/**************************************************** Chapter 3: Ratpup Analysis *****************************************************/ title "RatPup data"; data ratpup; infile "c:\temp\rat_pup.dat" firstobs = 2 dlm = "09"X; input pup_id weight sex $ litter litsize treatment $; if treatment = "High" then treat = 1; if treatment = "Low" then treat = 2; if treatment = "Control" then treat = 3; run; proc format ; value trtfmt 1 = "High" 2 = "Low" 3 = "Control"; run; proc means data = ratpup maxdec = 2; class treat sex; var weight; format treat trtfmt.; run; /* Boxplots for each level of treat and sex */ ods listing style=journal2; proc sgpanel data = ratpup; panelby treat / novarname noheaderborder columns = 3; vbox weight / category = sex; format treat trtfmt.; run; /* Generate RANKLIT variable to use for plots */ proc sort data=ratpup; by litsize litter; run; title "Ratpup2 data"; data ratpup2; set ratpup; by litsize litter; if first.litter then ranklit+1; label ranklit = "New Litter ID (Ordered by Size)"; run; /* Box plots of birth weights by treatment level and sex, ordered by litter size (RANKLIT) */ title; proc sgpanel data = ratpup2; panelby sex treat / novarname noheaderborder columns = 3 spacing=20; vbox weight / category = ranklit meanattrs=(size=6); format treat trtfmt.; colaxis display = (novalues noticks); run; /*************************************************** Model 3.1 ****************************************************/ title "Model 3.1"; proc glimmix data = ratpup order = internal method = RSPL; format treat trtfmt.; class treat sex litter; model weight = treat sex treat*sex litsize / solution ddfm = sat dist = normal; random int / subject=litter; covtest glm; run; /*************************************************** Model 3.2A ****************************************************/ title "Model 3.2A: Heterogeneous Residual Variances"; proc glimmix data = ratpup order = internal method=RSPL; class treat sex litter; model weight = treat sex treat*sex litsize / solution ddfm = sat dist = normal; random int / subject = litter; random residual / group = treat; covtest "Hypothesis 3.2" homogeneity / est; covtest "Hypothesis 3.3" general 0 1 -1 0 / est; format treat trtfmt.; run; /*************************************************** Model 3.2B ****************************************************/ title "RATPUP3 dataset"; data ratpup3; set ratpup2; if treatment in ("High", "Low") then trtgrp = 1; if treatment = "Control" then trtgrp = 2; run; proc format; value tgrpfmt 1 = "High/Low" 2 = "Control"; run; title "Model 3.2B"; proc glimmix data = ratpup3 order = internal method = RSPL; format treat trtfmt. trtgrp tgrpfmt.; class treat sex litter trtgrp; model weight = treat sex treat*sex litsize / solution ddfm = sat dist = normal; random int / subject = litter; random residual / group = trtgrp; covtest "Hypothesis 3.4" general 0 1 -1 / est; run; /*************************************************** Model 3.3 ML ****************************************************/ title "Model 3.3 using ML"; proc glimmix data = ratpup3 order = internal method = MSPL; format treat trtfmt. trtgrp tgrpfmt.; class treat sex litter trtgrp; model weight = treat sex litsize / solution ddfm = sat dist = normal; random int / subject = litter; random residual / group = trtgrp; run; /*************************************************** Model 3.3A ML Remove Fixed Treatment Effect. ****************************************************/ title "Model 3.3A using ML"; proc glimmix data = ratpup3 order = internal method = MSPL; format treat trtfmt. trtgrp tgrpfmt.; class treat sex litter trtgrp; model weight = sex litsize / solution ddfm = sat dist = normal; random int / subject = litter; random residual / group = trtgrp; run; title "P-value for Hypothesis 3.6"; data test; lrstat = 356.37 - 337.77; df = 2; pvalue = 1-probchi(lrstat,df); format pvalue 10.8; run; proc print data=test noobs; run; /*************************************************** Final Model 3.3 with diagnostics ****************************************************/ proc sort data=ratpup3; by litter; run; title; ods graphics on; ods rtf file = "c:\temp\ratpup_diagnostics.rtf" style = journal; proc glimmix data = ratpup3 order = internal method = RSPL plots=(residualpanel(blup) studentpanel(blup) boxplot(student blup)); format treat trtfmt. trtgrp tgrpfmt.; class treat sex litter trtgrp; model weight = treat sex litsize / solution ddfm = sat ; random int / subject = litter solution v=(3, 23) vcorr=(3,23) ; random residual / group = trtgrp ; lsmeans treat / adjust = tukey; output out = pdat pred(blup)=p_cond resid(blup)=r_cond student(blup)=s_cond pred(noblup)=p_marg resid(noblup)=r_marg student(noblup)=s_marg; ods output solutionR = eblupsdat; run; ods rtf close; ods graphics off; /* Model 3.3: Diagnostic Plots */ ods graphics on; ods listing style = journal2; title; /* Histograms of conditional raw residuals by treatment group */ proc sgpanel data=pdat; panelby trtgrp / novarname rows = 2; histogram r_cond; format trtgrp tgrpfmt.; run; /* Normal Q-Q plots of conditional raw residuals by treatment group */ proc univariate data=pdat normal; class trtgrp; var r_cond; qqplot / normal(mu=est sigma=est); format trtgrp tgrpfmt.; run; /* Plot of conditional raw residuals vs. predicted values */ proc sgpanel data = pdat; panelby trtgrp / novarname columns = 2 noheaderborder; scatter y=r_cond x=p_cond; format trtgrp tgrpfmt.; run; /* Boxplots of Studentized Residuals by Litter */ proc sgplot data = pdat ; vbox s_cond / category = litter; run; /* Histogram and Q-Q plot of Eblups */ proc univariate data = eblupsdat normal; var estimate; histogram; qqplot / normal(mu = est sigma = est); run; ods graphics off; ods rtf close; /**************************************************** Model 3.3 using Proc Mixed: To get Influence Diagnostics *****************************************************/ proc sort data = ratpup3; by pup_id; run; title; ods graphics on; ods exclude influence; proc mixed data = ratpup3 order = internal covtest plots =(studentpanel residualpanel boxplot influencestatpanel); class treat sex litter trtgrp; model weight = treat sex litsize / solution ddfm = sat outpred = pdat1 outpredm = pdat2 influence(iter = 5 effect = litter est) residual; random int / subject = litter solution v = 3,23 vcorr = 3,23; repeated / group = trtgrp; lsmeans treat / adjust = tukey; format treat trtfmt. trtgrp tgrpfmt.; run;