Suppose you have an unbiased, normally distributed estimator of a parameter B, and are testing the null hypothesis B=0 with a two tail test. For the present sample, the standard error of the estimator is 1.0

Suppose the true value of B=0.5. Professor X conducts the study and finds a positive and significant B. If you attempt to replicate Professor X's study in other data sets of the same size, what percentage of those data sets would you expect the sign of a significant B (that is, a B for which the null hypothesis is rejected) to flip from positive to negative?