It should be clear to you from what you have already learned that, in general:

	interpolation gives better predictions than extrapolation
	large samples give better predictions than small samples
	high correlation gives better predictions than low correlation

Whilst this is useful to know, you really need to be able to QUANTIFY the margin of error for a given prediction.  Provided that the assumptions of regression analysis hold, you can calculate:

Both intervals are centred on the value predicted by the regression equation, namely y₀ = a + bx₀, but the prediction interval has a bigger standard error.  The t-value is calculated from (n-2) degrees of freedom with tail area (a/2).

Use the spreadsheet to investigate how the intervals depend on the many different quantities involved.