Lecture 36: Three Factors - CRD Experiments Experiments The rest of the semester will look at different experiments using different combinations of factorial crossing and designs. Factorial Crossing Design CRD, RCBD, RLSq, Split Plot/Repeated Measures. 1 Agoraphobia 2 Agoraphobia Factor 1: Drug (4 levels) Factor 2: Psychotherapy (3 levels) Factor 3: Depression (2 levels) Response: Severity of agoraphobia (fear of public places). A higher score indicates more severe symptoms. 3 4 Comment Comment Drug and Psychotherapy are experimental factors that can be assigned at random to patients suffering from agoraphobia. Depression is an observational factor that is associated with a patient suffering from agoraphobia. 5 6 1 Lecture 36: Three Factors - CRD Treatments Randomization Select three depressed patients at random and assign one of the Drug, Psychotherapy combinations. Select three non-depressed patients at random and assign one of the Drug, Psychotherapy combinations. Do this until all 12 combinations are assigned to 3 depressed and 3 nondepressed patients. There are 12 combinations of Drug and Psychotherapy that can be assigned to depressed and non-depressed patients. There are 24 combinations of Depression, Drug and Psychotherapy. 7 JMP Table Drug Prozac Prozac Prozac Elavil Therapy Psychodynamic Psychodynamic Psychodynamic Psychodynamic Placebo Group JMP – Fit Model Y: Severity Construct Model Effects Depressed Severity No 11 No 12 No 7 No 13 Yes 8 Highlight Drug, Therapy, Depression Macros – Full Factorial 13 9 10 Response Severity Construct Model Effects Drug Therapy Drug*Therapy Depression Drug*Depression Therapy*Depression Drug*Therapy*Depression Summary of Fit RSquare 0.88434 RSquare Adj 0.82892 Root Mean Square Error 1.658312 Mean of Response 13.30556 Observations (or Sum Wgts) 72 Analysis of Variance Source Model Error C. Total DF 23 48 71 Sum of F Ratio Squares Mean Square 1009.2778 43.8816 15.9570 132.0000 2.7500 Prob > F 1141.2778 <.0001* Effect Tests Source Nparm Drug 3 Therapy 2 Drug*Therapy 6 Depression 1 Drug*Depression 3 Therapy*Depression 2 Drug*Therapy*Depression 6 11 DF 3 2 6 1 3 2 6 Sum of Squares 486.72222 33.36111 385.52778 40.50000 20.94444 31.08333 11.13889 F Ratio Prob > F <.0001* 58.9966 0.0045* 6.0657 <.0001* 23.3653 0.0004* 14.7273 0.0675 2.5387 0.0063* 5.6515 0.6703 0.6751 12 2 Lecture 36: Three Factors - CRD Drug Effect Drug Effect 2 F = 58.9966, P-value < 0.0001, statistically significant because of the small P-value. 2.66137 2.75 2 18 HSD = 2.66137(0.55277) = 1.47 13 Drug Effect Drug Placebo Xanax Elavil Prozac Therapy Effect Mean 16.333 A 15.278 A 11.667 B 9.944 F = 6/0657, P-value = 0.0045, statistically significant because of the small P-value. C Drugs not connected by the same letter are significantly different 15 Therapy Effect 16 Therapy Effect Therapy Psychodynamic Group Cog/Beh 2 L 14 2.01063 2.75 Mean 14.125 A 13.333 A 12.458 B B Therapies not connected by the same letter are significantly different LSD = 2.01063(0.47871) = 0.96 17 18 3 Lecture 36: Three Factors - CRD Depression Effect Depression Effect Depression No Yes F = 14.7273, P-value = 0.0004, statistically significant because of the small P-value. Mean 14.056 A 12.556 B Because the F statistic indicated a statistically significant difference, the mean for no depression is significantly different from the mean with depression. 19 20 Main (Factor) Effects Drug*Therapy So far we have only looked at the effects of individual factors. What about interactions? F = 23.3653, P-value < 0.0001. The small P-value indicates a statistically significant interaction between Drug and Therapy. 21 22 25 Elavil Placebo Prozac Xanax Severity LS Means 20 Interpretation 15 Prozac has about the same effect for each type of therapy. Elavil has about the same effect (not as good as Prozac) for each type of therapy. 10 5 Cog/Beh Group Psychodynamic Therapy 23 24 4 Lecture 36: Three Factors - CRD Interpretation Interpretation The effect of Xanax is better with Cog/Beh therapy and not so good with Group or Psychodynamic therapy. The effect of Placebo is not very good for Cog/Beh or Psychodynamic but it is effective with Group therapy. 25 26 Therapy*Depression 25 Cog/Beh Group Psychodynamic 20 Severity LS Means F = 5.6515, P-value < 0.0063. The small P-value indicates a statistically significant interaction between Therapy and Depression. 15 10 5 No Yes Depression 27 Interpretation 28 Interaction Effects There is no statistically significant interaction for For depressed patients all therapies give similar results. For non-depressed patients, Cog/Beh has lowest average severity followed by Group and Psychodynamic. Drug*Depression: F = 2.53878, P-value = 0.0675. Drug*Therapy*Depression: F = 0.6751, P-value = 0.6703. 29 30 5 Lecture 36: Three Factors - CRD Recommendation? Different Analysis? It is hard to make a recommendation based on the factors and interactions especially because of the interaction terms. Analyze the data from the 36 depressed patients separately from the data from the 36 non-depressed patients. 31 Not Depressed Source Not Depressed df Sum of Squares Mean Square F Ratio Prob>F Drug 3 334.778 111.5926 40.1733 <0.0001 Therapy Drug*Therapy 2 6 64.056 176.389 32.0278 29.3981 11.5300 10.5833 0.0003 <0.0001 2.7778 Error 24 66.667 C. Total 35 641.889 32 Elavil and Prozac are equally effective and significantly better than Placebo or Xanax. 33 Not Depressed 34 Not Depressed Cog/Beh is significantly better than Group or Psychodynamic therapies. 35 36 6 Lecture 36: Three Factors - CRD Recommendation Depressed For patients who are not depressed. Source You should get reasonably good results using either Prozac or Elavil and any of the Therapies Xanax and Cog/Beh or Placebo and Group are not much different. df Sum of Squares Mean Square F Ratio Prob>F Drug 3 172.889 57.6296 21.1701 <0.0001 Therapy Drug*Therapy 2 6 0.389 220.278 0.1944 36.7130 0.0714 13.4864 0.9313 <0.0001 2.7222 Error 24 65.333 C. Total 35 458.889 37 Depressed 38 Depressed Elavil and Prozac are equally effective and significantly better than Placebo or Xanax. There is no statistically significant effect of Therapy. 39 Depressed 40 Recommendation For patients who are depressed. You should get reasonably good results using Prozac and any of the Therapies. Xanax and Cog/Beh or Elavil and Psychodynamic are not much different. 41 42 7