ASA Connect

Hi Harvey, 

It is not clear to me what you mean by "ranked nominal variables" and also whether these variables are somewhat related to one another (e.g., some of them help measure an underlying latent construct).  

It would also help if you could share what research question you are trying to address, how you selected the subjects who provided you with data and how you collected the data. 

Thanks, 

Isabella

------------------------------
Isabella Ghement
Ghement Statistical Consulting Company Ltd.

Original Message:
Sent: 12-03-2015 01:59
From: Harvey Brown
Subject: Analyzing ranked factors

Hello all.

How do I analyze how several ranked factors affect an ordinal dependent variable? For example I want to analyze how 10 nominal factors ranked from 1 to 10 influence the level of motivation (assumed to be ordinal) to learn another language? The levels of motivation are (in increasing levels of motivation) are very unmotivated, a little unmotivated, neutral, a little motivated,  very motivated).

Should I transform the 10 ranked factors into an interval variable and then use regression analysis? Or can I just look at the correlation between the factors and the level of motivation?

Any suggestions are welcome.

Thanks in advance,

Harvey Brown

(1) Are all of the 10 nominal factors binary? If not, how many are not, and can you give us an example of one or two of them that are not?

(2) What criteria are used to rank the 10 nominal factors?

------------------------------
Eric Siegel
Biostatistician
Univ of Arkansas for Medical Sciences, Department of Biostatistics

Original Message:
Sent: 12-03-2015 01:59
From: Harvey Brown
Subject: Analyzing ranked factors

Hello all.

How do I analyze how several ranked factors affect an ordinal dependent variable? For example I want to analyze how 10 nominal factors ranked from 1 to 10 influence the level of motivation (assumed to be ordinal) to learn another language? The levels of motivation are (in increasing levels of motivation) are very unmotivated, a little unmotivated, neutral, a little motivated,  very motivated).

Should I transform the 10 ranked factors into an interval variable and then use regression analysis? Or can I just look at the correlation between the factors and the level of motivation?

Any suggestions are welcome.

Thanks in advance,

Harvey Brown

The ten factors are nominal e.g. want to get good grade in course or need to learn for job. These subjects had to then rank all ten of the factors from 1 to 10. Now I'm trying to see how a subject's ranking of the ten factors influences the level of motivation (assumed to be ordinal as well). I hope this clears up any confusion.

------------------------------
Harvey Brown

Original Message:
Sent: 12-05-2015 17:24
From: Eric Siegel
Subject: Analyzing ranked factors

(1) Are all of the 10 nominal factors binary? If not, how many are not, and can you give us an example of one or two of them that are not?

(2) What criteria are used to rank the 10 nominal factors?

------------------------------
Eric Siegel
Biostatistician
Univ of Arkansas for Medical Sciences, Department of Biostatistics

(1) I'm still unclear about the properties of the nominal factor "want to get good grade in course". If that nominal factor is an item on a questionnaire, then (A) can the subject answer it Yes or No? and (B) can the subject answer it in other ways? I have the same questions about the other nominal factor, "need to learn for job".

(2) When we say we're trying to see how a subject's ranking of the ten factors influences the level of motivation, that makes it sound like we have no interest in how the subject actually answers the nominal factor "want to get good grade in course", i.e., whether the subject answers Yes or No or I Don't Know is irrelevant to how the subject's response is scored. All that matters is the numerical rank that the subject gives the factor. Is that a correct understanding of what you mean? 

------------------------------
Eric Siegel
Biostatistician
Univ of Arkansas for Medical Sciences of Biostatistics

Original Message:
Sent: 12-09-2015 00:55
From: Harvey Brown
Subject: Analyzing ranked factors

The ten factors are nominal e.g. want to get good grade in course or need to learn for job. These subjects had to then rank all ten of the factors from 1 to 10. Now I'm trying to see how a subject's ranking of the ten factors influences the level of motivation (assumed to be ordinal as well). I hope this clears up any confusion.

------------------------------
Harvey Brown

I suspect that we are having some problems here by use of the term nominal.  Am I correct that you have 10 distinctly different independent variables that are ranks, i.e., the sum of the 10 ranked variables must equal 1+2+3+4+5+6+7+8+9+10=55, and one dependent variable (motivation) that is a simple ordinal rating?

If your ten independent variables are ranks such as I've described, then recognize that any regression-based solution would only allow you to use 9 of those variables, since the 10th is totally predetermined by the other 9.  And of course, all of this begs the question regarding the legitimacy of using ordinal data in regression models.

------------------------------
David Mangen

Original Message:
Sent: 12-12-2015 19:35
From: Eric Siegel
Subject: Analyzing ranked factors

(1) I'm still unclear about the properties of the nominal factor "want to get good grade in course". If that nominal factor is an item on a questionnaire, then (A) can the subject answer it Yes or No? and (B) can the subject answer it in other ways? I have the same questions about the other nominal factor, "need to learn for job".

(2) When we say we're trying to see how a subject's ranking of the ten factors influences the level of motivation, that makes it sound like we have no interest in how the subject actually answers the nominal factor "want to get good grade in course", i.e., whether the subject answers Yes or No or I Don't Know is irrelevant to how the subject's response is scored. All that matters is the numerical rank that the subject gives the factor. Is that a correct understanding of what you mean? 

------------------------------
Eric Siegel
Biostatistician
Univ of Arkansas for Medical Sciences of Biostatistics

Correct the 10 predictor variables are ordinal responses from 1 to 10. What I would like to measure is how one's ranking of these 10 factors influences one's level of motivation which is also a single ordinal response.

There are 7 different levels of motivation that a person can pick  from (from highly unmotivated to highly motivated). They are only allowed to pick one.

I hope this clears things up. Let me know if you have any questions.

------------------------------
Harvey Brown

Original Message:
Sent: 12-14-2015 11:38
From: David Mangen
Subject: Analyzing ranked factors

I suspect that we are having some problems here by use of the term nominal.  Am I correct that you have 10 distinctly different independent variables that are ranks, i.e., the sum of the 10 ranked variables must equal 1+2+3+4+5+6+7+8+9+10=55, and one dependent variable (motivation) that is a simple ordinal rating?

If your ten independent variables are ranks such as I've described, then recognize that any regression-based solution would only allow you to use 9 of those variables, since the 10th is totally predetermined by the other 9.  And of course, all of this begs the question regarding the legitimacy of using ordinal data in regression models.

------------------------------
David Mangen

Consider coding your ordinal independent variables using a scheme like that described in the following reference:

    Walter SD, Feinstein AR, Wells CK

        Coding ordinal independent variables in multiple regression analysis.

        American Journal of Epidemiology 1987;125(2):319-323.

Since your dependent variable is also ordinal, you can use several different logistic regression methods depending on the assumptions you want to make (for example, cumulative logit, adjacent categories, continuation-ratio).

------------------------------
Matthew Zack

Original Message:
Sent: 12-15-2015 05:56
From: Harvey Brown
Subject: Analyzing ranked factors

Correct the 10 predictor variables are ordinal responses from 1 to 10. What I would like to measure is how one's ranking of these 10 factors influences one's level of motivation which is also a single ordinal response.

There are 7 different levels of motivation that a person can pick  from (from highly unmotivated to highly motivated). They are only allowed to pick one.

I hope this clears things up. Let me know if you have any questions.

------------------------------
Harvey Brown