Figure 11-10 shows the standard results along with the robust
standard errors. Notice that both sets of results have identical
coefficients. However, the robust standard errors change the
t-statistics, confidence intervals, and p-values for the
coefficients. Some of the standard errors increase with the
heteroskedasticity correction and others decrease. In this
example, all the coefficients that were originally significant
remain significant with the heteroskedasticity correction.
However, the effect of tenure was originally insignificant, but
is marginally significant using the robust standard error.
Note: The goal isn’t to make all coefficients statistically
significant, but to obtain more accurate standard errors in the
presence of heteroskedasticity. If some standard errors
increase to the point of making some coefficients insignificant
that were previously significant, then you should accept this
as part of your correction and more legitimate results.