由买买提看人间百态

topics

全部话题 - 话题: h0
首页 上页 1 2 3 4 5 6 7 下页 末页 (共7页)
y********0
发帖数: 638
1
来自主题: ChineseMed版 - 再请教一个统计检验的问题
单纯从统计结果来讲,即使没有第二个实验结果。 这个p=0.0099属于边缘pvalue,下
结论的时候也要非常小心,不能随意得出推翻H0的结论。 这两个结果本身都没什么错
,属于统计结果,从某种程度上代表了得出某种结论的风险。
这个只是提供给决策者的其中一种依据,决策者面对的绝不不可能就这么一个单一的证
据。我们就说一个普通的医生吧,即使在pvalue不显著的情况下,他也可能根据其他的
依据,譬如统计上的effect size来判断某一种药的使用结果,当然也可能跟他承担的
后果不严重有关。
T*******I
发帖数: 5138
2
来自主题: ChineseMed版 - 统计的学问
“t-检验的零假设H0是两个总体的均值绝对相等,不能有丝毫差别。”这句话表明作者
甚至连假设检验在检验什么都没搞清楚。这个是在胡说八道。
假设检验中的t-检验是在检验两个样本均值的绝对差中由抽样变异引起的随机误差发生
的可能性是否显著,因为这个绝对差有且只有两个来源:一个是由于总体中个体的随机
变异而引起的抽样随机误差,另一个就是由定义两个总体的不同子系统属性所引起的系
统误差。但是,我们的问题是无法通过一次抽样直接测量这两类误差的绝对大小,这才
有了用概率来判断两类误差在此次抽样研究中发生可能性的假设检验问题。
还有就是拿关于疫苗回顾性比较研究的例子归纳出“文科统计”的概念,又是一个绝对
的无知或偏见。
i******d
发帖数: 4
3
来自主题: Computation版 - help needed badly! do appreciate.
I was trying use matlab to slove a complicated system of nonlinear algebraic
equation by function Solve. But it shows explicite solutions couldn't be
found. (for some parameters,it gives me solution in 5 minutes.but if i change
the parameter, it may takes forever for matlab or no explicit solution)
my equations are below. Unknowns are x, y and z, others are parameters.
Ssmith=solve('R*g*S*y/x^3-a*(1+0.5*R*g*y/x^3)/(b+R*g*y/x^3)-alpha*z/x^2=0',...
'0.0016*(647.68*z-1)*h0*x/(1+0.0024*(647.68*z
k***s
发帖数: 370
4
来自主题: EE版 - H.221 recommendation
H.221, Frame structure for a 64 to 1920 kbit/s channel in audiovisual teleservices
The purpose of this recommendation is to define a frame structure for audiovisual teleservices in single or multiple B or H0 channels or a single H11 or H12 channel which makes the best use of the characteristics and properties of the audio and video
encoding algorithms, of the transmission frame structure and of the existing recommendations. It offers several advantages:
It takes into account recommendations such
h****f
发帖数: 24
5
Consider the hypotheses
H0: q=0 H1: q =/= 0
and the observations
zi=q+wi i=1,...,n
with wi zero-mean jointly Gaussian but not independent.
denoting w=[w1,...,wn]'
one has the covariance matrix (assumed given)
E[ww']=P
For the above
1. specify the optimal hypothesis test for false alarm
probability a
2. solve explicitly for n=2,
P=[1 0.5] (this is a matrix)
[0.5 1 ]
and a=1%
Thanks!
t**i
发帖数: 688
6
来自主题: Mathematics版 - 请教一个Hypothesis Test的问题
H0: < 100;
H1: =100 or > 100.
m*****s
发帖数: 9
7
来自主题: Mathematics版 - ANOVA table是什么东西啊
家里的餐桌吧,两个菜以上,西红柿蛋,蛋番茄,番茄蛋,蛋西红柿,味道可能不一样,H0假
定idd高斯分布.
a********e
发帖数: 3771
w*s
发帖数: 25
9
来自主题: Pharmaceutical版 - 申请PK/PD咨询
可念的小宝宝








P堑6!v;雩x鶝<臱Q巰揜*6{
柶慆攨狫溣
窭狧Dj7
M8悾亜袻Bē=i1±搔砱g姁(ZlDzuf蔅跉莛;P喨伢H0:謭匍@
)(
9鉃憸鎿vD蕗s迶鷉彻hi蝳 NNi暍%i舩湫貞
B
P襉Z@LQ襚(4嵵奰(R))PE
)@QH僇E!墡6灖'爍S挟燞悱GT劼奲PK@聤)(R
%0
(
Z% )h (燗E Q@ E-)剽惷#Q@劌燾咵.) B)葱倰E
I!(:讯淪qNAE0
(藲R9c4凱QLQJ(訲iτB
)I@(燼E鮮:簸R (燗E
( (P0
(
( (燗EP0
(Q@聤((P0
(Q@聤((((((((((((JZ(JZJ(
(拃
(
(
(
(
EPEPEPEPEPE~Q@QL(((((((((厗
0Rp*`鮦慟S牞軍Hq7(wJB丐I
閬&碕8悾榠
8fI4斄巀)"E镹╡¨
1@mh塩晄
0)neu
p********a
发帖数: 5352
10
来自主题: Statistics版 - [合集] 问个test的问题
☆─────────────────────────────────────☆
OutofSky (天外) 于 (Tue Sep 18 10:16:13 2007) 提到:
有两组数据,分别在四种条件下
c1 c2 c3 c4
G1 x1 x2 x3 x4
G2 y1 y2 y3 y4
要测试H0:G1和G2没有显著区别
H1:。。。。有。。。。
想了几种方法都觉得不合适。
1,paired t-test: 要求normality,可x,y都不能保证
2,Wilcoxon Rank Sum: 好像是测试stochastically larger
的,不符合
3,Wilcoxon Signed Rank: 不知道x和y的分布情况,数据少
又不能近似normal
4,kai-square:不知道expected value怎么算
所以有点不知怎
m****d
发帖数: 331
11
Suppose that the it is a multiple linear regression model and constraints on
the estimated parameters are: beta1+beta2+...+beta(n) >=0
When I have done the regression, and want to do the hypothesis testing, what
statistics I should use?
H0: beta1=beta2...=beta(n)=0
H1: beta1+beta2+...+beta(n) >0
Now I use F test, comparing the unrestircted model and restricted model, to
test if the results are statistically significant...if the statistic >
critical value of F at 0.05, reject the null hypothesis.
z*********o
发帖数: 541
12
来自主题: Statistics版 - hypothesis test
let X1, X2,---X8 be a random sample of size n=8 from a Poisson distribution
with mean μ.Rejcet the simple null hypothesis H0:μ=0.5 and accept H1:μ>0.
5 if
the observed sum ?xi ≥ 8.
(b) find the power function γ(μ) of the test as a sum of Poisson
probabilities.
怎样求解power function? 谢谢
h*t
发帖数: 187
13
各位达人:
偶现在用R package gee 做一个GEE的regression,比如log(Y)=beta0+x*beta1,如果
我想测试H0:beta1=1,contrasts应该怎么写?
这个package里面没有给contrasts的例子,自己试了几次都不对。
另外,对同样的这组数据,如果用R做Generalized Linear Mixed-effected Model,用
哪个package的哪个命令比较好?
c********8
发帖数: 5
14
来自主题: Statistics版 - 各位大侠请帮忙
我有一个线性回归的问题,请高手教我。
Model; Y=β0 + β1X1+ β2X2 + ε.
I want to test the hypothesis:
H0: β1 + β2= 1 vs Ha: β1 + β2 not equal to 1.
What test statistic I should use and what is its distribution?
c********8
发帖数: 5
15
来自主题: Statistics版 - 各位大侠请帮忙
那和
test H0: β1=β2= 0 vs Ha: β1,β2 not equal to 0
有什么区别呢?F(df?,df?)
W**********E
发帖数: 242
16
来自主题: Statistics版 - 问个P-VALUE的问题
Likelihood ratio test example:
H0: var1=0 likelihood ratio chi-square test statistics is 6.6976 with df=1.
So 1-pchisq(6.697,1)= 0.0093.
However, this example gets a p-value of 0.0047 by dividing 0.0093 with 2.
I am not sure why we need to divide a 2 here. Variance is always positive
and Ha: var1>0 justifies a one tailed test only, right?
Any help will be appreciated.
p**5
发帖数: 2544
17
来自主题: Statistics版 - two special cases in paired T test with SPSS
I have two questions in paired sample T test in SPSS:
1. For single sample T test, I can input a test value, but for paired sample
T test, there is no such option.
If I want to test: H0: u1-u2 = 20, Ha: u1-u2 > 20, how can I specify 20 when
running this test?
2. SPSS only does two-tailed test; if I want to test one sided hypothesis
and want to calculate the 95% upper confidence bound, how to do it?
calculate a 90% C.I. and use the upper limit as 95% upper bound?
3. How to get p-value for one si
s*****n
发帖数: 2174
18
来自主题: Statistics版 - 什么是统计 - 兼谈找工作
我不是很喜欢conditional prob 那个定义.
我认为最好的p-value定义, 是把p-value定义成一个假设检验族其中否定H0的那个子集
里面Type II error 的下确界.
I***e
发帖数: 1136
19
来自主题: Statistics版 - 什么是统计 - 兼谈找工作
I don't understand your second interpretation: given a hypothesis test, p-
value is a function of the observed data X. However, both the rejection
region of H0 as well as the corresponding inf of type II error doesn't have
anything to do with X. They are a property of the test itself. How can that
give you the p-value, which is supposed to change for each sample you draw?
-iCare-
I***e
发帖数: 1136
20
来自主题: Statistics版 - 什么是统计 - 兼谈找工作
Of course, we have to have a test to talk about p-value. I meant p-value
depends on the observed data at least partially.
Inputs to p-value are:
1. The test hypothesis.
2. Rules of acceptance/rejection of H0.
3. Input X.
You still haven't answered why the interpretation you have doesn't depend on
X at all.
-iCare-

果换
X1)
的.
c******n
发帖数: 590
21
来自主题: Statistics版 - 什么是统计 - 兼谈找工作
啊,不算严格的conditional吧,因为condition的那一部分只涉及到假设的(H0)的参数
,和统计量无关。
W**********E
发帖数: 242
22

我觉得你可以考虑你的模型3
1。 人-随机150个
2。 人的眼睛(3个因素:左,右,双)固定
3。 每个位置5个时间点。固定
1>2>3
类似于一个SPLIT-SPLIT PLOT 设计。 人分3个位置,每个位置又分5个点。但你可
以用其他COVARIANCE STRUCTURE 来模拟WITHIN EYE POSITION的COVARIANCE。
3. proc mixed;
class eye time subid;
model score=eye time time*eye/ddfm=kr;
***subid*eye=error term for testing H0: eye=0
random subid subid*eye;
***subject=subid*eye is the unit within which time factor is nested;
repeated time/ type=ar(1) subject=subid*eye;
run;
i********f
发帖数: 206
23
来自主题: Statistics版 - 请教用R做factor analysis
chi-square不是越小越好,
越小的话,拒绝H0
当然chi-square只是一个参考
你这个Factor1好像下面的几个变量对应的值都比较大,所以factor1主要就是代表下面几
个变量
对应的Factor2主要代表上面的几个变量
c**********e
发帖数: 2007
24
来自主题: Statistics版 - How to compare two parameters.
The fitted models are
Y = a1 + b1*X + c1*Z + error
and
Y = a2 + b2*X + c2*W + error
I want to test H0: c1=c2.
D******n
发帖数: 2836
25
来自主题: Statistics版 - test whether it's LINEAR relationship
a lot of tests are like that , assume H0 is true and caculate the probabilit
y of seeing the deviation observed.
t**g
发帖数: 1164
26
来自主题: Statistics版 - 请教F test和t test区别?
小弟统计粗通皮毛
问个可能弱智点问题:
比如比较3个treatments是否有区别:
H0: theta1=theta2=theta3
t test和F test都可以比较
哪个比较好呢?
两个test区别在哪?
多谢!
p**********l
发帖数: 1160
27
What we learnt during a lecture was to use sphericity test to test if the
within subject variance-covariance matrix has a type H ( Huynh & Feldy)
structure, or others call it HF structure.
Covariance matrix is of type H iff its quadratic form with an orthogonal
contrast matrix.
H0: = sigma^2 I.
Ha:  = unstructured form.
If the test is nonsignificant, use univariate test for within-subject
effects and since they are more powerful then the multivariate test.
If the test is signifi
V*********n
发帖数: 198
28
来自主题: Statistics版 - 问一个统计的问题
我也觉得是poisson。
有很多信息已经包含在context里了,就是说除了方法不一样以外,其余的变量都一样
,没有
confounding variables。在这样的情况下,如何比较2种方法。
我觉得就是两个poisson distribution 的 lamda。
H0: lamda1 = lamda2
Ha: lamda1 != lamda2
至于test statistic是什么,你可能需要用normal approximation,然后就是two-
sample t
test了。
T*******I
发帖数: 5138
29
来自主题: Statistics版 - 问一个血压数据分析的问题
动物的血压之所以能被维持在一定比较稳定的水平上是由于体内有一个血压调节机制。
一个研究者要想知道高盐饮食喂养是否会显著地改变这个内在机制,必须使得这个影响
对既有机制产生足够的影响才能被统计测量到可能的期望结果。
我认为“两周的喂养时间是否确定地影响了老鼠的内在血压调节机制”是一个在“统计
上”不确定的命题。也就是说,如果高盐确实影响血压、而喂养时间足够长,则结果可
能就是阳性的;反之,如果喂养时间不够,即使已知高盐影响血压,但由于时间不够长
,很有可能也会导致统计测量的结果是阴性。当然,如果高盐不影响血压,则即使喂养
时间足够长,结果也会是阴性的。
LZ之所以期望一个“两周喂养”的阳性结果,是因为他仅仅在“两周喂养足够了”的
前提假设(assumption)下提出了自己的“高盐是否影响血压”检验假设(hypothesis)。 然而,问题是,如果结果是接受H0,则并不能说明“高盐不影响血压”,而只能是
“两周高盐喂养不足以影响血压”。
上述逻辑应该非常简单了吧。

study--I would assume that the blood pressure will be the end
h***i
发帖数: 634
30
来自主题: Statistics版 - 问个SAS GLM 和 Reg
Reg里面有以下语法来做一个F test H0:beta1=0 & beta4=0;
model y=x1-x5;
test x1=0,x4=0;
请问GLM里面对应的语法是什么啊
l**********9
发帖数: 148
31
想要检验H0:a'Ca=b,其中a和b已知,C是某分布的方差矩阵,一般应该用哪种检验方法
l***a
发帖数: 12410
32
来自主题: Statistics版 - how to test if two groups are same?
my understanding is paired t test is to test if two groups are different (e.
g. if a treatment works), by test to reject H0: a=b.
if I want to test if the two groups are the same, what kind of test should I
employ?
further more, t test is too sensitive if the sample size is real large (1mm
paired records). any test can overcome this over sensitivity issue?
h******3
发帖数: 190
33
I am confused about the two tests.
Does the degree of freedom of both the tests follow the following?
degree of freedom = number of unknown parameters in H1- number of unknown
parameters in H0
d******g
发帖数: 130
34
I think the large sample will make the test with -2log(likelihood ratio)
more valid. Since we take the log and double it to yield an approximate chi
-squared sampling distribution and we use LARGE sample to approximate this
chi-square distribution.
Also when H0 is false, the large the sample size, the larger the -2log(likelihood ratio).
v*****a
发帖数: 1332
35
来自主题: Statistics版 - 海量SAS data的处理
跟您老无法交流了。。
您老就报着那个无敌的“H0”和“FAIL TO REJECT”一辈子吧
千万别出去搞工程,祸害人。。

149500000000000毫米,可以写成1.495(亿公里)等。所以,简化纪录方式可以减小你
的那种超级超级超级超级超级超级超级……超级大的数据。
a*****3
发帖数: 601
36
来自主题: Statistics版 - 求助!! MEAN test 问题!
看了半天,到底想测啥? h0是什么?
D*********2
发帖数: 535
37
来自主题: Statistics版 - Google 面经
二面,没啥好说的,输的心服口服,实力不行。桑心那~准备专心写thesis了。准备tech面试真是体力活。
1. what are the modeling techniques you have used?
2. what's the difference btw fixed effect model and mixed model? fixed effect Vs. random effect, how would you choose in practice/in your projects.
3. what are the techniques you use in longitudinal models/ survival models.
4. lots of questions about GLM: error structure, link function, estimation method, etc.
5. experience about SAS and R
6. experience of data manipulation
7. say you built a ... 阅读全帖
l****u
发帖数: 529
38
来自主题: Statistics版 - 向版上大牛请教个问题
The null hypothesis of the contrast is slopeA1=slopeA2. the significant p_
value means the H0 does not hold.
R*********i
发帖数: 7643
39
请教分析方法:
一组约30个病人的数据,只有一种治疗方法,结果是阳性或阴性。阳性结果百分比是p
。H0: p=0.5, H1:p>0.5
因为样本小,一般情况下我们会用binomial分布,求exact p-value。现在的问题是这
30个结果中有的是同一个病人前后两次重复治疗的结果(共有约20个病人)。
我们计划的第一种方法是用GEE model(只含intercept)来解决一个病人重复取样点问
题。
下面是我的问题:
1.另外想请问一下有没有仍然用exact方法,但是根据重复取样的设计方案调整p-value
和confidence interval的?
2.看到一些关于Kish Design effect的方法,在这里适用吗?如果可行,ICC怎么估算?
问题可能挺笨的,因为Kish Design Effect以前没接触过。
先谢谢大家!
x**********0
发帖数: 163
40
来自主题: Statistics版 - Null Hypothesis
I have seen a similar one in a textbook.
There are two identically appearing bowls of jelly beans. Bowl 1 contains 60
red and 40 black jelly beans, and bowl 2 contains 40 red and 60 black jelly
beans. Therefore, the proportion of red jelly beans, p, for the two bowls
are
Bowl 1: p = 0.6
Bowl 2: p = 0.4.
One of the bowls is sitting on the table, but you do not know which one it
is.
You suspect that it is bowl 2, but you are not sure.
To test your hypothesis that bowl 2 is on the table, you sample... 阅读全帖
M******8
发帖数: 22
41
One suggested test for trend in a univariate population is as follows:
Group the unordered scores into nonoverlapping groups of 3 adjacent
observations. The test statistic D is the number of monotonic (either
increasing or decreasing) triples. Assume for now that N is always a
multiple of 3 (no partial triples).
Example (triples are underlined)
33, 16, 20 61, 44, 38 29, 40, 56, 22, 43, 31
no order decrease increase no order D = 2
H0: random ordering (independent, rando... 阅读全帖
p*******r
发帖数: 1951
42
来自主题: Statistics版 - 请教Reapeated measure mixed model
首先要明确你要做什么比较?看你说的意思你的null hypothesis是不是
H0: treatment effect = time2 - time1 = 0 ?
如果是,那你的dependent variable 完全可以用 effect = time2 - time1。而不是单
独的time1, time2。
其次,你的数据显示不同的subject有完全不同的时间点, 看你的SAS程序里面也根本
没有用到时间变量week。所以你想做的不是longitudinal data analysis?你只是想比
较不同treatment, pre-, post- response的差别。这样的话用repeated measure
ANOVA就成了。
很久没有用过mixed model了。说的不对的地方大家指教。
c**********4
发帖数: 12
43
来自主题: Statistics版 - two tail hypothesis about population mean
dummy question:
H0 = u
Ha <> u
sample mean x and test statistic z
why rejection region as to z(alpha/2)? I think that it should still be alpha.
z > z(alpha), 1-alpha confident that x>u
z< -z(alpha), 1-alpha confient that x
A*******s
发帖数: 3942
44
来自主题: Statistics版 - one question about variable selection in SAS
又学习新东西了,谢谢大牛
包子sent
不过突然有个疑惑,
比如说在logistic里面
我想test H0: beta2=0, beta3=0;
这个collection是直接做一个likelihood ratio test拉倒,还是做俩wald tests with
Bonferroni adjustments ?
a********s
发帖数: 188
45
来自主题: Statistics版 - Score test in Logistic regression
Just be curious if SAS has options to do Score test for the null hypothesis
(say, H0: beta2 = beta3 = 0) suppose the simple Logistic model includes
coefficients parameters beta0,beta1,beta2,and beat3.
f**s
发帖数: 185
46
需要应用3-equation simultaneous system在panel data的回归分析中,请教版上的大
牛。急求,愿意付报酬。SAS例如:
proc model data=tempdone;
instruments a b c d e f;
X=h0+h1*Y+h2*Z+h3*taadj+h4*leverage+h5*a+h6*b;
Y=g0+g1*X+g2*Z+g3*taadj+g4*leverage+g5*c+g6*d;
Z=b0+b1*Y+b2*X+b3*taadj+b4*leverage+b5*e+b6*f;
fit X Y Z/ no2sls gmm vdata=vblkdiag vardef=n nogengmmv;
但这个应该是对cross sectional的,有大牛能指点下如何做pandel data的吗?有
module或者codes吗? 还有一点,其中X是0,1变量; Y,Z是非连续数,如1,2,3...
..23(是个ordinal的数)。
z******n
发帖数: 397
47
对于第一个问题,从实用的角度来讲,是不是这样描述问题更有意义?
假设存在某个CTR的值,例如5%,超过这个值的广告被业界认为是成功的,然后作如下
检验:
H0:CTR_i >= 5%
这个检验很简单,大样本时可以用正态近似,小样本时用二项检验查表,每个广告算个
p-value,然后用bonferroni correction或者其他多重检验校正,这样可以得到一批“
成功”的广告列表
对于比较多个proportion间的差异,理论上可以用Marascuillo test,但是这个检验的
零假设太无聊,更遑论chi-sq test for homogeneity of proportion了
对于这个方法选出的广告,再做两两比较(还要做多重检验校正),这时的问题是得出
的结论可能是矛盾的,比如A比B好,B比C好,但A不比C好之类的。但其实问题不大。
p****e
发帖数: 165
48
哇,大牛出山了!感谢回答。
我有个疑问,就是H0假设后,用什么检验来求每个广告的p-value? t-test?
谢谢啦
p****e
发帖数: 165
49
在工作中经常会遇到做产品的AB test (or split test)来比较不同的结果,我准备用z
-test进行比较,H0设为Control, A,B,C组没有差别,以下为数据:
说明一下,要比较的内容为conversion rate (转化为注册用户的访客数量/访客数量)
,Control为原始网页设计,Test A/B/C分别为不同的测试网页
Treatment | # of Visitors | # of Registered Visitors | Conversion Rate |
Z-
score
Control | 182 | 35 | 19.23%| N/A
Test A | 180 | 45 | 25.00%| 1.33
Test B | 189 | 28 | 14.81%| -1.13
Test C | 188 | 61 | 32.45%| 2.94
Z-score = (Pa-Pc)/square root of [Pa(1-Pa)/Na + Pc(1-Pc)/Nc] (文本不知道怎
么输入数学公式,大家见附件中的截图吧)
这个例子就是Te... 阅读全帖
z****e
发帖数: 702
50
来自主题: Statistics版 - 假设检验的问题
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!!!
首页 上页 1 2 3 4 5 6 7 下页 末页 (共7页)