Running a regression
- Click Stepwise Regression on the Home tab to open the panel.
- Y range: select the dependent variable column, including its label cell, and click Use selection.
- X range: select the candidate predictor columns. Hold Ctrl (Cmd on Mac) to add non-adjacent columns. Every column must cover the same rows as Y.
- Leave First row contains labels ticked if your selections include headers.
- Pick the method and criterion, adjust the thresholds if needed, and click Run regression.
Methods
- Forward selection: starts with no predictors and adds the strongest candidate while it meets the entry threshold.
- Backward elimination: starts with every predictor and drops the weakest while it fails the removal threshold.
- Stepwise: adds like forward selection, then rechecks every term in the model and drops any that no longer qualify. Repeats until nothing changes.
Criteria
- T-Stat (default): a variable enters when |T-Stat| is at least 2.00 and is removed when |T-Stat| falls below 2.00.
- P-value: a variable enters when its p-value is below 0.05 and is removed when it rises above 0.10.
Both use the T-Stat of the variable's coefficient in the model being considered. For stepwise, the entry threshold must be at least as strict as the removal threshold, so a variable cannot cycle in and out.
Reading the report
- Selection steps: each variable entered or removed, with its T-Stat and p-value, and the model's R Square, Adjusted R Square, RMSE, AIC and BIC after that step.
- Final model: Regression Statistics, ANOVA and coefficient tables in the same layout as Excel's Data Analysis > Regression report. Standard Error of the regression is labeled RMSE.
- Variables not in the final model: the T-Stat and p-value each would have if it were added next.
Common problems
"Fewer than 3 complete numeric rows"
Check that the ranges cover your data and that the labels setting matches. Rows with a blank, text or error value in Y or any X are skipped, and the report shows how many were skipped.
"The model with every predictor cannot be estimated"
Backward elimination needs every predictor in the first model. A constant column, or one that is an exact combination of others (for example a total and its parts), makes that impossible. Remove the redundant column or use forward or stepwise.
The button is missing
Reinstall the add-in from the store or the add-in file (HOUMBA-28-Stepwise-Regression.xml) on the install page. In Excel for the web, add-ins uploaded from a file stay for that browser only.
Contact
Report a problem or suggest a change on the project's issue page.