Chapter 9. Correlation and Regression

Transcription

1 Chapter 9 Correlation and Regression

2 Lesson 9-1/9-2, Part 1 Correlation

3 Registered Florida Pleasure Crafts and Watercraft Related Manatee Deaths Year Boats in Ten Thousands Manatee Deaths

4 9-1 Overview This chapter introduces methods for making inferences based on sample data that come in pairs or bivariate data. Is there a relationship between the data? If so, what is the equation? Use that equation for prediction.

5 Correlation A correlation exists between two variables when one of them is related to the other in some way.

6 Scatter Diagram A scatterplot (or scatter diagram) is a graph in which the paired (x, y) sample data are plotted with a horizontal x-axis and a vertical y-axis. Each individual (x, y) pair is plotted as a single point.

7 Scatter Diagram Table 9-1: Registered Florida Craft (in tens of thousands) and Watercraft-Related Manatee Deaths Year x: Boats y: Manatee

8 Scatter Plots TI83

9 Manatee Deaths from Boats Scatter Plot Registered Boats (tens of thousands)

10 Dangers of Scatter Diagrams The horizontal or vertical scale of a scatter diagram can be set so that the scatter diagram misleads a reader.

11 Dangers of Scatter Diagrams Because of this danger you must use numerical summaries of paired data in conjunction with graphs in order to determine the type of relationship.

12 Positive Linear Correlations

13 Negative Linear Correlations

14 No Linear Correlations

15 Linear Correlation Coefficient The linear correlation coefficient (r ) measures the direction and strength of the linear relationship between paired x and y values in a sample.

16 Assumptions The sample of paired data (x, y) is a random sample. The pairs of (x, y) data have a bivariate normal distribution.

17 Notation for the Linear Correlation Coefficient n number of pairs of data present x the sum of all x-values 2 x x 2 the x-value should be squared and then those squares added. the sum of all x-values and the total then squared xy the x-value should be multiplied by its corresponding y-value r linear correlation coefficient for a sample ρ linear correlation coefficient for a population

18 r Formula for the Linear Correlation Coefficient n xy x y n x x n y y

19 Justification for r formula r x x y y n s s 1 x y x, y centriod Figure 9-7

20 Interpreting the Linear Correlation Coefficient 1) Linear correlation coefficient r is always between 1 and 1. 2) The closer r is to 1, the stronger the evidence of positive association between the two variables. 3) The closer r is to a 1, the stronger the evidence of negative relationship between two variables. 4) If r is closer to 0, does not rule out any strong relationship between x and y, there could still be a strong relationship but one that is not linear. 5) The units of x and y plays no role in the interpretation of r. 6) Correlation is strongly effected by outlying observations.

21 Example Page 511, #6 Use a scatter plot and the linear correlation r to determine whether there is a correlation between the two variables. x y r n xy x y n x x n y y x y x y x 2 y

22 Example Page 511, #6 r n xy x y n x x n y y x y xy x 2 y x = 16 y = 41 xy = 185 x = 70 x = 495

23 Example Page 511, #6 r n xy x y n x x n y y r 5(185) x = 16 y = 41 xy = 185 x 2 = 70 y 2 = 495

24 Example Page 511, #6 x = 16 y = 41 xy = 185 x 2 = 70 y 2 = 495

25 Example Page 511, #6

26 Example Page 511, #6 Using Table A-6, n = 5 and assuming = 0.05 Critical Value = 0.878; therefore r = indicates significant (positive) linear correlation.

27 Lesson 9-1/9-2, Part 2 Correlation

28 Interpreting the Linear Correlation Coefficient The absolute value of the computed value of r exceeds the value in Table A-6, conclude that there is significant linear correlation. Otherwise there is not significant evidence to support the conclusion of significant linear correlation.

29 1 r 1 Value of r does not change if all values of either variable are converted to a different scale. The r is not affected by the choice of x and y. Properties of the Linear Correlation Coefficient Interchange x and y and the value r will not change. r measures the strength of a linear relationship.

30 Interpreting r: Explain Variation The value of r 2 is the proportion of the variation in y that is explained by the linear relationship between x and y. r 2 is the square of the linear correlation coefficient r.

31 Example: Boats and Manatees A. Find the value of the linear correlation coefficient r. B. What proportion of the variation of the manatee deaths can be explained the variation in the number of boat registrations? Table 9-1: Registered Florida Craft (in tens of thousands) and Watercraft-Related Manatee Deaths Year x: Boats y: Manatee

32 Example: Boats and Manatees with r = 0.922, we get r 2 = We can conclude that (or about 85%) of the variation in manatee deaths can be explained by the linear relationship between the number of boat registrations and the number of manatee deaths from boats. This implies that 15% of the variation of manatee deaths cannot be explained by the number of boat registrations.

33 Hypothesis Test Method 1 We wish to determine whether there is a significant linear correlation between two variables. Let H o : = 0 No significant linear correlation Let H 1 : 0 Significant linear correlation Let H 1 : > 0 Significant positive linear correlation Let H 1 : < 0 Significant negative linear correlation

34 Test Statistic t r 1 r 2 n 2 Critical Values: Use table A-3 with df = n - 2

35 Example: Boats and Manatees Test the claim that there is a linear correlation between the number of registered boats and the number of manatee deaths from boats. Table 9-1: Registered Florida Craft (in tens of thousands) and Watercraft-Related Manatee Deaths Year x: Boats y: Manatee H o : = 0 H 1 : 0

36 Example: Boats and Manatees t r 2 1 r n p-value = 1.51E 4

37 Example: Boats and Manatees t 6.71 p value df n 2 8 Critical Values

38 Example: Boats and Manatees There is sufficient evidence to reject H o, since p-value = < = 0.05 and conclude that there is significant correlation between the number of registered boats and the number of manatee deaths from boats.

39 Hypothesis Test Method 2 We wish to determine whether there is a significant linear correlation between two variables. Test statistic is r Critical values refer to Table A-6 with no degrees of freedom.

40 Example Page 511, #2 Using data collected from the FBI and the Bureau of Alcohol, Tobacco, and Firearms, the number of registered automatic Weapons and the murder rate (in murders per 100,000 people) was obtained for each of eight randomly selected states. Statdisk was used to find the value of the linear correlation Coefficient is r = a) Is there a significant linear correlation between the number of registered automatic weapons and the murder rate? n r Use table A-6 assume = 0.05

41 Example Page 511, #2 critical values = Since the absolute value of is greater than there is significant positive correlation

42 Example Page 511, #2 b) What proportion of the variation in the murder rate can be explained by the linear relationship between the murder rate and the number of registered automatic weapons? r r (0.885) The proportion of the variation in murder rate that can be explained in terms of variation in registered automatic weapons is 78.3%.

43 Example Page 512, #8 Listed below are the numbers of fires (in thousands) and the acres that were burned (in millions) in 11 western states in year of the last decade (based on the data from USA Today. Is there a correlation? Fires Acres Burned

44 Example Page 512, #8 LinRegTTest Ho : ρ 0 p value H : ρ 0 1 α 0.05 r 2 r % There is sufficient evidence to fail to reject H o since the p-value > α and are unable to conclude that there is significant linear correlation between the number of fires and the numbers of acres burned.

45 Example Page 512, #8 The data were listed under a headline of Loggers seize on fires to argue for more cutting. Do the data support the argument that as loggers remove more trees, the risk of fire decreases because the forests are less dense? The results do not support any conclusion about the removal of trees affecting the risk of fires, because none of the variables addresses the removal/density of trees.

46 Common Errors Involving Correlation Causation: It is wrong to conclude that correlation implies causality. Averages: Averages suppress individual variation and may inflate the correlation coefficient. Linearity: There may be some relationship between x and y even when there is no significant linear correlation.

47 Example Page 515, #26 Describe the error in the stated conclusion. Given: There is significant linear correlation between personal income and years of education. Conclusion: More education causes a person s income to rise. The presence of a significant linear correlation does not necessarily mean that one of the variables causes the other. Correlation does not imply causality.

48 Example Page 515, #28 Describe the error in the stated conclusion. Given: There is significant linear correlation between state average tax burdens and state average incomes. Conclusion: There is significant linear correlation between Individual tax burdens and individual incomes rise. Averages tend to suppress variation among individuals, so a correlation among averages does not necessary mean that there is correlation among individuals.

49 Lesson 9-3, Part 1 Regression

50 Objective In lesson 9-2 we use a scatter diagram and linear correlation coefficient to indicate that a linear relation exists between two variables. In lesson 9-3 we are going to find a linear equation that describes the relation between the two variables.

51 Regression Equation The regression equation expresses a relationship between x (called the independent variable, predictor variable or explanatory variable), and y (called the dependent variable or response variable) y mx b The typical equation of a straight line is expressed in form yˆ bo b1x, where b o is the y-intercept and b 1 is the slope.

52 Assumptions We are investigating only linear relationships For each x-value, y is a random variable having a normal (bell-shaped) distribution. All of these y distributions have the same variance. For a given value of x, the distribution of y-values has a mean that lies on the regression line. Results are not seriously affected if departures from normal distributions and equal variances are not too extreme.

53 Regression Line Regression Equation Given a collection of paired data, the regression equation Algebraically describes the relationship between two variables. Regression Line yˆ b b x o 1 The graph of the regression equation is called the regression (or line of best fit, or least square line).

54 Notation for Regression Equation y-intercept of Regression equation Slope of regression equation Equation of the regression line Population Parameter Sample Statistic TI o b o a 1 b 1 b y = o + 1 x yˆ b b x o 1 y = a + bx

55 Formulas for b o and b 1 Formula 9-2 (Slope) b n xy x y n x x Formula 9-3 (y-intercept) b y b x o 1

56 Example: Boats and Manatee Table 9-1: Registered Florida Craft (in tens of thousands) and Watercraft-Related Manatee Deaths Year x: Boats y: Manatee Find the regression equation Enter the values in L1 and L2 and use LinRegTTest

57 Example: Boats and Manatee Find the equation of the regression line. y a bx yˆ x

58 Example: Boats and Manatee Regression Line

59 Example Page 527, #6 Find the equation of the regression line. Use the given data. x y Enter the data into L1 and L2 and use LinRegTTest

60 Example Page 527, #6 yˆ x

61 Using the Regression Equation for Predictions In predicting a value of y based on some given value of x If there is not a significant linear correlation, the best predicted y-value is ( y ). If there is significant linear correlation, the best predicated y-value is found by substituting the x-value into the regression equation.

62 Guidelines for Using The Regression Equation 1. If there is no linear correlation, don t use the regression to make predictions. 2. When using the regression equation for predictions, stay within the scope of the available sample data. 3. A regression equation on old data is not necessary valid now. 4. Don t make predictions about a population that is different from the population from which the sample data were drawn.

63 Example page 527, #2 In each of the following cases, find the best predicted value of y given that x = The given statistics are summarized from paired sample data. a. r y 8.00 n 30, and the equation of the regression line is yˆ x Step 1 Check for linear correlation Since the absolute value of is less than There is no linear correlation. yˆ y

64 Example page 527, #2 b. r y 8.00 n 30, and the equation of the regression line is yˆ x Step 1 Check for linear correlation Since the absolute value of is greater than There is linear correlation yˆ x yˆ (2.00) yˆ

65 Example Page 528, #8 Find the best predicted value for the number of acres burned given that there was 80 fires. Fires Acres Burned Enter data into L1 and L2 and use LinRegTTest to check for linear correlation.

66 Lesson 9-3, Part 2 Regression

67 Example Page 528, #8 Regression Equation: yˆ x P-value > α there is no significant linear correlation. Use y for predication when x 80.

68 Example Page 528, #8 Go to 2 nd STAT yˆ y

69 Interpreting the Regression Equation Marginal change is the amount that a variable changes when the other variable changes by exactly one unit. Outlier is a point lying far away from the other data points. Influential point strongly effects the graph of the regression line.

70 Example Page 530, #26 Refer to the sample data listed in Table 9-1. If we include another pair of values consisting of x = 120 (for 1,200,000 boats) and y = 10 (manatee deaths from boats), is this new point an outlier? Is it an influential point? Table 9-1: Registered Florida Craft (in tens of thousands) and Watercraft-Related Manatee Deaths Year x: Boats y: Manatee

71 Example Page 530, #26 Yes; the point is an outlier, since it is far from the other data points 120 is far from other boat values, and 10 is far from the other manatee death values. Yes; the point is an influential one, since it will dramatically alter the regression line the original regression line predicts yˆ (120) 159.4, and so the new regression line will have to change considerably to come close to (120, 10).

72 Residuals and Least-Square Property Residual For a sample paired (x, y) data, the difference between the observed sample y-value and y-value that is predicted by using the regression equation. Least-Square Property A straight line satisfies this property if the sum of the squares of the line residuals is the smallest sum possible. y yˆ

73 Example: Residuals x y yˆ 54x x yˆ

74 The Method Least-Square The line that best describes the relation between two variables is the one that makes the residuals as small as possible. The least-square regression line minimizes the square of the vertical distance between observed values and predicated values. Minimize (residuals) 2

75 Lesson 9-4, Part 1 Variation and Prediction Intervals

76 Different types of Variation We consider different types of variation that can be used for two major applications: To determine the proportion of the variation in y that can be explained by the linear relationship between x and y. To construct interval estimates of predicted y- values such intervals are called prediction intervals.

77 Total and Explained Deviations Total Deviation from the mean of the particular point (x, y) is the vertical distance y y which is the distance between the point (x, y) and the horizontal line passing through the sample mean y. Explained Deviation is the vertical distance ŷ y which is the distance between the predicted y-value and the horizontal line passing through the sample mean y.

78 Total and Explained Deviations Unexplained Deviation is the vertical distance y yˆ, which is the vertical distance between the point (x, y) and the regression line. The distance y yˆ, is also called residual, as defined by Lesson 9-3.

79 Deviations

80 Total Deviation (total deviation) = (explained deviation) + (unexplained deviation) y y ŷ y y yˆ y y 2 ŷ y 2 y yˆ 2 Formula 9-4

81 Coefficient of Determination The coefficient of determination measures the percentage of total variation in the y-value that is explained by the regression line. r 2 ŷ y y y 2 2 explained variation total variation

82 Example Page 538, #2 Use the value of the linear correlation coefficient r to find the coefficient of determination and the percentage of the total variation that can be explained by the linear relationship between the two variables. r r ( 0.6) 0.36 The portion of the total variation explained by the regression line is 36%

83 Standard Error of Estimate The standard error of estimate is a measure of the differences (or distances) between the observed sample y-values and the predicted y-values ŷ that are obtained using the regression equation. s e y yˆ y b y b xy n2 n2 2 2 o 1

84 Example Page 539, #10 Fourteen different second-year medical students took blood pressure measurements of the same patient and the results are listed below. Systolic Diastolic Enter the data into L1 and L2 and use LinRegTtest

85 Example Page 539, #10 2 A. Find the explained variation ŷ y. Step 1 Find the regression equation yˆ x

86 Example Page 539, #10 Step 2 Use L3 to find ŷ

87 Example Page 539, #10 Step 3 Use L4 to find y Use 2 nd Stat

88 Example Page 539, #10 Step 4 Use L4 to find ŷ y 2

89 Example Page 539, #10 Step 5 Find ŷ y 2 Explained Variation ŷ y Use 2 nd Stat

90 Example Page 539, #10 2 B. Find the unexplained variation y yˆ. 2 Step 1 Use L5 to find ( y yˆ )

91 Example Page 539, #10 Step 3 Find y yˆ 2 Unexplained Variation 2 y yˆ Use 2 nd Stat

92 Example Page 539, #10 C. Find the total variation y y 2 TV EV UV

93 Example Page 539, #10 D. Find the coefficient of determination. 2 r

94 Example Page 539, #10 E. Find the standard error of estimate s e

95 Lesson 9-4, Part 2 Variation and Prediction Intervals

96 Prediction Intervals Prediction Intervals are intervals constructed about the predicted value of y that are used to measure the accuracy of a single predicted value.

97 Predication Interval for an Individual y yˆ E y yˆ E E 2 1 n( x x) o t s 1 α e 2 n n x x 2 2 x o = the given value of x t /2 = has n 2 degree of freedom in table A-3

98 Example Page 540, #14 Refer to problem #10 and assume necessary conditions of normality and variance are met. Systolic Diastolic A. Find the predicted diastolic reading given that the systolic reading is 120. Enter the data into L1 and L2 and use LinRegTtest

99 Example Page 540, #14 Step 1 Find the regression equation yˆ x (120)

100 Example Page 540, #14 B. Find a 95% predication interval estimate of the diastolic reading given that the systolic reading is 120. Step 1 Identify the calculations needed for the margin of error formula n 14 x x x n x x

101 Example Page 540, #14 n 14 x x x n x x s e t ,0.025 E 2 1 n( x x) o t s 1 α e 2 n n x x ( )

102 Example Page 540, #14 Step 2 Use ŷ and margin of error yˆ E y yˆ E y y When x = 120, (78) we are 95% certain that the corresponding y-value is between and 98.22

103 Interpreting a Computer Display Minitab The regression equation is Nicotine = Tar Predictor Coef SE Coef T P Constant Tar S = R-Sq = 92.4% R-Sq(adj) = 92.1% Predicted values for New Observations New Obs Fit SE Fit 95.0% CI 95.0% PI (1.2107, ) (1.0731, )

104 Example Page 539, #6 and 8 6. Identify Total Variation What percent of the total variation in nicotine can be explained by the linear relationship between tar and nicotine? 92.4% 8. Finding the Prediction Interval For given tar amount 17mg. Identify the 95% confidence interval estimate of the amount of nicotine, and write a statement interpreting that interval. 1.1 < y < 1.4; we have 95% confidence that the limits 1.1 and 1.4 contain the true amount of nicotine.

105 Lesson 9-6 Modeling

106 Mathematical Modeling A mathematical model is a mathematical function that fits or describes real-world data

107 TI Generic Models y a bx 2 y ax bx c y a bln x y ab x y b ax 1 y c ae bx

108 TI Key Strokes Use the Stat key

109 Development of a Good Mathematics Model Look for a Pattern in the Graph Examine the graph of the plotted points and compare the basic pattern to the known generic graphs. Find and Compare Values of R 2 Select functions that result in larger values of R 2, because such larger values correspond to functions that better fit the observed points. Think Use common sense. Don t use a model that lead to predicted values known to be totally unrealistic.

110 Example Page 554, #2 Construct a scatterplot and identify the mathematical model that best fits the given data. Assume that the model is to be used only for the scope of the given data, and considered only linear, quadratic, logarithmic, exponential and power models. x y

111 Example Page 554, #2 GRAPH Zoom #9:Zoomstat

112 Example Page 554, #2 Step 3 Decide which mathematical model Linear: yˆ 5x2

113 Example Page 554, #4 Construct a scatterplot and identify the mathematical model that best fits the given data. Assume that the model is to be used only for the scope of the given data, and considered only linear, quadratic, logarithmic, exponential and power models. x y

114 Example Page 554, #4 GRAPH Zoom #9:Zoomstat

115 Example Page 554, #4 Step 3 Decide which mathematical model Power: yˆ 2x 1 2

116 Example Page 555, #6 Refer to the annual high values of the Dow-Jones Industrial Average that are listed in Data Set 25 in Appendix B. What is the best predicted value for the year 2001? Given that the Actual high value in 2001 was 11,350, how good was the predicted value? What does the pattern suggest about the stock market for investment purposes?

117 Example Page 555, #6

118 Example Page 555, #6 Decide which mathematical model Exponential: yˆ x

119 Example Page 555, #6 What is the best predicted value for the year 2001? Given that the actual high value in 2001 was 11,350, how good was the predicted value? yˆ x 22 ˆ ( ) 12,262 y It misses by a fairly large amount.