Anova, Regression, Correlation

Anova, Regression, Correlation

<p> 1</p><p>252regr 2/26/07 (Open this document in 'Outline' view!) Roger Even Bove</p><p>G. LINEAR REGRESSION-Curve Fitting 1. Exact vs. Inexact Relations 2. The Ordinary Least Squares Formula ˆ We wish to estimate the coefficients in Y   0  1 X   . Our ‘prediction’ will be Y  b0  b1 X and our error will be e  Y Yˆ so that Y  b0  b1 X  e . (See appendix for derivation)  XY  nXY b1  b0  Y  b1 X  X 2  nX 2</p><p>3. Example i Y X XY X 2 Y 2 1 0 0 0 0 0 2 2 1 2 1 4 3 1 2 2 4 1 4 3 1 3 1 9 5 1 0 0 0 1 6 3 3 9 9 9 7 4 4 16 16 16 8 2 2 4 4 4 9 1 2 2 4 1 10 2 1 2 1 4 sum 19 16 40 40 49 First copy n  10,  X  16, Y  19,  XY  40,  X 2  40 and Y 2  49. X 16 Y 19 Then compute means: X     1.60 Y     1.90 . n 10 n 10 2 2 2 Use these to compute ‘Spare Parts’: SS x   X  nX  40 101.60  14.40 2 2 2 SS y  Y  nY  49 101.9  12.90  SST (Total Sum of Squares) S xy   XY  nXY  40 101.61.9  9.60 . </p><p>Note that SS x and SS y must be positive, while S xy can be either positive or negative. We can compute the coefficients:</p><p>S xy XY  nXY 9.60 b      0.6667 b  Y  b X  1.90  0.6667 1.60  0.8333 1 2 2 0 1     SS x  X  nX 14.40 So our regression equation is Yˆ  0.8333 0.6667X or Y  0.8333  0.6667X  e . 2</p><p>4. R 2 , the Coefficient of Determination SST  SSR  SSE , SSR  b1S xy is Regression (Explained) Sum of Squares. 2 SSE  SST  SSR is the Error (Unexplained or Residual) Sum of Squares, and is defined as  Y Yˆ , a formula that should never be used for computation. 2 2 SSR b1S xy S xy 9.6 R 2      .4961 SST SSy SS x SS y 14.4012.90 b Y  b XY  nY 2 2 0  1  An alternate formula, if no spare parts have been computed, is R  Y 2  nY 2  2 2 R  r . The coefficient of determination is the square of the correlation. Note that SS x , b1, and r all have the same sign.</p><p>H. LINEAR REGRESSION-Simple Regression 1. Fitting a Line 2. The Gauss Markov Theorem OLS is BLUE ˆ 2 SSE Y Y  3. Standard Errors – The standard error is defined as s 2    . e n  2 n  2 2 2 SSE SST  SSR SS y  b1S xy SS y  b SS x se     . n  2 n  2 n  2 n  2 2 Y  b0 Y  b1 XY Or, if no spare parts are available, s 2     . e n  2 2 SS y 1 R  Note also that if R 2 is available s 2  . e n  2</p><p>Using data from G3, and using our spare parts SS x  14.40, SS y  12.90  SST, S xy  9.60 2 2  Y  nY  b1  XY  nXY  SS y  b1S xy 12.90  0.66679.60 s 2       0.8125 e n  2 n  2 8</p><p>4. The Variance of b0 and b1 . 1 X 2   1  s 2  s 2  s 2  s 2 b0 e   and b1 e   n SS x   SS x </p><p>I. LINEAR REGRESSION-Confidence Intervals and Tests</p><p>1. Confidence Intervals for b1 .</p><p>  b  t s df  n  2 n 1 1 2 b1 The interval can be made smaller by increasing either or the amount of variation in x . 3</p><p>2. Tests for b1 .</p><p>H 0 : 1  10 b1  10 To test use t  . Remember  is most often zero – and if the null hypothesis is  s 10 H1 : 1  10 b1 false in that case we say that 1 is significant.</p><p>2 2 S xy 9.6 To continue the example in G3: R 2    .4961 or SS x SS y 14.4012.90</p><p>2 2 b0 Y  b1 XY  nY 0.833319 0.666740101.90 R 2      .4961 SS y 12.90 SSR  b S  0.6667 9.60  6.400. 2 1 xy   We have already computed se  0.8125 , which implies that  1  0.8125 s 2  s 2   0.0564 and s  0.0564  0.2374 . b1 e   b1  SS x  14.40</p><p>H 0 : 1  0 The significance test is now  df  n  2  10  2  8 . Assume that   .05 , so that for a 2- H1 : 1  0 n2 8 sided test t  t  2.306 and we reject the null hypothesis if t is below –2.306 or above 2.306. Since 2 .025</p><p> b1  0 0.6667 t    2.809 is in the rejection region, we say that  is significant. A further test says that s 0.2374 1 b1</p><p>1 is not significantly different from 1.</p><p>  b  t s  0.6667  2.3060.2374  0.667  0.547 If we want a confidence interval 1 1 2 b1 . Note that this includes 1, but not zero. X 2  nX 2  1  s 2 2 SS x  2 2   e Note that since s   , sb  se  . This indicates that x 1  X 2  nX 2  n 1 s 2 n 1 n 1     x n x s 2 both the a large sample size, , and a large variance of will tend to make b1 smaller and thus decrease the size of a confidence interval for 1 or increase the size (and significance) of the t-ratio. To put it more negatively, small amounts of variation in x or small sample sizes will tend to produce values of b1 that are not significant. The common sense interpretation of this statement is that we need a lot of experience with what happens to y when we vary x to be able to put any confidence in our estimate of the slope of the equation that relates them. 4</p><p>3. Confidence Intervals and Tests for b0</p><p>H 0 :  0   00 b0   00 We are now testing with t  .  s H1 :  0   00 b0 1 X 2   1 1.602  s 2  s 2   0.8125   0.81250.2778  0.2256 . So s  0.2256  0.4749 . If b0 e     b0 n SS x  10 14.40 </p><p>H 0 :  0  0 b0  0 0.8333 we are testing t    1.754 . Since the rejection region is the same as in I2,  s 0.4749 H1 :  0  0 b0 we accept the null hypothesis and say that  0 is not significant. A confidence interval would be</p><p>  b  t s  0.8333 2.3060.4749  0.883 1.095 0 0 2 b0 Yˆ  0.8333  0.6667X A common way to summarize our results is, . The equation is written with the 0.4749 0.2374 standard deviations below the equation. For a Minitab printout example of a simple regression problem, see 252regrex1.</p><p>4. Prediction and Confidence Intervals for y  2  ˆ 2 2  1 X 0  X   The Confidence Interval is   Y  ts ˆ , where s ˆ  s  and the Prediction Interval Y0 0 Y Y e  n SS   x  2  1 X  X   is Y  Yˆ  ts , where s 2  s 2   0 1 . In these two formulas, for some specific X , 0 0 Y Y e  n SS  0  x  ˆ Y0  b0  b1 X 0 . For example, assume that X 0  5 so that for the results in G3, 2  X  X   2  ˆ 2 2  1  0    1 5 1.6  Y  0.8333  0.66675  4.168 . Then s ˆ  s   0.8125   0.733 and 0 Y e  n SS  10 14.40   x   </p><p> sYˆ  0.733  0.856 , so that the confidence interval is   Yˆ  ts  4.168  2.3060.856  4.168 1.974 Y0 0 Yˆ . This represents a confidence interval for the average value that Y will take when X  5 . For the same data  2   2  2 2  1 X 0  X   1 5 1.6 s  s  1  0.8125  1  1.545 and s  1.545  1.243 , so that the Y e  n SS  10 14.40  Y  x    ˆ prediction interval is Y0  Y0  tsY  4.168  2.3061.243  4.168  2.866 . This is a confidence interval for the value that Y will take in a particular instance when X  5 .</p><p>Ignore the remainder of this document unless you have had calculus! 5</p><p>Appendix to G2– Explanation of OLS Formula Assume that we have three points: X1,Y1 ,X 2 ,Y2  and X 3 ,Y3 . We wish to fit a regression line to these ˆ ˆ 2 points, with the equation Y  b0  b1 X and the characteristic that the sum of squares, SS  Y  Y  is a minimum. If we imagine that there is a 'true' regression line Y   0  1 X   we can consider b0 and b1 to be estimates of  0 and 1 . Let us make the definition e  Y  Yˆ . Note that if we substitute our equation for Yˆ , we find that ˆ e  Y  Y  Y  b0  b1 X , or Y  b0  b1 X  e . This has two consequences: First the sum of squares can be 2 written as ˆ 2 2 ; and second, that if we fit the line so that SS  Y  Y   Y  b0  b1 X   e e  0 or the mean of Y and Yˆ is the same we have Y  a  bX . Now if we subtract the equation for Y</p><p>Y  b0  b1 X  e from the equation for Y we find Y  b0  b1 X . Now let us measure X and Y as deviations </p><p>Y  Y  b1X  X   e ~ ~ ~ ~ from the mean, replacing X with X  X  X and Y with Y  Y Y . This means that Y  b1 X  e or ~ ~ e  Y  b1 X . If we substitute this expression in our sum of squares, we find that ~ ~ 2 2 . SS  e  Y  b1 X  Now write this expression out in terms of our three points and differentiate it to minimize SS with respect to b1 . To do this, recall that b1 is our unknown and that the X s and Y s are numbers (constants!), ~~ ~~ ~ ~ so that d b XY  XY and d b 2 X 2  2b X 2 . db1 1 db1 1 1 2 ~ ~ 2 ~ ~ 2 ~ ~ 2 ~ ~ 2 SS  e  Y  b1 X   Y1  b1 X1   Y2  b1 X 2   Y3  b1 X 3    . ~ 2 ~ ~ 2 ~ 2 ~ 2 ~ ~ 2 ~ 2 ~ 2 ~ ~ 2 ~ 2  (Y1  2b1 X1Y1  b1 X1 )  (Y2  2b1 X 2Y2  b1 X 2 )  (Y3  2b1 X 3Y3  b1 X 3 )</p><p>If we now take a derivative of this expression with respect to b1 and set it equal to zero to find a minimum, we find that: d ~ ~ ~ 2 ~ ~ ~ 2 ~ ~ ~ 2 db SS  0  2X 1Y1  2b1 X 1  0  2X 2Y2  2b1 X 2  0  2X 3Y3  2b1 X 3  1 . ~~ ~ 2 ~~ ~ 2    2XY  2b1 X  2 XY  b1 X  0 ~~ ~ 2 ~~ ~ 2 ~~ ~ 2 ~~ ~ 2 But if  2 XY  b1 X  0 , then XY  b1 X   0 or  XY  b1 X or  XY  b1  X , so ~~ XY  ~ ~ that if we solve for b1 , we find b1  ~ . But if we remember that X  X  X and Y  Y  Y , we can  X 2 X  X Y Y  XY  nXY b   b   . write this as 1 2 or 1  X  X   X 2  nX 2</p><p>Of course, we still need b0 , but remember that Y  b0  b1 X , so that b0  Y  b1 X .</p>

View Full Text

Details

  • File Type
    pdf
  • Upload Time
    -
  • Content Languages
    English
  • Upload User
    Anonymous/Not logged-in
  • File Pages
    5 Page
  • File Size
    -

Download

Channel Download Status
Express Download Enable

Copyright

We respect the copyrights and intellectual property rights of all users. All uploaded documents are either original works of the uploader or authorized works of the rightful owners.

  • Not to be reproduced or distributed without explicit permission.
  • Not used for commercial purposes outside of approved use cases.
  • Not used to infringe on the rights of the original creators.
  • If you believe any content infringes your copyright, please contact us immediately.

Support

For help with questions, suggestions, or problems, please contact us