Chapter 7: Correlation and regression

Transcription

1 Slide 7.1 Chapter 7: Correlation and regression Correlation and regression techniques examine the relationships between variables, e.g. between the price of doughnuts and the demand for them. Such analyses can be useful for formulating policies, e.g. if there is a positive relationship between the quantity of money and the price level, inflation may be curbed by stricter control of the money supply. Barrow, Statistics for Economics, Accounting and Business Studies, 5 th edition Pearson Education Limited 009

2 Slide 7. Positive correlation E.g. income and food expenditure Barrow, Statistics for Economics, Accounting and Business Studies, 5 th edition Pearson Education Limited 009

3 Slide 7.3 Negative correlation E.g. demand and price Barrow, Statistics for Economics, Accounting and Business Studies, 5 th edition Pearson Education Limited 009

4 Slide 7.4 Zero (absence of) correlation E.g. sales of cameras and the price of fish Barrow, Statistics for Economics, Accounting and Business Studies, 5 th edition Pearson Education Limited 009

5 Slide 7.5 The correlation coefficient Measures the strength of association between two variables, X and Y. -1 r +1 Positive correlation: r > 0 Negative correlation: r < 0 Zero correlation: r 0 Barrow, Statistics for Economics, Accounting and Business Studies, 5 th edition Pearson Education Limited 009

6 Slide 7.6 The correlation coefficient (continued) The closer r is to +1 (or -1), the closer the points lie to a straight line with positive (negative) slope. Slide 7.: r = 0.8 Slide 7.3: r = -0.7 Slide 7.4: r = 0 Barrow, Statistics for Economics, Accounting and Business Studies, 5 th edition Pearson Education Limited 009

7 Slide 7.7 Barrow, Statistics for Economics, Accounting and Business Studies, 5 th edition Pearson Education Limited 009 Formula for the correlation coefficient ) ) ( )( ) ( ( Y Y n X X n Y X XY n r Use either or, equivalently Y Y X X Y Y X X r

8 Slide 7.8 Why the formula works X X Y Y 0 X X Y Y 0 Y X X Y Y 0 X X Y Y 0 X More >0 than <0 points, hence r > 0 Barrow, Statistics for Economics, Accounting and Business Studies, 5 th edition Pearson Education Limited 009

9 Slide 7.9 Calculation of r between growth and birth rates Country Birth rate GNP growth Y X Y X XY Brazil Colombia Costa Rica India , Mexico , Peru , Philippines , Senegal , South Korea Sri Lanka Taiwan Thailand Total , ,139.7 Barrow, Statistics for Economics, Accounting and Business Studies, 5 th edition Pearson Education Limited 009

10 Slide 7.10 Calculation of r between growth and birth rates (continued) Using the second formula, r ( n X n XY X Y ( X ) )( n Y ( Y) ) we obtain r 11, ( )(11, ) 0.84 Barrow, Statistics for Economics, Accounting and Business Studies, 5 th edition Pearson Education Limited 009

11 Birth rate Slide 7.11 Chart of birth rate against growth rate Growth rate Barrow, Statistics for Economics, Accounting and Business Studies, 5 th edition Pearson Education Limited 009

12 Slide 7.1 Notes about r The correlation between Y and X is the same as between X and Y, i.e. it does not matter which variable is labelled X and which Y. r is independent of units of measurement. If the birth rate were measured as births per 100 population (3.0,.9,...) r would still be Correlation does not imply causality. Barrow, Statistics for Economics, Accounting and Business Studies, 5 th edition Pearson Education Limited 009

13 Slide 7.13 Is the result statistically significant? H 0 : r = 0 H 1 : r 0 The null asserts no genuine association between X and Y and the sample correlation observed is just due to (bad) luck. The test statistic is n t r n 1 r ~ t Barrow, Statistics for Economics, Accounting and Business Studies, 5 th edition Pearson Education Limited 009

14 Slide 7.14 Testing the hypothesis Choose a = 5%, hence t* 10 =.8 t r n 1 r Hence we reject H 0. There does seem to be some genuine association between X and Y Barrow, Statistics for Economics, Accounting and Business Studies, 5 th edition Pearson Education Limited 009

15 Slide 7.15 Regression We now assert X causes Y, i.e. the growth rate influences the birth rate (not vice versa) Regression measures the effect of X upon Y and whether it is statistically significant Regression also allows several explanatory variables to influence Y Barrow, Statistics for Economics, Accounting and Business Studies, 5 th edition Pearson Education Limited 009

16 Birth rate Slide 7.16 The regression line Y = a + bx Growth rate Barrow, Statistics for Economics, Accounting and Business Studies, 5 th edition Pearson Education Limited 009

17 Birth rate Slide 7.17 How to obtain the regression line Minimise the sum of squared errors, Se The error is the vertical distance between an observation and the regression line error, e Growth rate Barrow, Statistics for Economics, Accounting and Business Studies, 5 th edition Pearson Education Limited 009

18 Slide 7.18 Slope b Regression formulae n XY n X X Y ( X ) Intercept a Y bx Barrow, Statistics for Economics, Accounting and Business Studies, 5 th edition Pearson Education Limited 009

19 Slide 7.19 Slope Calculation of regression coefficients b 11, Intercept a (.700) Y i = X i + e i Barrow, Statistics for Economics, Accounting and Business Studies, 5 th edition Pearson Education Limited 009

20 Slide 7.0 Measuring goodness of fit Use the coefficient of determination, R R 0 R 1 RSS TSS RSS: regression sum of squares TSS: total sum of squares Barrow, Statistics for Economics, Accounting and Business Studies, 5 th edition Pearson Education Limited 009

21 Slide 7.1 The component parts of R 35 TSS component Y i Yˆi Y RSS component 30 3 Growth rate X.8 Barrow, Statistics for Economics, Accounting and Business Studies, 5 th edition Pearson Education Limited 009

22 Slide 7. Calculating sums of squares TSS = RSS + ESS Hence TSS Y Y Y ny ˆ ESS Y Y Y a Y b XY RSS 1, , , R Barrow, Statistics for Economics, Accounting and Business Studies, 5 th edition Pearson Education Limited 009

23 Slide 7.3 Summary Correlation measures the association between two variables Regression extends this by measuring the effect of X upon Y (the slope coefficient b) The regression line is found by minimising the sum of squared errors. It is the line of best fit A measure of goodness of fit is R Barrow, Statistics for Economics, Accounting and Business Studies, 5 th edition Pearson Education Limited 009