ECON 257D2 Lecture Notes - Lecture 3: National Institute Of Justice, Null Hypothesis, Cardiac Output
Document Summary
Goodness of (cid:133)t tests are designed to establish distributional properties of the data: such as do they come from a particular distribution: e. g. uniform, Poisson, normal, or whether in a bivariate or multivariate setting the di erent characteristics are independent. If the data is generated by a normal, then the distributions of the estimators, test statistics are known exactly in many cases (e. g. the sample mean has a normal distribution, the t-ratio is exactly student(cid:146)s t etc. ); there is no need to worry about how good the asymptotic approximation is. Violations of normality are common for (cid:133)nancial variables; incorrectly as- suming normality may lead to very misleading results. Kolmogorov-smirnov (ks) test is based on the supremum of the di erence between the (empirical) distribution function and the normal distribution func- tion. Another statistic is based on the comparison of moments.