1. Abstract
Ensuring that each explanatory indicator has the correct expected sign before running Predict prevents mis-specified relationships and unnecessary reruns. Confirming sign direction improves model validity, interpretability, and efficiency.
2. Context
Apply this best practice during Predict configuration, after selecting explanatory indicators but before launching the Predict job. This step is especially important when modeling economic relationships where directionality is well understood.
3. Content
3.1 Why It Matters
In Predict, each explanatory variable can be constrained to an expected sign (positive or negative). This setting influences how the model estimates coefficients.
If the expected sign is incorrect:
- The model may force an implausible relationship
- Predict may exclude an otherwise valid driver
- Users may need to rerun the entire job to correct the error
- Model interpretation may become confusing or misleading
Expected sign configuration ensures that statistical optimization aligns with economic logic. It acts as a safeguard against accidental inversion of cause-and-effect relationships.
For example, higher interest rates would typically be expected to reduce demand. Allowing a positive sign without review could produce a result that contradicts well-established economic theory.
3.2 How to Apply
Before running Predict:
- After selecting the model type and consistency exclusions, review the indicator selection screen.
- For each explanatory indicator, verify the “+” or “–” sign setting. Note that these will default to the sign of the correlation coefficient.
- Confirm that the selected sign matches theoretical expectations:
- Income → positive impact on sales
- Interest rates → negative impact on borrowing
- Prices → typically negative impact on demand
- Adjust the sign manually if necessary.
- Double-check sign settings before launching the Predict job.
- Document the reasoning for any constrained signs, particularly if they differ from the raw R-value direction.
3.3 Example
An analyst building a consumer demand model includes “Unemployment Rate” as an explanatory variable. The default R-value is slightly positive due to short-term volatility, but economic theory suggests unemployment should negatively impact demand. The analyst sets the expected sign to negative before running Predict, ensuring results align with economic intuition.
3.4 Common Pitfalls
- Leaving the default sign without review
- Assuming Predict automatically determines correct direction
- Forgetting to revisit expected signs after adding new indicators
- Constraining signs without economic justification
3.5 Expected Results
- Reduced need for rerunning Predict
- Coefficients aligned with economic theory
- Clearer model interpretation
- Stronger stakeholder confidence in results
