The Language of Statistical Decision Making Lecture 2 Section 1.3 Mon, Sep 5, 2005 Statistical Significance Statistically significant – The difference between the claim of the null hypothesis and the data is large enough that it cannot reasonably be attributed to chance. Possible Errors We might reject H0 when it is true. We might accept H0 when it is false. This is a Type I error. This is a Type II error. See Making Intelligent Errors, by Walter Williams. Decisions and Errors State of Nature H0 true H0 false Accept H0 Correct Type II Error Reject H0 Type I Error Correct Decision Example Consider the hypotheses: H0: A category 5 hurricane will not hit New Orleans in the next 25 years. H1: A category 5 hurricane will hit New Orleans in the next 25 years. What are the negative consequences of each type of error? Which type of error is more serious? Which should get the benefit of the doubt? Example We should probably reverse the roles: H0: A category 5 hurricane will hit New Orleans in the next 25 years. H1: A category 5 hurricane will not hit New Orleans in the next 25 years. Example Consider the hypotheses: H0: An asteroid will not hit New Orleans in the next 25 years. H1: An asteroid will hit New Orleans in the next 25 years. What are the negative consequences of each type of error? Which type of error is more serious? Which should get the benefit of the doubt? Example These are fine the way they are: H0: An asteroid will not hit New Orleans in the next 25 years. H1: An asteroid will hit New Orleans in the next 25 years. Significance Level Significance Level – The likelihood of rejecting H0 when it is true, i.e., the likelihood of committing a Type I error. – The likelihood of a Type I error. – The likelihood of a Type II error. That is, is the significance level. The quantity 1 – is called the power of the test. Significance Level The value of is determined by our criteria for rejecting the null hypothesis (which we haven’t talked about yet). If we demand overwhelming evidence against H0 before rejecting it, then will be small. If we demand little evidence against it, then will be large. Significance Level Generally speaking, as increases, decreases. That is, if we demand overwhelming evidence against H0 (i.e., for H1) before rejecting it, then will be large. If we demand little evidence against H0, then will be small. Let’s Do It! Let’s do it! 1.5, p. 13 – Which Error is Worse? Let’s do it! 1.6, p. 14 – Testing a New Drug. Porn can make you blind.