When to treat variables as continuous variables?

[Faiz Rasool]

Hi all,

 

My apologies for a slightly non-SPSS question. I'm a Blind student of Sociology. Unfortunately I'm learning statistics from a teacher who is not an expert at statistics. I've been trying to improve my understanding of what type of analysis to use in different situations. One of the main ideas that I've find it difficult to draw a conclusion about is when certain variables can be treated as a continuous variables. So techniques like regression and tests like T tests can be used. As a Sociology student I am mostly dealing with categorical variables measured using likert scale. I've  read that chi-square test is appropriate for finding out the relationship between two categorical variables. yet I observe in journal articles and on this list as well that regression is   used in such situations. The textbooks I've read seems to assume that the student is been informed about this kind of thing in the class. But unfortunate students like me are left to learn for themselves. So my question is that what rule of thumb should I try to follow and keep in my mind as I try to improve my understanding of statistical analysis as a Sociology student.

 

Thanks all, and sorry for a non-SPSS question.

Faiz.

 

 

[John F Hall]

Hi Faiz

 

I'm John Hall, and as a non-statistician used to teach basic statistics as part of a data analysis course (with SPSS) to sociology students in a way they could understand, using data that was relevant i.e. data from questionnaire surveys and administrative records, not standard deviations of the length of planks on a lorry, as the lecturer supplied by the Maths Dept did.  I even had a blind student for whom I prepared a braille examination paper, which she deferred for a year and then didn't take.  There was an official course in statistics, but my students said they learned more from mine.

 

I taught my last course in 1992 when I took early retirement.  So now what?  I'm in the middle of converting and uploading materials from my courses to my new website  

http://surveyresearch.weebly.com/.  For now I suggest you go through the accompanying Statistical Notes for the data analysis course and theh try working your way through the tutorials for SPSS/PASW 18  They're arranged in four blocks and proceed gently through every step from completing a short questionnaire, coding the responses and transferring the data to a special sheet, typing the data into the computer and then analysing the data from all the questionnaires completed by more than 600 students on six cycles of the course.  The tutorials then move on to some data from major surveys and progress from frequency counts on categorical variables, then scale variables and through crosstabulation to basic descriptive statistics in non-mathematical language.  The emphasis is always on the research question and the logic and process needed  to answer the quaestion from the available data.: SPSS just happens to be the package used.  They are all syntax-based in preference to drop-down menus, but many exercises are repeated using the menus, although I'm not sure how you'll manage with the screenshots!

 

There's a list of textbooks on my site, but I can recommend others if necessary.

 

Let me know how you get on with all this and I'll see if I can be of further assistance.

 

John Hall

 

 

----- Original Message -----

From: [hidden email]

To: [hidden email]

Sent: Thursday, July 29, 2010 5:33 PM

Subject: When to treat variables as continuous variables?

Hi all,

 

My apologies for a slightly non-SPSS question. I'm a Blind student of Sociology. Unfortunately I'm learning statistics from a teacher who is not an expert at statistics. I've been trying to improve my understanding of what type of analysis to use in different situations. One of the main ideas that I've find it difficult to draw a conclusion about is when certain variables can be treated as a continuous variables. So techniques like regression and tests like T tests can be used. As a Sociology student I am mostly dealing with categorical variables measured using likert scale. I've  read that chi-square test is appropriate for finding out the relationship between two categorical variables. yet I observe in journal articles and on this list as well that regression is   used in such situations. The textbooks I've read seems to assume that the student is been informed about this kind of thing in the class. But unfortunate students like me are left to learn for themselves. So my question is that what rule of thumb should I try to follow and keep in my mind as I try to improve my understanding of statistical analysis as a Sociology student.

 

Thanks all, and sorry for a non-SPSS question.

Faiz.

 

 

[Martin Holt]

Hi, conventional wisdom, to the point of being dogma, advises against categorization of a continuous variable. I'm sure that that is true..Professor Martin Bland says so. So I was most interested in the following paper when I came across it:

 

"The exaggerated relations between diet, body weight and mortality: the case for a categorical data approach"
Welch H. Gilbert, Schwartz LM, Woloshin S.

CMAJ, March 29, 2005; 172 (7)

 

This provides an argument for sometimes taking the categorical approah.

 

I distributed this reference to members of MedStats recently (I'm MedStats Group Owner) but it elicited no response, either for or against.

 

I think it will only be of use to you if/when people more expert than I evaluate it. But, at the least, it's worth a read. I hope this is helpful, rather than causing more indecision. The safest approach, without feedback on this article, is to stick to using continuous rather than categorical.

 

Best Wishes,

 

Martin Holt

 

Martin Holt
Medical Statistician

--- On Thu, 29/7/10, John F Hall <[hidden email]> wrote:

From: John F Hall <[hidden email]>
Subject: Re: When to treat variables as continuous variables?
To: [hidden email]
Date: Thursday, 29 July, 2010, 17:38

Hi Faiz

 

I'm John Hall, and as a non-statistician used to teach basic statistics as part of a data analysis course (with SPSS) to sociology students in a way they could understand, using data that was relevant i.e. data from questionnaire surveys and administrative records, not standard deviations of the length of planks on a lorry, as the lecturer supplied by the Maths Dept did.  I even had a blind student for whom I prepared a braille examination paper, which she deferred for a year and then didn't take.  There was an official course in statistics, but my students said they learned more from mine.

 

I taught my last course in 1992 when I took early retirement.  So now what?  I'm in the middle of converting and uploading materials from my courses to my new website  

http://surveyresearch.weebly.com/.  For now I suggest you go through the accompanying Statistical Notes for the data analysis course and theh try working your way through the tutorials for SPSS/PASW 18  They're arranged in four blocks and proceed gently through every step from completing a short questionnaire, coding the responses and transferring the data to a special sheet, typing the data into the computer and then analysing the data from all the questionnaires completed by more than 600 students on six cycles of the course.  The tutorials then move on to some data from major surveys and progress from frequency counts on categorical variables, then scale variables and through crosstabulation to basic descriptive statistics in non-mathematical language.  The emphasis is always on the research question and the logic and process needed  to answer the quaestion from the available data.: SPSS just happens to be the package used.  They are all syntax-based in preference to drop-down menus, but many exercises are repeated using the menus, although I'm not sure how you'll manage with the screenshots!

 

There's a list of textbooks on my site, but I can recommend others if necessary.

 

Let me know how you get on with all this and I'll see if I can be of further assistance.

 

John Hall

 

 

----- Original Message -----

From: Faiz Rasool

To: SPSSX-L@...

Sent: Thursday, July 29, 2010 5:33 PM

Subject: When to treat variables as continuous variables?

Hi all,

 

My apologies for a slightly non-SPSS question. I'm a Blind student of Sociology. Unfortunately I'm learning statistics from a teacher who is not an expert at statistics. I've been trying to improve my understanding of what type of analysis to use in different situations. One of the main ideas that I've find it difficult to draw a conclusion about is when certain variables can be treated as a continuous variables. So techniques like regression and tests like T tests can be used. As a Sociology student I am mostly dealing with categorical variables measured using likert scale. I've  read that chi-square test is appropriate for finding out the relationship between two categorical variables. yet I observe in journal articles and on this list as well that regression is   used in such situations. The textbooks I've read seems to assume that the student is been informed about this kind of thing in the class. But unfortunate students like me are left to learn for themselves. So my question is that what rule of thumb should I try to follow and keep in my mind as I try to improve my understanding of statistical analysis as a Sociology student.

 

Thanks all, and sorry for a non-SPSS question.

Faiz.

 

 

[David Greenberg]

I'd like to Martin Holt's message as an occasion for making a small plea on behalf of consideration for list subscribers. These messages are sent out to a folks with a wide range of backgrounds. Abbreviations that will be instantaneously recognizable by some will probably be unrecognizable by many others. I, for example, have no idea what the abbreviation CMAJ stands for. It would be considerate for posters to spell out journal titles or other references. Thank you, David Greenberg, Sociology Department, New York University

----- Original Message -----
From: Martin Holt <[hidden email]>
Date: Thursday, July 29, 2010 5:44 pm
Subject: Re: When to treat  variables as continuous variables?
To: [hidden email]

> Hi, conventional wisdom, to the point of being dogma, advises against
> categorization of a continuous variable. I'm sure that that is
> true..Professor Martin Bland says so. So I was most interested in the
> following paper when I came across it:
>
> "The exaggerated relations between diet, body weight and mortality:
> the case for a categorical data approach"
> Welch H. Gilbert, Schwartz LM, Woloshin S.
> CMAJ, March 29, 2005; 172 (7)
>
> This provides an argument for sometimes taking the categorical approah.
>
> I distributed this reference to members of MedStats recently (I'm
> MedStats Group Owner) but it elicited no response, either for or against.
>
> I think it will only be of use to you if/when people more expert than
> I evaluate it. But, at the least, it's worth a read. I hope this is
> helpful, rather than causing more indecision. The safest approach,
> without feedback on this article, is to stick to using continuous
> rather than categorical.
>
> Best Wishes,
>
> Martin Holt
>
>
> Martin Holt
> Medical Statistician
>
> --- On Thu, 29/7/10, John F Hall <[hidden email]> wrote:
>
>
> From: John F Hall <[hidden email]>
> Subject: Re: When to treat variables as continuous variables?
> To: [hidden email]
> Date: Thursday, 29 July, 2010, 17:38
>
>
>
>
>
> Hi Faiz
>
> I'm John Hall, and as a non-statistician used to teach basic
> statistics as part of a data analysis course (with SPSS) to sociology
> students in a way they could understand, using data that was relevant
> i.e. data from questionnaire surveys and administrative records, not
> standard deviations of the length of planks on a lorry, as the
> lecturer supplied by the Maths Dept did.  I even had a blind student
> for whom I prepared a braille examination paper, which she deferred
> for a year and then didn't take.  There was an official course in
> statistics, but my students said they learned more from mine.
>
> I taught my last course in 1992 when I took early retirement.  So now
> what?  I'm in the middle of converting and uploading materials from my
> courses to my new website
> http://surveyresearch.weebly.com/.  For now I suggest you go through
> the accompanying Statistical Notes for the data analysis course and
> theh try working your way through the tutorials for SPSS/PASW 18
> They're arranged in four blocks and proceed gently through every step
> from completing a short questionnaire, coding the responses and
> transferring the data to a special sheet, typing the data into the
> computer and then analysing the data from all the questionnaires
> completed by more than 600 students on six cycles of the course.  The
> tutorials then move on to some data from major surveys and progress
> from frequency counts on categorical variables, then scale variables
> and through crosstabulation to basic descriptive statistics in
> non-mathematical language.  The emphasis is always on the research
> question and the logic and process needed  to answer the quaestion
> from the available data.: SPSS just happens to be the package used.
> They are all
>  syntax-based in preference to drop-down menus, but many exercises are
> repeated using the menus, although I'm not sure how you'll manage with
> the screenshots!
>
> There's a list of textbooks on my site, but I can recommend others if
> necessary.
>
>
> Let me know how you get on with all this and I'll see if I can be of
> further assistance.
>
> John Hall
>
>
>
> ----- Original Message -----
> From: Faiz Rasool
> To: [hidden email]
> Sent: Thursday, July 29, 2010 5:33 PM
> Subject: When to treat variables as continuous variables?
>
>
> Hi all,
>
> My apologies for a slightly non-SPSS question. I'm a Blind student of
> Sociology. Unfortunately I'm learning statistics from a teacher who is
> not an expert at statistics. I've been trying to improve my
> understanding of what type of analysis to use in different situations.
> One of the main ideas that I've find it difficult to draw a conclusion
> about is when certain variables can be treated as a continuous
> variables. So techniques like regression and tests like T tests can be
> used. As a Sociology student I am mostly dealing with categorical
> variables measured using likert scale. I've  read that chi-square test
> is appropriate for finding out the relationship between two
> categorical variables. yet I observe in journal articles and on this
> list as well that regression is   used in such situations. The
> textbooks I've read seems to assume that the student is been informed
> about this kind of thing in the class. But unfortunate students like
> me are left to
>  learn for themselves. So my question is that what rule of thumb
> should I try to follow and keep in my mind as I try to improve my
> understanding of statistical analysis as a Sociology student.
>
> Thanks all, and sorry for a non-SPSS question.
> Faiz.
>
>

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Martin Holt-2]

I'm sorry, David, you're right. I googled CMAJ and got "Canadian Medical Association Journal". I've seen it often enough to think that it's a high-impact Journal.

 

Best Wishes,

 

Martin Holt

---

From: David Greenberg <[hidden email]>
To: [hidden email]
Cc: [hidden email]
Sent: Fri, 30 July, 2010 0:41:28
Subject: Re: When to treat variables as continuous variables?

I'd like to Martin Holt's message as an occasion for making a small plea on behalf of consideration for list subscribers. These messages are sent out to a folks with a wide range of backgrounds. Abbreviations that will be instantaneously recognizable by some will probably be unrecognizable by many others. I, for example, have no idea what the abbreviation CMAJ stands for. It would be considerate for posters to spell out journal titles or other references. Thank you, David Greenberg, Sociology Department, New York University

----- Original Message -----
From: Martin Holt <[hidden email]>
Date: Thursday, July 29, 2010 5:44 pm
Subject: Re: When to treat  variables as continuous variables?
To: [hidden email]

> Hi, conventional wisdom, to the point of being dogma, advises against
> categorization of a continuous variable. I'm sure that that is
> true..Professor Martin Bland says so. So I was most interested in the
> following paper when I came across it:
>  
> "The exaggerated relations between diet, body weight and mortality:
> the case for a categorical data approach"
> Welch H. Gilbert, Schwartz LM, Woloshin S.
> CMAJ, March 29, 2005; 172 (7)
>  
> This provides an argument for sometimes taking the categorical approah.
>  
> I distributed this reference to members of MedStats recently (I'm
> MedStats Group Owner) but it elicited no response, either for or against.
>  
> I think it will only be of use to you if/when people more expert than
> I evaluate it. But, at the least, it's worth a read. I hope this is
> helpful, rather than causing more indecision. The safest approach,
> without feedback on this article, is to stick to using continuous
> rather than categorical.
>  
> Best Wishes,
>  
> Martin Holt
>  
>
> Martin Holt
> Medical Statistician
>
> --- On Thu, 29/7/10, John F Hall <[hidden email]> wrote:
>
>
> From: John F Hall <[hidden email]>
> Subject: Re: When to treat variables as continuous variables?
> To: [hidden email]
> Date: Thursday, 29 July, 2010, 17:38
>
>
>
>
>
> Hi Faiz
>  
> I'm John Hall, and as a non-statistician used to teach basic
> statistics as part of a data analysis course (with SPSS) to sociology
> students in a way they could understand, using data that was relevant
> i.e. data from questionnaire surveys and administrative records, not
> standard deviations of the length of planks on a lorry, as the
> lecturer supplied by the Maths Dept did.  I even had a blind student
> for whom I prepared a braille examination paper, which she deferred
> for a year and then didn't take.  There was an official course in
> statistics, but my students said they learned more from mine.
>  
> I taught my last course in 1992 when I took early retirement.  So now
> what?  I'm in the middle of converting and uploading materials from my
> courses to my new website  
> http://surveyresearch.weebly.com/.  For now I suggest you go through
> the accompanying Statistical Notes for the data analysis course and
> theh try working your way through the tutorials for SPSS/PASW 18 
> They're arranged in four blocks and proceed gently through every step
> from completing a short questionnaire, coding the responses and
> transferring the data to a special sheet, typing the data into the
> computer and then analysing the data from all the questionnaires
> completed by more than 600 students on six cycles of the course.  The
> tutorials then move on to some data from major surveys and progress
> from frequency counts on categorical variables, then scale variables
> and through crosstabulation to basic descriptive statistics in
> non-mathematical language.  The emphasis is always on the research
> question and the logic and process needed  to answer the quaestion
> from the available data.: SPSS just happens to be the package used. 
> They are all
>  syntax-based in preference to drop-down menus, but many exercises are
> repeated using the menus, although I'm not sure how you'll manage with
> the screenshots!
>  
> There's a list of textbooks on my site, but I can recommend others if
> necessary.
>  
>
> Let me know how you get on with all this and I'll see if I can be of
> further assistance.
>  
> John Hall
>  
>  
>
> ----- Original Message -----
> From: Faiz Rasool
> To: [hidden email]
> Sent: Thursday, July 29, 2010 5:33 PM
> Subject: When to treat variables as continuous variables?
>
>
> Hi all,
>  
> My apologies for a slightly non-SPSS question. I'm a Blind student of
> Sociology. Unfortunately I'm learning statistics from a teacher who is
> not an expert at statistics. I've been trying to improve my
> understanding of what type of analysis to use in different situations.
> One of the main ideas that I've find it difficult to draw a conclusion
> about is when certain variables can be treated as a continuous
> variables. So techniques like regression and tests like T tests can be
> used. As a Sociology student I am mostly dealing with categorical
> variables measured using likert scale. I've  read that chi-square test
> is appropriate for finding out the relationship between two
> categorical variables. yet I observe in journal articles and on this
> list as well that regression is   used in such situations. The
> textbooks I've read seems to assume that the student is been informed
> about this kind of thing in the class. But unfortunate students like
> me are left to
>  learn for themselves. So my question is that what rule of thumb
> should I try to follow and keep in my mind as I try to improve my
> understanding of statistical analysis as a Sociology student.
>  
> Thanks all, and sorry for a non-SPSS question.
> Faiz.
>  
>  

<SCRIPT language=javascript><!-- bmi_SafeAddOnload(bmi_load,"bmi_orig_img",0);//--> </SCRIPT>

[Bruce Weaver]

Faiz Rasool wrote

Hi all,

My apologies for a slightly non-SPSS question. I'm a Blind student of Sociology. Unfortunately I'm learning statistics from a teacher who is not an expert at statistics. I've been trying to improve my understanding of what type of analysis to use in different situations. One of the main ideas that I've find it difficult to draw a conclusion about is when certain variables can be treated as a continuous variables. So techniques like regression and tests like T tests can be used. As a Sociology student I am mostly dealing with categorical variables measured using likert scale. I've  read that chi-square test is appropriate for finding out the relationship between two categorical variables. yet I observe in journal articles and on this list as well that regression is   used in such situations. The textbooks I've read seems to assume that the student is been informed about this kind of thing in the class. But unfortunate students like me are left to learn for themselves. So my question is that what rule of thumb should I try to follow and keep in my mind as I try to improve my understanding of statistical analysis as a Sociology student.

Thanks all, and sorry for a non-SPSS question.
Faiz.

Hi Faiz.  Here's a rule of thumb that I like:  If it is fair and honest to use means and SDs descriptively, then you won't go too far wrong using parametric methods to analyze the data.

With that in mind, let's distinguish between two situations that arise:

1. Individual items of the sort you describe, and
2. means (or sums) of several such items.

Here, I think it is clear that means and SDs are more appropriate for the second situation than the first.

HTH.

[Ryan]

Faiz,

I don't necessarily see a problem with fitting a GLM even if the
dependent variable has only 3 ordered categories, as long as you are
confident that the responses are ordered. Making this determintation
is not always easy (e.g. when the middle category is "neutral" or "I
don't know"). As Bruce suggested, you need to ask yourself if
reporting means and standard devations makes sense. Of course, one
should carefully examine the distribution--this recommendation goes
for practically any statistical test. If you have an extremely skewed
distrubtion, then you might be better off staying away from GLMs.
Another problem I've encountered with Likert items or Likert-type
items is that people tend to endorse the middle value, especially when
there are very few response options. This can make it challenging to
find a difference in means between groups.

Ryan

On Fri, Jul 30, 2010 at 7:33 AM, Bruce Weaver <[hidden email]> wrote:

> Faiz Rasool wrote:
>>
>> Hi all,
>>
>> My apologies for a slightly non-SPSS question. I'm a Blind student of
>> Sociology. Unfortunately I'm learning statistics from a teacher who is not
>> an expert at statistics. I've been trying to improve my understanding of
>> what type of analysis to use in different situations. One of the main
>> ideas that I've find it difficult to draw a conclusion about is when
>> certain variables can be treated as a continuous variables. So techniques
>> like regression and tests like T tests can be used. As a Sociology student
>> I am mostly dealing with categorical variables measured using likert
>> scale. I've  read that chi-square test is appropriate for finding out the
>> relationship between two categorical variables. yet I observe in journal
>> articles and on this list as well that regression is   used in such
>> situations. The textbooks I've read seems to assume that the student is
>> been informed about this kind of thing in the class. But unfortunate
>> students like me are left to learn for themselves. So my question is that
>> what rule of thumb should I try to follow and keep in my mind as I try to
>> improve my understanding of statistical analysis as a Sociology student.
>>
>> Thanks all, and sorry for a non-SPSS question.
>> Faiz.
>>
>>
>
> Hi Faiz.  Here's a rule of thumb that I like:  If it is fair and honest to
> use means and SDs descriptively, then you won't go too far wrong using
> parametric methods to analyze the data.
>
> With that in mind, let's distinguish between two situations that arise:
>
> 1. Individual items of the sort you describe, and
> 2. means (or sums) of several such items.
>
> Here, I think it is clear that means and SDs are more appropriate for the
> second situation than the first.
>
> HTH.
>
>
> -----
> --
> Bruce Weaver
> [hidden email]
> http://sites.google.com/a/lakeheadu.ca/bweaver/
>
> "When all else fails, RTFM."
>
> NOTE: My Hotmail account is not monitored regularly.
> To send me an e-mail, please use the address shown above.
>
> --
> View this message in context: http://spssx-discussion.1045642.n5.nabble.com/When-to-treat-variables-as-continuous-variables-tp2258468p2259433.html
> Sent from the SPSSX Discussion mailing list archive at Nabble.com.
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> [hidden email] (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD
>

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD