I want to see how Customer satisfaction is related to the following variables (survey metrics): Average response time, Drop off Rate, Click through rate, Number of Active users etc, Number of surveys filled in person etc. , so all numeric fields basically. I have come across different methods; linear regression, Logistic regression etc but not sure which to use. These predictor variables are correlated, for example the higher the Click through rate, the lower the average response time, and the lower the Drop off rate.
Also could I please get an approximation of how many training observations would generally be needed for such a model, I know it depends on the data but whats considered the bare minimum? Please help me pick the right approach, thanks!


closed as too broad by kjetil b halvorsen, mdewey, Michael Chernick, Jeremy Miles, StatsStudent Jan 12 at 0:23

Please edit the question to limit it to a specific problem with enough detail to identify an adequate answer. Avoid asking multiple distinct questions at once. See the How to Ask page for help clarifying this question. If this question can be reworded to fit the rules in the help center, please edit the question.

  • $\begingroup$ Show us what you have done so far. $\endgroup$ – user2974951 Jan 9 at 12:11
  • $\begingroup$ I do not actually have data for these metrics right now, (will do so in the near future when enough customers start using the platform), I wanted to get an idea of which method to use $\endgroup$ – Ray92 Jan 9 at 12:21
  • 1
    $\begingroup$ Without data we are just handwaiving here, anyway what model you use will depend on what the objective is, if satisfaction is a numeric variable then you would use linear regression. $\endgroup$ – user2974951 Jan 9 at 12:48
  • $\begingroup$ @user2974951 would you really want to use linear regression if the assumptions of linear regression weren't met even after remediation efforts? $\endgroup$ – StatsStudent Jan 12 at 0:22

The choice between linear and logistic regression depends on how the dependent variable is measured. Is it a Likert type scale? Then ordinal logistic is a good starting point. If it is continuous (or nearly so - say a 0 to 100 rating) then linear regression is the starting place.

The correlated independent variables may indicate coliearity. Without data there is no way to tell.

The number of observations needed depends on what you are trying to find out, how big your effect size is, what statistical method you wind up using, how many independent variables you have and so on. There are packages to try to figure this out for you - look up power analysis.

  • $\begingroup$ Hi thanks, if I am using a 0 to 100 rating, why should I stick to linear regression? for example wouldn't a Random forest regression also be suitable. And if I have 5 variables (with correlation between them, so they wouldn't be called independent?) approximately how many observations would be needed? I read somewhere that 10 observations per variable is sufficient. Is this correct, or do I need a minimum of a couple hundred observations for a reasonably accurate model? Thanks! $\endgroup$ – Ray92 Jan 10 at 14:28
  • $\begingroup$ As I said, the choice between linear and logistic (which you raised in the question) depends on your DV. The fact that there is relation among the independent variables doesn't change the term - they are predictors of your DV. You last point confuses overfitting (too many variables) with power analysis (can I find what I am looking for). $\endgroup$ – Peter Flom Jan 10 at 15:52

Not the answer you're looking for? Browse other questions tagged or ask your own question.