1. Motivation
In Classical Linear Model Assumptions, we make a strong assumption about the covariance matrix of error terms:
- Homoskedasticity (同方差): equal variance = the same value on diagonal
- No-autocorrelation (无自相关): zero covariance between different error terms = zero for non-diagonal values
If this assumption is violated, the OLS estimators remain unbiased. However, the expression for the estimator variance changes, rendering the values calculated using the original standard error formula incorrect. Hence, hypothesis tests conducted using these “SE” values become invalid. In this case, we will use robust standard errors instead.
2. How to Calculate
# Specify the model and Estimate
model <- ...
# t-test for coefficients using standard errors
summary(model)
# t-test for coefficients using robust standard errors "HC0" (i.e. White standard errors)
coeftest(model, vcov = vcovHC, type = "HC0")