 00:08:04.560
a full probability model.

00:08:04.560 --> 00:08:07.540
We can also provide similar
definitions of conditional

00:08:07.540 --> 00:08:10.460
PMFs for the case where we're
dealing with more than two

00:08:10.460 --> 00:08:12.120
random variables.

00:08:12.120 --> 00:08:16.080
In this case, notation is
pretty self-explanatory.

00:08:16.080 --> 00:08:21.930
By looking at this expression
here, you can probably guess

00:08:21.930 --> 00:08:26.230
that this stands for the
probability that random

00:08:26.230 --> 00:08:31.720
variable X takes on a specific
value, conditional on the

00:08:31.720 --> 00:08:36.380
random variables Y and
Z taking on some

00:08:36.380 --> 00:08:38.520
other specific values.

00:08:38.520 --> 00:08:41.429
Using the definition of
conditional probabilities,

00:08:41.429 --> 00:08:48.270
this is the probability that all
events happen divided by

00:08:48.270 --> 00:08:51.810
the probability of the
conditioning event, which, in

00:08:51.810 --> 00:08:56.930
our case, is the event that Y
takes on a specific value and

00:08:56.930 --> 00:09:01.280
simultaneously, Z takes another
specific value.

00:09:01.280 --> 00:09:07.080
In PMF notation, this is the
ratio of the joint PMF of the

00:09:07.080 --> 00:09:12.000
three random variables together,
divided by the joint

00:09:12.000 --> 00:09:19.670
PMF of the two random variables
Y and Z. As another

00:09:19.670 --> 00:09:24.280
example, we could have an
expression like this, which,

00:09:24.280 --> 00:09:28.370
again, stands for the
probability that these two

00:09:28.370 --> 00:09:35.120
random variables take on
specific values, conditional

00:09:35.120 --> 00:09:41.530
on this random variable taking
on another value.

00:09:41.530 --> 00:09:45.020
Finally, we can have versions of
the multiplication rule for

00:09:45.020 --> 00:09:47.460
the case where we're
dealing with more

00:09:47.460 --> 00:09:49.360
than two random variables.

00:09:49.360 --> 00:09:51.770
Recall the usual multiplication
rule.

00:09:51.770 --> 00:09:55.580
For three events happening
simultaneously, let's apply

00:09:55.580 --> 00:09:59.430
this multiplication rule for the
case where the event, A,

00:09:59.430 --> 00:10:04.490
stands for the event that the
random variable X takes on a

00:10:04.490 --> 00:10:06.050
specific value.

00:10:06.050 --> 00:10:11.160
Let B be the event that Y takes
on a specific value, and

00:10:11.160 --> 00:10:15.600
C be the event that the random
variable Z takes

00:10:15.600 --> 00:10:17.470
on a specific value.

00:10:17.470 --> 00:10:20.720
Then we can take this relation,
the multiplication

00:10:20.720 --> 00:10:23.820
rule, and translate it
into PMF notation.

00:10:23.820 --> 00:10:27.470
The probability that all three
events happen is equal to the

00:10:27.470 --> 00:10:32.690
product of the probability that
the first event happens.

00:10:32.690 --> 00:10:36.000
Then we have the conditional
probability that the second

00:10:36.000 --> 00:10:39.550
event happens given that the
first happened, times the

00:10:39.550 --> 00:10:43.270
conditional probability that
the third event happens--

00:10:43.270 --> 00:10:45.210
this one-- given that the first

00:10:45.210 --> 00:10:46.650
two events have happened.