1
$\begingroup$

I was performing Linear Regression which is based on E-Commerce Dataset. I was stuck with the following problem.

Assumption: In the dataset, I am taking store_purchase_event_count as a Dependent variable for predicting store_purchase_event_count using OLS Linear Regression.

Problem: I am trying to Normalize the Dependent variable but it contains more than 50-60% of zeroes. So, I was not able to figure out how I should move forward in making the dependent variable to follow the Normal distribution.

Solutions tried: 1. Added constant to each value of Y and then taking the log. 2. Taking the square root of each value.

None of the above solutions is making Y variable normal. Please suggest how to move forward

$\endgroup$

2 Answers 2

1
$\begingroup$

First, OLS regression does not require that the dependent variable be normal, it requires that the errors be normal. Since we can't determine the errors, we test the residuals.

Second, it is likely that the residuals in a model like this will not be normal. No transformation will make them normal and, in any case, transformations should be used for substantive rather than statistical purposes.

Third, you should, instead, use a method that doesn't rely on the assumption that the residuals are normal. Which method you should use depends on what you are trying to find out but some possibilities are quantile regression, robust regression, zero inflated normal models, regression trees (and related methods) and more.

$\endgroup$
0
$\begingroup$

Have you tried zero inflated poisson model? The inflated model should be more appropriate for OLS Linear for your problem.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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