Debiasing Panel Data

Panel-based datasets often carry geographic or behavioral biases based on how the data is sourced—whether from opt-in users, mobile devices, or submitted receipts. Panel Debiasing helps correct for these imbalances to produce more accurate, representative insights.

What Is Panel Debiasing?

Panel Debiasing reweights your framework output to align with the true distribution of users, behaviors, or transactions—typically using a trusted reference dataset. This process adjusts for over- or under-representation across geographies, cohorts, or channels.

If you're working with panel data such as credit card transactions or app usage, applying debiasing is critical—especially when benchmarking across time, comparing geos, or evaluating brand performance.

How It Works in Carbon Arc

Carbon Arc offers panel debiasing through the SDK with a dedicated function:

df_pb = ca.explorer.get_framework_panel_debias_data(
    framework_id=framework_id,
    insight_id=626,  # reference insight ID
    data_type="dataframe",
    fetch_all=True
)

This function:

Reweights your framework’s values against a reference insight
Outputs a cleaned, corrected dataset
Keeps your original schema (same metrics, improved accuracy)

How to Use Panel Debiasing

Build and purchase a framework using the Carbon Arc platform or SDK
Choose a reference insight ID (see guidance below)
Run the SDK function to retrieve debiased results:

from carbonarc import ExplorerAPIClient

client = ExplorerAPIClient(token="your_token")
debiased = client.get_framework_panel_debias_data(
    framework_id="your_framework_id",
    insight_id=626,  # reference
    data_type="dataframe"
)

Use the debiased DataFrame in your analysis—more accurate, more representative.

Function Reference

get_framework_panel_debias_data()

Argument	Description
`framework_id`	ID of your purchased framework
`insight_id`	The reference metric used to correct for skews
`data_type`	`"dataframe"` or `"timeseries"`
`fetch_all`	`True` to download full dataset

What Is a Reference Insight?

A reference insight is a benchmark metric (e.g., total card transactions) that represents the true distribution of the population. It acts as the "ground truth" used to adjust and correct your primary dataset.

An effective reference insight should be:
Population-representative (e.g., aligns with census-based card usage)
Geographically aligned (uses the same resolution—state, DMA, ZIP)
Time and topic matched (drawn from the same period and subject area as the primary dataset)

Example: Debiasing Card Spend at the State Level

"location_resolution": "us"

Then your debiased output will return metrics for the entire US volume of the insight.

FAQs

Question	Answer
Will results match Platform 1.0 debiased feeds?	No—debiasing logic has been updated. Reprocessing is recommended.
Can I get a list of valid Insight IDs for reference?	Yes with `get_valid_insights_for_framework_panel_debias"`
Can I view the reweighting factors?	Not yet—this will be available in a future release.
Is there an extra cost to use Panel Debias?	No—it's included with your SDK access after purchase.
Can I apply Panel Debias before purchasing a framework?	Not currently.
Can I return results as a timeseries?	Yes—set `data_type="timeseries"` in your request.
Can I debias multiple entities in a single call?	Not yet—currently limited to one entity per request.
Is Panel Debias only available for location-level data?	Yes—for now, only entity vs. location debiasing is supported.

What Is Panel Debiasing?​

How It Works in Carbon Arc​

How to Use Panel Debiasing​

Function Reference​

What Is a Reference Insight?​

Example: Debiasing Card Spend at the State Level​

FAQs​

Questions?​