Model M1a: uses prethk as continuous covariate 1 The GLIMMIX Procedure Model Information Data Set WORK.TV Response Variable thk Response Distribution Multinomial (ordered) Link Function Cumulative Logit Variance Function Default Variance Matrix Blocked By school Estimation Technique Maximum Likelihood Likelihood Approximation Gauss-Hermite Quadrature Degrees of Freedom Method Containment Number of Observations Read 1601 Number of Observations Used 1600 Response Profile Ordered Total Value thk Frequency 1 1 355 2 2 398 3 3 400 4 4 447 The GLIMMIX procedure is modeling the probabilities of levels of thk having lower Ordered Values in the Response Profile table. Dimensions G-side Cov. Parameters 1 Columns in X 20 Columns in Z per Subject 1 Subjects (Blocks in V) 28 Max Obs per Subject 137 Optimization Information Optimization Technique Dual Quasi-Newton Parameters in Optimization 11 Lower Boundaries 1 Upper Boundaries 0 Fixed Effects Not Profiled Starting From GLM estimates Quadrature Points 1 Model M1a: uses prethk as continuous covariate 2 The GLIMMIX Procedure Convergence criterion (GCONV=1E-8) satisfied. Fit Statistics -2 Log Likelihood 4233.05 AIC (smaller is better) 4255.05 AICC (smaller is better) 4255.21 BIC (smaller is better) 4269.70 CAIC (smaller is better) 4280.70 HQIC (smaller is better) 4259.53 Fit Statistics for Conditional Distribution -2 log L(thk | r. effects) 4195.95 Covariance Parameter Estimates Standard Cov Parm Subject Estimate Error Intercept school 0.07520 0.03896 Solutions for Fixed Effects Standard Effect thk cc tv Estimate Error DF t Value Pr > |t| Intercept 1 -1.1232 0.2040 24 -5.51 <.0001 Intercept 2 0.1219 0.2009 24 0.61 0.5496 Intercept 3 1.3025 0.2042 24 6.38 <.0001 cc 0 0.8076 0.2903 1566 2.78 0.0055 cc 1 0 . . . . tv 0 0.2974 0.2928 1566 1.02 0.3099 tv 1 0 . . . . cc*tv 0 0 0.02656 0.4224 1566 0.06 0.9499 cc*tv 0 1 0 . . . . cc*tv 1 0 0 . . . . cc*tv 1 1 0 . . . . prethk -0.2449 0.07446 1566 -3.29 0.0010 prethk*cc 0 -0.1826 0.1048 1566 -1.74 0.0817 prethk*cc 1 0 . . . . prethk*tv 0 -0.2534 0.1073 1566 -2.36 0.0184 prethk*tv 1 0 . . . . prethk*cc*tv 0 0 0.2313 0.1517 1566 1.53 0.1275 prethk*cc*tv 0 1 0 . . . . prethk*cc*tv 1 0 0 . . . . Model M1a: uses prethk as continuous covariate 3 The GLIMMIX Procedure Solutions for Fixed Effects Standard Effect thk cc tv Estimate Error DF t Value Pr > |t| prethk*cc*tv 1 1 0 . . . . Type III Tests of Fixed Effects Num Den Effect DF DF Chi-Square F Value Pr > ChiSq Pr > F cc 1 1566 15.12 15.12 0.0001 0.0001 tv 1 1566 2.18 2.18 0.1399 0.1401 cc*tv 1 1566 0.00 0.00 0.9499 0.9499 prethk 1 1566 108.79 108.79 <.0001 <.0001 prethk*cc 1 1566 0.78 0.78 0.3771 0.3772 prethk*tv 1 1566 3.28 3.28 0.0701 0.0703 prethk*cc*tv 1 1566 2.33 2.33 0.1273 0.1275 Model M1b: uses prethkord as continuous covariate 4 The GLIMMIX Procedure Model Information Data Set WORK.TV Response Variable thk Response Distribution Multinomial (ordered) Link Function Cumulative Logit Variance Function Default Variance Matrix Blocked By school Estimation Technique Maximum Likelihood Likelihood Approximation Gauss-Hermite Quadrature Degrees of Freedom Method Containment Number of Observations Read 1601 Number of Observations Used 1600 Convergence criterion (GCONV=1E-8) satisfied. Fit Statistics -2 Log Likelihood 4321.77 AIC (smaller is better) 4343.77 AICC (smaller is better) 4343.93 BIC (smaller is better) 4358.42 CAIC (smaller is better) 4369.42 HQIC (smaller is better) 4348.25 Fit Statistics for Conditional Distribution -2 log L(thk | r. effects) 4272.40 Covariance Parameter Estimates Standard Cov Parm Subject Estimate Error Intercept school 0.1265 0.05352 Solutions for Fixed Effects Standard Effect thk cc tv Estimate Error DF t Value Pr > |t| Intercept 1 -1.5480 0.2122 24 -7.29 <.0001 Intercept 2 -0.3449 0.2084 24 -1.66 0.1109 Intercept 3 0.7939 0.2092 24 3.80 0.0009 Model M1b: uses prethkord as continuous covariate 5 The GLIMMIX Procedure Solutions for Fixed Effects Standard Effect thk cc tv Estimate Error DF t Value Pr > |t| cc 0 0.5574 0.2939 1566 1.90 0.0580 cc 1 0 . . . . tv 0 0.2848 0.2937 1566 0.97 0.3324 tv 1 0 . . . . cc*tv 0 0 0.1006 0.4153 1566 0.24 0.8087 cc*tv 0 1 0 . . . . cc*tv 1 0 0 . . . . cc*tv 1 1 0 . . . . prethkord 0.005743 0.06190 1566 0.09 0.9261 prethkord*cc 0 -0.09931 0.08609 1566 -1.15 0.2489 prethkord*cc 1 0 . . . . prethkord*tv 0 -0.2783 0.08694 1566 -3.20 0.0014 prethkord*tv 1 0 . . . . prethkord*cc*tv 0 0 0.2226 0.1205 1566 1.85 0.0649 prethkord*cc*tv 0 1 0 . . . . prethkord*cc*tv 1 0 0 . . . . prethkord*cc*tv 1 1 0 . . . . Type III Tests of Fixed Effects Num Den Effect DF DF Chi-Square F Value Pr > ChiSq Pr > F cc 1 1566 8.60 8.60 0.0034 0.0034 tv 1 1566 2.62 2.62 0.1055 0.1057 cc*tv 1 1566 0.06 0.06 0.8087 0.8087 prethkord 1 1566 17.83 17.83 <.0001 <.0001 prethkord*cc 1 1566 0.04 0.04 0.8423 0.8423 prethkord*tv 1 1566 7.67 7.67 0.0056 0.0057 prethkord*cc*tv 1 1566 3.41 3.41 0.0648 0.0649 Model M1c: uses prethkord as factor 6 The GLIMMIX Procedure Model Information Data Set WORK.TV Response Variable thk Response Distribution Multinomial (ordered) Link Function Cumulative Logit Variance Function Default Variance Matrix Blocked By school Estimation Technique Maximum Likelihood Likelihood Approximation Gauss-Hermite Quadrature Degrees of Freedom Method Containment Number of Observations Read 1601 Number of Observations Used 1600 Convergence criterion (GCONV=1E-8) satisfied. Fit Statistics -2 Log Likelihood 4254.48 AIC (smaller is better) 4292.48 AICC (smaller is better) 4292.97 BIC (smaller is better) 4317.80 CAIC (smaller is better) 4336.80 HQIC (smaller is better) 4300.22 Fit Statistics for Conditional Distribution -2 log L(thk | r. effects) 4211.03 Covariance Parameter Estimates Standard Cov Parm Subject Estimate Error Intercept school 0.09994 0.04619 Solutions for Fixed Effects Standard Effect thk cc tv prethkord Estimate Error DF t Value Pr > |t| Intercept 1 -2.0564 0.3139 24 -6.55 <.0001 Intercept 2 -0.8272 0.3110 24 -2.66 0.0137 Intercept 3 0.3463 0.3103 24 1.12 0.2755 Model M1c: uses prethkord as factor 7 The GLIMMIX Procedure Solutions for Fixed Effects Standard Effect thk cc tv prethkord Estimate Error DF t Value Pr > |t| cc 0 0.1050 0.4086 1558 0.26 0.7973 cc 1 0 . . . . tv 0 -1.2457 0.4459 1558 -2.79 0.0053 tv 1 0 . . . . cc*tv 0 0 1.2752 0.5915 1558 2.16 0.0312 cc*tv 0 1 0 . . . . cc*tv 1 0 0 . . . . cc*tv 1 1 0 . . . . prethkord 0 0.1656 0.3310 1558 0.50 0.6169 prethkord 1 1.4965 0.3989 1558 3.75 0.0002 prethkord 3 0.5621 0.3120 1558 1.80 0.0718 prethkord 4 0 . . . . cc*prethkord 0 0 0.5453 0.4417 1558 1.23 0.2172 cc*prethkord 0 1 0.08049 0.5570 1558 0.14 0.8851 cc*prethkord 0 3 0.2748 0.4136 1558 0.66 0.5065 cc*prethkord 0 4 0 . . . . cc*prethkord 1 0 0 . . . . cc*prethkord 1 1 0 . . . . cc*prethkord 1 3 0 . . . . cc*prethkord 1 4 0 . . . . tv*prethkord 0 0 1.6090 0.4740 1558 3.39 0.0007 tv*prethkord 0 1 0.8337 0.5647 1558 1.48 0.1400 tv*prethkord 0 3 0.9091 0.4546 1558 2.00 0.0457 tv*prethkord 0 4 0 . . . . tv*prethkord 1 0 0 . . . . tv*prethkord 1 1 0 . . . . tv*prethkord 1 3 0 . . . . tv*prethkord 1 4 0 . . . . cc*tv*prethkord 0 0 0 -1.1669 0.6303 1558 -1.85 0.0643 cc*tv*prethkord 0 0 1 -0.9877 0.8159 1558 -1.21 0.2262 cc*tv*prethkord 0 0 3 -0.6876 0.6001 1558 -1.15 0.2521 cc*tv*prethkord 0 0 4 0 . . . . cc*tv*prethkord 0 1 0 0 . . . . cc*tv*prethkord 0 1 1 0 . . . . cc*tv*prethkord 0 1 3 0 . . . . cc*tv*prethkord 0 1 4 0 . . . . cc*tv*prethkord 1 0 0 0 . . . . cc*tv*prethkord 1 0 1 0 . . . . cc*tv*prethkord 1 0 3 0 . . . . cc*tv*prethkord 1 0 4 0 . . . . cc*tv*prethkord 1 1 0 0 . . . . cc*tv*prethkord 1 1 1 0 . . . . cc*tv*prethkord 1 1 3 0 . . . . cc*tv*prethkord 1 1 4 0 . . . . Model M1c: uses prethkord as factor 8 The GLIMMIX Procedure Type III Tests of Fixed Effects Num Den Effect DF DF Chi-Square F Value Pr > ChiSq Pr > F cc 1 1558 13.08 13.08 0.0003 0.0003 tv 1 1558 0.55 0.55 0.4577 0.4579 cc*tv 1 1558 2.77 2.77 0.0962 0.0964 prethkord 3 1558 72.17 24.06 <.0001 <.0001 cc*prethkord 3 1558 1.35 0.45 0.7181 0.7181 tv*prethkord 3 1558 12.36 4.12 0.0062 0.0064 cc*tv*prethkord 3 1558 3.71 1.24 0.2942 0.2946 Model M1d: same as M1a but drop prethk*cc*tv 9 The GLIMMIX Procedure Model Information Data Set WORK.TV Response Variable thk Response Distribution Multinomial (ordered) Link Function Cumulative Logit Variance Function Default Variance Matrix Blocked By school Estimation Technique Maximum Likelihood Likelihood Approximation Gauss-Hermite Quadrature Degrees of Freedom Method Containment Number of Observations Read 1601 Number of Observations Used 1600 Convergence criterion (GCONV=1E-8) satisfied. Fit Statistics -2 Log Likelihood 4235.37 AIC (smaller is better) 4255.37 AICC (smaller is better) 4255.51 BIC (smaller is better) 4268.70 CAIC (smaller is better) 4278.70 HQIC (smaller is better) 4259.45 Fit Statistics for Conditional Distribution -2 log L(thk | r. effects) 4197.52 Covariance Parameter Estimates Standard Cov Parm Subject Estimate Error Intercept school 0.07783 0.03963 Solutions for Fixed Effects Standard Effect thk cc tv Estimate Error DF t Value Pr > |t| Intercept 1 -1.0180 0.1931 24 -5.27 <.0001 Intercept 2 0.2276 0.1898 24 1.20 0.2420 Intercept 3 1.4077 0.1934 24 7.28 <.0001 Model M1d: same as M1a but drop prethk*cc*tv 10 The GLIMMIX Procedure Solutions for Fixed Effects Standard Effect thk cc tv Estimate Error DF t Value Pr > |t| cc 0 0.5854 0.2525 1567 2.32 0.0205 cc 1 0 . . . . tv 0 0.07122 0.2538 1567 0.28 0.7790 tv 1 0 . . . . cc*tv 0 0 0.4999 0.2897 1567 1.73 0.0846 cc*tv 0 1 0 . . . . cc*tv 1 0 0 . . . . cc*tv 1 1 0 . . . . prethk -0.3002 0.06558 1567 -4.58 <.0001 prethk*cc 0 -0.07198 0.07575 1567 -0.95 0.3421 prethk*cc 1 0 . . . . prethk*tv 0 -0.1381 0.07601 1567 -1.82 0.0694 prethk*tv 1 0 . . . . Type III Tests of Fixed Effects Num Den Effect DF DF Chi-Square F Value Pr > ChiSq Pr > F cc 1 1567 15.54 15.54 <.0001 <.0001 tv 1 1567 2.31 2.31 0.1286 0.1288 cc*tv 1 1567 2.98 2.98 0.0844 0.0846 prethk 1 1567 108.86 108.86 <.0001 <.0001 prethk*cc 1 1567 0.90 0.90 0.3420 0.3421 prethk*tv 1 1567 3.30 3.30 0.0692 0.0694 Model M1e: same as M1d but drop prethk*cc 11 The GLIMMIX Procedure Model Information Data Set WORK.TV Response Variable thk Response Distribution Multinomial (ordered) Link Function Cumulative Logit Variance Function Default Variance Matrix Blocked By school Estimation Technique Maximum Likelihood Likelihood Approximation Gauss-Hermite Quadrature Degrees of Freedom Method Containment Number of Observations Read 1601 Number of Observations Used 1600 Convergence criterion (GCONV=1E-8) satisfied. Fit Statistics -2 Log Likelihood 4236.28 AIC (smaller is better) 4254.28 AICC (smaller is better) 4254.39 BIC (smaller is better) 4266.27 CAIC (smaller is better) 4275.27 HQIC (smaller is better) 4257.94 Fit Statistics for Conditional Distribution -2 log L(thk | r. effects) 4198.21 Covariance Parameter Estimates Standard Cov Parm Subject Estimate Error Intercept school 0.07859 0.03982 Solutions for Fixed Effects Standard Effect thk cc tv Estimate Error DF t Value Pr > |t| Intercept 1 -0.9476 0.1785 24 -5.31 <.0001 Intercept 2 0.2960 0.1759 24 1.68 0.1054 Intercept 3 1.4765 0.1797 24 8.22 <.0001 Model M1e: same as M1d but drop prethk*cc 12 The GLIMMIX Procedure Solutions for Fixed Effects Standard Effect thk cc tv Estimate Error DF t Value Pr > |t| cc 0 0.4413 0.2023 1568 2.18 0.0293 cc 1 0 . . . . tv 0 0.07304 0.2541 1568 0.29 0.7738 tv 1 0 . . . . cc*tv 0 0 0.4944 0.2905 1568 1.70 0.0889 cc*tv 0 1 0 . . . . cc*tv 1 0 0 . . . . cc*tv 1 1 0 . . . . prethk -0.3368 0.05338 1568 -6.31 <.0001 prethk*tv 0 -0.1369 0.07611 1568 -1.80 0.0722 prethk*tv 1 0 . . . . Type III Tests of Fixed Effects Num Den Effect DF DF Chi-Square F Value Pr > ChiSq Pr > F cc 1 1568 22.50 22.50 <.0001 <.0001 tv 1 1568 2.29 2.29 0.1302 0.1304 cc*tv 1 1568 2.90 2.90 0.0887 0.0889 prethk 1 1568 108.58 108.58 <.0001 <.0001 prethk*tv 1 1568 3.24 3.24 0.0720 0.0722 Model M1f: same as M1e but drop prethk*tv 13 The GLIMMIX Procedure Model Information Data Set WORK.TV Response Variable thk Response Distribution Multinomial (ordered) Link Function Cumulative Logit Variance Function Default Variance Matrix Blocked By school Estimation Technique Maximum Likelihood Likelihood Approximation Gauss-Hermite Quadrature Degrees of Freedom Method Containment Number of Observations Read 1601 Number of Observations Used 1600 Convergence criterion (GCONV=1E-8) satisfied. Fit Statistics -2 Log Likelihood 4239.52 AIC (smaller is better) 4255.52 AICC (smaller is better) 4255.61 BIC (smaller is better) 4266.18 CAIC (smaller is better) 4274.18 HQIC (smaller is better) 4258.78 Fit Statistics for Conditional Distribution -2 log L(thk | r. effects) 4202.96 Covariance Parameter Estimates Standard Cov Parm Subject Estimate Error Intercept school 0.07318 0.03816 Solutions for Fixed Effects Standard Effect thk cc tv Estimate Error DF t Value Pr > |t| Intercept 1 -0.8212 0.1616 24 -5.08 <.0001 Intercept 2 0.4205 0.1594 24 2.64 0.0144 Intercept 3 1.5991 0.1641 24 9.74 <.0001 Model M1f: same as M1e but drop prethk*tv 14 The GLIMMIX Procedure Solutions for Fixed Effects Standard Effect thk cc tv Estimate Error DF t Value Pr > |t| cc 0 0.4580 0.1983 1569 2.31 0.0211 cc 1 0 . . . . tv 0 -0.1908 0.2035 1569 -0.94 0.3487 tv 1 0 . . . . cc*tv 0 0 0.4657 0.2842 1569 1.64 0.1016 cc*tv 0 1 0 . . . . cc*tv 1 0 0 . . . . cc*tv 1 1 0 . . . . prethk -0.4033 0.03886 1569 -10.38 <.0001 Type III Tests of Fixed Effects Num Den Effect DF DF Chi-Square F Value Pr > ChiSq Pr > F cc 1 1569 23.58 23.58 <.0001 <.0001 tv 1 1569 0.09 0.09 0.7662 0.7662 cc*tv 1 1569 2.68 2.68 0.1014 0.1016 prethk 1 1569 107.73 107.73 <.0001 <.0001 Model M1falt: reparameterization of m1f to facilitate interpretations 15 The GLIMMIX Procedure Model Information Data Set WORK.TV Response Variable thk Response Distribution Multinomial (ordered) Link Function Cumulative Logit Variance Function Default Variance Matrix Blocked By school Estimation Technique Maximum Likelihood Likelihood Approximation Gauss-Hermite Quadrature Degrees of Freedom Method Containment Number of Observations Read 1601 Number of Observations Used 1600 Convergence criterion (GCONV=1E-8) satisfied. Fit Statistics -2 Log Likelihood 4239.52 AIC (smaller is better) 4255.52 AICC (smaller is better) 4255.61 BIC (smaller is better) 4266.18 CAIC (smaller is better) 4274.18 HQIC (smaller is better) 4258.78 Fit Statistics for Conditional Distribution -2 log L(thk | r. effects) 4202.96 Covariance Parameter Estimates Standard Cov Parm Subject Estimate Error Intercept school 0.07318 0.03816 Solutions for Fixed Effects Standard Effect thk Estimate Error DF t Value Pr > |t| Alpha Lower Upper Intercept 1 -0.08835 0.1640 24 -0.54 0.5949 0.05 -0.4267 0.2500 Intercept 2 1.1534 0.1655 24 6.97 <.0001 0.05 0.8119 1.4949 Intercept 3 2.3320 0.1733 24 13.46 <.0001 0.05 1.9744 2.6896 Model M1falt: reparameterization of m1f to facilitate interpretations 16 The GLIMMIX Procedure Solutions for Fixed Effects Standard Effect thk Estimate Error DF t Value Pr > |t| Alpha Lower Upper cctv01 -0.2749 0.1975 1569 -1.39 0.1641 0.05 -0.6623 0.1125 cctv10 -0.9236 0.2038 1569 -4.53 <.0001 0.05 -1.3234 -0.5239 cctv11 -0.7329 0.1995 1569 -3.67 0.0002 0.05 -1.1242 -0.3416 prethk -0.4033 0.03886 1569 -10.38 <.0001 0.05 -0.4796 -0.3271 Type III Tests of Fixed Effects Num Den Effect DF DF Chi-Square F Value Pr > ChiSq Pr > F cctv01 1 1569 1.94 1.94 0.1639 0.1641 cctv10 1 1569 20.54 20.54 <.0001 <.0001 cctv11 1 1569 13.50 13.50 0.0002 0.0002 prethk 1 1569 107.73 107.73 <.0001 <.0001 Model M2: Same as m1falt but w/ class effect instead of school effect 17 The GLIMMIX Procedure Model Information Data Set WORK.TV Response Variable thk Response Distribution Multinomial (ordered) Link Function Cumulative Logit Variance Function Default Variance Matrix Blocked By class(school) Estimation Technique Maximum Likelihood Likelihood Approximation Gauss-Hermite Quadrature Degrees of Freedom Method Containment Number of Observations Read 1601 Number of Observations Used 1600 Convergence criterion (GCONV=1E-8) satisfied. Fit Statistics -2 Log Likelihood 4230.77 AIC (smaller is better) 4246.77 AICC (smaller is better) 4246.86 BIC (smaller is better) 4270.01 CAIC (smaller is better) 4278.01 HQIC (smaller is better) 4256.21 Fit Statistics for Conditional Distribution -2 log L(thk | r. effects) 4113.26 Covariance Parameter Estimates Standard Cov Parm Subject Estimate Error Intercept class(school) 0.1886 0.06370 Solutions for Fixed Effects Standard Effect thk Estimate Error DF t Value Pr > |t| Alpha Lower Upper Intercept 1 -0.07575 0.1466 131 -0.52 0.6063 0.05 -0.3658 0.2143 Intercept 2 1.1977 0.1485 131 8.06 <.0001 0.05 0.9038 1.4915 Intercept 3 2.4032 0.1579 131 15.22 <.0001 0.05 2.0908 2.7155 Model M2: Same as m1falt but w/ class effect instead of school effect 18 The GLIMMIX Procedure Solutions for Fixed Effects Standard Effect thk Estimate Error DF t Value Pr > |t| Alpha Lower Upper cctv01 -0.2059 0.1706 1462 -1.21 0.2277 0.05 -0.5405 0.1288 cctv10 -0.8613 0.1736 1462 -4.96 <.0001 0.05 -1.2018 -0.5208 cctv11 -0.7660 0.1719 1462 -4.46 <.0001 0.05 -1.1032 -0.4289 prethk -0.4148 0.03936 1462 -10.54 <.0001 0.05 -0.4920 -0.3376 Type III Tests of Fixed Effects Num Den Effect DF DF Chi-Square F Value Pr > ChiSq Pr > F cctv01 1 1462 1.46 1.46 0.2275 0.2277 cctv10 1 1462 24.62 24.62 <.0001 <.0001 cctv11 1 1462 19.86 19.86 <.0001 <.0001 prethk 1 1462 111.05 111.05 <.0001 <.0001 Model M3: Same as m1falt but w/ class and school effects 19 The GLIMMIX Procedure Model Information Data Set WORK.TV Response Variable thk Response Distribution Multinomial (ordered) Link Function Cumulative Logit Variance Function Default Variance Matrix Blocked By school Estimation Technique Maximum Likelihood Likelihood Approximation Laplace Degrees of Freedom Method Containment Number of Observations Read 1601 Number of Observations Used 1600 Convergence criterion (GCONV=1E-8) satisfied. Fit Statistics -2 Log Likelihood 4229.54 AIC (smaller is better) 4247.54 AICC (smaller is better) 4247.65 BIC (smaller is better) 4259.53 CAIC (smaller is better) 4268.53 HQIC (smaller is better) 4251.20 Fit Statistics for Conditional Distribution -2 log L(thk | r. effects) 4115.72 Covariance Parameter Estimates Standard Cov Parm Subject Estimate Error Intercept school 0.04475 0.04215 Intercept class(school) 0.1436 0.06242 Solutions for Fixed Effects Standard Effect thk Estimate Error DF t Value Pr > |t| Alpha Lower Upper Intercept 1 -0.09543 0.1683 24 -0.57 0.5760 0.05 -0.4429 0.2520 Intercept 2 1.1774 0.1699 24 6.93 <.0001 0.05 0.8266 1.5281 Model M3: Same as m1falt but w/ class and school effects 20 The GLIMMIX Procedure Solutions for Fixed Effects Standard Effect thk Estimate Error DF t Value Pr > |t| Alpha Lower Upper Intercept 3 2.3830 0.1781 24 13.38 <.0001 0.05 2.0153 2.7507 cctv01 -0.2374 0.2040 1462 -1.16 0.2446 0.05 -0.6375 0.1627 cctv10 -0.8850 0.2090 1462 -4.23 <.0001 0.05 -1.2951 -0.4750 cctv11 -0.7489 0.2048 1462 -3.66 0.0003 0.05 -1.1506 -0.3473 prethk -0.4086 0.03961 1462 -10.32 <.0001 0.05 -0.4863 -0.3309 Type III Tests of Fixed Effects Num Den Effect DF DF Chi-Square F Value Pr > ChiSq Pr > F cctv01 1 1462 1.35 1.35 0.2444 0.2446 cctv10 1 1462 17.92 17.92 <.0001 <.0001 cctv11 1 1462 13.38 13.38 0.0003 0.0003 prethk 1 1462 106.43 106.43 <.0001 <.0001 Tests of Covariance Parameters Based on the Likelihood Label DF -2 Log Like ChiSq Pr > ChiSq Note school var comp=0 1 4231.15 1.61 0.1020 MI class var comp=0 1 4239.52 9.98 0.0008 MI MI: P-value based on a mixture of chi-squares. Model M1falt-L: Same as m1falt but fit with Laplace method 21 The GLIMMIX Procedure Model Information Data Set WORK.TV Response Variable thk Response Distribution Multinomial (ordered) Link Function Cumulative Logit Variance Function Default Variance Matrix Blocked By school Estimation Technique Maximum Likelihood Likelihood Approximation Laplace Degrees of Freedom Method Containment Number of Observations Read 1601 Number of Observations Used 1600 Convergence criterion (GCONV=1E-8) satisfied. Fit Statistics -2 Log Likelihood 4239.52 AIC (smaller is better) 4255.52 AICC (smaller is better) 4255.61 BIC (smaller is better) 4266.18 CAIC (smaller is better) 4274.18 HQIC (smaller is better) 4258.78 Fit Statistics for Conditional Distribution -2 log L(thk | r. effects) 4202.96 Covariance Parameter Estimates Standard Cov Parm Subject Estimate Error Intercept school 0.07319 0.03817 Solutions for Fixed Effects Standard Effect thk Estimate Error DF t Value Pr > |t| Alpha Lower Upper Intercept 1 -0.08836 0.1640 24 -0.54 0.5949 0.05 -0.4268 0.2500 Intercept 2 1.1534 0.1655 24 6.97 <.0001 0.05 0.8119 1.4949 Intercept 3 2.3320 0.1733 24 13.46 <.0001 0.05 1.9743 2.6896 Model M1falt-L: Same as m1falt but fit with Laplace method 22 The GLIMMIX Procedure Solutions for Fixed Effects Standard Effect thk Estimate Error DF t Value Pr > |t| Alpha Lower Upper cctv01 -0.2749 0.1975 1569 -1.39 0.1641 0.05 -0.6623 0.1125 cctv10 -0.9237 0.2038 1569 -4.53 <.0001 0.05 -1.3235 -0.5239 cctv11 -0.7329 0.1995 1569 -3.67 0.0002 0.05 -1.1242 -0.3416 prethk -0.4033 0.03886 1569 -10.38 <.0001 0.05 -0.4796 -0.3271 Type III Tests of Fixed Effects Num Den Effect DF DF Chi-Square F Value Pr > ChiSq Pr > F cctv01 1 1569 1.94 1.94 0.1639 0.1641 cctv10 1 1569 20.53 20.53 <.0001 <.0001 cctv11 1 1569 13.50 13.50 0.0002 0.0002 prethk 1 1569 107.73 107.73 <.0001 <.0001 Model M2-L: Same as m2 but fit with Laplace method 23 The GLIMMIX Procedure Model Information Data Set WORK.TV Response Variable thk Response Distribution Multinomial (ordered) Link Function Cumulative Logit Variance Function Default Variance Matrix Blocked By class(school) Estimation Technique Maximum Likelihood Likelihood Approximation Laplace Degrees of Freedom Method Containment Number of Observations Read 1601 Number of Observations Used 1600 Convergence criterion (GCONV=1E-8) satisfied. Fit Statistics -2 Log Likelihood 4231.15 AIC (smaller is better) 4247.15 AICC (smaller is better) 4247.24 BIC (smaller is better) 4270.39 CAIC (smaller is better) 4278.39 HQIC (smaller is better) 4256.59 Fit Statistics for Conditional Distribution -2 log L(thk | r. effects) 4114.85 Covariance Parameter Estimates Standard Cov Parm Subject Estimate Error Intercept class(school) 0.1837 0.06238 Solutions for Fixed Effects Standard Effect thk Estimate Error DF t Value Pr > |t| Alpha Lower Upper Intercept 1 -0.07509 0.1461 131 -0.51 0.6081 0.05 -0.3640 0.2139 Intercept 2 1.1977 0.1480 131 8.09 <.0001 0.05 0.9049 1.4904 Intercept 3 2.4026 0.1574 131 15.27 <.0001 0.05 2.0913 2.7140 Model M2-L: Same as m2 but fit with Laplace method 24 The GLIMMIX Procedure Solutions for Fixed Effects Standard Effect thk Estimate Error DF t Value Pr > |t| Alpha Lower Upper cctv01 -0.2070 0.1696 1462 -1.22 0.2226 0.05 -0.5397 0.1258 cctv10 -0.8616 0.1726 1462 -4.99 <.0001 0.05 -1.2002 -0.5229 cctv11 -0.7659 0.1710 1462 -4.48 <.0001 0.05 -1.1013 -0.4306 prethk -0.4150 0.03935 1462 -10.55 <.0001 0.05 -0.4921 -0.3378 Type III Tests of Fixed Effects Num Den Effect DF DF Chi-Square F Value Pr > ChiSq Pr > F cctv01 1 1462 1.49 1.49 0.2224 0.2226 cctv10 1 1462 24.90 24.90 <.0001 <.0001 cctv11 1 1462 20.07 20.07 <.0001 <.0001 prethk 1 1462 111.22 111.22 <.0001 <.0001