WEBVTT

00:00:00.180 --> 00:00:01.940
- [Instructor] In a previous
video, we began to think about

00:00:01.940 --> 00:00:04.720
how we can use a regression
line and, in particular,

00:00:04.720 --> 00:00:08.090
the slope of a regression
line based on sample data,

00:00:08.090 --> 00:00:10.910
how we can use that in
order to make inference

00:00:10.910 --> 00:00:15.700
about the slope of the true
population regression line.

00:00:15.700 --> 00:00:17.960
In this video, what we're
going to think about,

00:00:17.960 --> 00:00:20.260
what are the conditions for inference

00:00:20.260 --> 00:00:22.610
when we're dealing with regression lines?

00:00:22.610 --> 00:00:24.900
And these are going to be, in some ways,

00:00:24.900 --> 00:00:27.280
similar to the conditions for inference

00:00:27.280 --> 00:00:30.320
that we thought about when we
were doing hypothesis testing

00:00:30.320 --> 00:00:33.920
and confidence intervals for
means and for proportions,

00:00:33.920 --> 00:00:36.890
but there's also going to
be a few new conditions.

00:00:36.890 --> 00:00:39.860
So to help us remember these conditions,

00:00:39.860 --> 00:00:44.860
you might want to think about
the LINER acronym, L-I-N-E-R.

00:00:46.950 --> 00:00:50.500
And if it isn't obvious to
you, this almost is linear.

00:00:50.500 --> 00:00:53.040
Liner, if it had an A, it would be linear.

00:00:53.040 --> 00:00:54.670
And this is valuable because, remember,

00:00:54.670 --> 00:00:57.140
we're thinking about linear regression.

00:00:57.140 --> 00:01:01.240
So the L right over here
actually does stand for linear.

00:01:01.240 --> 00:01:05.000
And here, the condition is, is
that the actual relationship

00:01:05.000 --> 00:01:08.620
in the population between
your x and y variables

00:01:08.620 --> 00:01:11.290
actually is a linear relationship,

00:01:11.290 --> 00:01:12.710
so actual

00:01:13.690 --> 00:01:14.750
linear

00:01:15.670 --> 00:01:16.853
relationship,

00:01:18.360 --> 00:01:19.310
relationship

00:01:20.230 --> 00:01:21.690
between,

00:01:21.690 --> 00:01:23.950
between x

00:01:23.950 --> 00:01:25.910
and y.

00:01:25.910 --> 00:01:28.920
Now, in a lot of cases, you
might just have to assume

00:01:28.920 --> 00:01:31.270
that this is going to be
the case when you see it on

00:01:31.270 --> 00:01:33.950
an exam, like an AP exam, for example.

00:01:33.950 --> 00:01:36.400
They might say, hey, assume
this condition is met.

00:01:36.400 --> 00:01:37.720
Oftentimes, it'll say assume all

00:01:37.720 --> 00:01:38.600
of these conditions are met.

00:01:38.600 --> 00:01:41.100
They just want you to maybe
know about these conditions.

00:01:41.100 --> 00:01:42.810
But this is something to think about.

00:01:42.810 --> 00:01:45.660
If the underlying
relationship is nonlinear,

00:01:45.660 --> 00:01:47.250
well, then maybe some of your

00:01:47.250 --> 00:01:50.150
inferences might not be as robust.

00:01:50.150 --> 00:01:53.290
Now, the next one is
one we have seen before

00:01:53.290 --> 00:01:55.560
when we're talking about general
conditions for inference,

00:01:55.560 --> 00:01:57.530
and this is the independence,

00:01:57.530 --> 00:01:59.960
independence condition.

00:01:59.960 --> 00:02:01.980
And there's a couple of
ways to think about it.

00:02:01.980 --> 00:02:04.070
Either individual observations

00:02:04.070 --> 00:02:05.830
are independent of each other.

00:02:05.830 --> 00:02:09.180
So you could be sampling with replacement.

00:02:09.180 --> 00:02:11.910
Or you could be thinking
about your 10% rule,

00:02:11.910 --> 00:02:13.430
that we have done when we thought about

00:02:13.430 --> 00:02:18.200
the independence condition
for proportions and for means,

00:02:18.200 --> 00:02:20.010
where we would need to feel confident

00:02:20.010 --> 00:02:23.710
that the size of our
sample is no more than 10%

00:02:23.710 --> 00:02:26.070
of the size of the population.

00:02:26.070 --> 00:02:28.140
Now, the next one is the normal condition,

00:02:28.140 --> 00:02:30.230
which we have talked about
when we were doing inference

00:02:30.230 --> 00:02:32.610
for proportions and for means.

00:02:32.610 --> 00:02:35.170
Although, it means something a
little bit more sophisticated

00:02:35.170 --> 00:02:37.580
when we're dealing with a regression.

00:02:37.580 --> 00:02:39.590
The normal condition, and, once again,

00:02:39.590 --> 00:02:42.160
many times people just
say assume it's been met.

00:02:42.160 --> 00:02:43.820
But let me actually
draw a regression line,

00:02:43.820 --> 00:02:44.880
but do it with a little perspective,

00:02:44.880 --> 00:02:46.670
and I'm gonna add a third dimension.

00:02:46.670 --> 00:02:48.410
Let's say that's the x-axis,

00:02:48.410 --> 00:02:50.500
and let's say this is the y-axis.

00:02:50.500 --> 00:02:54.810
And the true population
regression line looks like this.

00:02:54.810 --> 00:02:57.270
And so the normal condition tells us

00:02:57.270 --> 00:03:00.033
that, for any given x
in the true population,

00:03:00.870 --> 00:03:05.770
the distribution of y's that
you would expect is normal,

00:03:05.770 --> 00:03:06.603
is normal.

00:03:06.603 --> 00:03:08.810
So let me see if I can
draw a normal distribution

00:03:08.810 --> 00:03:10.910
for the y's,

00:03:10.910 --> 00:03:11.870
given that x.

00:03:11.870 --> 00:03:13.990
So that would be that
normal distribution there.

00:03:13.990 --> 00:03:16.860
And then let's say, for
this x right over here,

00:03:16.860 --> 00:03:21.300
you would expect a normal
distribution as well,

00:03:21.300 --> 00:03:23.460
so just like,

00:03:23.460 --> 00:03:24.530
just like this.

00:03:24.530 --> 00:03:25.380
So if we're given x,

00:03:25.380 --> 00:03:27.760
the distribution of y's should be normal.

00:03:27.760 --> 00:03:29.750
Once again, many times you'll just be

00:03:29.750 --> 00:03:32.470
told to assume that that has
been met because it might,

00:03:32.470 --> 00:03:34.390
at least in an introductory
statistics class,

00:03:34.390 --> 00:03:36.970
be a little bit hard to
figure this out on your own.

00:03:36.970 --> 00:03:38.810
Now, the next condition
is related to that,

00:03:38.810 --> 00:03:42.790
and this is the idea of
having equal variance,

00:03:42.790 --> 00:03:45.090
equal variance.

00:03:45.090 --> 00:03:46.390
And that's just saying that each

00:03:46.390 --> 00:03:48.670
of these normal distributions should have

00:03:48.670 --> 00:03:51.250
the same spread for a given x.

00:03:51.250 --> 00:03:52.870
And so you could say equal variance,

00:03:52.870 --> 00:03:54.520
or you could even think about them having

00:03:54.520 --> 00:03:56.360
the equal standard deviation.

00:03:56.360 --> 00:03:59.880
So, for example, if, for a
given x, let's say for this x,

00:03:59.880 --> 00:04:02.580
all of sudden, you had
a much lower variance,

00:04:02.580 --> 00:04:03.620
made it look like this,

00:04:03.620 --> 00:04:06.890
then you would no longer meet
your conditions for inference.

00:04:06.890 --> 00:04:10.430
Last, but not least, and this
is one we've seen many times,

00:04:10.430 --> 00:04:12.300
this is the random condition.

00:04:12.300 --> 00:04:14.600
And this is that the data comes from

00:04:14.600 --> 00:04:17.170
a well-designed random sample or

00:04:17.170 --> 00:04:19.200
some type of randomized experiment.

00:04:19.200 --> 00:04:23.040
And this condition we have
seen in every type of condition

00:04:23.040 --> 00:04:25.760
for inference that we
have looked at so far.

00:04:25.760 --> 00:04:27.140
So I'll leave you there.

00:04:27.140 --> 00:04:28.270
It's good to know.

00:04:28.270 --> 00:04:30.470
It will show up on some exams.

00:04:30.470 --> 00:04:32.960
But many times, when it
comes to problem solving,

00:04:32.960 --> 00:04:36.130
in an introductory statistics
class, they will tell you,

00:04:36.130 --> 00:04:38.720
hey, just assume the conditions
for inference have been met.

00:04:38.720 --> 00:04:40.910
Or what are the conditions for inference?

00:04:40.910 --> 00:04:42.970
But they're not going to
actually make you prove,

00:04:42.970 --> 00:04:46.010
for example, the normal or
the equal variance condition.

00:04:46.010 --> 00:04:47.040
That might be a bit much

00:04:47.040 --> 00:04:49.763
for an introductory statistics class.