Free Econometrics Tools Online: Robust SE, IV, Unit Root
The estimators and tests econometricians report: heteroskedasticity- and autocorrelation-robust standard errors, specification tests, instrumental variables, binary and count models with marginal effects, and the core time-series tests for stationarity, cointegration and Granger causality.
OLS with Robust Standard Errors and Diagnostics
Regression with HC0–HC3, clustered or Newey-West errors plus Breusch-Pagan, White, RESET, Durbin-Watson, Breusch-Godfrey and Jarque-Bera tests.
Replaces Stata · EViewsInstrumental Variables (2SLS) Regression
Two-stage least squares with first-stage F, partial R², Sargan over-identification and Wu-Hausman endogeneity tests.
Replaces Stata ivregressProbit and Logit with Average Marginal Effects
Binary choice models with average marginal effects, delta-method standard errors, odds ratios and fit statistics.
Replaces Stata marginsPoisson and Negative Binomial Regression
Count-data regression with incidence rate ratios, overdispersion tests and a model comparison.
Replaces Stata · SPSS GENLINUnit Root Tests: Augmented Dickey-Fuller and KPSS
Augmented Dickey-Fuller and KPSS stationarity tests in levels and first differences, with ACF, PACF and Ljung-Box.
Replaces EViews · StataEngle-Granger Cointegration Test
Engle-Granger two-step cointegration test with MacKinnon critical values, the long-run equation and residual plot.
Replaces EViews · StataGranger Causality Test
Test whether past values of one series predict another, in both directions, for each lag with AIC and BIC.
Replaces EViews · StataOLS with Robust Standard Errors and Diagnostics
Use it for any cross-section or time-series regression where you need to justify the standard errors and show that the model passes standard specification tests.
Instrumental Variables (2SLS) Regression
Use 2SLS when a regressor is correlated with the error term, because of omitted variables, simultaneity or measurement error, and you have instruments that affect the outcome only through that regressor.
Probit and Logit with Average Marginal Effects
Use probit or logit for binary outcomes. Report average marginal effects, because raw coefficients have no direct probability interpretation.
Poisson and Negative Binomial Regression
Use count models when the outcome is a non-negative integer. Negative binomial is preferred when the variance exceeds the mean, which is common.
Unit Root Tests: Augmented Dickey-Fuller and KPSS
Run unit root tests on every series before time-series regression, VAR, Granger causality or cointegration analysis, to avoid spurious regressions.
Engle-Granger Cointegration Test
Use it after unit root tests show that two or more series are I(1), to decide between an error-correction model (cointegrated) and a model in differences (not cointegrated).
Granger Causality Test
Use it to test lead-lag relationships between stationary time series, for example whether money growth predicts inflation.
Other categories
Data cleaning and preparation · Describe and explore · Compare means · Nonparametric tests · Correlation and regression · Bayesian statistics · Categorical data · Reliability and agreement · Multivariate analysis · Structural equation modeling · Panel data and causal inference · Marketing and decision research · Survival, clinical and meta-analysis · Study planning and calculators · PLS-SEM