Type of statistical test?

I am trying to measure if customer service or product quality is having an impact on performance. I am not sure what test to use? The questionnaire are all likert scale type questions with 5 level options like: strongly agree, agree, neutral etc. I was thinking of using chi square test since both are categorical variables.


Active Member
What does each row in your data set represent? How many observations (data points) do you have? Please list all the variables / question categories in your data set.
Each row represents each Participant. Total are 120 observations. My research question is to find out if customer service(5 point likert scale) or product quality(5 point likert scale) is having a negative impact on the performance(5 point likert scale) of the business? Independent variables are customer service and product quality and dependent variable is performance.


Active Member
So are you are saying that there are only 3 variables (columns) in your data set? What do you mean by "performance"? What exactly is going on?
Yes correct, only 3 columns. Okay performance is a broad name and refers to many aspects. When I say performance I mean financial performance of the business. I'm trying to see which one 'customer service' or ' product quality' is negatively impacting the profits( financial performance) of the firm.


Active Member
1] Play with modeling frameworks of the form

g(Performance) = B0 + B1 * Customer Service + B2 * Customer Service^2 + B3 * Product Quality + B4 * Product Quality^2 + B5 * Customer Service * Product Quality,

where g() is some non-linear function or g(y) = y. In each such modeling framework (for each function g), develop a model with only statistically significant terms using forward stepwise selection.

2] Compare different modeling frameworks and choose the best one using root-mean-square error and residual diagnostics.
Last edited:


Well-Known Member
PS, the Likert with 5 doesn't distribute normally, but the requirement is for a residuals' normality.
|Any way in your case you use a large sample size (120) so it shouldn't be a problem even if the distribution of the residuals skewed (not symmetrical)