HW2 2.6 Parameter Estimates Variable DF Parameter Standard t Value Pr > |t| 95% Confidence Limits Estimate Error Intercept 1 10.20000 0.66332 15.38 <.0001 8.67037 11.72963 transfer 1 4.00000 0.46904 8.53 <.0001 2.91839 5.08161 This table can help you do the data analysis. a. t(0.975; 8) = 2.306, b1 = 4.0, s{b1} = 0.469, so we get 4±2.306(0.469). Then we get the interval [2.918,5.082] b. H0: β1 = 0, Ha: β1 ≠0. t* = (4.0-0)/0.469=8.529. If |t*|≤2.306 conclude H0, otherwise Ha. Conclude Ha. P-value=0.00003 c. b0 = 10.20, s{b0} = 0.663, so we get 10.20±2.306(0.663), the interval is 8.671≤β0≤11.729 d. H0: β0 ≤ 9, Ha: β0 ≥ 9. t*= (10.20-9)/0.663 = 1.810. If t*≤ 2.306 conclude H0, otherwise Ha. Conclude H0. Pvalue=0.0053 e. H0: β1 = 0: δ= |2-0|/0.5 = 4, power =0.93 (or 0.94) H0: β0 ≤ 9: δ= |11- 9|/0.75 = 2.67, power =0.50304 Obs n sigb0 beta0 df alpha 1 10 0.75 11 delta t_c power 8 0.025 2.66667 2.75152 0.50304 2.8 a. H0: β1 = 3.0, Ha: β1≠3.0. t*= (3.57-3.0)/0.3470 = 1.643 ;t(.975; 23) = 2.069. If |t*|≤2.069 conclude H0, otherwise Ha. Conclude H0. b. δ=|0.5|/0.35=1.43, power=0.16+((1.43-1)/(2-1))(0.48-0.16)=0.297(or0.3) c. The test in part (a) is based on the slope=3, and the F* in the printout only explain the slope of the model that is fitted to the original dataset. Therefore F* is irrelevant to the test in (a) 2.10 a. Prediction b. Mean response c. Prediction 2.13 Output Statistics Obs ACT Dependent Predicted Std Error 95% CL Mean Variable Value Mean Predict 121 . 28 3.2012 0.0706 Residual 3.0614 3.3410 . a. ̂ ̂ ̂ 𝑌 ℎ = 3.2012, s {𝑌ℎ } = .0706, t(.975; 118) = 1.9803, 3.2012± 1.9803(.0706); 3.0614 ≤E{𝑌ℎ } ≤3.3410 (or use the SAS output above) b. Output Statistics Obs ACT Dependent Predicted Std Error 95% CL Predict Residual Variable Value Mean Predict . 121 28 3.2012 0.0706 1.9594 4.4431 b. s{pred} = .6271, 3.2012 ±1.9803(.6271), 1.9594 ≤ 𝑌ℎ(𝑛𝑒𝑤) ≤4.4430 (or get it directly from the table above) c. Yes, yes d. Obs n alpha dfn dfd 1 120 0.05 t w1 w alphat t_c 2 118 1.98027 6.14618 2.47915 0.014582 2.47915 W2 = 2F(.90; 2, 118) = 2(3.0731) = 6.1462, W = 2.4792, 3.2012±2.4792(0.07060), 3.0262 ≤β0 + β1Xh ≤ 3.3762, yes, yes 2.17 Greater, H0: β1 = 0 2.21 No, because R-sqr is only a relative reduction from SSTO and provides no information about absolute precision for estimation a mean response or prediction a new observation. (KNNL 76) 2.25 a. Analysis of Variance Source DF Sum of Squares Mean F Value Pr > F Square Model 1 160.00000 160.00000 Error 8 Corrected Total 9 177.60000 17.60000 72.73 <.0001 2.20000 SSM+SSE=SST, DFm+DFe=DFt, DFm+DFe+1=number of the total observations b. H0: β1 = 0, Ha: β1 ≠ 0 F* = 160.00/2.20 = 72.727, F(.95; 1, 8) = 5.32. If F* ≤ 5.32 conclude H0, otherwise Ha. Conclude Ha. c. t* = (4.00 - 0)/.469 = 8.529, (𝑡 ∗)2 = (8.529)*(8.529) = 72.7 = F* (You also can use the t* shows in the table below, t*=8.53) Parameter Estimates Variable DF Parameter Standard t Value Pr > |t| 95% Confidence Limits Estimate Error Intercept 1 10.20000 0.66332 15.38 <.0001 8.67037 11.72963 transfer 1 4.00000 0.46904 8.53 <.0001 2.91839 5.08161 d. 1.48324 R-Square 0.9009 Root MSE Dependent Mean 14.20000 Adj R-Sq 0.8885 10.44535 Coeff Var R-square = .9009, r = .9492, 90.09% 2.28 a. Output Statistics Obs age Dependent Predicted Std Error 95% CL Mean Residual Variable Value Mean Predict 39 79.0000 60 84.9468 1.0552 82.8347 87.0590 -5.9468 ̂ ̂ 𝑌 ℎ = 84.9468, s {𝑌ℎ } = 1.05515, t (.975 ; 58) = 2.00172, 84.9468 ± 2.00172(1.05515), 82.835≤E{𝑌̂ ℎ } ≤ 87.059 (also can get from the SAS output above) b. Output Statistics Obs age Dependent Predicted Std Error 95% CL Predict Residual Variable Value Mean Predict 61 60 . 84.9468 1.0552 68.4507 101.4430 . S{Yh(new)} = 8.24101, 84.9468 ± 2.00172(8.24101), 68.451 ≤ Yh(new) ≤101.443 (also can get from the SAS output above) c. Obs n alpha dfn dfd 1 60 0.05 2 t w1 w alphat t_c 58 2.00172 6.31186 2.51234 0.014797 2.00172 W2 = 2F(.95; 2, 58) = 2(3.15593) = 6.31186, W = 2.512342, 84.9468 ± 2.512342(1.05515), 82.296 ≤β0 + β1Xh ≤87.598, yes, yes SAS CODE 2.6/2.25 data new; input ampule transfer; datalines; (HERE IS THE DATASET) ; run; proc reg data=new; model ampule=transfer/clb p r; output out=new1 p=pred r=resid; run; (CALCULATE THE POWER) data new2; n=10; sigb0=0.75; beta0=11; df=n-2;alpha=0.025; delta=(beta0-9)/sigb0; t_c=tinv(1-alpha/2,df); power=1-probt(t_c, df,delta)+probt(-t_c,df,delta); output; proc print data=new2; run; 2.13 data new; input GPA ACT; datalines; (here is the dataset) ; //add a new observation data new1; ACT=28;output; data new; set new new1; run; //find the 95% mean inverval proc reg data=new; model GPA=ACT/clm; id ACT; run; //find the 95% predict interval proc reg data=new; model GPA=ACT/cli; id ACT; run; //find the bound data new; n=120; alpha=0.05; dfn=2; dfd=n-2; t=tinv(1-alpha/2,dfd); w1=2*finv(1-alpha,dfn,dfd); w=sqrt(w1); alphat=2*(1-probt(w,dfd)); t_c=tinv(1-alphat/2,dfd); output; proc print data=new; run; 2.28 data new; input mass age; datalines; (HERE IS THE DATASET) ; data new1; age=60; output; data new; set new new1; proc reg data=new; model mass=age/clm; id age; run; proc reg data=new; model mass=age/cli; id age; run; data new; n=60; alpha=0.05; dfn=2;dfd=n-2; t=tinv(1-alpha/2,dfd); w1=2*finv(1-alpha,dfn,dfd); w=sqrt(w1); alphat=2*(1-probt(w,dfd)); t_c=tinv(1-alpha/2,dfd); output; proc print data=new; run;