0

I am currently working on a problem in which I have survey response basis a survey conducted by a market research agency. The survey measured perception of coverage about the services of the product. Scale of survey: 0-100. Sample size 4K.

The task at hand is to find correlation between the respondents survey response and their spending with the company, i.e. to say that is the spend of high perception customers high and vice-versa.

My approach:
As the scale was large, firstly I scaled it down to 1-10, i.e 0-10% in 1, 11-20% in 2... and so forth. Then I used uni-variate linear regression on the new scale and spend.
I treated the survey scale as continuous after scaling.

Questions:
1) Is the assumption to treat the scale (after scaling to 1 -10) continuous right or wrong?

2) Is there a need to normalize? When I normalize the data the coefficients cannot be interpreted as dollar values which makes more sense to business people. What would be the impact if I run the analysis without normalizing?

3) Also, will normalization be correct here given one is survey response and other is spend?

Matheus Lacerda
  • 5,983
  • 11
  • 29
  • 45
user3674232
  • 21
  • 2

1 Answers1

0

Questions: 1) Wrong Usually when a continuous variable is recoded it's to make it discrete. Then the linear regression does not fit amymore for your case.

2) The normalization is done to reduce the impact of your outliers in your dataset. By performing your analyse without normalizing the data you are taking your extremes values as an information for your model.

3) It Depends of what you want to do afterwards.... I would say it's always better to keep Every Thing Also Equal when doing data analysis.

Haalanam
  • 5
  • 3
  • I framed the first question wrong. I meant to say that can i treat the scale(1-10) as interval scale and run linear regression.? – user3674232 Jan 30 '18 at 15:58
  • Bad idea ! If you want to run the model in scaled variable as an input you will have to use the logistics regression. You'll have also to regroup it again at the 1-5 scale in order to have a explainable model. – Haalanam Jan 30 '18 at 16:23