1. What is the order in model configuration?
In the first-order Markov chain the probability of changing the state from one to another depends solely on the current state.
If te current state is C, the probability of change to D or E is the same whether the preceding state was A or B (order 1).
A higher order Markov chain is a chain with "memory". In this case, the probability of changing the state depends not just on the current state, but also on the previous states.
The order (2, 3, 4) denotes how "deep" the memory is. Order 2 means that the probability depends on the current state and the previous state.
For example, if te current state is C, the probability of change to D or E is different, depending on whether the preceding state was A or B.
Order 3 means that the probabilities depend on the current state as well as on the two preceding states.
Order 4 means that the probabilities depend on the current state as well as on the three preceding states.
2. What is one- and multi-channel paths data split?
For one-channel paths, we know that the given channel produced a number of conversions and there is no reason to distribute these conversions among other paths.
For example, a path with just a single Facebook click followed by conversion, generated 5 conversions, the attribution is obvious. Facebook should be attributed with 5 conversions.
However, in the “pure” Markov model, one-channel paths may be underestimated and part of the one-channel conversions is distributed among other paths.
The solution is to remove the one-channel paths from the graph and calculate the attribution of the multi-channel paths first, and then add the conversions attributed to one-channel paths.
If you wish to do so, select “Split data for one- and multi-channel paths” (YES)