Binary Variables Input and R² Discrepancy

[shantisb]

Hi!

I’d like to include binary variables within the "non_media_treatments" where a value of 1 indicates that a promotion took effect, and 0 indicates that it did not. Can Meridian handle this input without any issues?

Additionally, I noticed a discrepancy between the R² values shown in the model diagnostics (national level) and those in the summarizer’s model fit metrics. Could you clarify the difference between these two R² values? Just for context I'm running my model at a national level.

[cpulavarthi]

Hello @shantisb,

Thank you for contacting us!

You can definitely use binary variables with Meridian. You can in fact even perform dummy encoding for categorical variables to pass them as inputs to the model. Please check Discussion #473 which answers a similar query.

I am unable to reproduce the issue of discrepancy in R² values using the two methods you suggested on national level data. I am using the following code along with sample national data available in the GitHub repository.

model_diagnostics = visualizer.ModelDiagnostics(meridian)
model_diagnostics.predictive_accuracy_table()

filepath = '/content/drive/MyDrive'
start_date = '2021-01-25'
end_date = '2024-01-15'
summarizer.output_model_results_summary('summary_output.html', filepath, start_date, end_date)
IPython.display.HTML(filename='/content/drive/MyDrive/summary_output.html')

Please share the following information to help identify the root cause of this issue:

1. Are you using holdout sets?
2. Can you also share the code used to compute R² values using the two methods you mentioned?

Feel free to reach out if you have any further queries.

Thank you

Google Meridian Support Team