Discussion: View Thread

  • 1.  Regression Modelling

    Posted 09-21-2011 12:40
    Dear All,
    If we have some predictor variables with many levels for instance, sometimes individuals are asked to rate their pain on a scale of 0 -100 . What is the best way to enter such kind of predictors in a regression model?
    Thank you.
    Best Regards,
    Tasneem


  • 2.  RE:Regression Modelling

    Posted 09-21-2011 12:55

    Some more detail needed.

    Your pain variable, is it   a baseline variable? What's your dependent (y) variable?
      Is the pain scale  a validated measure?

    You mentioned "On a scale of 0 to 100". How is it recorded, on a so called "thermometer" scale or some other way, for example, as an integer value?


     
    -------------------------------------------
    Chris Barker, Ph.D.
    President - San Francisco Bay Area Chapter of the American Statistical Association
    www,barkerstats.com
    -------------------------------------------








  • 3.  RE:Regression Modelling

    Posted 09-21-2011 14:12
    Hello,

    If at all possible, treat the pain scale as continuous. Treating this as discrete levels will use many degrees of freedom in building your model, particularly is you are considering including interaction terms.


    -------------------------------------------
    Patrick Spagon
    -------------------------------------------








  • 4.  RE:Regression Modelling

    Posted 09-21-2011 14:22
    Break it into subgroups so you can find out whether you have a linear, dose-response, threshold, etc.
    relationship; pain scales can be quite non-linear.

    Ray

    -------------------------------------------
    Raymond Hoffmann
    Professor
    Medical College of Wisconsin
    -------------------------------------------






    Original Message


  • 5.  RE:Regression Modelling

    Posted 09-26-2011 15:23
    Dear All, 

    Thank you for responding to my post. It's much appreciated. 

    I am sorry for the late response to all your posts regarding my query. Actually, I started working on this project recently and wanted to get some more details from my  supervisor before I got back to you. 

    Dependent variable is binary variable (1 if visited ER due to the chronic disease under study else 0). The pain variable is actually a score calculated based on patients response indicating his control over the chronic disease after using preventive measures. They are measured on thermometer scale and are baseline measures. Some of the other predictor  variables are  individual's income, age group etc.  

    The patients are     clustered under different physicians. For every patient they obtain two measurements on the independent and explanatory variables, one at the beginning and the other at the end of study that is after three months period.  I was wondering      whether the response { ER hospitalization indicator (ER_Hosp)} at baseline (at the beginning of the study)  be considered as a response or as a covariate.

    In general, in epidemiological cohorts, the conditions in which responses are collected at baseline and during the follow-up are often different. The first response is often considered as a covariate but this choice has to be confirmed in view of the analysis of dependency between the responses. However, here we only have two time points one start and one three months after the study.

    I would like to thank all of  you once again for your time and valuable comments and suggestions.

    Best Regards,

    -------------------------------------------
    [Tasneem] [Zaihra]
    [Assistant Professor]
    [Concordia University]
    [Montreal]
    [QC]
    [Canada]
    -------------------------------------------






    Original Message