Cause: Perfect multicollinearity in auxiliary regression Solution: Check for duplicated or linearly dependent predictors in the auxiliary regression. Removing or combining offending variables usually resolves the issue.
Cause: Negative values in variable used for log transformation Solution: Ensure the variable is strictly positive before applying a log transformation or add a small constant to shift the data.
For very large data sets consider using
performWhiteTestStreaming() or running diagnostics on a
representative sample to reduce computation time.
runHeteroTests() runs the streaming tests when it is given
a finite chunk_threshold_mb, such as
chunk_threshold_mb = 100; by default it runs the exact
tests whatever the size of the data.
Numerical issues may arise with extreme multicollinearity or poorly scaled variables. Rescaling predictors or using robust optimisation methods can help.
No single test is uniformly most powerful. Examine residual plots and consider the nature of your data when diagnostics give conflicting results.
Some tests have low power in small samples. Simulation via
simulate_power() can help determine the best approach for a
given situation.