Basic Idea Vary the independent variable x Vary the dependent variable y We are trying to find out Does y depend on x Sometimes this is clear Other times it is not Is there a trend here Probably We need a statistical parameter to determine if there is a trend n rxy 1 r xy x x y i 1 i i mean x y 1 2 2 2 x x y y i i i 1 i 1 n n mean y 1 But how do we interpret this rxy 1 Indicates a perfect linear relationship with a positive slope rxy 1 Indicates a perfect linear relationship with a negative slope rxy 0 Indicates no linear correlation between x and y Unchanged by multiplying each value of a variable by a constant Unchanged by adding a constant to each variable Unchanged by interchanging x and y For a given sample size we need to determine whether a given rxy is significant or a result of pure chance For practical problems we can consult a table General procedure Calculate rxy from the measured data Determine level of significance a required a gives the probability that an experimental value of rxy will be greater by pure chance a 0 05 rt value at which 5 chance due to random effects Compare rt to rxy if rxy rt then confidence level is confirmed 95 confidence often used in engineering Say we have a sample size of 18 and we get a correlation coefficient of 0 49 Does a linear relationship exist between our variables 95 Yes our correlation coefficient is greater than that required for 95 Confidence see table It s thought that the lap times for a race car depend on the ambient temperature The following data for the same car with the same driver were measured at different races Ambient temperature F 40 47 55 62 66 88 Lap time s 65 3 66 5 67 3 67 8 67 66 6 Does a linear relationship exist between the two variables 68 Lap Time s Let s look at the plot 67 66 65 40 50 60 70 80 Temperature F 90 Ambient temperature F 40 47 55 62 66 88 Lap time s 65 3 66 5 67 3 67 8 67 66 6 Now we calculate the correlation coefficient n rxy x x y i 1 i i y 1 2 2 2 x x y y i i i 1 i 1 n n rxy 0 413 Lets look for a 95 confidence level this means that a 1 95 0 05 rxy 0 413 rt 0 811 Regression Analysis Used to establish a functional relationship between the independent variable and the dependent variable Can be for a single variable or multiple variables y f a y f a Often we are looking for predictive capability li Let s say we hang weights xi on the spring The resulting spring length is li 0 1 xi unloaded length spring constant 0 Now we do an experiment Apply known weights Measure the total length yi li i 0 1 xi i Regression coefficients li error 0 li 0 1 xi What do we want to get Unloaded length 0 Spring constant 1 uncertainties To get the coefficients we can find a line that fits the data the best in the least squares sense Choose a line y i 0 1 xi Our measured values are off this line by ei yi y i Residual for point xi yi n ei 2 i 1 For the best fit minimize the sum of the squares of the residuals n n n K ei yi y i 2 i 1 2 i 1 i 1 yi 0 1 xi 2 K K 0 0 0 1 Solve for the regression coefficients 2 Equations 2 unknowns n 1 x x y y i i 1 n i 2 x x i 0 y 1 x li 0 i 1 y i 0 1 xi prediction estimate of true slope y i 0 1 xi estimate of true intercept Don t extrapolate outside of the range of the data used for the fit Don t use the least squares line when the data is not linear The slope of the best fit line is related to the correlation coefficient sy 1 r sx A measure of the goodness of fit is given by the Coefficient of Determination r2 n r2 n 2 y y y y i i i 2 i 1 i 1 n 2 y y i i 1 n 1 2 y y i i i 1 n y i 1 y 2 i Square of the correlation coefficient
View Full Document
Unlocking...