Where \(\hat{a}_1\) and \(\hat{a}_2\) are two probability % and \(n_1\) and \(n_2\) are their effective sample sizes and \(se_1\) and \(se_2\) are their standard errors, the test statistic is:
\[\begin{align} t=\frac{\hat{a}_1-\hat{a}_2}{\sqrt{se^2_1 + se^2_2}} \end{align},\]
where:
\(p = 2\Pr(t_v \ge |t|),\)
\(v = \frac{(se^2_1+se^2_2)^2}{se^4_1/(n_1-b)+se^4_2/(n_2-b) },\)
\(b\) = Bessel's correction.