Just take the (unstandardized) residuals from your regression and divide them by their standard deviation. It doesn't matter if they are normally distributed or otherwise.

What Dragan suggests is by far the easiest way to reach the criteria you want. There are other variations on this method especially for regression analysis where the variance of the residuals isn't constant. To correct for this you could look into what are known as studentized residuals.

Thanks guys! Yes it's a regression study, but you see a model only once; and variance is not constant. Let me see if I can find any formula for the latter.