Presentation is loading. Please wait.

Presentation is loading. Please wait.

1 Statistical Analysis Professor Lynne Stokes Department of Statistical Science Lecture 5 Complete Factorial Experiments, Completely Randomized Designs,

Similar presentations


Presentation on theme: "1 Statistical Analysis Professor Lynne Stokes Department of Statistical Science Lecture 5 Complete Factorial Experiments, Completely Randomized Designs,"— Presentation transcript:

1 1 Statistical Analysis Professor Lynne Stokes Department of Statistical Science Lecture 5 Complete Factorial Experiments, Completely Randomized Designs, Main Effects & Interactions

2 2 Complete Factorial Experiments Does not require that each combination occur an equal number of times Does not state what type of statistical design is being used All combinations of the factor levels appear in the design at least once All combinations of the factor levels appear in the design at least once “Factorial Design” is not a correct statistical term

3 3 Completely Randomized Design Can be any type of complete or fractional factorial experiment Randomize the sequence of test runs, assignment to experimental units Randomize the sequence of test runs, assignment to experimental units

4 4 Randomization Inexpensive Insurance Bias due to changes in uncontrollable Factors or in experimental conditions Bias due to premature termination of the experiment Bias due to machine drift, fatigue, wear Validates key assumptions (Independence, Randomization Distributions) Validates key assumptions (Independence, Randomization Distributions)

5 5 Changes in Experimental Conditions : Oil Viscosity Tests Experiment :Two Oils One Test Run Per Day Six Test Runs Per Week DayWeek #1Week #2 1 56.4 60.6 2 44.7 54.5 3 67.5 62.8 4 76.6 65.2 5 71.3 78.6 6 72.8 80.3 Average 64.9 67.0 t-Test: averages are not significantly different (A) (B)

6 6 Changes in Experimental Conditions : Oil Viscosity Tests DayWeek #1Week #2 1 56.4 65.6 2 44.7 59.5 3 67.5 67.8 4 76.6 70.2 5 71.3 83.6 6 72.8 85.3 Average 64.9 72.0 Average viscosity for Oil B is significantly greater than for Oil A Environmental or equipment change between weeks : add 5 unit bias to all measurements in Week #2 (A) (B)

7 7 Changes in Experimental Conditions : Oil Viscosity Tests DayWeek #1Week #2 1 80.3 67.8 2 67.5 77.8 3 56.4 70.2 4 54.5 76.3 5 76.6 65.6 6 78.6 49.7 Average 67.4 69.5 Randomize the test sequence, 3 oils each week, add 5 unit bias to all measurements in Week #2 (B) (A) (B) (A) (B) (A) (B) (A) (B) (A) Original Biased Randomized Oil B - Oil A 2.1 7.1 2.1

8 8 Premature Termination : Oil Viscosity Tests Oil TypeLubricantViscosity A #154.6 54.3 #244.7 47.0 #367.5 67.2 #476.6 77.4 B #160.6 #2 #3 #4 Equipment Failure

9 9 Background Noise Time Test Runs ResponseDrift Figure 4.7 Influence of machine drift; test runs indicated by arrows. 50 Gallon Drum of Chemicals

10 10 Torque Study Goal :Investigate the effects of three factors on torque forces on rotating shafts FactorLevels Shaft AlloySteel, Aluminum Sleeve MetalPorous, Nonporous Lubricant TypeLub 1, Lub 2, Lub 3, Lub 4 Stationary SleeveRotating Shaft MGH Figure 5.1 Lay Out the Design Lay Out the Design

11 11 Construction of Completely Randomized Designs List the factor-level combinations All combinations, if complete factorial; only those to be tested, if a fractional factorial Include repeat tests, if any Number the combinations (including repeats) from 1 to N Obtain one or more random number sequences of numbers from 1 to N Randomize the test run sequence, if testing is performed sequentially (one after another) Randomize the assignment of factor-level combinations to experimental units, if any

12 12 Completely Randomized Design for Torque Study Random Number Sequence : 8, 13, 4, 7, 5, 1, 11, 15 9, 3, 12, 10, 6, 14, 16, 2 MGH Table 5.1

13 13 Completely Randomized Design for Torque Study Repeats : Same Procedure MGH Table 5.2

14 14 Completely Randomized Design for Torque Study Randomly Selected Repeat Tests cf. Table 5.3 CRD with 2 Repeats

15 15 Lubricant Deposit Study What are the Main Factor Effects ?

16 16 Notation (One-Factor Experiment) 123k... Factor Level Average Data: y ij i = Factor Level j = Repeat Overall Average

17 17 Factor Main Effects Factor Level12...k Mean  1  2...  k Average... Are the means different ? Are the averages significantly different ? (Main Effects) Change in the average response due to changes in levels of one factor Change in the average response due to changes in levels of one factor

18 18 Factor Main Effects Fixed Effects : Constant (mean) changes (Pre-selected levels, systematic changes) Random Effects : Random changes (Standard Deviation > 0) (Effects sampled from a probability distribution)

19 19 Factor Main Effects Models One factor model Averages Assumption E(e ij ) = 0 Conventions  i =  +  i   i = 0 Assumption E(e ij ) = 0 Conventions  i =  +  i   i = 0 Cell Means Model Effects Model

20 20 Factor Main Effects Models Main Effects Note:  i =  j  i =  j Main Effects Theoretical: changes in factor-level means Empirical : changes in factor-level averages Note:  i =   i = 0

21 21 Pilot Plant Experiment 160180 Temperature 20 40 Concentration C1 C2 Catalyst MGH Figure 5.3 6072 52 5468 83 4580

22 22 Pilot Plant Experiment 160180 Temperature 20 40 Concentration C1 C2 Catalyst 6072 52 5468 83 4580

23 23 Pilot Plant Experiment 160180 Temperature 20 40 Concentration C1 C2 Catalyst 6072 52 5468 83 4580

24 24 Pilot Plant Experiment 160180 Temperature 20 40 Concentration C1 C2 Catalyst 6072 52 5468 83 4580

25 25 Pilot Plant Experiment Main Effects Main effects do not measure joint factor effects Main effects are averaged across levels of the other factors Main effects do not measure joint factor effects Main effects are averaged across levels of the other factors Change in the average response due to changes in levels of one factor Change in the average response due to changes in levels of one factor (High – Low)

26 26 Interactions Effects of the levels of one factor on the response depend on the levels of one or more other factors

27 27 No Interaction Response Factor #2 Factor #1 Change in average response for factor #1 is constant for all levels of factor #2 Level #1 Level #2 Change

28 28 Pilot Plant Experiment 160180 Temperature 20 40 Concentration C1 C2 Catalyst 6072 52 5468 83 4580 Change with concentration roughly the same for each catalyst 66 61 67.5 62.5

29 29 Pilot Plant Experiment 160180 Temperature 20 40 Concentration C1 C2 Catalyst 6072 52 5468 83 4580 Change with temperature greater for catalyst C2 than for catalyst C1 5770 48.5 81.5

30 30

31 31 Pilot Plant Experiment Interaction Plot 50 55 60 65 70 75 Average Yield (%) 2040 Concentration (%) MGH Figure 5.4 Catalyst C1 Catalyst C2 No Interaction

32 32 Pilot Plant Experiment Interaction Plot 50 55 60 65 70 75 Average Yield (%) 160180 Temperature (deg C) MGH Figure 5.4 80 Catalyst C1 Catalyst C2 Interaction

33 33 Cutting Tool Life 5006007008009001000 Lathe Speed (rpm) 0 10 20 30 40 50 Tool Life (hrs) Tool Type A Tool Type B 15 Hours } No Interaction MGH Figure 5.2

34 34 Plasticity Study Purpose : Study “Plastic-Like” Properties of Friction-Reducing Lubricants and Additives Stationary Platform Moveable Test Sample Lubricant + Additive Responses: Plastic Viscosity Gel

35 35 Plasticity Experiment Design Factors Lubricant Acme XLT, Monarch1, Standard Additive None, 1%, 5% Design Factorial Experiment 2 Repeat Tests / Combination Completely Randomized Design

36 36 Factor Effects on Plastic Viscosity Effects of Lubricants Acme XLT - Standard : 6.22 - 4.81 = 1.41 Monarch 1 - Standard : 4.79 - 4.81 = -0.02 Main Effects for Lubricants Effects of Additives 1% - None : 4.04 - 5.39 = -1.35 5% - None : 6.39 - 5.39 = 1.00 Main Effects for Additives

37 37 Factor Effects on Plastic Viscosity None1 %5 %Additive Amount Average Plastic Viscosity 0 2 4 6 8Acme XLT Monarch1 Standard Primary Conclusion Acme XLT has a greater mean plastic viscosity, especially with 5% concentration Primary Conclusion Acme XLT has a greater mean plastic viscosity, especially with 5% concentration If Verified by statistical analysis Weak, if any, interaction effects

38 38 Factor Effects on Gel Effects of Lubricants Acme XLT - Standard : 11.45 - 10.01 = 1.44 Monarch 1 - Standard : 10.10 - 10.01 = 0.09 Main Effects for Lubricants Effects of Additives 1% - None : 8.49 - 15.70 = -7.21 5% - None : 7.37 - 15.70 = -8.33 Main Effects for Additives

39 39 Factor Effects on Gel None1 %5 %Additive Amount Average Gel 0 5 10 15 20 Acme XLT Monarch1 Standard Primary Conclusion Acme XLT has a lower mean gel with no additive, greater with additives Primary Conclusion Acme XLT has a lower mean gel with no additive, greater with additives If verified by statistical analysis Strong Interaction Effects

40 40 Completely Randomized Design for Torque Study

41 41 Interaction Effects : Torque Study 60 65 70 75 80 85 Average Torque (in-oz) Lubricant Type 1234 Porous Sleeve Aluminum Shaft

42 42 Interaction Effects : Torque Study 60 65 70 75 80 85 Average Torque (in-oz) Lubricant Type 1234 Porous Sleeve Nonporous Sleeve Aluminum Shaft

43 43 Interaction Effects : Torque Study 60 65 70 75 80 85 Average Torque (in-oz) Lubricant Type 1234 Porous Sleeve Nonporous Sleeve Aluminum Shaft 60 65 70 75 80 85 Lubricant Type 1234 Porous Sleeve Nonporous Sleeve Average Torque (in-oz) Steel Shaft

44 44 Interaction Effects Interaction effects cannot be properly evaluated if the design does not permit their estimation Interaction effects cannot be properly evaluated if the design does not permit their estimation Complete factorials permit the evaluation of all main effects and all interaction effects


Download ppt "1 Statistical Analysis Professor Lynne Stokes Department of Statistical Science Lecture 5 Complete Factorial Experiments, Completely Randomized Designs,"

Similar presentations


Ads by Google