R-Squared Plummets from 0.98 to 0.51: The Geo-Level vs National-Level Mystery in Google Meridian

When analyzing data with Google Meridian, including geographical data yields an R-squared of 0.89 for geo-level models and 0.98 for national-level models. However, excluding the geographical data causes the national-level R-squared to drop dramatically to 0.51. This unexpected discrepancy raises questions about how Google Meridian processes and aggregates data when geographical information is included or excluded. The significant difference in R-squared values suggests that the platform might apply some form of weighting or aggregation when geo data is present, which impacts the national-level results. This finding highlights the importance of understanding how statistical models handle different levels of data aggregation.

Source: stackoverflow.com