m****e
发帖数: 255 | 1 General you want X to be roughly normal. But in reality as the residuals are
normal, you are fine. Otherwise, transform, categorical or nonparametric
regression. |
|
o******6
发帖数: 538 | 2 如果MLR中含INTERACTION TERM的话，VIF肯定大，我知道CENTER THE VARIABLES的话可
以REDUCE MULTICOLLINEARITY问题，ORIGINAL MODEL是：
E(Y|X,Z)=b0+b1*X+b2*Z+b3*X*Z;where X*Z is the interaction term;
我现在要用standardized multiple linear regression,那么：
YC=(Y-MEAN OF Y)/Sy;
XC=(X-MEAN OF X)/Sx;
ZC=(Z-MEAN OF Z)/Sz；
interxz=(X*Z-MEAN OF X*Z)/Sxz
MODEL应该用一下哪一个（我觉得应该是2）?
1) E(YC)=b0+b1*XC+b2*ZC+b3*XC*ZC
2）E(YC)=b0(INTERCEPT,WILL BE 0)+b1*XC+b2*ZC+b3*interxz
那么下面的公式对不对:
E(X*Z)=corr(X,Z)*Sx*Sz+E(X)*E(Z)（这样就可以从SAS中得到，我不认为应该是直接
计算X*Z然后用P |
|
o******6
发帖数: 538 | 3 我还是决定不用standardized regression model了，因为我只想处理
multicollinearity问题，center the variables就可以达到这个目的了，没必要再
standardize，而且model中如果有interaction term的话SRM有些局限性,standard
error of the coefficient for interaction term不是很对。 |
|
o******6
发帖数: 538 | 4 Multiple Regression: Testing and Interpreting Interactions by Aiken and West
(1991)
谢谢。 |
|
l******1
发帖数: 292 | 5 在sas 里面，poisson regression model里面，response variable可以是binary吗？
我用了proc genmod option/dist/poisson,不是很清楚如果dependent variable是
binary (0,1)可以用/poisson还是用别的选项？菜鸟，等大虾指点！ |
|
z******a
发帖数: 3 | 6 If your response follows Poisson distribution, you may use poisson
regression. If your response is Binomial, you may consider using Proc GIMMIX
and using the proper Link function. |
|
w****h
发帖数: 1 | 7 有人说seber的linear regression analysis 不错，到底如何~或者别的好书，评论评
论呗 |
|
|
q**********0
发帖数: 335 | 9 What stastistic method used in correlation regression analysis in software
GraphPad Prism 5? Thanks a lot !!! |
|
s**********e
发帖数: 63 | 10 Used pooled time series data, Split the 40 years by 10 for each period,
create an period var called "period' with value 1- 4, then use panel
regression by period in STATA, e.g.
tsset period
xtreg y x1 x2, reg
|
|
x*******u
发帖数: 500 | 11 如果目的是predict， 请问linear regression model是否还需要满足variance
normality和constant variance assumption？
谢谢 |
|
i*****r
发帖数: 1302 | 12 【 以下文字转载自 Quant 讨论区 】
发信人: iambear (我是熊), 信区: Quant
标 题: matlab里怎么做residual是自回归的regression?
发信站: BBS 未名空间站 (Thu Dec 17 11:49:12 2009, 美东)
相当于SAS中的autoreg
y = b*x + e
e(t)=a*e(t-1)...
matlab的统计toolbox确实很烂啊,连generalized least squared都没有 |
|
b*******e
发帖数: 3324 | 13 在做LOGISTIC REGRESSION 分析的时候，ORDINAL 的自变量可以当作CONTINUOUS 来处
理么，
非常感谢 |
|
B****n
发帖数: 11290 | 14 One way is to compare parametric fit (nonlinear regression) with
nonparametric fit. |
|
f****r
发帖数: 1140 | 15 把学过的那些课程看看基本就差不多了吧。
logistic, regression, experiment design, survival, time series, mixed
model, prob.
也不一定都问，看职位吧。 |
|
a*******k
发帖数: 25 | 16 Like the title:
I am using SAS. I wonder how to write the code of getting Pseudo R2 in
ordered probit regression.
Thanks. |
|
n*****x
发帖数: 9 | 17 Does anyone know how to calculate odds ratios of every knots in spline
regression? I know there is a SAS macro could do this. However, I am not
able to find on the website.
Thanks |
|
s****u
发帖数: 1811 | 18 Applied Regression Analysis, bu Draper and Smith, John Wiley and Sons
包子答谢～ |
|
y*********4
发帖数: 76 | 19 那我们用LOGISTIC REGRESSION的时候，就是说假定有这么一个线性关系了，这个假定
合理么？ |
|
l***a
发帖数: 12410 | 20 with 24 variables I only got a R-sqr of 0.23. is it too low?
I know for linear regressions, I will feel confident if R-sqr is over 0.9 or
sometimes 0.8 or even 0.7.
normally how big R-sqr will make you feel comfortable? |
|
l***a
发帖数: 12410 | 21 thanks. the data is not too big (about 45k records) so I am afraid when I
regress on another data file with 4mil records, the r-sqr might get even
worse. plus I want to control the number independent var to under 20....
besides r-sqr, what other statistics you normally rely on measuring the fit
of the logistic model? some people like AIC but the AIC is just an absolute
value to me. i think relative number like xx% makes more sense.... |
|
z**k
发帖数: 378 | 22 I believe ROC is better ... never touched the R-sq for logistic regression
before ...
btw, is that a film camera? looks like a fisheye?
0.9 or |
|
j*****e
发帖数: 182 | 23 To TNEGIETNI, what you said is not true.
In experimental design/response surface analysis, you set X to be at an
array of different levels and measure up Y.
Regression analysis and ANOVA are basically the same thing. You are modeling
the conditional expectation of Y given X. It doesn't matter whether X is
random or fixed. X can also be continuous, or discrete.
As a matter of fact, in a lot of situations, people fix X to get the best
estimate of model parameters. |
|
c****o
发帖数: 69 | 24 谢谢你。
MERGE变量我想说的是比如本来一个变量有好5个LEVEL的，然后用了4个DUMMY VARIABLE
，但是如果MERGE其中的几个LEVEL，那么就会少用一些DUMMY VARIABLE了。。。
我想知道,MSE变小又能说明什么？？？
对于REGRESSION来说，具体的意思在哪里？
期望 |
|
t********y
发帖数: 469 | 25 logistic regression
PCA
CLUSTER ANALYSIS
NEURAL NETWORK
etc. |
|
f********g
发帖数: 989 | 26 logistic regression和CLUSTER ANALYSIS有没有快一点能入门的材料可以看的？什么
是PCA? 多谢！ |
|
j*****e
发帖数: 182 | 27 You can sure oversample the rare event. This is known as the case-control
study.
But, only the slope estimate is meaningful, the intercept estiamte is not.
Unless you know the marginal probability of the rare event, you wouldn't be
able to predict the binary outcome.
There are a little bit of discussion given by Agresti's book. You can also
read Hosmer and Lemoshow's Applied logistic regression for more. |
|
o******6
发帖数: 538 | 28 ☆─────────────────────────────────────☆
Emilyxue (雪) 于 (Thu May 15 14:12:39 2008) 提到:
急问用SAS 分析LOGISTIC regression, 有fix variable &random variable应该用什么
MODEL?非常感谢
☆─────────────────────────────────────☆
statcompute (statcompute) 于 (Thu May 15 21:22:46 2008) 提到:
there is nothing called "fix variable & random variable" in logistic reg.
Instead, there are "fixed effect" and "random effect". This type of model is
also called "generalized linear mixed model", which can be analyzed with "
pr |
|
a********a
发帖数: 346 | 29 Normally the logistic regression is two categories for the dependent
variable, what if there are more than two categories. Can I still use proc
logistic is SAS?
Thanks |
|
h******e
发帖数: 1791 | 30 intercept is useless in this case. what you need to pay attention is that
the assumption of mulinomial logistic regression when you use proc logistic. |
|
A*******s
发帖数: 3942 | 31 请教各位前辈一个问题：我最近做的一个project是关于regression tree的，因为俺的
R水平很一般，所以准备用sas做。问题是：
1. 虽然拜PharmD同学之福，我搞到了SAS EM。但是我的tree 的split criteria是
maximum likelihood的，我没找到EM里的tree有这个option。
2. google了一把，好像也没啥macro是和CART相关的。
嗯，前辈们支个招吧。万分感谢！小辈无以为报，唯有包子答谢。 |
|
l***a
发帖数: 12410 | 32 if there are many potential candidate predictors, do you always prefer a
simple regression over time series? |
|
s********e
发帖数: 68 | 33 请问在stata里面，做一个简单的poisson regression， 如何做goodness of fit test
来检验
adequency of the model。
stata output好像没有给出这个output。
谢谢各位 |
|
s*********e
发帖数: 1051 | 34 follow is copied from one of my paper and hope it helpful.
"
To evaluate the goodness-of-fit of a regression for count data, the most
popular but somewhat problematic practice is to compare the predicted and
observed values of the dependent variable. However, a measure of
goodness-of-fit solely based upon the expected value is unable to address
the improvement achieved by a model with less restrictive variance
assumption. A better alternative is to compare the predicted and observed
prob |
|
s*******e
发帖数: 226 | 35 I'm working on a random intercept model. I wonder why the model gives all
significant covariates if I exclude a constant intercept, but the results
become insignificant when I include the constant intercept.
I know in most cases an intercept term should be included in a regression
even it's not significant. But is there a theory about why to include
intercept or not?
Thanks in advance for any input. |
|
s*******e
发帖数: 226 | 36 This is a good point. In my model, the dependent variable is of firm
risk, and the covariates are a series of firm characteristics and market
conditions. Strictly speaking, when all those covariates are zero, which
means the firm is out of market or no longer exists, the firm's risk
should also be zero. In this sense, the intercept does not have any
meaningful effect.
I'm not sure about the intercept issue only because it's very common in
finance literature to include an intercept in regression |
|
x********u
发帖数: 64 | 37 这2个是一样的么？
还是有什么区别
好像log-linear model，是用proc genmod，dist=poisson
模型是
log (E(y))=ax
而poisson regression呢？ |
|
l*********s
发帖数: 5409 | 38 poisson regression with all categorical covariates are usually termed as log
-linear model |
|
l*****l
发帖数: 4170 | 39 sas support里面有这样的例子http://support.sas.com/kb/25/007.html
因为我的logistic regression 里用了ridging=absolute的option，不然不能converge
, 所以不能直接套用proc qlim。但是用例子里面第二个，就用他们自己的data,后面的
marginal effects 竟然是俩个新变量而不是俩个确定数值。有兴趣的请把他们的例子
copy paste, run一下，就知道。难道要去取这两个变量的平均值做答案？ |
|
A*******s
发帖数: 3942 | 40 我记得有个non parametric regression的algorithm是这样的，一时想不起名字 |
|
J*y
发帖数: 271 | 41 Linear regression line 是通过 minimizes the sum of the SQUARE of the
vertical distances from the points to the line.
为什么不是那条 Sum of the vertical distances 最小的线？ |
|
z**k
发帖数: 378 | 42 当然可以了，只是Loss function不同而已。如果你查Robust Regression，会有更多不
同的规
则。 |
|
c****s
发帖数: 63 | 43 I don't know what the divergence is either. Hope somebody can answer that.
Also, could some one tell me whether KS statistics, gini co-efficient or
divergence can be used in logistic regression model, or say dichotomous
outcome model? |
|
j*****e
发帖数: 182 | 44 This approach is correct. But you would need a relative large validation
data set.
KS statistic is not constructed to examine the goodness of fit of logistic
regression, no matter what people do in the real world.
calculate |
|
o******e
发帖数: 1001 | 45 在linear regression 中用categorical data, 发现有些categorical data的levels是
significant, 而有些是不significant.在这种情况下，是不是只能把不significant归
到一类，然后重新作linear model?谢谢！ |
|
x*******i
发帖数: 1791 | 46 自变量是categorical data？
如果用linear regression, 不存在有的level不显著的问题。因为只test mean
parameters.
如果是anova, 那有overall significant 和各level的问题。用contrast 解决。 |
|
o******e
发帖数: 1001 | 47 谢谢各位，我的意思是做anova的时候，发现categorical factor（比如Ａ）是
significant的，但是把Ａ放到linear regression model里，A的levels,比如a1,a2,a3
,a4,a1是default的level,根据t-test, 其中a2,a3是significant的，但是a4是不
significant的，那这个时候我是不是应该把a4和a1归为一个level,再重新做linear
model？还是这样直接就可以了？ |
|
T*******I
发帖数: 5138 | 48 I am comfused with your question. The standard deviation of the residual
error is a single point measure for the dataset in the regression analysis.
How could you get more than one standard deviations with your dataset? |
|
o******e
发帖数: 1001 | 49 对，TNEGIETNI,我目前的估计是这个standard deviation会随IV指数增长，而同时这个
IV又是linear regression中的变量。
sir提到的Heteroscedasticity可能是一种解决方案，我还没有仔细看。
independent |
|
s*r
发帖数: 2757 | 50 see page 12 of
Transformation and weighting in regression
By Raymond J. Carroll, David Ruppert
people have tried to explicitly model the sd as a function of IV
independent |
|
