Search All of the Math Forum:

Views expressed in these public forums are not endorsed by NCTM or The Math Forum.

Notice: We are no longer accepting new posts, but the forums will continue to be readable.

Topic: multiple comparison with ambiguous ranks
Replies: 9   Last Post: Mar 13, 2013 4:46 PM

 Messages: [ Previous | Next ]
 Art Kendall Posts: 200 Registered: 12/7/04
Re: multiple comparison with ambiguous ranks
Posted: Mar 12, 2013 6:22 PM

Perhaps you want to consider CATREG i.e.,categorical regression. (one
version is in SPSS).
If your categories are measured at least on the ordinal level (are
_ratings_), you might try running CATREG using 3 levels of
measurement(nominal, ordinal, interval) for the 5 "group values". If
Interval or ordinal is better than nominal level you have evidence for
the validity of ordering.

A further exploration, you might also use the 3 integer values (1,2,3)
the same way and getting the model based on only those cases. Then see
what the predicted value would be for the cases in the 1.5 and 2.5
groups when you use the other measured as predictors.

If you are just trying to establish this measure of "condition" you
should consider using more values in your response scale. (In general,
it is better to use as fine a grained measurement at as high a
measurement level as practical considerations will allow.)

Please clarify what you mean by "evaluation".
Art Kendall
Social Research Consultants

On 3/12/2013 2:58 PM, Dave Dan wrote:
> It is a regression/classification problem. The data is like this:
> A set of variables describe patient's age, body mass index, etc.
> These variables are used to "predict" (using regression or

classification) the health condition, graded as "1", "2", "3", "Either 1
or 2", "Either 2 or 3", in total 5 labels/ranks.
> According to the doctor, predicting a patient with "Either 1 or 2" as
"1" is considered a correct prediction.
> According to the doctor, predicting a patient with "Either 1 or 2" as
"2" is also considered a correct prediction.
>
> I once considered using 1.5 to replace "Either 1 or 2". This can be a

valid approach during training of a regressor, but my original post is
concerned with the "evaluation" problem. I do NOT want separate
categories of 1.5 and 2.5 during "evaluation".
>
> Hope its clearer.
>
>

Art Kendall
Social Research Consultants

On 3/12/2013 12:39 PM, Art Kendall wrote:
> Please explain you data in more detail.
> One variable is condition (an at least ordinal variable).
> You want to relate condition to some other variables.
> What are the other variables and how are they measured?
>
> What questions are you trying to answer with the data?
>
>
> Art Kendall
> Social Research Consultants
>
> On 3/12/2013 11:19 AM, Da wrote:

>> For a clinical study, I have patient's health condition labeled as
>> grade 1, 2, or 3. Some patients are labeled as "Either 1 or 2" by the
>> doctor, meaning the doctor considers both grade 1 and 2 as valid
>> descriptions of the health condition.
>>
>> When performing say a Kruskal Wallis test, I don't want to consider
>> such patients as a separate group. What I wish to do is to run K-W
>> test multiple times, each time using either 1 or 2 as rank, and report
>> the final result based on the test with the lowest p-value.
>>
>> However, doing this would be slow, as I have 20 patients with such
>> label, so I need to run K-W Test 2^20 times.
>>
>> Any suggestions?
>> Thanks
>>

Date Subject Author
3/12/13 Art Kendall
3/12/13 Art Kendall
3/12/13 Richard Ulrich
3/12/13 Ray Koopman
3/13/13 Richard Ulrich
3/13/13 Ray Koopman
3/13/13 Bruce Weaver
3/13/13 Art Kendall
3/13/13 Bruce Weaver