1. Agresti/Franklin Statistics, 1 of 88 Chapter 11 Analyzing Association Between Quantitative Variables: Regression Analysis Learn…. To use regression analysis to explore the association between two quantitative variables

2. Agresti/Franklin Statistics, 2 of 88  Section 11.1 How Can We “Model” How Two Variables Are Related?

3. Agresti/Franklin Statistics, 3 of 88 Regression Analysis The first step of a regression analysis is to identify the response and explanatory variables We use y to denote the response variable We use x to denote the explanatory variable

4. Agresti/Franklin Statistics, 4 of 88 The Scatterplot The first step in answering the question of association is to look at the data A scatterplot is a graphical display of the relationship between two variables

5. Agresti/Franklin Statistics, 5 of 88 Example: What Do We Learn from a Scatterplot in the Strength Study? An experiment was designed to measure the strength of female athletes The goal of the experiment was to find the maximum number of pounds that each individual athlete could bench press

6. Agresti/Franklin Statistics, 6 of 88 Example: What Do We Learn from a Scatterplot in the Strength Study? 57 high school female athletes participated in the study The data consisted of the following variables: x: the number of 60-pound bench presses an athlete could do y: maximum bench press

7. Agresti/Franklin Statistics, 7 of 88 Example: What Do We Learn from a Scatterplot in the Strength Study? For the 57 girls in this study, these variables are summarized by: x: mean = 11.0, st.deviation = 7.1 y: mean = 79.9 lbs, st.dev. = 13.3 lbs

8. Agresti/Franklin Statistics, 8 of 88 Example: What Do We Learn from a Scatterplot in the Strength Study?

9. Agresti/Franklin Statistics, 9 of 88 The Regression Line Equation When the scatterplot shows a linear trend, a straight line fitted through the data points describes that trend The regression line is: is the predicted value of the response variable y is the y-intercept and is the slope

10. Agresti/Franklin Statistics, 10 of 88 Example: Which Regression Line Predicts Maximum Bench Press?

11. Agresti/Franklin Statistics, 11 of 88 Example: What Do We Learn from a Scatterplot in the Strength Study? The MINITAB output shows the following regression equation: BP = 63.5 + 1.49 (BP_60) The y-intercept is 63.5 and the slope is 1.49 The slope of 1.49 tells us that predicted maximum bench press increases by about 1.5 pounds for every additional 60-pound bench press an athlete can do

12. Agresti/Franklin Statistics, 12 of 88 Outliers Check for outliers by plotting the data The regression line can be pulled toward an outlier and away from the general trend of points

13. Agresti/Franklin Statistics, 13 of 88 Influential Points An observation can be influential in affecting the regression line when two thing happen: Its x value is low or high compared to the rest of the data It does not fall in the straight-line pattern that the rest of the data have

14. Agresti/Franklin Statistics, 14 of 88 Residuals are Prediction Errors The regression equation is often called a prediction equation The difference between an observed outcome and its predicted value is the prediction error, called a residual

15. Agresti/Franklin Statistics, 15 of 88 Residuals Each observation has a residual A residual is the vertical distance between the data point and the regression line

16. Agresti/Franklin Statistics, 16 of 88 Residuals We can summarize how near the regression line the data points fall by The regression line has the smallest sum of squared residuals and is called the least squares line

17. Agresti/Franklin Statistics, 17 of 88 Regression Model: A Line Describes How the Mean of y Depends on x At a given value of x, the equation: Predicts a single value of the response variable But… we should not expect all subjects at that value of x to have the same value of y Variability occurs in the y values

18. Agresti/Franklin Statistics, 18 of 88 The Regression Line The regression line connects the estimated means of y at the various x values In summary, Describes the relationship between x and the estimated means of y at the various values of x

19. Agresti/Franklin Statistics, 19 of 88 The Population Regression Equation The population regression equation describes the relationship in the population between x and the means of y The equation is:

20. Agresti/Franklin Statistics, 20 of 88 The Population Regression Equation In the population regression equation, α is a population y-intercept and β is a population slope These are parameters In practice we estimate the population regression equation using the prediction equation for the sample data

21. Agresti/Franklin Statistics, 21 of 88 The Population Regression Equation The population regression equation merely approximates the actual relationship between x and the population means of y It is a model A model is a simple approximation for how variable relate in the population

22. Agresti/Franklin Statistics, 22 of 88 The Regression Model

23. Agresti/Franklin Statistics, 23 of 88 The Regression Model If the true relationship is far from a straight line, this regression model may be a poor one

24. Agresti/Franklin Statistics, 24 of 88 Variability about the Line At each fixed value of x, variability occurs in the y values around their mean, µ y The probability distribution of y values at a fixed value of x is a conditional distribution At each value of x, there is a conditional distribution of y values An additional parameter σ describes the standard deviation of each conditional distribution

25. Agresti/Franklin Statistics, 25 of 88 A Statistical Model A statistical model never holds exactly in practice. It is merely a simple approximation for reality Even though it does not describe reality exactly, a model is useful if the true relationship is close to what the model predicts

26. Agresti/Franklin Statistics, 26 of 88 Find the predicted fertility for Vietnam, which had the highest value of x = 91. a.5.25 b.469.2 c.1.196 d.10.73 For recent data on several nations, the prediction equation relating y = fertility rate to x = female economic activity (the female labor force as a percentage of the male labor force) is:

27. Agresti/Franklin Statistics, 27 of 88 Find the residual for Vietnam, which had y = 2.3. a.-2.136 b.1.104 c.-1.104 d.2.136 For recent data on several nations, the prediction equation relating y = fertility rate to x = female economic activity (the female labor force as a percentage of the male labor force) is:

28. Agresti/Franklin Statistics, 28 of 88  Section 11.2 How Can We Describe Strength of Association?

29. Agresti/Franklin Statistics, 29 of 88 Correlation The correlation, denoted by r, describes linear association The correlation ‘r’ has the same sign as the slope ‘b’ The correlation ‘r’ always falls between -1 and +1 The larger the absolute value of r, the stronger the linear association

30. Agresti/Franklin Statistics, 30 of 88 Correlation and Slope We can’t use the slope to describe the strength of the association between two variables because the slope’s numerical value depends on the units of measurement

31. Agresti/Franklin Statistics, 31 of 88 Correlation and Slope The correlation is a standardized version of the slope The correlation does not depend on units of measurement

32. Agresti/Franklin Statistics, 32 of 88 Correlation and Slope The correlation and the slope are related in the following way:

33. Agresti/Franklin Statistics, 33 of 88 Example: What’s the Correlation for Predicting Strength? For the female athlete strength study: x: number of 60-pound bench presses y: maximum bench press x: mean = 11.0, st.dev.=7.1 y: mean= 79.9 lbs., st.dev. = 13.3 lbs. Regression equation:

34. Agresti/Franklin Statistics, 34 of 88 Example: What’s the Correlation for Predicting Strength? The variables have a strong, positive association

35. Agresti/Franklin Statistics, 35 of 88 The Squared Correlation Another way to describe the strength of association refers to how close predictions for y tend to be to observed y values The variables are strongly associated if you can predict y much better by substituting x values into the prediction equation than by merely using the sample mean y and ignoring x

36. Agresti/Franklin Statistics, 36 of 88 The Squared Correlation Consider the prediction error: the difference between the observed and predicted values of y Using the regression line to make a prediction, each error is: Using only the sample mean, y, to make a prediction, each error is:

37. Agresti/Franklin Statistics, 37 of 88 The Squared Correlation When we predict y using y (that is, ignoring x), the error summary equals: This is called the total sum of squares

38. Agresti/Franklin Statistics, 38 of 88 The Squared Correlation When we predict y using x with the regression equation, the error summary is: This is called the residual sum of squares

39. Agresti/Franklin Statistics, 39 of 88 The Squared Correlation When a strong linear association exists, the regression equation predictions tend to be much better than the predictions using y We measure the proportional reduction in error and call it, r 2

40. Agresti/Franklin Statistics, 40 of 88 The Squared Correlation We use the notation r 2 for this measure because it equals the square of the correlation r

41. Agresti/Franklin Statistics, 41 of 88 Example: What Does r 2 Tell Us in the Strength Study? For the female athlete strength study: x: number of 60-pund bench presses y: maximum bench press The correlation value was found to be r = 0.80 We can calculate r 2 from r: (0.80) 2 =0.64 For predicting maximum bench press, the regression equation has 64% less error than y has

42. Agresti/Franklin Statistics, 42 of 88 Correlation r and Its Square r 2 Both r and r 2 describe the strength of association ‘r’ falls between -1 and +1 It represents the slope of the regression line when x and y have been standardized ‘r 2 ’ falls between 0 and 1 It summarizes the reduction in sum of squared errors in predicting y using the regression line instead of using y

43. Agresti/Franklin Statistics, 43 of 88 Find the predicted math SAT score for a student who has the verbal SAT score of 800. a.250 b.500 c.650 d.750 All Students who attend Lake Woebegone College must take the math and verbal SAT exams. Both exams have a mean of 500 and a standard deviation of 100. The regression equation relating y = math SAT score and x = verbal SAT score is:

44. Agresti/Franklin Statistics, 44 of 88 Find the r-value. a..5 b..25 c.1.00 d..75 All Students who attend Lake Woebegone College must take the math and verbal SAT exams. Both exams have a mean of 500 and a standard deviation of 100. The regression equation relating y = math SAT score and x = verbal SAT score is:

45. Agresti/Franklin Statistics, 45 of 88 Find the r 2 value. a..5 b..25 c.1.00 d..75 All Students who attend Lake Woebegone College must take the math and verbal SAT exams. Both exams have a mean of 500 and a standard deviation of 100. The regression equation relating y = math SAT score and x = verbal SAT score is:

46. Agresti/Franklin Statistics, 46 of 88  Section 11.3 How Can We make Inferences About the Association?

47. Agresti/Franklin Statistics, 47 of 88 Descriptive and Inferential Parts of Regression The sample regression equation, r, and r 2 are descriptive parts of a regression analysis The inferential parts of regression use the tools of confidence intervals and significance tests to provide inference about the regression equation, the correlation and r-squared in the population of interest

48. Agresti/Franklin Statistics, 48 of 88 Assumptions for Regression Analysis Basic assumption for using regression line for description: The population means of y at different values of x have a straight-line relationship with x, that is: This assumption states that a straight-line regression model is valid This can be verified with a scatterplot.

49. Agresti/Franklin Statistics, 49 of 88 Assumptions for Regression Analysis Extra assumptions for using regression to make statistical inference: The data were gathered using randomization The population values of y at each value of x follow a normal distribution, with the same standard deviation at each x value

50. Agresti/Franklin Statistics, 50 of 88 Assumptions for Regression Analysis Models, such as the regression model, merely approximate the true relationship between the variables A relationship will not be exactly linear, with exactly normal distributions for y at each x and with exactly the same standard deviation of y values at each x value

51. Agresti/Franklin Statistics, 51 of 88 Testing Independence between Quantitative Variables Suppose that the slope β of the regression line equals 0 Then… The mean of y is identical at each x value The two variables, x and y, are statistically independent: The outcome for y does not depend on the value of x It does not help us to know the value of x if we want to predict the value of y

52. Agresti/Franklin Statistics, 52 of 88 Testing Independence between Quantitative Variables

53. Agresti/Franklin Statistics, 53 of 88 Testing Independence between Quantitative Variables Steps of Two-Sided Significance Test about a Population Slope β: 1. Assumptions: The population satisfies regression line: Randomization The population values of y at each value of x follow a normal distribution, with the same standard deviation at each x value

54. Agresti/Franklin Statistics, 54 of 88 Testing Independence between Quantitative Variables Steps of Two-Sided Significance Test about a Population Slope β: 2. Hypotheses: H 0 : β = 0, H a : β ≠ 0 3. Test statistic: Software supplies sample slope b and its se

55. Agresti/Franklin Statistics, 55 of 88 Testing Independence between Quantitative Variables Steps of Two-Sided Significance Test about a Population Slope β: 4. P-value: Two-tail probability of t test statistic value more extreme than observed: Use t distribution with df = n-2 5. Conclusions: Interpret P-value in context If decision needed, reject H 0 if P-value ≤ significance level

56. Agresti/Franklin Statistics, 56 of 88 Example: Is Strength Associated with 60-Pound Bench Press?

57. Agresti/Franklin Statistics, 57 of 88 Example: Is Strength Associated with 60-Pound Bench Press? Conduct a two-sided significance test of the null hypothesis of independence Assumptions: A scatterplot of the data revealed a linear trend so the straight-line regression model seems appropriate The scatter of points have a similar spread at different x values The sample was a convenience sample, not a random sample, so this is a concern

58. Agresti/Franklin Statistics, 58 of 88 Example: Is Strength Associated with 60-Pound Bench Press? Hypotheses: H 0 : β = 0, H a : β ≠ 0 Test statistic: P-value: 0.000 Conclusion: An association exists between the number of 60-pound bench presses and maximum bench press

59. Agresti/Franklin Statistics, 59 of 88 A Confidence Interval for β A small P-value in the significance test of H 0 : β = 0 suggests that the population regression line has a nonzero slope To learn how far the slope β falls from 0, we construct a confidence interval:

60. Agresti/Franklin Statistics, 60 of 88 Example: Estimating the Slope for Predicting Maximum Bench Press Construct a 95% confidence interval for β Based on a 95% CI, we can conclude, on average, the maximum bench press increases by between 1.2 and 1.8 pounds for each additional 60-pound bench press that an athlete can do

61. Agresti/Franklin Statistics, 61 of 88 Example: Estimating the Slope for Predicting Maximum Bench Press Let’s estimate the effect of a 10-unit increase in x: Since the 95% CI for β is (1.2, 1.8), the 95% CI for 10β is (12, 18) On the average, we infer that the maximum bench press increases by at least 12 pounds and at most 18 pounds, for an increase of 10 in the number of 60-pound bench presses

62. Agresti/Franklin Statistics, 62 of 88  Section 11.4 What Do We Learn from How the Data Vary Around the Regression Line?

63. Agresti/Franklin Statistics, 63 of 88 Residuals and Standardized Residuals A residual is a prediction error – the difference between an observed outcome and its predicted value The magnitude of these residuals depends on the units of measurement for y A standardized version of the residual does not depend on the units

64. Agresti/Franklin Statistics, 64 of 88 Standardized Residuals Standardized residual: The se formula is complex, so we rely on software to find it A standardized residual indicates how many standard errors a residual falls from 0 Often, observations with standardized residuals larger than 3 in absolute value represent outliers

65. Agresti/Franklin Statistics, 65 of 88 Example: Detecting an Underachieving College Student Data was collected on a sample of 59 students at the University of Georgia Two of the variables were: CGPA: College Grade Point Average HSGPA: High School Grade Point Average

66. Agresti/Franklin Statistics, 66 of 88 Example: Detecting an Underachieving College Student A regression equation was created from the data: x: HSGPA y: CGPA Equation:

67. Agresti/Franklin Statistics, 67 of 88 Example: Detecting an Underachieving College Student MINITAB highlights observations that have standardized residuals with absolute value larger than 2:

68. Agresti/Franklin Statistics, 68 of 88 Example: Detecting an Underachieving College Student Consider the reported standardized residual of -3.14 This indicates that the residual is 3.14 standard errors below 0 This student’s actual college GPA is quite far below what the regression line predicts

69. Agresti/Franklin Statistics, 69 of 88 Analyzing Large Standardized Residuals Does it fall well away from the linear trend that the other points follow? Does it have too much influence on the results? Note: Some large standardized residuals may occur just because of ordinary random variability

70. Agresti/Franklin Statistics, 70 of 88 Histogram of Residuals A histogram of residuals or standardized residuals is a good way of detecting unusual observations A histogram is also a good way of checking the assumption that the conditional distribution of y at each x value is normal Look for a bell-shaped histogram

71. Agresti/Franklin Statistics, 71 of 88 Histogram of Residuals Suppose the histogram is not bell- shaped: The distribution of the residuals is not normal However…. Two-sided inferences about the slope parameter still work quite well The t- inferences are robust

72. Agresti/Franklin Statistics, 72 of 88 The Residual Standard Deviation For statistical inference, the regression model assumes that the conditional distribution of y at a fixed value of x is normal, with the same standard deviation at each x This standard deviation, denoted by σ, refers to the variability of y values for all subjects with the same x value

73. Agresti/Franklin Statistics, 73 of 88 The Residual Standard Deviation The estimate of σ, obtained from the data, is:

74. Agresti/Franklin Statistics, 74 of 88 Example: How Variable are the Athletes’ Strengths? From MINITAB output, we obtain s, the residual standard deviation of y: For any given x value, we estimate the mean y value using the regression equation and we estimate the standard deviation using s: s = 8.0

75. Agresti/Franklin Statistics, 75 of 88 Confidence Interval for µ y We estimate µ y, the population mean of y at a given value of x by: We can construct a 95 %confidence interval for µ y using:

76. Agresti/Franklin Statistics, 76 of 88 Prediction Interval for y The estimate for the mean of y at a fixed value of x is also a prediction for an individual outcome y at the fixed value of x Most regression software will form this interval within which an outcome y is likely to fall This is called a prediction interval for y

77. Agresti/Franklin Statistics, 77 of 88 Prediction Interval for y vs Confidence Interval for µ y The prediction interval for y is an inference about where individual observations fall Use a prediction interval for y if you want to predict where a single observation on y will fall for a particular x value

78. Agresti/Franklin Statistics, 78 of 88 Prediction Interval for y vs Confidence Interval for µ y The confidence interval for µ y is an inference about where a population mean falls Use a confidence interval for µ y if you want to estimate the mean of y for all individuals having a particular x value

79. Agresti/Franklin Statistics, 79 of 88 Example: Predicting Maximum Bench Press and Estimating its Mean

80. Agresti/Franklin Statistics, 80 of 88 Example: Predicting Maximum Bench Press and Estimating its Mean Use the MINITAB output to find and interpret a 95% CI for the population mean of the maximum bench press values for all female high school athletes who can do x = 11 sixty- pound bench presses For all female high school athletes who can do 11 sixty-pound bench presses, we estimate the mean of their maximum bench press values falls between 78 and 82 pounds

81. Agresti/Franklin Statistics, 81 of 88 Example: Predicting Maximum Bench Press and Estimating its Mean Use the MINITAB output to find and interpret a 95% Prediction Interval for a single new observation on the maximum bench press for a randomly chosen female high school athlete who can do x = 11 sixty-pound bench presses For all female high school athletes who can do 11 sixty-pound bench presses, we predict that 95% of them have maximum bench press values between 64 and 96 pounds

82. Agresti/Franklin Statistics, 82 of 88  Section 11.5 Exponential Regression: A Model for Nonlinearity

83. Agresti/Franklin Statistics, 83 of 88 Nonlinear Regression Models If a scatterplot indicates substantial curvature in a relationship, then equations that provide curvature are needed Occasionally a scatterplot has a parabolic appearance: as x increases, y increases then it goes back down More often, y tends to continually increase or continually decrease but the trend shows curvature

84. Agresti/Franklin Statistics, 84 of 88 Example: Exponential Growth in Population Size Since 2000, the population of the U.S. has been growing at a rate of 2% a year The population size in 2000 was 280 million The population size in 2001 was 280 x 1.02 The population size in 2002 was 280 x (1.02 )2 … The population size in 2010 is estimated to be 280 x (1.02 )10 This is called exponential growth

85. Agresti/Franklin Statistics, 85 of 88 Exponential Regression Model An exponential regression model has the formula: For the mean µ y of y at a given value of x, where α and β are parameters

86. Agresti/Franklin Statistics, 86 of 88 Exponential Regression Model In the exponential regression equation, the explanatory variable x appears as the exponent of a parameter The mean µ y and the parameter β can take only positive values As x increases, the mean µ y increases when β>1 It continually decreases when 0 < β<1

87. Agresti/Franklin Statistics, 87 of 88 Exponential Regression Model For exponential regression, the logarithm of the mean is a linear function of x When the exponential regression model holds, a plot of the log of the y values versus x should show an approximate straight-line relation with x

88. Agresti/Franklin Statistics, 88 of 88 Example: Explosion in Number of People Using the Internet

89. Agresti/Franklin Statistics, 89 of 88 Example: Explosion in Number of People Using the Internet

90. Agresti/Franklin Statistics, 90 of 88 Example: Explosion in Number of People Using the Internet

91. Agresti/Franklin Statistics, 91 of 88 Example: Explosion in Number of People Using the Internet Using regression software, we can create the exponential regression equation: x: the number of years since 1995. Start with x = 0 for 1995, then x=1 for 1996, etc y: number of internet users Equation:

92. Agresti/Franklin Statistics, 92 of 88 Interpreting Exponential Regression Models In the exponential regression model, the parameter α represents the mean value of y when x = 0; The parameter β represents the multiplicative effect on the mean of y for a one-unit increase in x

93. Agresti/Franklin Statistics, 93 of 88 Example: Explosion in Number of People Using the Internet In this model: The predicted number of Internet users in 1995 (for which x = 0) is 20.38 million The predicted number of Internet users in 1996 is 20.38 times 1.7708