1 The Two-Factor Mixed Model Two factors, factorial experiment, factor A fixed, factor B random (Section 13-3, pg. 495) The model parameters are NID random variables, the interaction effect is normal, but not independent This is called the restricted model

2 Testing Hypotheses - Mixed Model Once again, the standard ANOVA partition is appropriate Relevant hypotheses: Test statistics depend on the expected mean squares: H o is rejected if F o > F ,a-1,(a-1)(b-1) F o > F ,b-1,ab(n-1) F o > F ,(a-1)(b-1), ab(n-1)

3 Estimating the Variance Components – Two Factor Mixed model Use the ANOVA method; equate expected mean squares to their observed values: Estimate the fixed effects (treatment means) as usual

4 Example 13-3 (pg. 497) The Measurement Systems Capability Study Revisited Same experimental setting as in example 13-2 Parts are a random factor, but Operators are fixed Assume the restricted form of the mixed model Minitab can analyze the mixed model The variance components can also be estimated as

5 Example 13-3 (pg. 497) Minitab Solution – Balanced ANOVA Source DF SS MS F P Part Operator Part*Operator Error Total Source Variance Error Expected Mean Square for Each Term component term (using restricted model) 1 Part (4) + 6(1) 2 Operator 3 (4) + 2(3) + 40Q[2] 3 Part*Operator (4) + 2(3) 4 Error (4)

6 Example 13-3 Minitab Solution – Balanced ANOVA There is a large effect of parts (not unexpected) Small operator effect No Part – Operator interaction Negative estimate of the Part – Operator interaction variance component Fit a reduced model with the Part – Operator interaction deleted This leads to the same solution that we found previously for the two-factor random model

7 The Unrestricted Mixed Model Two factors, factorial experiment, factor A fixed, factor B random (pg. 498) The random model parameters are now all assumed to be NID. is no longer assumed – unrestricted model

8 Testing Hypotheses – Unrestricted Mixed Model The standard ANOVA partition is appropriate Relevant hypotheses: Expected mean squares determine the test statistics:

9 Estimating the Variance Components – Unrestricted Mixed Model Use the ANOVA method; equate expected mean squares to their observed values: The only change compared to the restricted mixed model is in the estimate of the random effect variance component Which model to use? oThey are fairly close in many cases oThe restricted model is slightly more general oThe restricted model is mostly preferred

10 Example 13-4 (pg. 499) Minitab Solution – Unrestricted Model Source DF SS MS F P Part Operator Part*Operator Error Total Source Variance Error Expected Mean Square for Each Term component term (using unrestricted model) 1 Part (4) + 2(3) + 6(1) 2 Operator 3 (4) + 2(3) + Q[2] 3 Part*Operator (4) + 2(3) 4 Error (4)

11 Sample Size Determination with Random Effects Consider a single-factor random effects model Power = 1 –  P(Reject H o  H o is false)  P(F o > F ,a-1,N-a  H o is false) F o = MS Treatments /MS E (dofs are needed to determine the OC curve) The operating characteristic curves (Chart VI, Appendix) can be used The curves plot the probability of type II error against the parameter

12 Sample Size Determination with Random Effects – Example 13-5 Five treatments randomly selected (a = 5) Six observations per treatment (n = 6)  = 0.05, a – 1 = 4 (v 1 ), N – a = 25 (v 2 ) Assume that Then   0.20

13 Sample Size Determination with Random Effects Use the percentage increase in the standard deviation of an observation If the treatments are homogeneous,  If the treatments are different, P is the fixed percentage increase in the standard deviation Then

14 Sample Size Determination with Random Effects – Two Factors Table 13-8

15 Finding Expected Mean Squares Obviously important in determining the form of the test statistic In fixed models, it’s easy: Can always use the “brute force” approach – just apply the expectation operator Straightforward but tedious Rules on page [due to Cornfield and Tukey (1956)] work for any balanced model Rules are consistent with the restricted mixed model

16 Approximate F Tests Sometimes we find that there are no exact tests for certain effects

17 Approximate F Tests One possibility: assume that certain interactions are negligible – needs conclusive evidence If we cannot assume that certain interactions are negligible, then use an approximate F test (“pseudo” F test) Test procedure is due to Satterthwaite (1946), and uses linear combinations of the original mean squares to form the F-ratio For example: MS’ = MS r + …+ MS s MS’’ = MS u + …+ MS v The mean squares are chosen so that E(MS’) – E(MS’’) is a multiple of the effect considered in the null hypothesis F is distributed approximately as F p,q

18 Approximate F Tests The linear combinations of the original mean squares are sometimes called “synthetic” mean squares Adjustments are required to the degrees of freedom Refer to Example 13-7, page 505 Minitab will analyze these experiments, although their “synthetic” mean squares are not always the best choice