The crux of my question is as follows: Would a higher-order Markov model produce a different result than a first-order Markov model when used for Channel Attribution modelling?

Once the transition matrix is constructed/estimated using the given data, Removal Effects are calculated to understand the importance of each channel in the data. Removal Effects are the percentage decrease that would occur if a particular channel is removed. Basically what this means is all the incoming and outgoing edges for a given channels would be removed.

Now in case of a k-order Markov model, even though the transition matrix would be much larger than its first-order counterpart, the removal effects would be calculated by eliminating a particular channel, say, A. This, however, means that every sequence of channels of length k that contains A would be removed.

I think due to this the removal effects of a first-order and higher-order Markov Chains would be almost similar.

And since Removal Effects are the ultimate result of Markov Chain attribution, is it worth it to implement a higher-order Markov model for attribution modelling?

P.S. - My question is simply based on my intuition and has no numerical data to back it up. Apologies if its too wordy and thanks in advance!



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.