[程式] SAS proc genmod的警告

[tokyo291 (工口工口)]

[軟體程式類別]:

SAS

[程式問題]:

由於為sparse data 產生以下警告訊息

WARNING: The generalized Hessian matrix is not positive definite. Iteration
will be terminated


[軟體熟悉度]:

高(1年以上)


[問題敘述]:

再跑GEE模型分析的時候，由於資料在某個時間為sparse data

SAS無法估計，有試過加改maxiter修改收斂次數，可是依然出現警告訊息

想請問是否有其他的option可以調整，以解決此問題

[程式範例]:

PROC GENMOD data=patient2 DESCEND ;

   CLASS group (ref="2") sex SN PreVac40_B(ref="0");

   MODEL SR_all= Age Kt_V   group PreVac40_B  BUN   Ferritin time /Dist=bin

link=logit maxiter=100000 ;

   REPEATED SUBJECT=SN/type=exch;

run;

--
※ 發信站: 批踢踢實業坊(ptt.cc), 來自: 140.116.52.59
※ 文章網址: http://www.ptt.cc/bbs/Statistics/M.1412419735.A.5EA.html
→ bmka: outcomes in some subgroups are all zero or one.
→ bmka: or the subgoups defined by a covariate are subsets of
→ bmka: those defined by another covariate. You might want to
→ bmka: pay special attention to PreVac40_B (I suppose this
→ bmka: indicates that a subject's baseline antibody titer is
→ bmka: >=1:40), and other covariates that may affect antibody
→ bmka: level, including gender and age.
感謝您的回覆,您提的這兩個問題我們都有想過,

現階段想嘗試解決第一個可能的問題-sparse data

不過查過許多SAS mannual和其他人的code都找不到解決方案...
※ 編輯: tokyo291 (140.116.52.59), 10/05/2014 19:55:01
→ bmka: 如果大部份的observation都是0,為什麼硬是要fit這個model
→ bmka: Consider reduce the number of covariates.