Skip to main content

Regression workflow

End-to-end: from raw data to a published-ready regression table.

1. Inspect & clean​

  • Analyze → Descriptives over all numeric variables to spot ranges and missing.
  • Analyze → Missing Data → Patterns to see what's missing where.
  • If missingness is non-trivial, Analyze → Missing Data → MICE (multiple imputation).

2. Check assumptions before fitting​

  • Analyze → Diagnostics → Normality on the outcome
  • Analyze → Diagnostics → Multicollinearity (VIF) on the candidate predictors

3. Fit​

Analyze → Regression. Pick OLS, choose dependent, predictors. Check "Robust standard errors (HC3)" if Breusch-Pagan suggests heteroscedasticity.

For variants the same form covers (no second dialog needed):

  • Ridge / LASSO / Elastic Net if multicollinearity is severe
  • Robust (Huber) if outliers
  • Quantile for median or other quantile regression
  • WLS with a weights variable
  • IV-2SLS with endogenous + instrument pickers

4. Diagnose the fit​

After fitting, run:

  • Analyze → Diagnostics → Regression Diagnostics (Breusch-Pagan, White, Goldfeld-Quandt, Durbin-Watson, Ljung-Box, RESET)
  • Analyze → Diagnostics → Influence (Cook's distance, leverage, DFBETAS)

5. Output​

Analyze → Publication Tables → Regression Table → APA/HTML/LaTeX-ready table.