ADM 2304 Lecture Notes - Lecture 9: Central Limit Theorem, Sampling Distribution, Statistical Hypothesis Testing
Document Summary
> the hypothesis testing outlined earlier depends on sampling distribution of the statistic being approximately normal. The sample size is large enough (central limit theorem) Suppose we have observations from a symmetric (non-normal) distribution. Compute di = xi - 0 where xi"s are the actual observations. Delete all zero differences and let the sample size n be the number of remaining di"s. Take the absolute values of the di"s and rank them (1 for smallest and n for largest) If there are ties, give the rank as the average of the possible ranking positions. Add up the ranks of those di"s that were originally negative and call the sum r- Add up the ranks of those di"s that were originally positive and call the sum r+ If the null hypothesis is correct then one would expect r- and r+ to be fairly equal. > data should be evenly spread out around the median.