1. Multivariate Data

2. Descriptive techniques for Multivariate data In most research situations data is collected on more than one variable (usually many variables)

3. Graphical Techniques The scatter plot The two dimensional Histogram

4. The Scatter Plot For two variables X and Y we will have a measurements for each variable on each case: x i, y i x i = the value of X for case i and y i = the value of Y for case i.

5. To Construct a scatter plot we plot the points: ( x i, y i ) for each case on the X-Y plane. ( x i, y i ) xixi yiyi

6. Data Set #3 The following table gives data on Verbal IQ, Math IQ, Initial Reading Acheivement Score, and Final Reading Acheivement Score for 23 students who have recently completed a reading improvement program InitialFinal VerbalMathReadingReading StudentIQIQAcheivementAcheivement 186941.11.7 21041031.51.7 386921.51.9 41051002.02.0 51181151.93.5 6961021.42.4 790871.51.8 8951001.42.0 9105961.71.7 1084801.61.7 1194871.61.7 121191161.73.1 1382911.21.8 1480931.01.7 151091241.82.5 161111191.43.0 1789941.61.8 18991171.62.6 1994931.41.4 20991101.42.0 2195971.51.3 221021041.73.1 23102931.61.9

8. (84,80)

10. Some Scatter Patterns

13. Circular No relationship between X and Y Unable to predict Y from X

16. Ellipsoidal Positive relationship between X and Y Increases in X correspond to increases in Y (but not always) Major axis of the ellipse has positive slope

18. Example Verbal IQ, MathIQ

20. Some More Patterns

23. Ellipsoidal (thinner ellipse) Stronger positive relationship between X and Y Increases in X correspond to increases in Y (more freqequently) Major axis of the ellipse has positive slope Minor axis of the ellipse much smaller

25. Increased strength in the positive relationship between X and Y Increases in X correspond to increases in Y (almost always) Minor axis of the ellipse extremely small in relationship to the Major axis of the ellipse.

28. Perfect positive relationship between X and Y Y perfectly predictable from X Data falls exactly along a straight line with positive slope

31. Ellipsoidal Negative relationship between X and Y Increases in X correspond to decreases in Y (but not always) Major axis of the ellipse has negative slope slope

33. The strength of the relationship can increase until changes in Y can be perfectly predicted from X

39. Some Non-Linear Patterns

42. In a Linear pattern Y increase with respect to X at a constant rate In a Non-linear pattern the rate that Y increases with respect to X is variable

43. Growth Patterns

46. Growth patterns frequently follow a sigmoid curve Growth at the start is slow It then speeds up Slows down again as it reaches it limiting size

47. Review the scatter plot

48. Some Scatter Patterns

50. Non-Linear Patterns

51. Measures of strength of a relationship (Correlation) Pearson’s correlation coefficient (r) Spearman’s rank correlation coefficient (rho,  )

52. Assume that we have collected data on two variables X and Y. Let ( x 1, y 1 ) ( x 2, y 2 ) ( x 3, y 3 ) … ( x n, y n ) denote the pairs of measurements on the on two variables X and Y for n cases in a sample (or population)

53. From this data we can compute summary statistics for each variable. The means and

54. The standard deviations and

55. These statistics: give information for each variable separately but give no information about the relationship between the two variables

56. Consider the statistics:

57. The first two statistics: are used to measure variability in each variable they are used to compute the sample standard deviations and

58. The third statistic: is used to measure correlation If two variables are positively related the sign of will agree with the sign of

59. When is positive will be positive. When x i is above its mean, y i will be above its mean When is negative will be negative. When x i is below its mean, y i will be below its mean The product will be positive for most cases.

61. This implies that the statistic will be positive Most of the terms in this sum will be positive

62. On the other hand If two variables are negatively related the sign of will be opposite in sign to

63. When is positive will be negative. When x i is above its mean, y i will be below its mean When is negative will be positive. When x i is below its mean, y i will be above its mean The product will be negative for most cases.

64. Again implies that the statistic will be negative Most of the terms in this sum will be negative

65. Pearsons correlation coefficient is defined as below:

66. The denominator: is always positive

67. The numerator: is positive if there is a positive relationship between X ad Y and negative if there is a negative relationship between X ad Y. This property carries over to Pearson’s correlation coefficient r

68. Properties of Pearson’s correlation coefficient r 1.The value of r is always between –1 and +1. 2.If the relationship between X and Y is positive, then r will be positive. 3.If the relationship between X and Y is negative, then r will be negative. 4.If there is no relationship between X and Y, then r will be zero. 5.The value of r will be +1 if the points, ( x i, y i ) lie on a straight line with positive slope. 6.The value of r will be -1 if the points, ( x i, y i ) lie on a straight line with negative slope.

69. r =1

70. r = 0.95

71. r = 0.7

72. r = 0.4

73. r = 0

74. r = -0.4

75. r = -0.7

76. r = -0.8

77. r = -0.95

78. r = -1

79. Computing formulae for the statistics:

81. To compute first compute Then

82. Example Verbal IQ, MathIQ

83. Data Set #3 The following table gives data on Verbal IQ, Math IQ, Initial Reading Acheivement Score, and Final Reading Acheivement Score for 23 students who have recently completed a reading improvement program InitialFinal VerbalMathReadingReading StudentIQIQAcheivementAcheivement 186941.11.7 21041031.51.7 386921.51.9 41051002.02.0 51181151.93.5 6961021.42.4 790871.51.8 8951001.42.0 9105961.71.7 1084801.61.7 1194871.61.7 121191161.73.1 1382911.21.8 1480931.01.7 151091241.82.5 161111191.43.0 1789941.61.8 18991171.62.6 1994931.41.4 20991101.42.0 2195971.51.3 221021041.73.1 23102931.61.9

85. Now Hence

86. Thus Pearsons correlation coefficient is:

87. Thus r = 0.769 Verbal IQ and Math IQ are positively correlated. If Verbal IQ is above (below) the mean then for most cases Math IQ will also be above (below) the mean.

88. Is the improvement in reading achievement (RA) related to either Verbal IQ or Math IQ? improvement in RA = Final RA – Initial RA

89. The Data Correlation between Math IQ and RA Improvement Correlation between Verbal IQ and RA Improvement

90. Scatterplot: Math IQ vs RA Improvement

91. Scatterplot: Verbal IQ vs RA Improvement

92. Spearman’s rank correlation coefficient  (rho)

93. Spearman’s rank correlation coefficient  (rho) Spearman’s rank correlation coefficient is computed as follows: Arrange the observations on X in increasing order and assign them the ranks 1, 2, 3, …, n Arrange the observations on Y in increasing order and assign them the ranks 1, 2, 3, …, n. For any case (i) let ( x i, y i ) denote the observations on X and Y and let ( r i, s i ) denote the ranks on X and Y.

94. If the variables X and Y are strongly positively correlated the ranks on X should generally agree with the ranks on Y. (The largest X should be the largest Y, The smallest X should be the smallest Y). If the variables X and Y are strongly negatively correlated the ranks on X should in the reverse order to the ranks on Y. (The largest X should be the smallest Y, The smallest X should be the largest Y). If the variables X and Y are uncorrelated the ranks on X should randomly distributed with the ranks on Y.

95. Spearman’s rank correlation coefficient is defined as follows: For each case let d i = r i – s i = difference in the two ranks. Then Spearman’s rank correlation coefficient (  ) is defined as follows:

96. Properties of Spearman’s rank correlation coefficient  1.The value of  is always between –1 and +1. 2.If the relationship between X and Y is positive, then  will be positive. 3.If the relationship between X and Y is negative, then  will be negative. 4.If there is no relationship between X and Y, then  will be zero. 5.The value of  will be +1 if the ranks of X completely agree with the ranks of Y. 6.The value of  will be -1 if the ranks of X are in reverse order to the ranks of Y.

97. Example x i 25.033.916.737.424.617.340.2 y i 24.338.713.432.128.012.544.9 Ranking the X’s and the Y’s we get: r i 4516327 s i 3625417 Computing the differences in ranks gives us: d i 1-1-11-110

99. Computing Pearsons correlation coefficient, r, for the same problem:

101. To compute first compute

102. Then

103. and Compare with

104. Comments: Spearman’s rank correlation coefficient  and Pearson’s correlation coefficient r 1.The value of  can also be computed from: 2.Spearman’s  is Pearson’s r computed from the ranks.

105. 3.Spearman’s  is less sensitive to extreme observations. (outliers) 4.The value of Pearson’s r is much more sensitive to extreme outliers. This is similar to the comparison between the median and the mean, the standard deviation and the pseudo-standard deviation. The mean and standard deviation are more sensitive to outliers than the median and pseudo- standard deviation.

106. Scatter plots

107. Some Scatter Patterns

109. Non-Linear Patterns

110. Measuring correlation 1.Pearson’s correlation coefficient r 2.Spearman’s rank correlation coefficient 

111. Simple Linear Regression Fitting straight lines to data

112. The Least Squares Line The Regression Line When data is correlated it falls roughly about a straight line.

113. In this situation wants to: Find the equation of the straight line through the data that yields the best fit. The equation of any straight line: is of the form: Y = a + bX b = the slope of the line a = the intercept of the line

114. a Run = x 2 -x 1 Rise = y 2 -y 1 b = Rise Runx2-x1x2-x1 = y2-y1y2-y1

115. a is the value of Y when X is zero b is the rate that Y increases per unit increase in X. For a straight line this rate is constant. For non linear curves the rate that Y increases per unit increase in X varies with X.

116. Linear

117. Non-linear

118. Age Class30-4040-5050-6060-7070-80 Mipoint Age (X) 3545556575 Median BP (Y) 114124143158166 Example: In the following example both blood pressure and age were measure for each female subject. Subjects were grouped into age classes and the median Blood Pressure measurement was computed for each age class. He data are summarized below:

119. Graph:

120. Interpretation of the slope and intercept 1.Intercept – value of Y at X = 0. –Predicted Blood pressure of a newborn (65.1). –This interpretation remains valid only if linearity is true down to X = 0. 2.Slope – rate of increase in Y per unit increase in X. –Blood Pressure increases 1.38 units each year.

121. The Least Squares Line Fitting the best straight line to “linear” data

122. Reasons for fitting a straight line to data 1.It provides a precise description of the relationship between Y and X. 2.The interpretation of the parameters of the line (slope and intercept) leads to an improved understanding of the phenomena that is under study. 3.The equation of the line is useful for prediction of the dependent variable (Y) from the independent variable (X).

123. Assume that we have collected data on two variables X and Y. Let ( x 1, y 1 ) ( x 2, y 2 ) ( x 3, y 3 ) … ( x n, y n ) denote the pairs of measurements on the on two variables X and Y for n cases in a sample (or population)

124. Let Y = a + b X denote an arbitrary equation of a straight line. a and b are known values. This equation can be used to predict for each value of X, the value of Y. For example, if X = x i (as for the i th case) then the predicted value of Y is:

125. For example if Y = a + b X = 25.2 + 2.0 X Is the equation of the straight line. and if X = x i = 20 (for the i th case) then the predicted value of Y is:

126. If the actual value of Y is y i = 70.0 for case i, then the difference is the error in the prediction for case i. is also called the residual for case i

127. If the residual can be computed for each case in the sample, The residual sum of squares (RSS) is a measure of the “goodness of fit of the line Y = a + bX to the data

128. X Y=a+bX Y (x 1,y 1 ) (x 2,y 2 ) (x 3,y 3 ) (x 4,y 4 ) r1r1 r2r2 r3r3 r4r4

129. The optimal choice of a and b will result in the residual sum of squares attaining a minimum. If this is the case than the line: Y = a + bX is called the Least Squares Line

136. The equation for the least squares line Let

137. Computing Formulae:

138. Then the slope of the least squares line can be shown to be:

139. and the intercept of the least squares line can be shown to be:

140. The following data showed the per capita consumption of cigarettes per month (X) in various countries in 1930, and the death rates from lung cancer for men in 1950. TABLE : Per capita consumption of cigarettes per month (X i ) in n = 11 countries in 1930, and the death rates, Y i (per 100,000), from lung cancer for men in 1950. Country (i)X i Y i Australia4818 Canada5015 Denmark3817 Finland11035 Great Britain11046 Holland4924 Iceland236 Norway259 Sweden3011 Switzerland5125 USA13020

143. Fitting the Least Squares Line

144. First compute the following three quantities:

145. Computing Estimate of Slope and Intercept

146. Y = 6.756 + (0.228)X

147. Interpretation of the slope and intercept 1.Intercept – value of Y at X = 0. –Predicted death rate from lung cancer (6.756) for men in 1950 in Counties with no smoking in 1930 (X = 0). 2.Slope – rate of increase in Y per unit increase in X. –Death rate from lung cancer for men in 1950 increases 0.228 units for each increase of 1 cigarette per capita consumption in 1930.

148. Age Class30-4040-5050-6060-7070-80 Mipoint Age (X) 3545556575 Median BP (Y) 114124143158166 Example: In the following example both blood pressure and age were measure for each female subject. Subjects were grouped into age classes and the median Blood Pressure measurement was computed for each age class. He data are summarized below:

149. Fitting the Least Squares Line

150. First compute the following three quantities:

151. Computing Estimate of Slope and Intercept

152. Graph:

153. Relationship between correlation and Linear Regression 1.Pearsons correlation. Takes values between –1 and +1

154. 2.Least squares Line Y = a + bX –Minimises the Residual Sum of Squares: –The Sum of Squares that measures the variability in Y that is unexplained by X. –This can also be denoted by: SS unexplained

155. Some other Sum of Squares: –The Sum of Squares that measures the total variability in Y (ignoring X).

156. –The Sum of Squares that measures the total variability in Y that is explained by X.

157. It can be shown: (Total variability in Y) = (variability in Y explained by X) + (variability in Y unexplained by X)

158. It can also be shown: = proportion variability in Y explained by X. = the coefficient of determination

159. Further: = proportion variability in Y that is unexplained by X.

160. Example TABLE : Per capita consumption of cigarettes per month (X i ) in n = 11 countries in 1930, and the death rates, Y i (per 100,000), from lung cancer for men in 1950. Country (i)X i Y i Australia4818 Canada5015 Denmark3817 Finland11035 Great Britain11046 Holland4924 Iceland236 Norway259 Sweden3011 Switzerland5125 USA13020

161. Fitting the Least Squares Line First compute the following three quantities:

162. Computing Estimate of Slope and Intercept

163. Computing r and r 2 54.4% of the variability in Y (death rate due to lung Cancer (1950) is explained by X (per capita cigarette smoking in 1930)

164. Y = 6.756 + (0.228)X

165. Comments Correlation will be +1 or -1 if the data lies on a straight line. Correlation can be zero or close to zero if the data is either –Not related or –In some situations non-linear

166. Example The data

167. One should be careful in interpreting zero correlation. It does not necessarily imply that Y is not related to X. It could happen that Y is non-linearly related to X. One should plot Y vs X before concluding that Y is not related to X.