p******s
发帖数: 229 | 1 好像只能给出 beta ，还得手动把beta换成odds ratio.
哪位高手有经验， 如何在 proc nlmixed 加个option,让输出含odds ratio,而不只有
beta?
拜谢！ |
|
b**********i
发帖数: 1059 | 2 Proc mixed 有个weight statement 可以minimize weighted residual sum of
squares. 现同样要用到weight但是却是non-linear mixed effect analysis. 请教有
什么方法可以实现? proc nlmixed 没有weight statement，只有replicate statement
，两个好像不是一回事。
还有什么sas user forum比较好？准备在上面发个贴征求一下意见。多谢多谢 |
|
S******J
发帖数: 30 | 3 You can try to specify a variance function in nlmixed to solve your problem. |
|
A*******s
发帖数: 3942 | 4 sorry i don't have any specific modeling problem at this point of time, but
just would like to learn some more complicated examples. I went through the
GLIMMIX and NLMIXED examples on SAS website. Could you please recommend more
materials with good examples?
Many thanks!! |
|
z**********i
发帖数: 12276 | 5 Besides the free SAS documents, I have a SAS notes for NLMIXED.
If you need, I would drop an email to you.
Dale is a big niu on these areas. |
|
b**********i
发帖数: 1059 | 6 Good, I requested one. I've been using glimmix and nlmixed in a way of 活囵
吞枣.
him. |
|
z**********i
发帖数: 12276 | 7 我硕士论文用的NLMIXED,有一年多拉,也就是知道一点皮毛. :-( |
|
A*******s
发帖数: 3942 | 8 问问大牛们, 你们搞nlmixed的时候容易收敛么？
不用说别的复杂问题，就sas自己的例子，很豪放，不容易收敛
http://support.sas.com/documentation/cdl/en/statug/63033/HTML/d
这个初始值要是我自己猜的话怎么都不行... 搞grid search来选初始值也不行。
比如说我把sample code改了parms statement
parms beta1=-3.01 to 3 by 0.5 beta2=-3.02 to 3 by 0.5 beta3=-3.03 to 3 by 0.
5
s2b1 =0.1 to 2 by 0.1 cb12 =-1 to 1 by 0.1 s2b2 =0.1 to 2 by
0.1 s2=0.1 to 2 by 0.1/best=50;
结果出来
ERROR: Quadrature accuracy of 0.000100 could not be achieved with 31 points.
The achieved
accuracy ... 阅读全帖 |
|
P****D
发帖数: 11146 | 9 如果nlin算出来的初始值不好用的话，就放弃nlmixed改用glimmix，完毕。 |
|
A*******s
发帖数: 3942 | 10 Thanks for all replies. Basically big bulls agree on three points:
1) estimate the fixed effects first;
2) grid search would help;
3) GLIMMIX is better than NLMIXED
Thanks again. |
|
i*******g
发帖数: 2 | 11 假设data set 有5000条记录，对每1000条记录进行一次proc nlmixed，如何用一个
loop进行这5次循环呢。
我加了一个variable,simur=1 到5 表示相应的第1到第5个1000条记录。
试了以下的code，不成功。macro nlmixed 里就是一个Proc nlmixed.
%macro nlmixed(dset) ;
....
%mend nlmixed;
%macro tt;
%do i=1 %to 5;
data data12a;set data12;where simur=i;
%nlmixed(dset=data12a);
%end;
%MEND tt;
%tt;
Thanks a lots! |
|
j*****e
发帖数: 182 | 12 I have never used NLMIX macro. But I have used PROC NLMIX a lot. It is a
likelihood based approach, not GEE. You mentioned normal distribution. Are
you referrring to the random effectis treated normal? NLMIX can only
tolerate one random effect in the model. If you have multiple random effects
, you have to use GEE. Is NLMIX macro different from the PROC?
In general linear model, the random statement controls the design matrix and
the repeated statement control the residuals/sigma. In generalized |
|
h*****5
发帖数: 14 | 13 NLINMIX macro 可以fit gee model而proc nlmixed 不能。但NLINMIX macro只用于正
态分布，proc nlmixed 适用于更多的分布。
我已经proc nlmixed做了random effect zero-inflated model,现在想做gee model.找
不到现成的方法。 |
|
z**********i
发帖数: 12276 | 14 经过这段时间的学习,得到如下收获.
这是个COUNT DATA,OVERDISPERSED, ZERO-INFLATED.
如下方法我需要比较:
POISSON, NEGATIVE BINOMIAL, ZERO-INFLATED POISSON, ZERO-INFLATED NEGATIVE
BINOMIAL.
我面临的问题:我们的数据只能用SAS9.1
没有现成的ZERO-INFLATED MODEL在9.1上.
我需要用NLMIXED PROCEDURE自己写点程序.NLMIXED没法处理REPEATED,只能放到RANDOM
里考虑.
希望能继续得到大家的帮助和建议.
多谢!!
a
year? |
|
z**********i
发帖数: 12276 | 15 最近一直在用NLMIXED,看到你有些探讨NLMIXED的贴子.原来是PAYPAL的EMAIL.
your
we
fields
on |
|
s*********e
发帖数: 1051 | 16 The syntax of proc model is very similar to proc nlmixed. You should be able
to find plenty examples on line how to model panel data with proc nlmixed |
|
s*********e
发帖数: 1051 | 17 does anyone here have experience in simplex distribution?
I drafted a piece of nlmixed code based upon the density function but am not
sure it is correct or not.
proc nlmixed data = tmp1 technique = NRRIDG ;
parms b0 = 0 b1 = 0.1 sigma = 5;
bounds sigma > 0;
u = 1 / (1 + exp(-(b0 + b1 * x)));
dd = (y - u) ** 2 / (y * (1 - y) * u ** 2 * (1 - u) ** 2);
lh = (2 * constant('pi') * sigma * (y * (1 - y)) ** 3) ** (-0.5) * exp(-(2
* sigma) ** (-1) * dd);
ll = log(lh);
model y ~ general(l... 阅读全帖 |
|
A*******s
发帖数: 3942 | 18 the saddest thing is not that your NLMIXED cannot finish, but rather that
the NLMIXED finished after 20 hours and then you realized you
overparameterized the model... DAMN!!!!! :) |
|
A*******s
发帖数: 3942 | 19 proc nlmixed seems to be the only way u can do it.
glimmix does not support self defined likelihood function, so u cannot do
tobit.
qlim does not support tri-variate model fitting.
I recall someone posted the code using nlmixed for Heckman model in SAS-L
forum. you may have some ideas after checking it. |
|
P****D
发帖数: 11146 | 20 他这么关心算法，给我感觉是他要用手算……用SAS的，往NLIN或NLMIXED里一扔，哪里
会去关心NLIN用的是least squares而NLMIXED用的是maximum likelihood。还是我太不
求甚解了？
Besides, SPSS=Stupid Package for Stupid Students. :-D |
|
c**i
发帖数: 234 | 21 多谢楼上，model里面有不少category variable。如果用proc nlmixed好像不太方便。
SAS 9.1没法用proc genmod去fit ZINB。请问有没有什么proc去fit。如果用proc
nlmixed，怎么看model好不好。感谢 |
|
s****s
发帖数: 368 | 22 Use Proc NLMIXED to write out the likelihood and do the estimation.
★ 发自iPhone App: ChineseWeb 7.7 |
|
o****o
发帖数: 8077 | 23 【 以下文字转载自 Economics 讨论区 】
发信人: oloolo (戒网几天), 信区: Economics
标 题: any one knows if SAS could do Fractional Logit Model fit?
发信站: BBS 未名空间站 (Tue Feb 20 15:16:45 2007)
anyone knows if SAS could do the quasi-ML estimate of fractional Logit Model
of Woodridge etc '96?
I currently use PROC NLMIXED to fit it, looks good, but not the original quasi-
ML, rather the ML estimation |
|
j*****e
发帖数: 182 | 24 Use Proc Nlmixed to write your own likelihood function. |
|
m********1
发帖数: 368 | 25 Thank you very much. Your hint is very useful.
I have just noticed that, using PROC NLMIXED, in the "model" statement we
can use general(ll) to specify a general log likelihood function that we
construct using SAS programming statements.
Are there any other solutions for this problem? |
|
m********1
发帖数: 368 | 26 I think the suggestion of jsdagre on the #4 floor above is better.
My original idea is exactly same as yours, however, it has a problem: \alpha
_0=g(\pi_0) is already given. So the only parameter to estimate is \alpha_1=
\beta_1.
To use fisher-scoring and find a solution to maximize the defined log likelihood is
more straight forward, I think. Only need to define the log likelihood function and PROC NLMIXED will solve it for you. |
|
o****o
发帖数: 8077 | 27 不过这种方法你就可以试试offset的技巧with no intercept， 这样就不用自己写似然
方程了
alpha
1=
likelihood is
function and PROC NLMIXED will solve it for you. |
|
x**g
发帖数: 807 | 28 来自主题: Statistics版 - 雨天哭求。 If it is a nonlinear mixed effects model with normal assumptions, you can
try PROC NLMIXED. |
|
t**i
发帖数: 688 | 29 In the SAS dataset, create a new variable with values 1, 2, ..., 5,
respectively.
Then
Proc nlmixed ...... ;
by newvar;
... |
|
x**g
发帖数: 807 | 30 这些模型是analytical equivalent的，由于用于不同领域，所以有了不同的名称。一
般用SAS PROC MIXED， PROC NLMIXED， WinBugs， R 都可以做的。 |
|
c*******o
发帖数: 3829 | 31 ESTIMATE "Odds Ratio" exp(beta); |
|
p******s
发帖数: 229 | 32 thank you so much, capriccio! |
|
p******s
发帖数: 229 | 33 capriccio,
输出list里面，为什么odds ratio 不exactly 等于exp(beta), 只是很接近于 exp(
beta)？ 还有，oods ratio里面的p-value 总是《0.0001， 也和beta estimation 里
面的不一致、
麻烦您答疑一下，不胜感激！ |
|
o****o
发帖数: 8077 | 34 GLIMMIX里面的等效repeated statement是通过random statement来实现的，读读手册
另外你可以用NLMIX/GLIMMIX的macro，那个支持repeated statement
SAS |
|
P****D
发帖数: 11146 | 35 1. Feed your data and the relationship to SAS proc nlin. That proc is
usually helpful in finding initial values for estimates. After you have good
initial values you can improve them using nlmixed if needed.
2. Good luck... God bless you... |
|
o****o
发帖数: 8077 | 36 做算法和理论的，肯定要C和R了；如果做偏实证和应用统计分析的，似乎还是SAS+R为主
SAS的PROC MIXED， PROC NLMIXED和PROC NLIN还是很powerful的
另外，manipulate data其实也是做应用分析的基本功，以后工作公司都会要求自己做
数据的I/O和变换
SAS |
|
o****o
发帖数: 8077 | 37 'if then' programming statements are valid in many PROCs, such as PHREG,
NLMIXED, NLIN, to name a few |
|
p******s
发帖数: 229 | 38 有时，看到 p < 0.05，但 OR 的95% CI cross 1.
这种情况时常发生，大家怎么处理的？谢谢！ |
|
b*******r
发帖数: 152 | 39 it could happen, computationally. if the variable is very large and the
estimate of the parameter is very small, but still significant(means it's
not zero), and you will get ci comes close to or on '1'. |
|
p******s
发帖数: 229 | 40 variable is very large = observation number in the data is large or variable
number in the model is large?
the observation in my data is usually >10k.
Thanks! |
|
l*******n
发帖数: 19 | 41 As long as you can write the explicit models for the Pop PK, you can use
statistical software to fit models so Nonmem is not necessary. For example,
there is an example for One-Compartment Pharmacokinetic Model in SAS manual
for sas nlmixed. One advantage of Nonmem is that it contains some standard
modules (like macros) for PK.
In my view, Pop-PK may serve two purposes depending on the data used: if
early stage data are used, Pop-PK may provide guidance for later study
designs; if Phase I |
|
z**********i
发帖数: 12276 | 42 数据不normal,绝大多数的数据都是不是0.8-1,所以我用the patients number who not
received measure,这样,很多数成为0.
我试了transformation,没有明显的作用.
现在用的是generalized model.
还有一点麻烦的是,我们的数据只能用SAS9.1,只有9.2里增加了zero-inflated
analysis.
有人建议用NLMIXED,还要自己写点程序.
我试着装了个一个hotfix,或许proc countreg能起点作用,但只有9.2才正式加到里面,9
.1是试用品. |
|
s********n
发帖数: 69 | 43 talkstats.com has a sas forum. |
|
b**********i
发帖数: 1059 | 44 Thanks, I emailed sas support ,they are suprisingly helpful, well beyond my
expectation. |
|
b**********i
发帖数: 1059 | 45 写了个sas simulation，主要用nlmixed procedure。simulated datset 有160万个
record（1000次simulation合一起），10几个variable。昨天test了100次simulation
（也就是16万个record），只花了半个多小时。昨天走前设成1000次sim一早来程序还
在慢慢爬。问问先进是不是因为sas到了瓶颈了。 我的机器Q6600 2G RAM， winxp.
想着改写成个loop 让sas一次simulate一个 dataset 1600 个record,应该会快。不过
还不不清楚sas里怎么实现。 用别的语言的话在一个loop里面run就可以了。
R的话类似
result<-matrix(...).
for (ndat in 1:1000){
...
result[ndat,]<-...} 就可以了
用sas的话 datastep 或者procedure怎么在loop里面实现？是用macro吗，哪位大虾给
个意见？
谢谢 |
|
z**********i
发帖数: 12276 | 46 请用过的同学给点经验吧。
我有10几个CATEGORICAL VARIABLE，2-4个CONTINUOUS VARIABLES，想放到POISSON，
NEGATIVE BINOMIAL，ZERO-INFLATED MODEL里比较一下。
用在SAS 9.1中。
多谢指教！！ |
|
|
|
|
p******s
发帖数: 229 | 50 病人，医生，医院，3级资料。
outcome是catagorical。想控制clustering effect at 医生和医院。
以前2-级的资料，我用nlmixed, glimmix, genmod。
这3级别的，有什么SAS proc可用吗？STATA有合适的命令吗？
谢谢！ |
|
