1. 1 Statistical Analysis Professor Lynne Stokes Department of Statistical Science Lecture 6 Solving Normal Equations and Estimating Estimable Model Parameters

2. 2 Regression Models Residuals Least Squares Solution: Solve the Normal Equations Model Sum of Squared Residuals Sum of Squared Residuals

3. 3 Regression Solution Under usual assumptions, the least squares estimator is Unique Unbiased Minimum Variance Consistent Known sampling distribution Universally used

4. 4 Analysis of Completely Randomized Designs Fixed Factor Effects Factor levels specifically chosen Inferences desired only on the factor levels included in the experiment Systematic, repeatable changes in the mean response

5. 5 Flow Rate Experiment MGH Fig 6.1 Fixed or Random ?

6. 6 Flow Rate Experiment ABCD 0.30 0.25 0.20 Average Flow Rate Conclusion ? Filter Type 0.35

7. 7 Statistical Model for Single-Factor, Fixed Effects Experiments ResponseOverall Mean (Constant) Main Effect for Level i Error Model y ij =  +  i + e ij i = 1,..., a; j = 1,..., r i  i : Effect of Level i = change in the mean response

8. 8 Statistical Model for Single-Factor, Fixed Effects Experiments Cell Means Model y ij =  i + e ij i = 1,..., a; j = 1,..., r i Effects Model y ij =  +  i + e ij i = 1,..., a; j = 1,..., r i Fixed Effects Models Connection:  i =  +  i Fixed Effects Models Connection:  i =  +  i

9. 9 Solving the Normal Equations Single-Factor, Balanced Experiment y ij =  +  i + e ij i = 1,..., a j = 1,..., r n = ar Matrix Formulation y = X  + e y = (y 11 y 12... y 1r... y a1 y a2... y ar )’

10. 10 Solving the Normal Equations Residuals Least Squares Solution: Solve the Normal Equations

11. 11 Solving the Normal Equations Normal Equations Check

12. 12 Solving the Normal Equations Normal Equations Check Linearly Dependent a + 1 Parameters, a Linearly Independent Equations Infinite Number of Solutions

13. 13 Solving the Normal Equations Normal Equations One Solution

14. 14 Solving the Normal Equations Normal Equations Another Solution

15. 15 Solving the Normal Equations Another Solution Normal Equations

16. 16 Solving the Normal Equations Solutions are not estimates Estimable Functions All solutions provide one unique estimator Estimators are unbiased All solutions to the normal equations produce the same estimates of “estimable functions” of the model means All solutions to the normal equations produce the same estimates of “estimable functions” of the model means

17. 17 Solving the Normal Equations Two-Factor, Balanced Experiment Matrix Formulation y = X  + e y ijk =  ij + e ijk =  +  i +  j + (  ) ij + e ijk i = 1,..., a j = 1,..., b k = 1,..., r X = [ 1 : X A : X B : X AB ]  1  a  1  b  11  ab  n = abr

18. 18 Solving the Normal Equations Two-Factor, Balanced Experiment Matrix Formulation y = X  + e y ijk =  ij + e ijk =  +  i +  j + (  ) ij + e ijk i = 1,..., a j = 1,..., b k = 1,..., r X = [ 1 : X A : X B : X AB ] Number of Parameters 1 + a + b + ab rank( X ) < 1+a+b+ab n = abr  1  a  1  b  11  ab 

19. 19 Solving the Normal Equations Normal Equations Check

20. 20 Solving the Normal Equations Matrix Linear Dependencies One Solution 1 n None X A 1 : Columns of X A Sum to 1 n  a  = 0 Eliminates a column From X A Eliminates a column From X A a – 1 “degrees of freedom”

21. 21 Solving the Normal Equations Matrix Linear Dependencies One Solution 1 n None X A 1 : Columns Sum of X A to 1 n  a  = 0 X B 1 : Columns Sum of X B to 1 n  b = 0 Eliminates a column From X B Eliminates a column From X B b – 1 “degrees of freedom”

22. 22 Solving the Normal Equations Matrix Linear Dependencies One Solution 1 n None X A 1 : Columns sum to 1 n  a  = 0 X B 1 : Columns sum to 1 n  b = 0 X AB 1 + (a - 1) + (b - 1) : Sum over all columns = 1 n (  ) ab = 0 Eliminates a column from X AB Eliminates a column from X AB

23. 23 Solving the Normal Equations Matrix Linear Dependencies One Solution 1 n None X A 1 : Columns Sum to 1 n  a  = 0 X B 1 : Columns Sum to 1 n  b = 0 X AB 1 + (a - 1) + (b - 1) : Sum over all columns = 1 n (  ) ab = 0 Sums of columns over each i = 1,...,a-1 & each j = 1,...,b-1 (  ) ib = 0 equal one of the remaining i=1,...,a-1 columns of X A and X B (  ) aj = 0 j=1,...,b-1 (a – 1)(b – 1) “degrees of freedom”

24. 24 Solving the Normal Equations Matrix Linear Dependencies One Solution X A 1 : Columns sum to 1 n  a  = 0 X B 1 : Columns sum to 1 n  b = 0 X AB 1 + (a - 1) + (b - 1) : Sum over all columns = 1 n (  ) ab = 0 Sums of columns over each i = 1,...,a-1 & each j = 1,...,b-1 (  ) ib = 0 equal one of the remaining i=1,...,a-1 columns of X A and X B (  ) aj = 0 j=1,...,b-1 Constraints : 1 + 1 + {1 + (a - 1) + (b - 1)} = a + b + 1 Degrees of Freedom : (1 + a + b + ab) - (a + b + 1) = ab = 1 + (a - 1) + (b - 1) + (a - 1)(b - 1)

25. 25 Solving the Normal Equations Check

26. 26 Solving the Normal Equations Check Another Solution

27. 27 Flow Rate Experiment MGH Fig 6.1 Fixed or Random ?

28. 28 Quantifying Factor Effects Effect Change in average response due to changes in factor levels 123k... Factor Level Average Overall Average Effect of Level t : -

29. 29 Quantifying Factor Effects Effect Change in average response due to changes in factor levels 123k... Factor Level Average Overall Average Effect of changing from Level s to Level t :...

30. 30 Quantifying Factor Effects Main Effects for Factor A Change in average response due to changes in the levels of Factor A Main Effects for Factor B Change in average response due to changes in the levels of Factor B Interaction Effects for Factors A & B Effect of Level i of Factor A at Level j of Factor B Effect of Level i of Factor A

31. 31 Quantifying Factor Effects Main Effects for Factor A Main Effects for Factor B Interaction Effects for Factors A & B Change in average response due to changes in the levels of Factor A Change in average response due to changes in the levels of Factor B Change in average response due joint changes in Factors A & B in excess of changes in the main effects

32. 32 Two-Level Factors Common to Use Note: If r 1 = r 2, Effect of Level 1: Effect of Level 2:

33. 33 Factors at Two Levels Most common choice for designs involving many factors Many efficient fractional factorial and screening designs available Can use p two-level factors in place of factors whose number of levels is 2 p

34. 34 Calculating Two-Level Factor Effects: Pilot Plant Study Main Effect Difference between the average responses at the two levels M(Temp) = Average @ 180 o - Average @ 160 o = 75.8 - 52.8 = 23.0 M(Conc) = Average @ 40% - Average @ 20% = 61.8 - 66.8 = -5.0 M(Catalyst) = Average @ C 2 - Average @ C 1 = 65.0 - 63.5 = 1.5 BHH Section 10.3 MGH Section 5.3

35. 35 Calculating Two-Level Factor Effects Two-Factor Interaction Effect Half the difference between the main effects of one factor at each level of the second factor M(Conc @ C 2 )= Average @ 40%&C 2 - Average @ 20%&C 2 = 62.5 - 67.5 = -5.0 M(Conc @ C 1 ) = Average @ 40%&C 1 - Average @ 20%&C 1 = 61.0 - 66.0 = -5.0 I(Conc,Cat)= {M(Conc @ C 2 ) - M(Conc @ C 1 )} / 2 = 0 BHH Section 10.4 MGH Section 5.3

36. 36 Calculating Two-Level Factor Effects Two-Factor Interaction Effect Half the difference between the main effects of one factor at each level of the second factor M(Temp @ C 2 )= Average @ 180 o &C 2 - Average @ 160 o &C 2 = 81.5 - 48.5 = 33.0 M(Temp @ C 1 ) = Average @ 180 o &C 1 - Average @ 160 o &C 1 = 70.0 - 57.0 = 13.0 I(Temp,Cat)= {M(Temp @ C 2 ) - M(Temp @ C 1 )} / 2 = (33.0 - 13.0) / 2 = 10.0

37. 37 Cell Means and Effects Model Estimability Three-Factor Balanced Experiment y ijkl =  ijk + e ijkl i = 1,..., a ; j = 1,..., b ; k = 1,..., c ; l = 1,..., r  ijk =  +  i +  j +  k + (  ) ij + (  ) ik + (  ) jk + (  ) ijk

38. 38 Cell Means Models: Estimable Functions All cell means are estimable

39. 39 Cell Means Models: Estimable Functions All cell means are estimable All linear combinations of cell means are estimable Does not depend on parameter constraints Does not depend on parameter constraints

40. 40 Cell Means Models: Estimable Functions All cell means are estimable Some linear combinations of cell means are uninterpretable Some linear combinations of cell means are essential

41. 41 Cell Means and Effects Models Imposing parameter constraints simplifies the relationships; makes the parameters more interpretable

42. 42 Parameter Equivalence: Effects Representation & Cell Means Model Parameter constraints Means and mean effects

43. 43 Contrasts Contrasts often eliminate nuisance parameters; e.g., 

44. 44 Contrasts Main Effects Interactions Show