f*********e
发帖数: 1144 | 1 Backgroud:
GLMM model: y = a + bX + cZ + e
bX: fixed effect (between-group variations: control vs treatment)
cZ: random effect (within-group variations)
Dist: poisson
===========================================================================
Quesetion 1): I have two levels of within-group variations for random
effects, specifically: biological variations and technicl variations. How do
I incorporate these two levels into one single random effect term? I am
using "glmer" from R-programming
Quest... 阅读全帖 |
|
s*********e
发帖数: 1051 | 2 1) although i don't know much about "glmer", 2-level random effect is common
and very doable in sas glimmix procedure. Do take a look at the "random"
statement in this procedure.
2) GLMM "assume the data points are NOT independently distributed" is not a
correct statement. Instead, it ought to be "GLMM does NOT have to assume the
data points are independently distributed". GLMM can take care of this
dependency by various co-variance structure in either G-side matrix or R-
side matrix. |
|
A*******s
发帖数: 3942 | 3 One friend asked me this question but i know nothing about GLMMs. Can any
big bull give some examples about GLMMs' practices on risk management/
marketing/insurance? |
|
a****m
发帖数: 693 | 4 GLM are extension of LM to cases where data are independent and standard
linear model assumptions are violated, and GLMM just incorporate another
extra random effect.
for Q1, you can not separate those biological and technical variation in the
random effect
For Q2, for independent assumption, you can easily solve those parameter
analytically using ML, however this is not doable in GLMM, you may use some
numerical method to get optimal value of parameter, like pseudo-likelihood
approach. |
|
f*********e
发帖数: 1144 | 5 anybody familiar with the application of GLMM to model spectral count data? |
|
s*********t
发帖数: 3 | 6 有个关于Generalized Linear Mixed Models(GLMMs）的问题：
请问在R或SAS中能否得到 covariance matrix between fixed effect estimators (\
beta) and variance component estimator for random effect(\sigma)
在R或SAS中可以得到fixed effect estimators的variance matrix，也可以得到
variance component estimator(\sigma)的variance。但是能否输出它们两者的
covariance呢？
多谢～ |
|
|
f*********e
发帖数: 1144 | 8 Thanks a lot for the reply!!!!! esp in the holidays!
But my background is in bio/chem, I have marginally stat background.....
I only know R-programming a little bit....
As for 1), thanks for the explanatioin, at least I know it is doable,
although I don't know how to implement it.
As for 2), when u say "GLMM can take care of the dependency by......", do
you mean those two matrix are already embeded in the functions/packages so
that I don't have to modify anything?
common
a
the |
|
s********1
发帖数: 54 | 9 ______________________________________________________________________
In terms of the variance structure
______________________________________________________________________
Normal structure depend on each individual,Spatial structure depends on the
distances between two points and compound, etc.
______________________________________________________________________
In terms of the likelihood:
______________________________________________________________________
The PL method is based on Wol... 阅读全帖 |
|
h****s
发帖数: 16779 | 10 Both GEE and Linear Mixed Effects Model (LMM) can be viewed as special cases
of the Generalized Linear Mixed Effects Model (GLMM):
GEE is the population averaged (PA) marginal model of GLMM, see Zeger, Liang
, Albert, 1988, Biometrics, 1049-1060.
LMM is the GLMM with normal distribution and identity link function, see SAS
help. |
|
s*********e
发帖数: 1051 | 11 just want to clarify your question.
are you specifically interested in the difference in estimation method
between gee and glmm?
or are you interested in the conceptual difference between marginal model
referred to gee and conditional model referred to glmm?
also, since the difference between gee and glmm is less pronounced in
gaussian models, I assume you are referring to non-gaussian models.
in any sense, this is an interesting topic. |
|
h*t
发帖数: 187 | 12 1. missing的问题
（1）检测是MCAR还是MAR，如果是Monotonic missing，可以用logistic regression
来检测
（2）IPW的方法一般也是在MMDP的情况下好用
2. model选择的问题
（1）link function的选择
如果是count data，可以试试log link；如果ceiling现象很明显，可以作为
binomial data，试试logit link。
（2）用GEE的话，weight GEE是把IPW的方法扩展到GEE里面
用GLMM的话，因为是likelihood based的方法，对missing更加robust一些。
3.软件方面
如果用GLMM又用logit link的话，推荐不用SAS Procedure GLIMMIX，而用NLMIXED，
原因可参见Statistics in Medicine 30:2562-2572, 2011 |
|
c*****1
发帖数: 115 | 13 我的看法是既然你的data observations are not independent，你就不能用Poisson
GLM了，而要用GLMM了，在SAS里就是Proc GLIMMIX,因为你要定义你的covariance
structure。
可以说GLMM一点也不容易，作为alternative，你看是不是可以用Poisson GAM，将时间
设为Additive。 |
|
h*t
发帖数: 187 | 14 ···原创，原创，版主请发钱···
引子：在俺家漂洋过海第二代八刀还在娘胎的时候，就想着写点什么来纪念这个小崽子
来到花花世界，便开始构思这篇文章了。其后一直瞎忙，一边在尿布和奶瓶之间打转，
一边在GEE,GLMM,Cox- model的世界里扑腾，直到今日，八刀要迎来他的第一个圣诞节
之日，才抽出时间把它写完。
翻看祖国历史，几十度的兴衰罔替，数千年的云起云散，各路英雄你方唱罢我登台，有
人落得荣华梦一场，有人得到功名纸半张。有人二十年尘土征衫，却叹韩彭未央；有人
不让胡马度阴山，临终前也只“但愿生入玉门关”；又有人“心在天山，身老沧州”；
也有人年轻时为功名走遍天下路，年老后却“长恨此身非吾有”，只要“小舟从此逝，
江海寄余生”，整个一东邪黄药师的情怀。
罗贯中觉得他看透了历史，是非成败转头空，青山依旧在，几度夕阳红。可是辩证唯物
主义者的眼里，历史当然不是虚空。几百年后，老毛又在琢磨：问苍茫大地，谁主沉浮
？是啊，谁主沉浮？在猴霸天的中学《社会发展简史》课本上写着“人民，只有人民，
才是历史的创造者。”可是霸天当年一直纳闷，人民数目庞大，众口难调，又据说素质
不高，还不能投票搞民 |
|
w******t
发帖数: 16937 | 15 分特，想看专业的？
看这个。声明：因为网络安全原因，我删去了一些必须删去的内容。
http://schema.org/WebPage">Google
|
|
