Binary Logistic Regression - Evaluation of Survey Responses

[abhipatel]

Am trying to evaluate the survey responses from a group (Yes or no answers to questions 1-4) and rated scale answers (scale of 1-5 (Poor- Excellent) for questions 5-7).

So am running a binary logistic regression for first 4 questions (individually in minitab for each question) and ordinal regression for questions 5-7.

hmmm...issue is I do not know how to interpret much from the results.

Can anyone assist or guide me in interpreting the results and what they reflect?

File is attached (results in seperate tabs).

Thanks

 

[reynald]

Re: Binary Logistic Regression

Hi abhipatel,
I think i don't understand yet what you are trying to accomplish.
Logistic Regression's goal is to create a model where in you can quantify the probability of having "yes" answers given a set of predictors. That is you model the Probability Pr(Answer) = Function of (Predictor Variables).
In your excel file which factors are you considering as the independent variable(s).
Is it the questionaire number such that you are trying to model
Pr(Answer) = Function of (Questionaire #)?
Please let us know,
Regards,
Reynald

 

[abhipatel]

Re: Binary Logistic Regression

Hi,

Thanks! appreciate the response..maybe i should have done some additional research....anyways we have a survey questonnaire posted out for operators to fill in regarding one of our auditing process system.

Based on their answers we want to determine how effective it is and what counter measures we can take to improve based on their feedback.

A sample question was :How effective do you think your current auditing system is? (Rank : 1-5 (Poor - Excellent)). We want to gauge the perception of the system's user.

Is there a better way to statistically determine and analyze the results? I was under the impression that logisitic regression could do it but based on your answer I think I grossly misunderstood.

Kindly assist with any alternate suggestions if you have one.

 

[reynald]

Hi,
Maybe a simple confidence interval would do the trick.
It can be interpreted as "I am 95% confident that the percentage of people who would answer yes to Question#1 is between 46.27% and 76.81%.".
The special consideration here is if the interval includes 50%, then you can say that the response if about 50%-50%. If the inteval is higher that 50%, then you can say there more than 50% is infavor of "yes", and os on.
I have re-attached the excel file with a formula for the confidence interval. I only used nornal approxiation though. If you want an 'exact' (more precise, more reliable) you may have to run it in minitab or some other stat software.

Hope that helps,
Reynald

 

[abhipatel]

Thanks Rey,

What would I do for the ordinal scale data in the second tab? How do I interpret those? They are ranked data from 1-5.

Is there any way I could link it to y = f(x) for regression with y being my audit effectiveness and the function of x or predictor variables being the set of questions?

I mean I want to know based on responses how effective mfg thinks their audits are.....what would be the model if the survey is a combination of binary (yes/no answers) and ordinal questions?

Thanks and really appreciate your responses for the confidence interval theory. Hope you could teach me more stuff

 

[reynald]

Thanks Rey,

What would I do for the ordinal scale data in the second tab? How do I interpret those? They are ranked data from 1-5.

oopps, i did not notice the second tab.
For that case you can extend the confidence interval into multiple comparisons of proportions. The method is known as the Analysis of Means
Unfortunately it is not easy to do it in excel, so you need to use a stat software for this.
Here is sample output of your data (as well as in attached)
You may look at it this way, "categories 2,4, and 5 have equal proportions of about 0.05 - 0.34. Category 1 has lower proportion than that while category 3 has a higher proportion"
View attachment ANOMeans.bmp

Is there any way I could link it to y = f(x) for regression with y being my audit effectiveness and the function of x or predictor variables being the set of questions?

I mean I want to know based on responses how effective mfg thinks their audits are.....what would be the model if the survey is a combination of binary (yes/no answers) and ordinal questions?

I think thier objectives are not aligned. What you did was to gather data to assess the effectiveness of your audits. If you will use regression then the goal is to gather data to Predict the future effectiveness of audit based on the answers on Q1-Q7 (which will not make sense since audit in itself is concerned with past&present activities).

Thanks and really appreciate your responses for the confidence interval theory. Hope you could teach me more stuff

I keep on learning as well from the great folks here. They are the really cool guys.