Thread: What type of statistical test should I use?

For some data in a survey I work on, I have two sets of current year-
prior year ratios. I have a set of current-year prior year ratios for
numbers that ended up being changed from what was originally reported
by the data respondent (around 500 numbers in this set) and another
set of current-year prior year ratios for numbers that were not
changed from what was originally reported (around 800,000 numbers in
this set). I wanted to do a statistical test to determine if the
average current year-prior year ratio of the numbers that were changed
on the survey are significantly different from the average current
year prior year ratio of numbers that were not changed. What kind of
test should I use? Also wanted to know if it might be better to
separate the current-year prior year ratios into ratios that increased
from the prior year vs. decreased from the prior year and run tests on
each group?? Thanks for any advice...

Julie

Originally Posted by wardnine

For some data in a survey I work on, I have two sets of current year-
prior year ratios. I have a set of current-year prior year ratios for
numbers that ended up being changed from what was originally reported
by the data respondent (around 500 numbers in this set) and another
set of current-year prior year ratios for numbers that were not
changed from what was originally reported (around 800,000 numbers in
this set). I wanted to do a statistical test to determine if the
average current year-prior year ratio of the numbers that were changed
on the survey are significantly different from the average current
year prior year ratio of numbers that were not changed. What kind of
test should I use? Also wanted to know if it might be better to
separate the current-year prior year ratios into ratios that increased
from the prior year vs. decreased from the prior year and run tests on
each group?? Thanks for any advice...

Julie

Forget about them being ratios, what you have are two sets of numbers
and you want to know if they are from the same distribution. Looks like
a chi-squared test is what is needed here (the Kolmogorov-Smirnov test
is also probably appropriate).

RonL