BookmarkSubscribeRSS Feed
Quartz | Level 8

Hi everyone,


I  want to come up with the return on investment of the marketing campaigns of a company.


Let's say that I have 5 years of weekly data for a handful of region with the following variables

  • # of product sold (the response variable)
  • TV GRP (Gross Rating Points)
  • Radio GRP
  • # of unaddressed mail delivered

The product is sold through the year, but sales show a seasonal pattern.  



Here are the three questions I want to answer:

1- % of  sales from each year were from "base", and % were incremental from TV / Radio / unaddressed mail

2- The ROI of a dollar spent in all three marketing campaign  (knowing the total cost of TV , radio and mail campaigns and the profit from each sales).

3- The shape of the (S-shaped)  media response curve so that we can see at what GRP levels the effect of tv campaign begins to taper off


My best find so far is this paper from an older SESUG :


I have what he calls aggregated data (total sales by week) and he says that regression methods ( PROC REG) or time series regression methods (proc autoreg) are what is mostly used these days.  I can't see myself using proc reg since it doesnt account for the fact that I am using time serie data, so I would probably use proc AUTOREG if I dont get a better suggestion.


My question is how can I answer my questions after running  PROC AUTOREG?  Here are my first ideas..


For question #1 (about the incremental sales from each campaign), I was thinking I could just predict the total number of sales for the past 5 years with all campaigns spending set to 0, then predict again the past 5 years with one campaign spending set to its actual value.  The difference in the number of sales is the incremental number of sales from that campaign.  Does that make sense, or would I be violating too many assumptions?


For question #2, I would use that incremental number of sale, multiply it by the dollar value of each sale and divide by the dollar cost of the campaign.


For question #3, I am at a loss...









Lapis Lazuli | Level 10

you could try adding second and third-order polynomial effects for GRP to the linear term in the model specification and if signifiacnt use calculus on the significant polynomial GRP coefficients to identify points of minimum and maximum curvature and thus characterise the sigmoid response.

Calcite | Level 5

You can easily automate the process that you wish to work on by using one of the available softwares like MassTer from MASS analytics.

You can also find some really insightful content on the educational YouTube channel.



Registration is open! SAS is returning to Vegas for an AI and analytics experience like no other! Whether you're an executive, manager, end user or SAS partner, SAS Innovate is designed for everyone on your team. Register for just $495 by 12/31/2023.

If you are interested in speaking, there is still time to submit a session idea. More details are posted on the website. 

Register now!

What is Bayesian Analysis?

Learn the difference between classical and Bayesian statistical approaches and see a few PROC examples to perform Bayesian analysis in this video.

Find more tutorials on the SAS Users YouTube channel.

Get the $99 certification deal.jpg



Back in the Classroom!

Select SAS Training centers are offering in-person courses. View upcoming courses for:

View all other training opportunities.

Discussion stats
  • 2 replies
  • 3 in conversation