3
$\begingroup$

I prefer this model in R

We are capturing sales data by time series (Month by month). Some of items have commissions and some have Discounts and others have both commissions and discounts. Is it Commissions or Discounts or commissions + Discounts have impact on my sales growth? Or is it my sales are growing because of those commissions or discounts or discounts +commissions Can you suggest me best model to solve my use case? I am thinking multiple regression. But I want to double check with experts like you.

Thanks for your all your help

Sample Data set: (5 variables)

Year-Month -Product -Sales  -Commission -Discounts
2013-01 Milk    300 No  Yes
2013-02 Milk    400 No  Yes
2013-03 Milk    200 No  Yes
2013-04 Milk    150 No  Yes
2013-05 Milk    500 No  Yes
2013-01 Bread   800 Yes No
2013-02 Bread   879 Yes No
2013-03 Bread   790 Yes No
2013-04 Bread   459 Yes No
2013-05 Bread   600 Yes No
2013-01 Cheese  400 Yes Yes
2013-02 Cheese  350 Yes Yes
2013-03 Cheese  600 Yes Yes
2013-04 Cheese  590 Yes Yes
2013-05 Cheese  720 Yes Yes
$\endgroup$
3
$\begingroup$

Multiple regression sounds appropriate in this case. Real question is what variables to use. Definitely, you should include indicator variables commisions and discounts (and possibly their combination - commisions_and_discounts - as a separate variable). Date and time information may play important role as well, though treating them is a little bit harder. At the very least, it's worth to include year (because there may be a global trend), month and day_of_week (many patterns are repeated periodically).

Product type is a little bit more complicated. Obviously, some products will always have higher sales than others, so first idea is to include a set of product_type_x dummy variables. But product type may affect sales not additively, but instead multiplicatively, i.e. not as

sales ~ beta_0 + beta_1*product_type_x + ...

but instead as

sales ~ beta_1 * prodcut_type_x * (beta_0 + ...)

In this case better solution would be to create separate regression models for each product type.

$\endgroup$
  • $\begingroup$ @ ffriend Thanks for your feedback.Product name is important for us because some products have discounts and others have no discounts. Our objective is measure the significance of discounts on product growth. Do you think I can use real commissions/Discounts values in dollars or Zero dollars instead of Yes or NO? How R interpret Date? do I need to convert the data into time series first? Is it Multiple regression is good model for this? $\endgroup$ – Murali May 16 '15 at 20:23
  • $\begingroup$ Can you measure significance of discounts per product? As for real discount values vs. "yes/no", try both! It's fairly easy to build several alternative models and estimate how much they fit your data (e.g. using RSS). Regarding date, you probably don't need it per se, but instead convert it into several variables like month and date_of_week (see 1st paragraph of my answer). And yes, multiple regression looks fine here. $\endgroup$ – ffriend May 16 '15 at 23:38
  • $\begingroup$ Also, please, don't cross-post questions on several stackexchange sites. You can always migrate your question if you feel like original site is a bad fit for it, but making cross-posting makes it harder for others to find good answers to the same question, since answers become distributed between several places. $\endgroup$ – ffriend May 16 '15 at 23:42
  • $\begingroup$ @friend Thank you for your help. I am new and I am not aware that all forums are connected. I won't do that next time. $\endgroup$ – Murali May 17 '15 at 0:41
  • $\begingroup$ No problem. Also note, that on your profile page (e.g. this for current site) you can easily reach all your accounts in StackExchange network. $\endgroup$ – ffriend May 17 '15 at 0:56

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.