A Data Science Central Community
In market mix models, my sales(y) depends on 4 promotions(p1--p4). Let (b11, b21, b31, b41 ) be the coefficients of p1--p4 in one iteration and
(b12, b22, b32, b42 ) be the coefficients in the second iteration. I check the validity of the model based on some general guidelines:(i)b1--b4 should be positive and
significant (ii)Adstocking uses reasonable lags(Eg: TV has high lag of more than 60%) (ii)sales contributions should not vary much from last year provided corresponding
spends remain same (iii)sales contributed by p1--p4 may vary within +/-20% w.r.t last year.
Both the sets of coefficients follow guidelines (i), (ii) & (iii). And I have many such sets with lag of a promotion varying(within logical range) across different sets of coefficients.
How can I be confident enough that a particular set with corresponding sales contributions matches with the actual sales contributions realized in reality.For convenience sake I have taken 4 promotions. My model has many more promotions and it further worsens the confidence .
I would like to know if there are any additional points to consider to improve the level of confidence in showcasing the final results.
And, if these remain the guidelines for model validity, then what additional insights is market mix model providing us. Wouldn't it be almost as good as using the last year results. Please clarify.