Confidence
Intervals
Sometimes,
rather than computing one point estimate, such as a mean, we want to
calculate a range of potential values for whatever statistic we are testing.
Why? Because we know there is chance of error in our sampling, so our
point estimate is likely to "be off" somewhat. A confidence interval
is the range of plausible values, within some level of error, for your
statistic (point estimate). We can conduct hypothesis tests on a
confidence interval, similar to that of a statistic.
Ex. We want to know what the range is for the average difference
in drinks between men and women. We
know the average difference is 3 in our sample, but what would be a likely
range for any potential sample we would pull from that population using the
same research methods.
For 95%
confidence, the interval would be: Statistic +/- .025
Formula:
See board
Always divide alpha in half. CI's are always two tailed.
The Steps to Computing Confidence
Intervals
1.
First you choose your level of
confidence: 90, 95, 99
Ex. You
will say you are 95% that the population mean falls between those two values.
2. Divide error in half. Look up critical t for that value.
3. Then use the formula to compute the interval.
n=100,
mean difference =3, alpha=.95, s=5
t
critical = 1.98
3+/-.99
Confidence
interval = 2.01-3.99
We are
95% confident that in the population the average difference in weekly alcohol
consumption between men and women is 2.01 to 3.99 drinks.
The
higher the confidence, the wider the interval.
We
can also use the confidence interval to test your hypothesis:
If the
null hypothesis value is included in the interval than you must accept the null
hypothesis.
In our
null hypothesis we state “no difference” or group1-group2 =0.
So if 0 was in the interval, than there was 5% chance or higher that
there could be no difference between men and women’s drinking.
If
the null value, in this case 0, is not included in the interval, than you can
reject the null. In this case, the
interval does not include 0, so we can conclude that men drink more. On
average they drink 3 glasses more.
What if
the null hypothesis said men drink an average of 0-3 drinks more per week
than women, and the research hypothesis said men drink more than that per week
than women:
3 is contained in the interval. So we must accept the null hypothesis.