|
V32
Hypothesis Testing Welcome to part eight of our
video series in support of hypothesis testing. In this video, we are going to
review the chi square test statistic involving the
population variance, sigma squared. We're going to do a visual review of the chi
square distribution. We will also review the F test statistic
involving the ratio of two population variances from independent populations, and also do a visual review of the F
distribution. I'm Renee Clark from the Swanson School of Engineering at the
University of Pittsburgh. Okay, for inference on one
variance, sigma squared, okay, we know that a point estimate for sigma squared
is s squared, or the sample variance. Okay, so, if
we take a sample out of a normally distributed population, then this
statistic or random variable, which is n minus 1 * the sample variance over
the population variance, has a chi squared distribution, which we write as chi
squared with n minus 1 degrees of freedom. Okay, or sample size minus one degrees of freedom. Note that our population must be
normally distributed when using the Chi Square distribution for this
inference. Okay, let's do a visual review of the chi square distribution. Okay,
recall that the chi square distribution, as shown here, is… it's a continuous
distribution. It is skewed to the right because it has a long right tail to
the right. Okay, if I were to identify this point right here along the x-axis,
that's a particular critical value of the chi square distribution that is actually labeled as chi^ 2 sub alpha / 2. The reason being
is because this particular critical value is showed
as having alpha over 2 area to the right in the picture,
and remember that the subscript on the critical value by convention
refers to area to the right of it under the curve. Okay, so, likewise… wise,
this value right here in red along the x-axis is designated as chi^ 2 sub 1 -
alpha / 2, and the reason for that is you will notice that, in the picture,
it has alpha over 2 area to the left of it under the curve. That means it
must have area 1 - alpha /2 to the right, since total area under the curve
must equal to one. Okay, so, when we are doing
inference with one variance, this is our null hypothesis versus the two-sided
alternative that the variance is not equal to some hypothesized variance. Okay,
but, to proceed with a proof
by contradiction in order to prove the alternative, we have to assume the
null to be true. Therefore, we will insert the null hypothesized value for sigma
squared into our chi square test statistic. Okay, and, this quantity becomes our chi square random variable
or test statistic for performing inference on the variance, sigma squared. So,
we have a hypothesized value for sigma squared in the denominator. Okay, let's next… next talk about
inference for the ratio of two variances from two independent populations, or
sigma 1^ 2 over Sigma 2^ 2. Okay, so, we know a point estimator. From the
past, we know the point estimator for sigma 1^ 2/ sigma 2^ 2 is s1 ^2 / s 2^2,
okay, or the ratio of the two sample variances from those populations. Okay,
so, if we take a sample from each of two normally distributed independent
populations, okay, then this particular statistic
here, okay, which is the ratio of the sample to the population variance in
the first pop… population, divided by the ratio of the sample variance to the
population variance of the second population. That particular
random variable or statistic has an F distribution, right, and this
can actually be written in that way as well just by some simple rearrangement
of the… of the terms. Okay, this random variable..able
has n1 -1 and n2 -1 degrees of freedom- so two different degrees of freedom. But,
like the chi square distribution, each population, from which… from which
we're sampling to do this inference, must be normally distributed. So, when
using the F distribution in this way, your populations must be normally
distributed. Okay, so, again, let's do a
visual review of the F distribution. This point right here along the x-axis,
which is a particular critical value of the F distribution, would be labeled,
or should be labeled as F sub alpha / 2 because it is shown as having area of
alpha over 2 to the right, and you'll remember the subscript on the critical
value by convention represents area to the right. Okay, so, likewise, this particular critical value of F at that point along the x
axis should be labeled as F sub 1 minus alpha / 2. Okay, the reason being is
because it is shown as having alpha over 2 to the
left of it under the curve. So, if it's got Alpha over 2 area
to the left, that means it must have 1 - alpha over 2 area to the right of it
under the curve, right, because the total area under the F curve must equal
one. Okay, in performing inference for
the ratio of two variances, sigma 1^ 2 to sigma 2^ 2, okay, this is our
typical null hypothesis- that the two population variances are equal. Okay,
another way to write that is that the ratio of sigma 1^ 2 to sigma 2^ 2 equal
to 1, simply by dividing by sigma 2^ 2 on each side of the equality. Or,
equivalently, we could say… we could write this as sigma 2^ 2 to sigma 1^ 2 equal
to 1. Either way, in which case you would be dividing
by sigma 1^2 on each side of the equality. Okay, so, in proceeding with a proof by contradiction, what we are going to do is we
are going to insert the null hypothesized information into our test statistic.
Okay, so, we saw on the previous page that this is our test statistic
distributed according to the F distribution involving our two population
variances. Okay, it can be written either way, but what we're going to do is,
when we substitute this information in about the ratio of the two population
variances, we're going to get a cancellation, okay? If that hypothesized
ratio is one that's going to cancel or, alternatively, if you're going to
look at it the other way, it's going to cancel. So, what you're ultimately
left with is that the F test statistic can be simplified to the ratio of the
sample variances, s1^ 2 over s2 ^2- probably the simplest test statistic
we've seen thus far. Okay, this, of course, will have n1 -1 degrees of
freedom and n2 - 1 degrees of freedom- two separate degrees of freedom. We wish to thank the National
Science Foundation under Grant 233582 for supporting our work. Thank you for watching. |