eakl
commented
7 years ago Original post by ya6n
Remember that advertising impact spans across multiple weeks. I would recommend computing scores using more than 1 week as your time window, it will increase the number of observations and limit low sample size noise.
Original post by andrei_const
My question is related to the Mars case, and more precisely to the AVI score. Our group managed to calculate an AVI score for each copy.
But the results we get are sometimes quite weird for a few campaigns, with very low and very high scores. Especially for non tv advertising campaigns. And we are pretty confident that our calculations are correct.
So we are looking for a rigorous method to exclude non relevant campaigns.
An explanation of why some results would be absurd is that we do not have enough households on which to compute a relevant AVI score. If, for instance a copy only reaches 40 households per week, that means that we compute its AVI score for that week only on 40 households. And if, among those 40 households, none has purchased the relevant brand during the week, which happens pretty often in this non chocolate loving chocolate population, the AVI score for that particular week will be null, which will drive the mean AVI score for all the campaign down.
Because we compute the AVI score on a weekly basis, we thought that we needed to analyse more precisely the exposure by copy by week to determine if for a given campaign, the reach is sufficient for us to compute an AVI score.
We indeed managed to compute the average reach per week for each copy.
Here are our questions. Is our method correct? And if so, what would be a good threshold to determine the relevant campaigns from the non-relevant ones which do not have enough reach? We thought of the value of 250, approximately 5% of the population. What do you think about it? Is there another method to point out non-relevant campaigns?
Thank you for your time