2.8 Take-Home Points

• We may create an exact sampling distribution or simulate a bootstrap sampling distribution in simple situations or if we have a lot of computing power.

• For a bootstrap sampling distribution, we need about 5,000 bootstrap samples from our original sample.

• An exact sampling distribution can only be used with categorical variables.

• We can often approximate the sampling distribution of a sample statistic with a known theoretical probability distribution.

• Approximations only work well under conditions, which we have to check.

• Conditions usually involve the size of the sample, sample type (independent vs. dependent/paired), and the shape or variance of the population distribution.

• If these conditions are not met or we do not have a theoretical approximation to the sampling distribution, we use bootstrapping or exact tests.

• Samples are independent if, in principle, we can draw a sample for one group without taking into account the sample for another group of cases. Otherwise, the samples are dependent or paired.