Hello Community, I am using Joint Marginalized Random Effect model and I have received these messages: NOTE: Convergence criterion (GCONV=1E-8) satisfied. NOTE: At least one element of the gradient is greater than 1e-3. The result for my second joint variable does NOT seem reasonable. All estimated p-values for the second parameter are larger than 0.9! I do not really know, what is the reason of this outcome. I created the following code segments: 281 proc nlmixed data=original noad ; 282 parms a1=-2.2133 a2=-1.2087 Time_a=-0.0178 group_a=0.0885 time_ICU_a=0.0551 282! LASIXICU_a=-0.0046 CVP_a=0.0313 BE_a=-0.0764 Hgb_24h_after_a=-0.0004 283 b1=-4.5876 b2=-2.1619 Time_b=-0.0286 group_b=0.4285 time_ICU_b=0.1637 283! LASIXICU_b=0.0324 CVP_b=0.1275 BE_b=-0.1555 Hgb_24h_after_b=-0.0008 284 s1=0.2 s2=1.96 s12=-0.107; 285 286 *y1 = PH_ordinal; 287 if repeat in (1,3,5,7,9,11,13)then do; 288 eta1 = Time_a*Time + group_a*group + time_ICU_a*time_ICU + LASIXICU_a*LASIXICU+ CVP_a* 288! CVP + BE_a* BE + Hgb_24h_after_a* Hgb_24h_after ; 289 delta11= sqrt(1+s1**2) * PROBIT( exp(a1+eta1)/(1+exp(a1+eta1))); 290 delta12= sqrt(1+s1**2) * PROBIT( exp(a2+eta1)/(1+exp(a2+eta1))); 291 if ordinal=0 then do; 292 lik= PROBNORM (delta11+u1); 293 end; 294 if ordinal=1 then do; 295 lik= 1- PROBNORM (delta12+u1); 296 end; 297 if (lik > 1e-10) then loglik = log(lik); 298 else loglik = 1e100; 299 end; 300 *y2 = Lactate_ordinal; 301 if repeat in (2,4,6,8,10,12,14)then do; 302 eta2 = Time_b*Time + group_b*group + time_ICU_b*time_ICU + LASIXICU_b*LASIXICU+ CVP_b* 302! CVP + BE_b* BE + Hgb_24h_after_b* Hgb_24h_after ; 303 delta21 = sqrt(1+s2**2) * PROBIT( exp(b1+eta2)/(1+exp(b1+eta2))); 304 delta22 = sqrt(1+s2**2) * PROBIT( exp(b2+eta2)/(1+exp(b2+eta2))); 305 if ordinal=0 then do; 306 lik = PROBNORM(delta21+u2); 307 end; 308 if ordinal=1 then do; 309 lik= 1- PROBNORM(delta22+u2); 310 end; 311 if (lik > 1e-10) then loglik = log(lik); 312 else loglik = 1e100; 313 end; 314 model ordinal ~ general(loglik); 315 random u1 u2 ~ normal ([0,0] , [s1**2,s12,s2**2]) subject=id; 316 estimate "Rho" s12/(s1*s2); 317 run; The parameter estimates of this coding is available in attachment. Hope some one can help me out. TNX
... View more
