z****e 发帖数: 702 | 1 两个二项分布:X - B(p1,n1),Y - B(p2,n2);
H0是p1=p2.置信水平是alpha;
要求用统计量D=X/n1-Y/n2,和S=(X+Y)/(n1+n2),
头一个用正态分布即可: |D/sqrt(p1(1-p1)/n1 + p2(1-p2)/n2)|>Z_\alpha/2。对吧?
另外问这个置信水平是exact还是asymptotic这个怎么回答?
还有用S的话该用什么分布?
Thx!!! |
|
t**c 发帖数: 539 | 2 Based on CLT,both D and S can be approximated using normal distribution when
n1, n2 is large. So the CIs are asymptotic.
They have the same variance; mean(D) = 0 (if H0 is true); mean(S) = p1 + p2.
So I am confused here how you can use S to test H0. |
|
B****n 发帖数: 11290 | 3 statistical inference
probability theory (measure theory based)
如果有開asymptotic statistics 也可以聽聽
其他的就要看你自己研究的方向了
相對於數學來說 統計的理論比較雜 這是被統計這門學科是被data的形式所影響而決定
要發展什麼統計方法或是理論 是要看有什麼實際問題要分析 因此具體用到什麼數學
其實很難說 |
|
w****r 发帖数: 748 | 4 nonparametric还是比较理论的,至少要探讨各种estimation的asymptotic properties
吧。 |
|
s******h 发帖数: 539 | 5 如果你问的是square root of sample variance of A_i/(\bar B), 那么可以直接用
sample variance的公式。
但我相信你问的可能不是这个,而是var(A_i/(\bar B) 的 estimator. 特别是你只有
10个数据的情况下,asymptotic results may not hold (otherwise, since \bar B \
to E(B_i) w.p.1, you can show that var(A_i/(\bar B_n)) \to var(A_i)/E^2(B_i)
).
For finite sample, If A and B are independent, then var(A_i/\bar B)
= var(A_i)*E(1/\bar B^2). You can estimate var(A_i) using its sample
variance and for E(1/\bar B^2), unless you have distribution assumptions (e.
g.... 阅读全帖 |
|
|
t*******t 发帖数: 633 | 7 http://en.wikipedia.org/wiki/Logrank_test
According to WIKI, the distribution of the test statistic is asymptotic
normal, which means it's symmetric, so the one sided p-vlaue=two sided p-
value/2. |
|
s******h 发帖数: 539 | 8 I assume that your subjects are randomly assigned to the two designs (i.e.,
[A B] and [B A]). If the sample size of the two groups are the same, say n,
then 1) and 2) are essentially equivalent, imo.
To see this, assume that X1, ..., Xn are the Bernoulli responses of choosing
A in design [A B], and Y1, ..., Yn are the response of choosing A in design
[B A]. Assume that E(X1) = P1 and E(Y1) = P2. You want to test if P1 = P2.
So I think 2) makes more sense.
Method 1) is then {(X1 + ... + Xn) + (n... 阅读全帖 |
|
w******e 发帖数: 142 | 9 水木上看见的推荐列表,小多,呵呵。
入门:
Pattern Recognition And Machine Learning
Christopher M. Bishop
Machine Learning : A Probabilistic Perspective
Kevin P. Murphy
The Elements of Statistical Learning : Data Mining, Inference, and Predictio
n
Trevor Hastie, Robert Tibshirani, Jerome Friedman
Information Theory, Inference and Learning Algorithms
David J. C. MacKay
All of Statistics : A Concise Course in Statistical Inference
Larry Wasserman
优化:
Convex Optimization
Stephen Boyd, Lieven Vandenberghe
Numerical Opti... 阅读全帖 |
|
c********h 发帖数: 330 | 10 Asymptotic equivalent吧
一个应该是score test,另一个是likelihood ratio test taylor展开,渐进分布都一样 |
|
c********h 发帖数: 330 | 11 getdown2的大部分我都同意,不过我觉得可能关系不是linear的,因为冬天需要暖气,
夏天需要空调吧,所以很高温度和很低温度,能耗应该都比较大。而且我觉得可能两个
tail都会有个asymptotic limit,比如说夏天气温高于某个threshold了,空调就会7am
-7pm一直开着?这个跟你数据有关系。
可以先画图看看是啥关系,如果观察就是linear,那linear regression就可以了。如
果有很明显的nonlinear pattern,就再考虑。 |
|
h*********n 发帖数: 278 | 12 Used the method from below and obtained Kolmogorov-Smirnov Test:
KS Two-Sample Test (Asymptotic):
KS 0.25
KSa 227.28
D 0.5
Pr > KSa: <.0001
Well below 0.85. I guess my model may even be described as not
fitting well at all? |
|
h*********n 发帖数: 278 | 13 我前面有个贴里有run出来的ks,是0.25. 跟0.5-0.6好像差得有点远,这个是怎么回事?
另外,如果从auc就能算出来ks,那为啥还要同时看这俩呢?
KS Two-Sample Test (Asymptotic):
KS 0.25
KSa 227.28
D 0.5
Pr > KSa: <.0001 |
|
j******4 发帖数: 6090 | 14 log-likelihood就是把likelihood function 用log转一下,方便计算。
至于likelihood么,是一种inference的方法。
大概可以认为likelihood越大越好,有一种著名的estimator就是基于maximum
likelihood来的,就是maximum likelihood estimator (MLE),这种estimator有非
常好的属性,比如asymptotically unbiased,invariance等等。 |
|
t********y 发帖数: 166 | 15 我是指数理统计里的某些领域。
nonparametric asymptotics, inference on processes, empirical processes...没
有一定的数学基础,这些方向真做不动。 |
|
w*******9 发帖数: 1433 | 16 I was too quick to say any number beyond c(150, 5) is not helpful -- it's
wrong.
Method 3) in the paper you refer to is the most rigorous approach to
calculate the minimum sample size needed in order to achieve certain length
in confidence interval, or equivalently, certain power in hypothesis test.
See for example http://sphweb.bumc.bu.edu/otlt/mph-modules/bs/bs704_power/BS704_Power4.html
If I understand your question correctly, you're trying to estimate the
probability (say p) that the mean of... 阅读全帖 |
|
m****s 发帖数: 18160 | 17 ☆─────────────────────────────────────☆
baiji (白暨豚) 于 (Mon Apr 26 00:21:55 2010, 美东) 提到:
[相关信息]:请问申请前是否已阅读:1)《站规》2)《版务操作简易手册》
3)《版务操作注意事项》?是
[申请ID]:baiji
[申请版面]:Running版
[申请职务]:版主
[版务经验]:有 (非mitbbs)
[申请纲领]:活跃气氛,维持秩序,整理精华,方便交流
☆─────────────────────────────────────☆
julianzhong (Training started) 于 (Mon Apr 26 01:15:18 2010, 美东) 提到:
支持。
☆─────────────────────────────────────☆
wolfstar76 (天狼星) 于 (Mon Apr 26 01:28:21 2010, 美东) 提到:
support
☆─────────────────────────────────────☆
t... 阅读全帖 |
|
r******n 发帖数: 351 | 18 非常感谢你的提醒。这是从wiki里摘的一段话:
http://en.wikipedia.org/wiki/Binomial_proportion_confidence_int
The central limit theorem applies poorly to this distribution with a sample
size less than 30 or where the proportion is close to 0 or 1. The normal
approximation fails totally when the sample proportion is exactly zero or
exactly one.
这里就是 sample proportion is exactly zero 的情况。 可以用exact或着wilson,
但是不能直接用Asymptotic normal。 我猜这是面试官想考察的内容。不知道如果用
normal近似的话怎么计算。
性。 |
|
h*****7 发帖数: 6781 | 19 EM没有理论论证全局最优的概率,to the best of my knowledge
记住一条,EM这类方法,不属于概率论范畴,属于随机过程,因为它用了指示器(可参
考随机过程计算方法算法导论之类的),所以很难给出理论确界和最优解概率。一般说
无限趋近最优解。
在GMM条件下,倒是有人系统测试过EM的解和最优解有多远
另外MLE本来就没道理的,就是通俗说法的屁股决定脑袋。就算解出全局最优,对参数
估计也是imperfect solution,至于后续Wilk's theorem,也是asymptotic的,所以LZ
就别要求太高了
LZ这种情况,没法用时序或者频域作分析,用EM是比较理想的
靠,说了一堆,回头看发现对LZ没啥帮助,还是疑似我老马甲的YueJia讲得好 |
|
|
|
Z**r 发帖数: 346 | 22 他们俩已经asymptotically converge to 100% MALE le
but you, i am confused......... |
|