Multiple Linear Regression with Nominal, Ordinal and Interval Data

Hi,
I am hoping you can help me. We did a study using a 19 page
questionnaire containing a demographic questionnaire (which contains
categorical questions such as race, mar status etc) and likert scaled
ordinal items. The other questionnaires are various validated
assessment tools such as Barkley and Murphey's ADHD Screening Tool and
Rosenberg's Self Esteem Scale, which after they are scored translate to
interval data. The dependent variable for this study is also interval
and is the student's final grade in the course ranging from 48-100.
The goal is to develop one prediction model which evaluates all the
IV's at the same time. I have dummy coded the categorical data, the
dichotomous variables such as gender are coded with 0's and 1's but the
question remains - what do I do with the likert scale data? I am not
comfortable treating it as interval data as some researchers are, so
that is not an option. Some articles and text books I have read
suggested dichotomizing the scale (1-3 (disagree or negative response)
= 0 and (4-5 agree or positive response) = 1. Others believe you
"lose" data that way and opt for a 3 pt recoding such as 1,2= -1, 0=
neutral, and 4,5=+1. If you agree that the scale can be combined in
one of the manners mentioned above, what is the explanation to someone
who believes you are losing data when you collapse the scales?

What other statistical tests are available to me with this likert data?
Tests that may help describe the population's responses to the likert
scales in a technical report?

One last question:
We did the regression with just the interval IV's bc we had a deadline
and needed to turn something in, so we had not included the demographic
questionnaire yet.... One of the variables came up predictive in a
stepwise linear regression model however it was not correlated. Is
this possible? What is the explanation? Can a variable be "causal"
but not have a significant relationship?

Thank you very much for your help and feedback!

.

Relevant Pages

  • Re: Multiple Linear Regression with Nominal, Ordinal and Interval Data
    ... By "Likert data" to you mean previously developed summative scales built upon a set of items with a Likert response scale? ... If you have single items with the kind of response scale that a Likert scale would use, you would need to make a judgment about whether you think the underlying construct can be well represented. ... questionnaire containing a demographic questionnaire (which contains ...
    (sci.stat.consult)
  • Re: Multiple Linear Regression with Nominal, Ordinal and Interval Data
    ... Thank you for your response. ... How do I construct scale scores and test ... construct scale scores, test their internal consistency in you sample, ... questionnaire containing a demographic questionnaire (which contains ...
    (sci.stat.consult)
  • Re: Stats Help
    ... Firstly ALL participants complete a 19 item questionnaire (responses ... for scale are parametric), participants are then allocated into the ... effect either as a result of the passage of text OR from the initial ...
    (sci.stat.edu)
  • Re: Multiple Linear Regression with Nominal, Ordinal and Interval Data
    ... questionnaire containing a demographic questionnaire (which contains ... categorical questions such as race, mar status etc) and likert scaled ... Rosenberg's Self Esteem Scale, which after they are scored translate to ...
    (sci.stat.consult)
  • Re: Multiple Linear Regression with Nominal, Ordinal and Interval Data
    ... questionnaire containing a demographic questionnaire (which contains ... Rosenberg's Self Esteem Scale, which after they are scored translate to ... interval data. ... The scale values that are assigned to the categories on a Likert scale ...
    (sci.stat.consult)