Something that I find shocking is that I saw nowhere (and it is not you, but nowhere else), any mention of causality and stating clearly what the model is. Hence, MMM is dangerous and might lead to costly mistakes. Behind the fancy MMM terms, the model is actually very basic in statistics, it is mostly a multiple linear regression or a ridge/lasso regression (e.g. Robyn). Those models do not measure causal effect unless you have no endogeneity issues. To make is super simple, unless you capture absolutely every variable that affect your outcome and your explanatory variable. This is hardly the case as the list of potential variable might be very large and most importantly, it might be impossible to get all those data (e.g. granular data on advertisment spending on different channel for every competitor). Moreover, note that even if you capture everything it might be also a problem to assess causality (e.g. what we call in econometrics bad control or in causal inference colider should not be controlled for). In order to measure causal effects you need an actual experimental setup (e.g. A/B testing if the model is well setup) or quasi-experimental setup. If you want to know more about causality, you can read the free ebook by matheus facure "Causal inference for the brave and true", or "the Mixtape" by Scott Cuningham or watch my TEDx or read my articles on Towards Data Science called "The Science and Art of causality"
Hi Cassandra, we will see what we can create :) do you have any specific questions on MMM that you'd like answered? We also have some extra blog resources on it if you're curious funnel.io/blog/what-is-marketing-mix-modeling-and-how-does-it-work
Loved football analogy.. excellent content, eye catchy graphics and well presented...subbed!
Finally! I keep hearing about MMM all around, but never quite understood it. That is, until now.
Thanks!
Love it! Always hearing about it, nice to get a breakdown.
Something that I find shocking is that I saw nowhere (and it is not you, but nowhere else), any mention of causality and stating clearly what the model is. Hence, MMM is dangerous and might lead to costly mistakes. Behind the fancy MMM terms, the model is actually very basic in statistics, it is mostly a multiple linear regression or a ridge/lasso regression (e.g. Robyn). Those models do not measure causal effect unless you have no endogeneity issues. To make is super simple, unless you capture absolutely every variable that affect your outcome and your explanatory variable. This is hardly the case as the list of potential variable might be very large and most importantly, it might be impossible to get all those data (e.g. granular data on advertisment spending on different channel for every competitor). Moreover, note that even if you capture everything it might be also a problem to assess causality (e.g. what we call in econometrics bad control or in causal inference colider should not be controlled for).
In order to measure causal effects you need an actual experimental setup (e.g. A/B testing if the model is well setup) or quasi-experimental setup. If you want to know more about causality, you can read the free ebook by matheus facure "Causal inference for the brave and true", or "the Mixtape" by Scott Cuningham or watch my TEDx or read my articles on Towards Data Science called "The Science and Art of causality"
You rock! Great breakdown. Keep it up.
How can we learn mmm?
amazing, well explained
Boa esse conteúdo não tem no brasil
I just watched a 1M-Sub-Quality video from a channel that has 986 subs. Sub #987 over here, remind me when this channel blows up, soon. Keep it up!
Oh wow - thank you so much for that feedback!
Ehi can you share the deepdive on MMM?
Hi Cassandra, we will see what we can create :) do you have any specific questions on MMM that you'd like answered? We also have some extra blog resources on it if you're curious funnel.io/blog/what-is-marketing-mix-modeling-and-how-does-it-work
Is this basically just running a random forest regression?
Seriously ? Football?
Are you into Handegg more?