我查到若是将预设值method=reml改为method=MIVQUE0就可以解决问题 proc mixed noclprint covtest method=MIVQUE0 data=zzzz ic; class smoking gender ht dm medica; model a=CO age bmi2 smoking gender ht dm medica tem rh /s ddfm=betwithin cl; repeated / subject=num type=un rcorr r; run; 想请问版友 1.是否是因为MIVQUE0比REML或ML好的部分是不需假设估计值为常态分布? 2.要采用这个方法我的资料需先有什麽样的处理? 谢谢大家 ※ 引述《minicharco (小茶)》之铭言： 我主要是想看一群人某个健康变项跟空气污染之间的关系 手边有这群人重复测量的资料 为了进一步分析在冷暖季的空气污染影响 而将资料分成两群组 两组样本数一样多 可是程式码一组可以跑一组出现了以下的警告而无法跑出结果 WARNING: Stopped because of too many likelihood evaluations. WARNING: Did not converge. 我找到别人回覆某人问题的解答 可是光从信件中我不知道要如何修改程式 想请问板上的高手有没有人可以帮忙，谢谢 我的程式码如下 proc mixed noclprint covtest data=zzzz ic; class smoking gender ht dm medica; model a=CO age bmi2 smoking gender ht dm medica tem rh /s ddfm=betwithin cl; repeated / subject=num type=un rcorr r; run; 信件连结 http://www.listserv.uga.edu/cgi-bin/wa?A2=ind0207a&L=sas-l&P=17502 主要内文如下 There are any number of different approaches which you might use to address your problem. First, you could just change the number of likelihood evaluations which SAS performs before stating that convergence could not be obtained. The option MAXFUNC allows you to specify the number of likelihood evaluations the MIXED procedure should perform before stopping. Note that this is different from the number of iterations which should be performed before stopping. The number of iterations to perform is controlled by the option MAXITER. These options are specified on the PROC MIXED invocation. You can also change the convergence criterion. The default convergence criterion is a Hessian convergence with tolerance 1E-8. You can change the tolerance to something larger, or you can specify a different convergence criterion altogether. The various criteria are also options to the PROC MIXED statement. Rather than my reciting the manual here, I'll let you look them up. In addition to changing the convergence criterion and/or allowing PROC MIXED to run longer, you can employ the PARMS statement to evaluate the likelihood at various values for the variance and covariance of the random effects. You can then plot the likelihood surface as a function of the variances. This can be instructive about problems which you might encounter in estimating the variance and covariance terms. The PARMS statement also allows SAS to start the iterative process from points which are potentially nearer to the solution than the initial values of 0. 麻烦你们了，谢谢 --

※ 发信站: 批踢踢实业坊(ptt.cc) ◆ From: 140.112.4.183 ※ 编辑: minicharco 来自: 114.37.89.98 (06/02 23:23)