Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss
nominal scale - comparison of populations
16 views
Skip to first unread message
czyta...@gmail.com
Dec 23, 2012, 4:42:53 PM12/23/12
to
Hi,
Using a short questionnaire I have interviewed two populations. For majority of questions the answers are of the nominal type only (there are more than two possible answers which are self excluding and complete). Now, I would like to check if these populations are/are not the same with respect to the frequencies given to answers. Therefore, the every possible answer was coded as a distinct integer number. This gives me a probability distribution. Can I use statistical tests to compare these distributions? If yes - would be the outcome of a test robust to the way in which the numbers are adjusted?
Best,
Gruppo
Using a short questionnaire I have interviewed two populations. For majority of questions the answers are of the nominal type only (there are more than two possible answers which are self excluding and complete). Now, I would like to check if these populations are/are not the same with respect to the frequencies given to answers. Therefore, the every possible answer was coded as a distinct integer number. This gives me a probability distribution. Can I use statistical tests to compare these distributions? If yes - would be the outcome of a test robust to the way in which the numbers are adjusted?
Best,
Gruppo
Paul Hightower
Dec 23, 2012, 6:47:19 PM12/23/12
to
<czyta...@gmail.com> wrote in message
news:49a8c832-2239-487d...@googlegroups.com...
---------------------------------------------------------------------------------------------------------
I really have no business commenting here, I am merely an introductory level
tutor. but your question might be elementary -- you have categorical data
such as (to take an example using automobiles) color, style, etc. For data
encoding purposes you may designate blue = 1, red =2, white = 3, etc. But to
treat this data as a single variable with ordinal or cardinal rank is
absurd. In what sense would white be 50% more than red and three times blue,
or even white > red > blue ? OK, you didn't mention rank, so presumably you
are aware of that.
Next, you can deal with proportions -- are there differences in colors among
different body stayles? I would attack such questions using chi-square
methods or perhaps ANOVA. Look up "test of homogeneity" in a standard stats
text, for example. Does that help?
news:49a8c832-2239-487d...@googlegroups.com...
I really have no business commenting here, I am merely an introductory level
tutor. but your question might be elementary -- you have categorical data
such as (to take an example using automobiles) color, style, etc. For data
encoding purposes you may designate blue = 1, red =2, white = 3, etc. But to
treat this data as a single variable with ordinal or cardinal rank is
absurd. In what sense would white be 50% more than red and three times blue,
or even white > red > blue ? OK, you didn't mention rank, so presumably you
are aware of that.
Next, you can deal with proportions -- are there differences in colors among
different body stayles? I would attack such questions using chi-square
methods or perhaps ANOVA. Look up "test of homogeneity" in a standard stats
text, for example. Does that help?
Ray Koopman
Dec 23, 2012, 9:02:21 PM12/23/12
to
the corresponding 2 x K contingency table, where K = the # of possible
answers to that question. If you do that for several questions, you
should adjust your threshhold for declaring a result "significant".
The simplest procedure is the so-called "Bonferroni correction": a
test is significant only if the product of its p-value times the
number of tests is less than the usual threshhold (e.g., .05).
czyta...@gmail.com
Dec 30, 2012, 4:14:30 PM12/30/12
to
On Monday, December 24, 2012 3:02:21 AM UTC+1, Ray Koopman wrote:
> To compare the groups on a single question, do a Pearson chi-square on
> the corresponding 2 x K contingency table, where K = the # of possible
> answers to that question. If you do that for several questions, you
> should adjust your threshhold for declaring a result "significant".
> The simplest procedure is the so-called "Bonferroni correction": a
> test is significant only if the product of its p-value times the
> number of tests is less than the usual threshhold (e.g., .05).
Thanks a lot!
> To compare the groups on a single question, do a Pearson chi-square on
> the corresponding 2 x K contingency table, where K = the # of possible
> answers to that question. If you do that for several questions, you
> should adjust your threshhold for declaring a result "significant".
> The simplest procedure is the so-called "Bonferroni correction": a
> test is significant only if the product of its p-value times the
> number of tests is less than the usual threshhold (e.g., .05).
0 new messages