I have a dataset with at Customer-Date level. I want to fit a line on the data estimating spend of a customer on a certain date. One of the covariates I am using in the model is historical sales of the customers. For example, For a customer at time (t + x) the dependent variable is the sales at time (t + x), however, one independent variable is sales total sales until time (t + x), which includes sales at time (t). Now at time (t) in the data, the sale is used as a dependent variable. Hence, there is a circularity in the data, the dependent variable of a record may constitute the independent variable of another record.

I am struggling to understand the nuances of this model, both intuitively and technically. Is this even a valid way to fit a model? If not, why?

  • 1
    $\begingroup$ this is essentially OK - you should look up timeseries models which will give you better ideas ( eg what is "stationary" year on year sales etc) $\endgroup$ – seanv507 Jul 4 '16 at 13:07

Seems questionable to me. Why don't you just recalculate it so that you have total sales till time [t-1]?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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