Regression involving likert scale response

#1
Hi,

I am trying to carry out regression analysis. I would like my dependent variable to be a Yes/No type, and my independent variables to be likert scales type (1-5).

I want to see for examples if those respondents that selected Yes tended to rate cost more highly.
 
#3
From your last sentence it sounds like you want the yes/no to be the independent variable?
Yes. I would like my independent variable to be a Yes/No answer. I added the data to the data view and changed the values so that Yes=1 and No=0. Do you know what analysis method I can use?
 

noetsi

Fortran must die
#4
If your dependent variable is cost (which is normally interval) than you can use linear regression . If your dependent variable is coded Yes/NO than you should use logistic regression.
 
#5
My dependent variable is: Did you meet the deadline? Yes=1 and No=0
My independent variable is: Was lack of cost an issue? Rate on a scale of 1 to 5 with 1 being strongly disagree to 5 being strongly agree.

Should i go for binary logistic then?
 

hlsmith

Less is more. Stay pure. Stay poor.
#6
That would be a good option then.

What is your sample size and how many people do you have in each of your dependent variable groups as well as the independent variable categories.
 
#8
Hi, can anyone help me carry out binary logistic analysis and interpret the results. I have my yes/no response question as my dependent and 'Was lack of cost an issue?Rate on a scale of 1 to 5 with 1 being strongly disagree to 5 being strongly agree.' as my coveriate. I am not sure where to go from here or what the output data means.
 
#9
There is a statistical paradigm which is absolutely perfect for your application, called Optimal Data Analysis (ODA). This paradigm weds the fields of operations research (in which best solutions are called "optimal") and exact non-parametric statistics ("data analysis"). In the ODA paradigm, the dependent variable is called the class variable. The independent variable is called an attribute. ODA finds a threshold value on the attribute which is used to make a classification model. For example, the ODA model might be: if rating on attribute is 3 or less then predict class=0; otherwise predict class=1. Some or all of these predictions are correct: observations predicted to be from class ) will be from class 0, and some predicted 1's will be actual 1's. Sometime the predicted class membership and the actual class membership is incorrect. ODA finds the threshold value which results in the greatest number of correct predictions. When the "confusion table" showing actual class as rows, and predicted class as columns, is constructed, a Fisher's randomization procedure is used to determine the Type I error rate. Classification performance is summarized using traditional indices such as sensitivity, specificity, and positive and negative predictive value, as well as a normed index called ESS, on which 0 is the classification accuracy expected by chance, and 100 is perfect, errorless classification. Thus, models can be directly compared in terms of ESS, regardless of the underlying geometry (e.g., N, number of attributes, etc.). And, bootstrap, jackknife and hold-out validity analyses are available.

Here is a link to a free article which introduces early development of the paradigm, including exactly your problem (binary class variable, ordered attribute): http://optimalprediction.com/files/pdf/V1A2.pdf

And, the seminal introduction to the ODA paradigm, which comes with software for Windows (the only software which can accomplish this analysis), is widely available in college libraries, and is available new and used at an economical cost. Here is a link to the book/software: http://www.apa.org/pubs/books/4316000.aspx