My question pertains to the case where a dependent variable is functionally related to an independent variable in a multiple linear regression. Here is a fictional scenario suppose that $Sales_A,Sales_B,Sales_C,Sales_D,Sales_E$ represent the sales resulting from products A,B,C,D,E and Total Sales = $Sales_A +Sales_B +Sales_C + Sales_D + Sales_E$. Consider the following model: $$Sales_D = Sales_I + Total Sales + \text{other control variables}$$

Assuming the assumptions for linear regression are met and there is no multicollinearity between the independent variables. If the goal is to determine how increases in $Sales_I$ effect changes in $Sales_D$ would the above model work or does the fact that the dependent variable is "part" of one of the independent variables cause problems?

  • $\begingroup$ why not put totalsales - sales_d to the model isntead of total sales to avoid this problem? $\endgroup$
    – rep_ho
    Sep 29, 2021 at 23:26
  • $\begingroup$ Wouldn't controlling for (totalsales - sales_d) be different than controlling for total_sales? In the latter the interpretation of the coefficient for Sales_I would be holding total sales constant the effect is \beta_1 which is different than holding (totalsales - sales_d) constant. $\endgroup$ Sep 29, 2021 at 23:35
  • $\begingroup$ also I'm curious if there are any technical reasons as to why it won't work as opposed to a workaround so the question is more theoretical in nature. $\endgroup$ Sep 29, 2021 at 23:37

1 Answer 1


I don't see an issue as formulated it looks like the model would look at how additional sales from product I are cannibalizing or complementing sales from product D assuming observations are stores or something similar. Others more well informed in statistics may be able to pinpoint where the model is incorrect.


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.