l*********d
发帖数: 254 | 1 Really thank you!
The help says:
The criterion used is
AIC = - 2*log L + k * edf,
where L is the likelihood and edf the equivalent degrees of freedom (i.e.,
the number of free parameters for usual parametric models) of fit.
For linear models with unknown scale (i.e., for lm and aov), -2log L is
computed from the deviance and uses a different additive constant to logLik
and hence AIC. If RSS denotes the (weighted) residual sum of squares then
extractAIC uses for - 2log L the formulae RSS/s - n (c |
|
d*********a
发帖数: 255 | 2 Applied Change Point Problems in Statistics
E. Carlstein, Nonparametric changepoints estimation, Ann. Statist. 16 (1988),
J. Chen and A.K. Gupta, Parametric Statistical Change Point Analysis,
Birkhauser, Berlin (2000).
B.E. Brodsky and B.S. Darkhovsky, Nonparametric Methods in Changepoint
Problems, Kluwer, Dordrecht (1993). |
|
n****c
发帖数: 78 | 3 The scored between left and right knee are correlated within the same
subject, so two sample t-test is no appropriate. But my sample is so small 8
patients, 32 obs. Parametric method is ok? Thanks |
|
q********i
发帖数: 795 | 4 It is a clustered data, parametric mixed model should work, try
proc mixed;
class id knee condition;
model score = knee condition;
random id;
run; |
|
h*******e
发帖数: 226 | 5 有6组数据做比较，每组的sample size 不是相同的(ie,unbalanced)
想要比较他们的mean是不是不同，做anova 和posthoc
做anova要满足，normal distribution and homogeneity of variance
我要先测试这两个条件满不满足。
我的问题是
1) test normality的时候是对全部六组的数据一起做，还是单独的每一组数据来分别
test
2）homogeneity of variance 是比较这六组的各自的variance是不是相等么？
3）如果我一套数据里针对某一些变量是normal distributed and equal variance ，
但另一些不是，那么我在处理这一套数据的里面是不是要根据满不满足条件而选择
parametric或者nonparametric的方法进行处理？这样做对于一整套数据来说会不会有
问题呢？ |
|
i**f
发帖数: 1195 | 6 我不是说因为要check the normality麻烦所以用非参
我是说可以用非参，也可以用log transformation then go for t-test,however you
still neet to check the normality.
你是学校的project还是real study?
前者的话两种方法都可以试试
如果后者的话，是要你写SAP? 你没有protocol可以follow?
the most common methods for handling non-normal data are:
sub group
transforming data
non-parametric |
|
i**f
发帖数: 1195 | 7 我做过的大部分都是用非参,这个很难讲那个方法更好的，如果是我我会说
it is hard to tell which one is superior to the other one. i will prefer non
parametric method... so you dont have to worry about the normality and
homogeneity assumption... you know...it is more robust...balah balah... |
|
y***6
发帖数: 46 | 8 If the data don't follow normality, you can do the t-test using non-
parametric technique. In SAS, try proc npar1way for more details. |
|
m*****s
发帖数: 156 | 9 可以try non-parametric test |
|
R********n
发帖数: 519 | 10 平时做research（computer vision, machine learning）的时候，很多时候需要不时
review一下统计的知识，经常感到要是有一本好的参考书就好了。
不是特别熟悉，但希望cover的内容有，各种distribution的详细介绍（包括多元），
各种testing，estimation的性质和分析，Bayesian and non-parametric Bayesian（
带有probability graphical model最好），理论上的一些收敛性分析
谢谢！ |
|
m*****s
发帖数: 156 | 11 removing outliers should be fine, but you need to be careful and reasonable.
You might try some non-parametric method such as wilcoxon rank test.
T |
|
s**c
发帖数: 1247 | 12 sample size太小
用non-parametric方法，ex：Signed rank test |
|
B****n
发帖数: 11290 | 13 One way is to compare parametric fit (nonlinear regression) with
nonparametric fit. |
|
b*****n
发帖数: 685 | 14 有missing data你的ANOVA就不好用了。这个是longitudinal data的hypothesis test
问题，有点麻烦，不过parametric和nonpara的方法应该都有，查查paper了，比如
Jianqing Fan的。 |
|
j*****e
发帖数: 182 | 15 There are many theoretical distributions that can be used to model a
positive population. For example,log-normal, log-logistic, weibull, gamma.
In survival analysis, modeling parameters of a weibull distribution may
correspond to a proportional hazard model. This parametric approach is
realized in SAS PROC LIFEREG. There are comparable models in econ and
engineer. You have to read more on what's the common choice of distribution
and how to model the corresponding parameters. |
|
j*****e
发帖数: 182 | 16 If you want to do survival analysis on clinical data, try this book:
Survival Analysis by John P. Klein and Melvin L. Moeschberger.
Read the most recent edition, since there are many mistakes in the old ones.
This book is not the greatest. But it is easy to understand and focus on the
application. Also be aware that it is mostly on semiparametric/
nonparametric approch. You may want some other books if you want to learn
the parametric approach.
Personal opinion. |
|
g******h
发帖数: 266 | 17 Thank you!
Does 'focus on application' means it gives the SAS sample codes?
Anyone knows what book will teach the parametric approach?
ones.
the |
|
j*****e
发帖数: 182 | 18 Survival analysis is also needed in engineering. I remember a example given
by Klein's book. Say, you want to estimate the lifetime of light bulbs. You
start with 1000 light bulbs and stop when, say, the fifth bulb burns out.
In my impression, engineers like to use the parametric approach. |
|
s*******e
发帖数: 1385 | 19 我的data是这个格式的
dead censor day
10 5 1
8 5 2
3 1 3
8 0 4
.....
我想用lifereg fit parametric的model
model statement应该怎么写？
谢谢大家了 |
|
s*******e
发帖数: 1385 | 20 这个不是病人，是生物实验，确切的说是虫子，我每两天检查一次，每次记录死亡的个
数，还有missing的个数，每个group里面的虫子，genetic是一样的。我要看的是group
之间的差异，例如两个survival function是不是有显著性差异。
想fit几个parametric和nonparametric 的model比较一下
不知道这样应该怎么写model statement，一定要是一个个体一行的数据才行吗？ |
|
c*********n
发帖数: 87 | 21 我觉得我犯了一个大错，你在1楼提到使用lifereg对吧，但是我把lifereg和lifetest
混用了，因为lifetest是nonparametric的，常用的方法包括KM和life table （2个
group），log rank test对多组，但是lifereg是parametric的，所以可能要改成：
proc lifereg ;
class trt;
model time*status(0) = trt / dist=???;
run;
你得specify一个survial time的distribution，常用的有exponential，gamma等等，
你看看这个网页：http://support.sas.com/documentation/cdl/en/statug/63033/HTML/default/statug_lifereg_sect011.htm
run完后的output里有Analysis of Parameter Estimates，你可以看到对trt effect，
看他们有没有作用，但是貌似没有pairwise的test。如果是我的话我就用l |
|
s*******e
发帖数: 1385 | 22 已经很感谢了，parametric 和 nonparametric我都用，就是为了看看哪个更好一些。
lifetest |
|
|
p*********8
发帖数: 1039 | 24 Mine is practical nonparametric statistics by W.J. Conover |
|
D*********2
发帖数: 535 | 25 all about nonparametric statistics |
|
|
h*****m
发帖数: 25 | 27 I think it is called ``All of Nonparametric Statistics''. |
|
C*********e
发帖数: 587 | 28 non parametric kernel plus large margin idea 基本上就是SVM了
SVM中 的 regularization 就是为了更好的 generalize ability，某种程度上是对
samples过少有帮助的。这个时候可以不用non linear kernel，因为维度已经足够
高了
data
your |
|
A*******s
发帖数: 3942 | 29 我记得有个non parametric regression的algorithm是这样的，一时想不起名字 |
|
m*******7
发帖数: 244 | 30 Why is Cox Proportional Hazard Model semi-parametric method? |
|
s*****n
发帖数: 2174 | 31 There is no easy way to test independence as far as I am aware of.
Pearson's test, Spearmean Rho, Kendall's Tau are all testing correlation.
The definition of independence is too strong for any parametric and
nonparametric approaches. The best approach I can think about is to
partition the X and Y range into small pieces and test as discrete variables
(such as Chisq test). Of course, this will require large sample size.
try |
|
c******n
发帖数: 380 | 32
而如果一个样
本的size达到了基本统计分析的要求，结论自然就在其中，因此，任何permutation也
不能提供更多
的意义。
这句话只是个客观的事实，没有任何其他意思，您想的太多了。
在缺少information of underlying distribution的情况下，用permutation test比拍脑
袋用其它parametric tests的优势显而易见，怎么不能提供更多意义？ |
|
Y******Y
发帖数: 8753 | 33 cox ph model is a semi-parametric model, it makes no assumption on the
distribution of the baseline hazard.
the paramatric ph models fitted by survreg assume the hazard function
follows a specific distrubution such as exponential, weibull, etc. |
|
f***a
发帖数: 329 | 34 试了几个学校的图书馆，死活找不到，请好心人帮忙
Gelfand, A.E. (with S. Sahu and B. Carlin) 1996. Efficient Parametrization
for Generalized Linear Mixed Models. Bayesian Statistics 5 Eds J.
Bernardo, et al. Clarendon Press, Oxford, 165-180.
“Noncentral Parameterizations for Hierarchical Models and Data
Augmentation” In Bayesian Statistics 7, Eds: J. Bernardo, et al. |
|
f**********m
发帖数: 157 | 35 Here is the question:
Use the software to calculate CI for 2X2 table.
Non-Parametrics
Two Independent Binomials
Unconditional procedures
CI on Ratio of Proportions
I can surely use the package to get CI for one single table.
If there are a bunch of 2X2 tables, can I do it on multiple tables?
Thanks. |
|
i**4
发帖数: 4 | 36 You can probably use Kruskal-Wallis non-parametric test (multiple-group
extension of Man-WHittney test), since each ingredient distributions are not
normal. The test should be fairly powerful due to the large sample size.
This is an essentially rank-based test. |
|
d*******1
发帖数: 293 | 37 Kruskal-Wallis non-parametric test did not work for my case.
What I want to do is like following.
A sample with 2500 students, normal distribution for exam scores.
students come from 10 regions. For each region, the score distribution may
be not normal. Each region have different number of candidates.
there is a mean and SD for total sample. I want to know for the mean of each
region, how far away from total sample mean?
Can I just use z test or t-test, and use standard error of total sample fo... 阅读全帖 |
|
y****d
发帖数: 432 | 38 ★★★★★★★★★★★★★★★★★★★★★★★★★★★★★
前面说明：
需要的童鞋请到我的签名档的博客查找！谢谢！发E-mail太累了！
觉得有价值的话可以顶一下，以便更多的人看到！谢谢！
★★★★★★★★★★★★★★★★★★★★★★★★★★★★★
详细介绍
1. Advanced Calculus with Applications in Statistics
2.A History of Probability and Statistics and Their Applications before 1750
3.Markov Decision Processes: Discrete Stochastic Dynamic Programming
4.Probability and Statistical Inference
5.Continuous Univariate Distributions, Vol. 1
6.Continuous Univariate Distributions, Vol. 2
7.The Theory of Measures and Integratio... 阅读全帖 |
|
d*******1
发帖数: 293 | 39 对于heavy tailed distribution的数据，该如何计算mean 的CI呢？
知道用non-parametric one sample可以估算median的CI， 但是不知道有什么好的方法
可以估算mean的CI？ |
|
j*****e
发帖数: 182 | 40 You can convert left censored data to right censored date by defining a new
variable, say time_new, such that time_new=l-time_old, where l is a very
large aritifical value. Time_new is then right censored.
As to interval censoring, SAS gives a self-consistent estimator of the
survival function (non-parametric). But that is it, no proportionall hazard
model. |
|
j********t
发帖数: 40 | 41 没错！！！为什么sas公司的人不开发一个呢？感觉这个实际project中用到interval
或者left censored 的不会没有吧。。。还是因为这种用right censored的方法估计出
来的区别也不大。。。parametric的model有局限性啊。 |
|
a*********r
发帖数: 139 | 42 Introduction to Time Series and Forecasting by Peter J. Brockwell and
Richard A. Davis.
Very readable, plenty of examples, excellent exercises at the end of each
chapter (all are very doable) and teaches you how to use the time series
package ITSM2000.
Nonlinear Time Series: Nonparametric and Parametric Methods by Jianqing Fan
and Qiwei Yao should NOT be the first reading for you. |
|
r*****y
发帖数: 199 | 43 i think you can use bootstrap method which is non-parametric, works quite
well for sample samples, no equal variance assumption needed, no distribution assumption needed. |
|
m******v
发帖数: 25 | 44 为什么要迷信统计？为什么non-parametric 就可以？
统计最重要的目的是解决问题。解决问题的第一步是搞清楚问题。搞清楚问题了，有了
数据，有了统计，还有更重要的是理解解释统计结果。
statistical significance sometimes means nothing, completely nothing.
数据有限，靠各种方法，搞到一个statistical significant at 5% 的结果，然后兴高
采烈？这个5%对你的问题的意义就那么大？为什么不能是20%，15%？
个人觉得有些人对统计的期望有点太高了。
test |
|
|
|
a****g
发帖数: 8131 | 47 value freq-1 freq_2
3 12 11
4 4 3
5 6 15
6 7 13
7 9 11
8 11 4
第一个column是数字，第二和第三column是frequency
如果我要比较第二column和第三column的mean的大小，该如何处理？我用SAS
Thanks a lot.
*****************************************
修改一下，我并不是单纯需要找到mean
我需要做1-way anova or t-test or non-parametric比较这两个group |
|
a****g
发帖数: 8131 | 48 修改一下，我并不是单纯需要找到mean
我需要做1-way anova or t-test or non-parametric比较这两个group |
|
a****g
发帖数: 8131 | 49 修改一下，我并不是单纯需要找到mean
我需要做1-way anova or t-test or non-parametric比较这两个group
thanks |
|
c*****t
发帖数: 1712 | 50
spearman 应该是 non-parametric 版本的 pearson |
|
