Apr. 27 Statistic for the day: Americans eat, on average, 15 quarts of ice cream per year. Power of a test The power of a statistical test refers to its ability to detect when something is happening – in other words, to reject the null hypothesis when it should be rejected. Assignment: Take practice exams Stop by to pick up your old exams Power as it relates to type 2 errors Influences on power If the null hypothesis is false: The closer the observed statistic is to the null value, the smaller the power. The larger the sample, the larger the power. Rejecting it is a good thing. Failure to reject is a type 2 error. The probability of rejecting is the power. Thus, high power corresponds to low probability of type 2 error. Eliminating one type of error: A couple of really stupid tests Test A: Always reject the null hypothesis, no matter what the data are. This test will never make a type 2 error! It has perfect power! (But it’s stupid.) Test B: Never reject the null hypothesis, no matter what the data are. This test will never make a type 1 error! But it has no power, and it’s stupid. Return to chi-squared statistics Suppose we are interested in the following research question: Is there a significant difference between men and women in STAT 100 with respect to the proportion who have smoked marijuana? According to the survey for this class, 58.8% of women (out of 114) versus 61.4% of men (out of 101) have smoked marijuana. 1 Rows: sex How to measure the distance between what the research advocate observes in the table and what the skeptic expects: Columns: marijuana top lines are observed counts bottom lines are expected counts No 47 45.60 Yes 67 68.40 All 114 114.00 Male 39 40.40 62 60.60 101 101.00 All 86 86.00 129 129.00 215 215.00 Female Add up the following for each cell: χ2 = (obs − exp) 2 exp (47 − 45.6) 2 (67 − 68.4) 2 (39 − 40.4) 2 (62 − 60.6) 2 + + + = 0.152 45.6 68.4 40.4 60.6 Chi-squared distribution with 1 degree of freedom: 2.0 How about a p-value for the marijuana test? The key is to take the square root of the chi-squared statistic and treat that as the standardized score! 1.0 1.5 If chi-squared statistic is larger than 3.84, it is declared large and the research advocate wins. Null: No difference between men & women 0.5 Cutoff=3.84 Alternative (2-sided): A difference exists 5% on this side Test statistic: 0.152 = 0.39 0.0 95% on this side 0 1 2 3 4 5 6 But our chi-squared is 0.152, so the research advocate does not win. There is NOT a statistically significant difference between men and women. 2-sided p-value: 2×.35 = .70 Decision: WE HAVE NO EVIDENCE OF ANY DIFFERENCE IN THE PERCENTAGE WHO HAVE SMOKED MARIJUANA. 2
© Copyright 2026 Paperzz