Post by account_disabled on Feb 25, 2024 11:27:08 GMT 5.5
The in Google Ads and Analytics uses the Shapley value in its algorithm. Unfortunately, it also has limitations. One of the most frequently raised objections is that it ignores the order of interactions. For example, the Google Facebook and Facebook Google paths are combined into one Google + Facebook combination. So, at the very beginning of data processing, we blur information that intuitively seems important from the point of view of analysis. Moreover, the computational complexity of the Shapley value increases exponentially n with the number of channels involved in the conversion process, which makes its calculation much more difficult for larger channel granularities.
When calculating the Shapley value, conversion rates are primarily taken into account, so this measure rewards channels with higher conversion rates and significantly underestimates those that do not generate conversions on their own such as Latvia WhatsApp Number List remarketing. The Shapley value is very sensitive to random values with little statistical significance, which it treats on an equal footing with reliable data. Combined with the fact that computing a large number of channels will be very resource-intensive, we can conclude that the Shapley value is best calculated for interactions grouped into a small number of channels, each containing data of appropriate statistical significance.
Finally, it's worth noting that Shapley's value is powered by marginal contribution data calculated from observed funnel conversion rates. This is not an empirical measurement. This is one of the reasons why the interpret contributions from interactions such as clicks on search results for self-branded keywords. These interactions are often associated with purchase intention and, consequently, a high conversion rate, which is interpreted as a high contribution to conversion, even though in fact these interactions should not be assigned a higher value. To sum up, the Shapley value is an interesting analytical tool supporting attribution modeling, but it is certainly not a panacea for all problems related to the proper assessment of the value.
When calculating the Shapley value, conversion rates are primarily taken into account, so this measure rewards channels with higher conversion rates and significantly underestimates those that do not generate conversions on their own such as Latvia WhatsApp Number List remarketing. The Shapley value is very sensitive to random values with little statistical significance, which it treats on an equal footing with reliable data. Combined with the fact that computing a large number of channels will be very resource-intensive, we can conclude that the Shapley value is best calculated for interactions grouped into a small number of channels, each containing data of appropriate statistical significance.
Finally, it's worth noting that Shapley's value is powered by marginal contribution data calculated from observed funnel conversion rates. This is not an empirical measurement. This is one of the reasons why the interpret contributions from interactions such as clicks on search results for self-branded keywords. These interactions are often associated with purchase intention and, consequently, a high conversion rate, which is interpreted as a high contribution to conversion, even though in fact these interactions should not be assigned a higher value. To sum up, the Shapley value is an interesting analytical tool supporting attribution modeling, but it is certainly not a panacea for all problems related to the proper assessment of the value.