Or How to make the numbers say whatever you want.课程.ppt

Earnings Even when the scale is fine, graphs can still be deceptive Even when the scale is fine, graphs can still be deceptive As you can see, Bob earns substantially more than Joe Joe’s income Bob’s Income Is it really a two to one ratio that’s being displayed? Percent of US income going to federal taxes * Or How to make the numbers say whatever you want. Free Pizza Lunch with Faculty – Jordan Hall 4th floor lounge Congrats to the four individuals who were successful in the syllabus challenge “There are lies, damned lies, and statistics” -- Disraeli Anecdotal evidence is unreliable Why does the phone always ring when you’re in the shower? Determining the difference between chance and real effects Data = Signal + Noise Signal = What we’re trying to measure Noise = Error in our measurement If noise is random, then as the sample size increases, noise tends to cancel, leaving only signal. Mean – arithmetic average = ?x/n Median – the halfway point Mode – the most common answer 1 4 3 4 3 2 2 2 27 Mean = 4.8 Median = 3 Mode = 2 0 6 5 27 4 3 2 1 27 Mean = 7.5 Median = 4 Mode = 27 1 24 23 26 2 2 1 1 27 Mean = 10.7 Median = 2 Mode = 1 Range – Overall difference between the highest and lowest scores. Variance – Average difference from the mean. 1 9 9 9 9 9 11 11 11 11 11 19 1 1 1 1 1 1 19 19 19 19 19 19 Identical Range 1 9 9 9 9 9 11 11 11 11 11 19 6 6 6 6 6 7 13 14 14 14 14 14 Identical Variance The Normal Curve Central Limit Theorem Any time you have a measure which is created by summing several individual trials of data (Signal + Noise) you will end up with a normal curve Regression to the mean: On any given free throw, Bob has an 80% chance of making it. Bimodal Distributions Skewed Distributions Uniform distributions P value -- The probability that a particular set of data was gained by chance alone. Statistical Significance – Any P value of less than .05 Correlation measures the strength of a relationship between two variables. Positive Correlation 0 R 1 N