June 3, 2008Stat Lecture 5 - Correlation1 Exploring Data Numerical Summaries of Relationships between Statistics Lecture 5

June 3, 2008Stat Lecture 5 - Correlation2 Administrative Notes Homework 2 due Monday, June 8 th Start the homework tonight –It’s long –There are things that will confuse you You will need JMP for the homework –Download it (link on my website) –Use a Wharton computer (link to the Wharton account information on website)

June 2, 2008Stat Lecture 5 - Correlation3 Outline Correlation Linear Regression Linear Prediction

June 2, 2008Stat Lecture 5 - Introduction4 Course Overview Collecting Data Exploring Data Probability Intro. Inference Comparing Variables Relationships between Variables MeansProportionsRegressionContingency Tables

June 2, 2008Stat Lecture 5 - Correlation5 Examples from Last Class 60 US Cities: Education vs. mortality Vietnam draft lottery: draft order vs. date New Orleans Baltimore San Jose York Lancaster Denver

June 2, 2008Stat Lecture 5 - Correlation6 How two variables vary together We have already looked at single-variable measures of spread: We now have pairs of two variables: We need a statistic that summarizes how they are spread out “together”

June 2, 2008Stat Lecture 5 - Correlation7 Correlation Correlation of two variables: We divide by standard deviation of both X and Y, so correlation has no units

June 2, 2008Stat Lecture 5 - Correlation8 Correlation Correlation is a measure of the strength of linear relationship between variables X and Y Correlation has a range between -1 and 1 r = 1 means the relationship between X and Y is exactly positive linear r = -1 means the relationship between X and Y is exactly negative linear r = 0 means that there is no linear relationship between X and Y r near 0 means there is a weak linear relationship between X and Y

June 2, 2008Stat Lecture 5 - Correlation9 Measure of Strength Correlation roughly measures how “tight” the linear relationship is between two variables

June 2, 2008Stat Lecture 5 - Correlation10 Examples Education and Mortality r = Draft Order and Birthday r = -0.22

June 2, 2008Stat Lecture 5 - Correlation11 Cautions about Correlation Correlation is only a good statistic to use if the relationship is roughly linear Correlation can not be used to measure non-linear relationships Always plot your data to make sure that the relationship is roughly linear!

June 2, 2008Stat Lecture 5 - Correlation12 Correlation with Non-linear Data A high correlation does not always mean a strong linear relationship A low correlation does not always mean no relationship (just no linear relationship)

June 2, 2008Stat Lecture 5 - Correlation13 Linear Regression If our X and Y variables do show a linear relationship, we can calculate a best fit line in addition to the correlation Y = a + b·X The values a and b together are called the regression coefficients a is called the intercept b is called the slope

June 2, 2008Stat Lecture 5 - Correlation14 Example: US Cities Mortality = a + b·Education How to determine our “best” line ? ie. best regression coefficients a and b ?

June 2, 2008Stat Lecture 5 - Correlation15 Finding the “Best” Line Want line that has smallest total “Y-residuals” Must square “Y-residuals” and add them up: Y-residuals

June 2, 2008Stat Lecture 5 - Correlation16 Best values for slope and intercept Line with smallest total residuals is called the “least-squares” line need calculus to find a and b that give “least-squares” line

June 2, 2008Stat Lecture 5 - Correlation17 Example: Education and Mortality Mortality = · Education Negative association means negative slope b

June 2, 2008Stat Lecture 5 - Correlation18 JMP Output Mortality = · Education Note that JMP gives correlation in terms of r 2 Need to take square root and give correct sign r = -0.51

June 2, 2008Stat Lecture 5 - Correlation19 Interpreting the regression line The slope coefficient b is the average change you get in the Y variable if you increased the X variable by one Eg. increasing education by one year means an average decrease of ≈ 38 deaths per 100,000 The intercept coefficient a is the average value of the Y variable when the X variable is equal to zero Eg. an uneducated city (Education = 0) would have an average mortality of 1353 deaths per 100,000

June 2, 2008Stat Lecture 5 - Correlation20 Example: Vietnam Draft Order Draft Order = · Birthday Slightly negative slope means later birthdays have a lower draft order

June 2, 2008Stat Lecture 5 - Correlation21 Example: Crack Cocaine Sentences Relationship seems to be non-linear

June 2, 2008Stat Lecture 5 - Correlation22 Sentence = · Quantity Extra gram of crack means extra 0.27 months (about 8 days) sentence on average Intercept is not always interpretable: having no crack gets you an 90 month sentence!

June 2, 2008Stat Lecture 5 - Correlation23 Prediction The regression equation can be used to predict our Y variable for a specific value of our X variable Eg: Sentence = · Quantity Person caught with 175 grams has a predicted sentence of · 175 =137.5 months Eg: Mortality = · Education City with only 6 median years of education has a predicted mortality of · 6 = 1127 deaths per 100,000

June 2, 2008Stat Lecture 5 - Correlation24 Problems with Prediction Quality of predictions depends on truth of assumption of a linear relationship Eg: crack sentences for large quantities

June 2, 2008Stat Lecture 5 - Correlation25 Problems with Prediction We shouldn’t extrapolate predictions beyond the range of data for our X variable Eg: Sentence = · Quantity Having no crack still has a predicted sentence of 90 months! Eg: Mortality = · Education A city with median education of 35 years is predicted to have no deaths at all!

Break! Take 5 June 3, 2008Stat Lecture 5 - Correlation26

How to Use JMP Scatterplots Correlation Regression –Hand calculating (using Distribution) –Have JMP calculate (using Fit Y by X) June 3, 2008Stat Lecture 5 - Correlation27

June 3, 2008Stat Lecture 5 - Correlation28 Next Class Probability –This is the hardest part of the course. Moore, McCabe and Craig: Section