K****n 发帖数: 5970 | 1 【 以下文字转载自 Computation 讨论区 】
发信人: KeeVan (Kevin), 信区: Computation
标 题: probit regression一问
发信站: BBS 未名空间站 (Fri Aug 22 21:53:32 2008)
请问有没有现成的教材把maximum likelihood的导数求出来的? 我想对一下,网上居然
google不出来... 我不太放心matlab里的glm方程之类的,那个training的时候震荡比较
大.
另外如果对probit方程的参数设一个gaussian prior,然后求bayesian的
P(data)=Integrate(P(data|parameter)*P(parameter),over parameter)
好像这里用probit方程作P(data|parameter),用Gaussian作P(parameter),在optimize
bayeisan likelihood的时候比较好算?不知道有没有人已经算过?又google不出来...
谢谢! |
|
w********h 发帖数: 17 | 2 其实在binary data regression中还有其他的links:最常用的要数
identity: Y=\beta_0+\beta_1 X
Probit: Probit(Y)=\beta_0+\beta_1 X
和
Complementary log-log:log(-log(1-p))=\beta_0+\beta_1 X.
不同的links中对\beta_1的解释不同,用处也不同,要小心区别.Identity显然
不是一个好的link,为什么呢?(Ronna,北大数学系的学长给你布置的第一道家
庭作业,免得你在北大班猛灌 :))Probit link常用在bioassay的数据分析,而
C-log-log则常用于传染病的数据分析. |
|
x********u 发帖数: 411 | 3 【 以下文字转载自 Economics 讨论区 】
发信人: xiaoxiaoyu (雨潇潇~玉暖香融), 信区: Economics
标 题: probit 为什么给出的时 z statistic?
发信站: BBS 未名空间站 (Sun Apr 19 19:14:51 2009), 转信
问个很白痴的问题~ 为什么在 stata 里 probit y x,给出的是 z statistic ;而如
果我用 reg y x,给出的就是 t statistic ?
正在处理一个简单的问卷调查的数据,需要用到 probit 模型,如果 stata 给的是 z
statistic,那是不是说决定 significance level 的时候用 z statistic 就可以了?
不知道说清楚了没有,先谢谢~~ :P |
|
x********u 发帖数: 411 | 4 【 以下文字转载自 Economics 讨论区 】
发信人: xiaoxiaoyu (雨潇潇~玉暖香融), 信区: Economics
标 题: probit 为什么给出的时 z statistic?
发信站: BBS 未名空间站 (Sun Apr 19 19:14:51 2009), 转信
问个很白痴的问题~ 为什么在 stata 里 probit y x,给出的是 z statistic ;而如
果我用 reg y x,给出的就是 t statistic ?
正在处理一个简单的问卷调查的数据,需要用到 probit 模型,如果 stata 给的是 z
statistic,那是不是说决定 significance level 的时候用 z statistic 就可以了?
不知道说清楚了没有,先谢谢~~ :P |
|
g*******i 发帖数: 258 | 5 The difference is minor. That's why in real applications, logistic is
dominent, except some econometricians would choose probit, which I don't
understand.
For multinomial case, things become different. The probit setting is more
flexible, but its estimation is much more time consuming. This is the reason
why you still don't see many real applications are done with probit. |
|
g*******i 发帖数: 258 | 6 are you saying probit with Bayesian estimation proposed by Chib?
The biggest problem with probit is it needs to deal with truncated normal,
which bothers researchers for decades. Even now truncated normal takes lots
of time. In real applications, people rarely use probit, even though it is
free from the independence of irrelevant alternatives restriction. Most
people, including some well developed business packages, use logistic and
ignore this IIA assumption in trade of efficiency.
convenient |
|
K****n 发帖数: 5970 | 7 嗯,谢谢。
我把bishop的那本machine learning翻过了,没有现成的。自己推导了还是想和标准的
答案对对,因为我觉得那个probit的ML解应该是早被好多人做过了,可是网上居然找不
到,自己有点儿心虚。
Bayesian probit好像用gaussian prior就可以,不过本来还是想找现成答案:P 因为我
只是要用,推导似乎应该是有牛人专门搞?
哈哈,我自己求好了。 |
|
K****n 发帖数: 5970 | 8 请问有没有现成的教材把maximum likelihood的导数求出来的? 我想对一下,网上居然
google不出来... 我不太放心matlab里的glm方程之类的,那个training的时候震荡比较
大.
另外如果对probit方程的参数设一个gaussian prior,然后求bayesian的
P(data)=Integrate(P(data|parameter)*P(parameter),over parameter)
好像这里用probit方程作P(data|parameter),用Gaussian作P(parameter),在optimize
bayeisan likelihood的时候比较好算?不知道有没有人已经算过?又google不出来...
谢谢! |
|
x********u 发帖数: 411 | 9 问个很白痴的问题~ 为什么在 stata 里 probit y x,给出的是 z statistic ;而如
果我用 reg y x,给出的就是 t statistic ?
正在处理一个简单的问卷调查的数据,需要用到 probit 模型,如果 stata 给的是 z
statistic,那是不是说决定 significance level 的时候用 z statistic 就可以了?
不知道说清楚了没有,先谢谢~~ :P |
|
k*m 发帖数: 6 | 10 为什么不报告se呢
问个很白痴的问题~ 为什么在 stata 里 probit y x,给出的是 z statistic ;而如
果我用 reg y x,给出的就是 t statistic ?
正在处理一个简单的问卷调查的数据,需要用到 probit 模型,如果 stata 给的是 z
statistic,那是不是说决定 significance level 的时候用 z statistic 就可以了?
不知道说清楚了没有,先谢谢~~ :P |
|
f*******r 发帖数: 257 | 11 The reason that probit or logit reports z stat is that they are maximum
likelihood estimators, the asymptotic distribution of it is normal. "reg y
x" is an ols model. When the assumptions of OLS hold, the distribution of
the stat (b/se(b)) conforms to t distribution. However, if the assumptions
do not hold, then that statistic does not conform to t distribution anymore.
Despite that, stats packages still report them as t stat. The conclusion
is: you are safe to use z-stat for probit or t-s |
|
L*****2 发帖数: 66 | 12 面世时被问到这个问题, 除了link funciton 不同之外,还那些不同,
问到什么时候用logit什么时候用probit,特别是如果response是非常rare的事件,应该
用probit还是logit?
谢谢 |
|
f*******r 发帖数: 257 | 13 In practice, they are equally good. In theory, if there are a lot of
observations in the tail, it may be better to use logit... The coefficient
estimated by probit is about .6 of that by logit.
Rare event case should be estimated by something else, neither logit nor
probit is appropriate:
http://www.stanford.edu/~tomz/software/software.shtml |
|
p********a 发帖数: 5352 | 14 ☆─────────────────────────────────────☆
killniu (killniu) 于 (Mon Nov 21 14:05:46 2011, 美东) 提到:
多变量 单变量不都是一个proc么?
☆─────────────────────────────────────☆
pepper1982 (pepper1982) 于 (Mon Nov 21 14:11:03 2011, 美东) 提到:
it's not simple multivariate regression.
There are more than one probit regression and the disturbances are
correlated.
☆─────────────────────────────────────☆
killniu (killniu) 于 (Mon Nov 21 14:23:18 2011, 美东) 提到:
repeated measure啊
proc genmod 带 repeated 选项
这个是你要的么?
☆... 阅读全帖 |
|
c********d 发帖数: 253 | 15 That depends on how you treat your error term. Probit model's error term is
normally distributed. logistic model can also be written as the latent
variable form of probit model, but its error term is logistic distributed. |
|
l*g 发帖数: 46 | 16 用于categorical outcome,所以先用了logit link,用的multinom这个function,然
后需要用newdata predict probability,但是在试probit link的时候发现不太对,首
先multinom没有probit选项,然后试了mlogit,非要transfer成long format,可是
predict的又不对,发现newdata非得包含outcome variable(y),如果y都有了还
predict什么呢,不明白了。。。然后又试了mnp,估计因为是有random sample的原因
,每次出来的probability都不一样,也不知我理解的对不对,请赐教!!!谢谢各位
大神 |
|
l*g 发帖数: 46 | 17 谢谢回复!这个页面我也看了,可惜没有具体解释probit,作业要求比较logit和
probit的细微差别,所以还是得搞出来。。。Anyway,谢谢! |
|
e******e 发帖数: 274 | 18 I am using a Bayesian method to estimate a multinomial probit model through
data augmentation. All the priors are conjugate priors. Covariance matrix
follows wishart. Normalize the first element of covariance matrix to be 1. I
find that other elements of the covariance matrix are very easy to explode.
They are very sensitive to the starting value and the parameter of the priors.
Can any Da Xia tell me why those elements explode and how to deal with them?
Thank you very much. |
|
e********I 发帖数: 693 | 19 ☆─────────────────────────────────────☆
reshaping (0818283848) 于 (Tue May 13 16:43:39 2008) 提到:
发信人: reshaping (0818283848), 信区: Statistics
标 题: 初级probit regression问题请教
发信站: BBS 未名空间站 (Tue May 13 15:44:49 2008), 转信
如果DV里的1太多0太少,estimate结果会受到什么样的影响?
谢谢
☆─────────────────────────────────────☆
mmandroy (xiaoqing) 于 (Tue May 13 23:42:07 2008) 提到:
large variance.
☆─────────────────────────────────────☆
reshaping (0818283848) 于 (Tue May 13 23:43:13 2008) 提到:
就是比较不容易得到significance是么?
☆ |
|
f*******n 发帖数: 588 | 20 理论上,当样本n<30的时候,t分布能更加准确的捕捉可能出现的
fat tails.而在n>=30的时候,t分布和z分布几乎趋同。
对于普通的regression,我们可以证明它的coefficient estimate
是屈从于t(df=n-number of coefficients)分布的。因此,你用
reg y x 的时候看到的是t-statistic。
但我不清楚probit regression coefficient estimate背后的分布
理论。如果它可以被证明是屈从于z分布的,那么stata自然就会给出
z-statistic而非t-statistic了。可以去统计版找专业人士求证一下。
z |
|
x********u 发帖数: 411 | 21 谢谢回答~ 听经济学版上的一个人讲,说 probit 好像是用到了maximum likelihood
的方法,所以才用 z statistics 。。。。 我其实也不大懂这些东西,就只会用 。。
。。 |
|
L*****2 发帖数: 66 | 22 freerider,
Thank you very much for your help!
Can you explain more why it may be better to use logit if there are a lot of
observations in the tail. I know logistic curve is slightly flatter than
probit curve. Is that related to your answer?
Thanks again |
|
a*******k 发帖数: 25 | 23 Like the title:
I am using SAS. I wonder how to write the code of getting Pseudo R2 in
ordered probit regression.
Thanks. |
|
a*******k 发帖数: 25 | 24 各位高手,ordered probit regression, 请问如何写SAS程序得到 Pseudo R^2?
谢谢!! |
|
a*******k 发帖数: 25 | 25 Under ordered probit regression, when the DV is defined into 0 to 5, how to
explain the results of marginal effects?
Thanks a lot! |
|
D*********Y 发帖数: 3382 | 26 Is there any sample code showing how to do the fixed effects in probit model?
Thanks! |
|
|
i*********e 发帖数: 783 | 28 binay dependent variables.
The predictor includes: categorical variables and numeric variables.
I use logistic regression?
Can I also use probit regression?
Any difference? |
|
i*********e 发帖数: 783 | 29 logistic model can also be written as the latent
variable form of probit model, but its error term is logistic distributed.
What does this mean? |
|
c********d 发帖数: 253 | 30 Probit model有conjugate prior, logistic model没有。 |
|
|
K****n 发帖数: 5970 | 32 有Probit function,就是Cumulative Gaussian,作概率P(x|w),w是probit function的
一个参数,另一个已知。
然后用一个Gaussian当prior prior(w),其中Gaussian的参数都是常数
我又知道w只有从1到100这100个整数的可能
现在有一堆sample,用X表示, likelihood是一堆probit function P(x|w)的乘积L(X|w)
那 posterior(w|X) 正比于 L(X|w)*prior(w)
这个虽然没有analytical解,但是我是不是只要把1到100一个一个试一遍,找出那个让
posterior
最大的就算解完这道题了。。。
谢谢指导! |
|
g****g 发帖数: 1828 | 33 In probability theory, the normal (or Gaussian) distribution, is a
continuous probability distribution that is often used as a first
approximation to describe real-valued random variables that tend to cluster
around a single mean value. The graph of the associated probability density
function is “bell”-shaped, and is known as the Gaussian function or bell
curve:[nb 1]
f(x) = \tfrac{1}{\sqrt{2\pi\sigma^2}}\; e^{ -\frac{(x-\mu)^2}{2\sigma^2}
},
where parameter μ is the mean (location of the pe... 阅读全帖 |
|
t****g 发帖数: 715 | 34 No big difference between logit and probit, however, you choose one of them
based on your preference. For instance, it may make your life easier to
choose probit, as its error is normal, which can either fit your model
better or make your presentation more reasonable for people outside
econometrics(you know, sometimes it is hard to explain to dumb guys why you
need a logistic distribution instead of normal). Historically, logit was
thought easier to compute during the period when computing distr |
|
c*****a 发帖数: 16 | 35 【 以下文字转载自 Statistics 讨论区 】
发信人: cdsdata (CDS), 信区: Statistics
标 题: Heckman two-step
发信站: BBS 未名空间站 (Sun Jan 20 10:31:48 2013, 美东)
First step: run a probit equation of participation using all the
observations. The estimates of from this probit model are then used to
construct consistent estimates of the inverse Mills ratio term.
Second step: Include the inverse Mills ratio and run the original regression
equation.
Question: do these two equations have to include the SAME control variables
(exce... 阅读全帖 |
|
k***g 发帖数: 7244 | 36 呵呵,是啊,突然想起来原来上计量经济学的时候这类问题都是用 Multinomial Logit
Model或是Multinomial Probit Model来做的,不能用OLS, 对于罢工还是游行的选择都是
categorical response, 不一定存在任何order或是rank的,如果你给他们数值表示,那
你假定了不同的选择之间存在着顺序,这种假定的顺序显然是不合理的,其实就算这里存
在着rank,也应该用odered probit 或是 ordered logit, 因为归根结底,你的dependent
variable 是qualititve的而不是quantative,呵呵,好像是酱紫,学久了用不到都忘了:
)
intention
形
更
明
这
是
如果这么算的话,许多independent
但是,如果我把愿意游行作为1,愿意罢工作为2,相加之后,我发现原来显著的好多inde
行 |
|
J**Y 发帖数: 34 | 37 The only difference between logit and probit is that we assume different
distributions for the random term. In logit model, we assume the random term
is Gumbel distribution. In probit model, we assume the term is normal
distribution. |
|
c********d 发帖数: 253 | 38 Propensity score will create large bias when data is not monotone missing.So
I don't recommend that approach. A lot of methods can be used in your case
if you only have one missing variable, such as hot-deck, predictive mean
matching using a logistic model. You can also use multivariate probit model
for your case since race is nominal. By using multivariate probit model, it'
s easy to develop MCMC algorithm to do multiple imputation. |
|
D*********Y 发帖数: 3382 | 39 快要抓狂了。
我现在有三个equations,一个probit,俩tobit,根本没有找到code
发现前人有做过一个OLS,一个TOBIT
或者一个OLS,一个PROBIT。
但是也是没有code可寻。
更不要说像我这种情况了。
想用PROC syslin,全部设OLS,我有fixed effects,该如果code呢?谢谢! |
|
A*******s 发帖数: 3942 | 40 非常有趣,读lz的blog受益良多。
logit和probit实质上没有什么差别,因为standard normal distribution和scaled
logistic distribution with unit variance的形状是很相近的。logit和probit的系
数的关系大概是1.6至1.8的关系,但是对应的std dev也是同样的倍数关系(因为
standard normal和standard logit的square root of variance也是同样的倍数关系)
,所以两者的t statistics和p-value非常接近。marginal effect是相同应该也是同样
道理吧。
其他model为啥marginal effect相同我就不清楚原因了。
is |
|
c*****a 发帖数: 16 | 41 First step: run a probit equation of participation using all the
observations. The estimates of from this probit model are then used to
construct consistent estimates of the inverse Mills ratio term.
Second step: Include the inverse Mills ratio and run the original regression
equation.
Question: do these two equations have to include the SAME control variables
(except IV)? |
|
b****l 发帖数: 23606 | 42 decision tree是什么我都不知道。
logistic regression就是对付离散数据的,logit model
跟probit差不多,就是一个假设exponential error,一个假设正太error |
|
b****l 发帖数: 23606 | 43 我不是跟你说我不知道什么是decision tree了么?
至于什么是logistic reg
自己看维基去。
关于logit model 和probit model的区别,恐怕你从来没学过吧?
Logistic regression
From Wikipedia, the free encyclopedia
Jump to navigation
Jump to search
"Logit model" redirects here. It is not to be confused with Logit function.
In statistics, the logistic model (or logit model) is a statistical model that is usually taken to apply to a binary dependent variable. In regression analysis, logistic regression or logit regression is estimating the parameters of a... 阅读全帖 |
|
b****l 发帖数: 23606 | 44 不牛逼,不过刚好我手头有这本书,而这本书是个常用教材。
这一章就是讲离散数据响应的,讲了logistic reg和probit model |
|
b****l 发帖数: 23606 | 45 md,你大概根本就没看懂这两个model到底怎么来的,就是用R run了run回归而已。
logit 和probit的最大的区别就是 porbit是正太的,logit不是。 |
|
b****l 发帖数: 23606 | 46 再笑。脸都打肿了,还笑。
Logistic regression measures the relationship between the categorical dependent variable and one or more independent variables by estimating probabilities using a logistic function, which is the cumulative logistic distribution. Thus, it treats the same set of problems as probit regression using similar techniques, with the latter using a cumulative normal distribution curve instead. Equivalently, in the latent variable interpretations of these two methods, the first assumes a stand... 阅读全帖 |
|
s***h 发帖数: 487 | 47 想起马克思同志说过的一句话:在统计学家们钻 correlation 的牛角尖时,但上班更
重要的是 build a classifier!
: 我不是跟你说我不知道什么是decision tree了么?
: 至于什么是logistic reg
: 自己看维基去。
: 关于logit model 和probit model的区别,恐怕你从来没学过吧?
: Logistic regression
: From Wikipedia, the free encyclopedia
: Jump to navigation
: Jump to search
: "Logit model" redirects here. It is not to be confused
with Logit
function.
: In statistics, the logistic model (or logit model) is a
statistical
model that is usually taken to appl... 阅读全帖 |
|
b****l 发帖数: 23606 | 48 这个弱鸡,就是学了点回归分析就出来卖弄
Logistic regression measures the relationship between the categorical dependent variable and one or more independent variables by estimating probabilities using a logistic function, which is the cumulative logistic distribution. Thus, it treats the same set of problems as probit regression using similar techniques, with the latter using a cumulative normal distribution curve instead. Equivalently, in the latent variable interpretations of these two methods, the first assumes ... 阅读全帖 |
|
b****l 发帖数: 23606 | 49 p统计phd,统计phd不知道logit model和probit model的区别?
统计phd会说出样本无穷大就等于总体这么荒谬的话来?
我相信他是phd,但估计是生物,教育之类的phd,上过一两门
regression analysis之类的课程而已。 |
|
发帖数: 1 | 50 生物统计
[在 bushel (失乐园) 的大作中提到:]
:p统计phd,统计phd不知道logit model和probit model的区别?
:统计phd会说出样本无穷大就等于总体这么荒谬的话来?
:我相信他是phd,但估计是生物,教育之类的phd,上过一两门
:regression analysis之类的课程而已。 |
|