permutation/randomization test
generates a null distribution for the association between two variables by repeatedly and randomly re arranging the values
what are the assumptions for permutation?
-observations are randomly selected and independent
-the measure of association describes the pattern of interest
-if comparing means/medians the distributions have similar shapes
when can we do permutation ?
1-when the normality cannot be fixed by transformation
2-balanced design with similar variances (when the data is not too crazy)
what should we know for permutation?
the data will get shuffle, without replacement. the Ho for it is that there is no association
bootstrapping standard errors and CI
estimate SE and CI when we cannot assume normality (measures the precision)
uses the data itself to get an estimate of the confidence
resample the sample with replacement
