I'm using Google Meridian to analyze a dataset containing dates, countries, channels, and a KPI. When I run the model including the geo (country) data, I get two R-squared values:

• R-squared for geographical level data: 0.89
• R-squared for national level data: 0.98

However, when I run the model without the geographical data (excluding the geo column), the R-squared drops significantly to 0.51.

I don't understand why the national-level R-squared is 0.98 when using geo data, but drops to 0.51 when I exclude it. Shouldn't the national-level R-squared be more consistent between these two scenarios?

I tried running the model both with and without the geo (country) column. I expected the R-squared for national-level data to remain similar in both cases, as I thought removing the geo information wouldn't drastically affect the national-level model.

Instead, the national-level R-squared was 0.98 when including geo data, but only 0.51 when excluding it. I’m wondering if Google Meridian applies some kind of aggregation or weighting when using geo-level data that affects the national-level results.

Why does this discrepancy exist, and how Meridian handles geo vs national data?