|
l*****n
发帖数: 17 | 2 我做研究需要借阅这篇文章，可是在学校的文献检索中无法拿到全文。如果哪位同学能
拿到此文章的电
子版本，是否能发给我b******[email protected]？谢谢！
这篇文章发表在2011年2月的Statistics in Medicine上。
Generalized propensity score for estimating the average treatment effect
of multiple treatments |
|
x**m
发帖数: 941 | 3 18个月，可惜不是perm.
Job Description:
Design, develop, implement and maintain targeted marketing campaigns and
analytic programs.
* Ensure integrity of campaigns and analyses through rigorous attention to
detail and validation of data.
* This individual will execute direct marketing campaigns (including: direct
mail, electronic leads, e-mail, web site, ATM, etc.) for customers and
prospects.
* Additionally, this role will involve conducting analytics, creating
reports, and improving processes.
* Lever... 阅读全帖 |
|
p********a
发帖数: 5352 | 4 propensity scoring to select control group |
|
d******9
发帖数: 134 | 5 A实验是一个prospective randomized clinical trial,比较两种疗法的作用，trial是
在A医院做的。
B实验replicate了A实验(指使用了同样的inclusion/exclusion criteria, 同样的两种
疗法，收集了同样的variable，做了同样的analyses...)。不过，B是一个
retrospective study,病人数据是现有的，来自B医院，用propensity score matching
调整了一下selection bias.
再对B医院的数据独立做了analyses之后，我还想比较一下A与B实验的结果，想看
retrospective和prospective study的结果会否有区别，请问可以用什么方法？
本来想用meta-analysis,但是这两个实验不是随机抽取的，我也不需要整合它们的结果
，只是想比较。于是觉得meta-analysis可能不合适，对不对？想请教大家应该用什么
方法来做？
刚开始学统计，很多东西不是很懂，查了很多文章也找不到这种情况下一般常用的方法
是啥，先谢谢各位了~ |
|
s*********e
发帖数: 1051 | 6 google "propensity score" |
|
l****u
发帖数: 529 | 7 it is a observational experiment. Propensity score are usually used that is
P(x=1/y). |
|
a*********n
发帖数: 1331 | 8 因为data是secondary data，不是自己收集的，sampling不可能。只能用统计方法减少
bias的影响。
统计上基本上三种方法（如果还有其它好方法，请指正）：
1、Heckman Two Steps
2、Propensity Scoring
3、Simultaneous Equation Modeling
各自适合什么情况，都有哪些优点、缺点？
有书/文章推荐吗？ |
|
d*********u
发帖数: 8521 | 9 假设在某产品市场里，80%的消费者选择A品牌，12%选择B品牌，8%选择C品牌。
所有消费者的各个变量已知，包括性别，年龄，种族，婚否，教育水平，工资收入（连
续变量）等。
如何选择正确方法分析三类消费者某些变量之间是否有显著差异？
做方差分析？不同消费者的case数量差别巨大，且不能满足正态分布与方差齐次……
三组数据之间对不同变量两两做卡方检验？
多元逻辑回归，设定80%的A消费者为参照组，比较B和C的不同？
有人介绍了个Propensity Score Matching方法，应该挺适合，不过一来现学估计来不
及，二来不考虑这些消费者是某产品消费者所以肯定和市场上的其他人不同，不符合大
数法则啥的
，就以这些消费者为全体，来分析三个类别之间某些变量是否有显著区别，怎么做？
多谢！ |
|
m******u
发帖数: 277 | 10 LOCF
propensity score
predictive mean matching
^_^ |
|
|
c********d
发帖数: 253 | 12 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. |
|
c**********e
发帖数: 2007 | 13 Can it handle arbitrary missing pattern? |
|
|
|
h***i
发帖数: 3844 | 16 propensity score to adjust selection bias? |
|
g********3
发帖数: 123 | 17 We don't have old score to compare with.....Is propensity score method an
academic way? |
|
t*****2
发帖数: 94 | 18 您好，小弟是FRESH GRADATE, 最近在工作，在面试的时候很多时候被问到MISSING
VALUE的问题。我看到你经常在这里解答别人的问题，而且很专业。希望能得到您的答
案。
for example: how to deal with missing value so that it can be used as input
for model? what if 80% of the data are missing?
我就回答了： a)test the pattern of missing value (MCAR/MAR/MNAR)
test some assumptions (eg. normality, because some datasets
are assumed to be normally distributed)
b) Solution: Multiple Imputation... 阅读全帖 |
|
y*****w
发帖数: 1350 | 19 By "matched sample", you are right, paired data analysis should be applied.
However, this is still not a strictly matched sample by any means. LZ would
have to prove that the samples were matched through some mechanism (age and
gender matching, as a simple example), not arbitrarily.
LZ can check out the following article by Peter Austin. It's about
propensity score matching, with some contents on survival analysis (K-M
estimates and Cox models) in the context of matched pairs. Check out Page
113... 阅读全帖 |
|
z********n
发帖数: 710 | 20 谢谢楼上的。我的design里面还涉及到propensity matching to get control group
using more than 10 covariates. 另外，在用logistic model去预测一个dichinomous
dependent variable。那么我到底怎么估计effect size和variability呢？另外我现
在都不知道要怎么选定statistical test了。 |
|
e**p
发帖数: 4259 | 21 如题，有些迷惑！正在读相关的文章，还没有来得及读完，希望了解这两个方法的朋友
讨论一下 |
|
s*********e
发帖数: 1051 | 22 Our client is seeking qualified applicants for the role of Research
Statistician in its Predictive Analytics team. They have been distinguished
by Fortune Magazine as one of the world's most admired companies for 2013.
This position is part of the a department, whose mission is to extend
knowledge, derive insights, and support a strategic framework to promote
innovation, balanced decision-making, and coordinated action across the
corporation. In this role you will be developing, interpreting and... 阅读全帖 |
|
w*******9
发帖数: 1433 | 23 两者都不是，关于这个model selection有很多文章，但是没有一个统一的定论。最后
的结论是1：包括的covariates越少越好。2如果你选择了某个看起来promising的model
，结果某个variable match的不好，就再把那个variable放进去。3要多考虑functions
of covariates like logx, x^2。4评价model的最终标准是match的怎样，一个t-test
是不够的，KStest也要考虑，而且you also need to match any functional forms of
the covariates。所以这是一个back and forth很烦人的过程。
logistic |
|
|
c*****1
发帖数: 2460 | 25 Use Bayesian logistic regression instead. |
|
y*****w
发帖数: 1350 | 26 1. Use those covariates that are related to both the grouping variable and
the outcome variable in the study. For example, if your study is to analyze
whether a drug treatment is highly associated with mortality, and the
demographics and baseline characteristics are not balanced between the
treatment groups, then you would need to (1) run univariate logistic
regressions to identify those covairates that are highly associated with
drug treatment, (2) run univariate Cox regression models to identi... 阅读全帖 |
|
h**********1
发帖数: 155 | 27 恩。这些题目还蛮有意思的。我来试着回答下
Q2: run一个regression看coefficient？
Q6: 先按一个开关10分钟，然后开另外一个。然后马上去房间里看，根据灯泡的热度马
上就可以知道哪个对应哪个
Q10：中间
Q16： matching X or match propensity score or reweight. there is a paper
about it.
Q17: bagging?
Q19: this is a binomial dist'n, I think. p = 1/2000 ?
Q20: |
|
h**********1
发帖数: 155 | 28 恩。这些题目还蛮有意思的。我来试着回答下
Q2: run一个regression看coefficient？
Q6: 先按一个开关10分钟，然后开另外一个。然后马上去房间里看，根据灯泡的热度马
上就可以知道哪个对应哪个
Q10：中间
Q16： matching X or match propensity score or reweight. there is a paper
about it.
Q17: bagging?
Q19: this is a binomial dist'n, I think. p = 1/2000 ?
Q20: |
|
c***z
发帖数: 6348 | 29 【 以下文字转载自 DataSciences 讨论区 】
发信人: chaoz (面朝大海，吃碗凉皮), 信区: DataSciences
标 题: [Data Science Project Case] Bias Correction - third try
发信站: BBS 未名空间站 (Tue Feb 11 18:26:40 2014, 美东)
Dear all, thank you so much for your earlier inputs! Now I am able to put my
thoughts together and understand the project better.
Let me write down the thing again. Any comments are extremely welcome!
Project name: Bias correction
Business objective: We have a panel of 25M users’ shopping cart information
, we want to in... 阅读全帖 |
|
w*******9
发帖数: 1433 | 30 propensity score matching？ |
|
y*****w
发帖数: 1350 | 31 Not necessarily matching. Could be propensity score weighting (IPTW). Often
times weighting is better than matching, because (1) by matching, you are
gonna lose quite a lot subjects, and (2) with the matched data, you have to
take into account its paired nature when conducting analysis, such as using
paired t-test or McNemar test, and sometimes it would be hard to perform
such analysis, for example it's hard to find a counterpart of mixed model
under the paired nature of the matched data. |
|
B******e
发帖数: 70 | 32 可以把 clustering 设为 random effect，用 GEE 或 GLIMMIX 吗？
或者，adjust clustering后算propensity score，从新match ？ |
|
y*****w
发帖数: 1350 | 33 Need to use specific analysis methods for paired data as you pointed out,
because the matched pairs are not independent anymore. Would be
complicated if you have to run mixed or survival models. Check out
Peter Austin's papers.
Would suggest using the weighting method (IPTW) instead of whatever matching. |
|
m***c
发帖数: 118 | 34 一个数据有三个变量pct1,pct2,pct3，均是工人工作完成率0 - 1，obs大概有1500个,
pct2和pct3分别有10多个missing values。已经检测过这三个变量均不是normal。
data a;
input pct1 pct2 pct3;
cards;
0.2345 0.2657 0.3410
0.8009 0.7011 0.6945
1.0000 0.5699 0.8940
0.7109 . 0.6945
0.5470 0.9999 1.0000
0.8901 0.5557 .
0.4522 0.9672 0.6012
。。。。。
；
run;
问题：
1. 一把情况下，如何比较pct1,pct2,pct3是不是significantly same or different?
2.如果后2个变量oct2,pct3的工人是通过第一个pct1的工人PSM（propensity score
matching）找出的结果， 比如，用工人1的年龄技术经验，，来找到匹配的工人2（工
人1和工人2具有很高的相似性），工人2的完成率就是pct2，同样以工人1再... 阅读全帖 |
|
a**u
发帖数: 59 | 35 The Mid America Heart Institute at Saint Luke’s Hospital is seeking a PhD
biostatistician to join its nationally recognized cardiovascular research
program. This non-academic position will:
• Provide statistical support for research & publication activities
across various studies & registries
• Actively collaborate with researchers and study teams on new
studies, including design, planning, study operations, data validation and
preparation, analysis and presentation of results... 阅读全帖 |
|
l********8
发帖数: 668 | 36 本人心理统计学博士，本科读的是心理学。博士毕业后一直在一个毒品滥用研究所工作
(non-profit H1B).侧重longitudinal observational data,treatment evaluation
的统计分析,主要用一些growth curve models, propensity score adjustment,
Markov transition models, Cox models with frailty or competing risks. 自己还
写过几
篇latent variable model，structural equation model的发展创造的文章。如有合适
的机会的话，请多多推荐。
简历和CV随时可以寄到您的帐内，多谢！ |
|
l******d
发帖数: 168 | 37 帮顶！
[在 laozhu8888 (laozhu) 的大作中提到：]
：本人心理统计学博士，本科读的是心理学。博士毕业后一直在一个毒品滥用研究所工
作(non-profit H1B).侧重longitudinal observational data,treatment evaluation
：的统计分析,主要用一些growth curve models, propensity score adjustment,
：........... |
|
Q*****T
发帖数: 558 | 38 背景 是这样的，我是公共卫生（public health）的phd，博士期间做的东西很杂，有
epidemiology，有小型的clinical trial，生统的东西用的相对较多，简历上也写了（
吹嘘？？）自己的skill有biostatistics，最近在找工作，过了第一轮技术面（面试内
容偏epidemiology），早上得知被安排在下周一进行第二轮电话面试，对方是public
health牛校的biostatistics博士，毕业后的工作前几年title都是biostatistician。
他来现在这个公司以后，开始是associate，现在的title是manager。
我一直觉得自己的biostat水平接近一个biostat master level的学生，跟biostat的
phd当然还差得很远。具体做工作的时候，用的东西倒是都知道，会用，但是一旦触及
深层次的问题，就很迷糊。我处在有时候觉得自己生统懂得还不少，多数时候觉得自己
水的不行、没啥自信的状态。
我面的这个工作是技术类的consulting公司，title是associate。组里的人是
epidemiolog... 阅读全帖 |
|
a*******g
发帖数: 80 | 39 三组病人 follow up长短严重不同 最长一组比最短一组平均值多三倍 而outcome是某
种并发症 出现时间在几周到几年不等 因此 这个随访时间的不同会严重影响结论 这是
retrospective study，三组病人还很不balance 第二 三组病人分别只有 20几个。
我考虑以下几种办法：1 propensity score matching，然后用matched data 做
analysis。问题是matching后一大半病人去掉了，对于我们综述这个病不利。
2做 time to event analysis 这样可以部分修正因为follow up 不同造成的bias 问题
是对于太短的那组病人 也无法有准确的estimate 几乎都censor 了
3 设一个cutoff 比如一年 把这个时间点后面的event都当作censored 其实这个和
time to event那个有点像 纠正bias 还不如2 好处是可以用全部病人 并且比较直观好
理解
还有没有别的好办法 这三个方法那个稍微好点，或者是有什么问题 请大家指教 |
|
l**********8
发帖数: 305 | 40 楼主来了，每周一个面试，实在强度有点大
话说之前2次电面，非常简单，问了简历上提到的工作经验，还有就是问了
什么是propensity score，什么是GLM ，有没有用过。
然后就是onsite，纯聊天，见了组里3个人。感觉挺有戏。话说HR写来邮件说SVP下周要
再给我个follow up conversation . 不知道聊啥。前面有人说的对，这家公司就是
underpay，出价有点低，用来保底吧。 |
|
n**********0
发帖数: 66 | 41 选你最感兴趣以及必不可少的的characteristics，用propensity score去match |
|
s*******d
发帖数: 132 | 42 我用的是R matchIt
我需要match on disease status， 但是现在用所有变量match， 结果显示disease st
atus not balanced。
是否有办法exact match on disease status，然后 coarse match on other variable
s？ 使用什么package？
多谢！ |
|
y**3
发帖数: 267 | 43 library(Hmisc)
w <- find.matches(case, ctrl, tol=rep(0, ncol(ctrl))) |
|
C******n
发帖数: 284 | 44 可以manually match —— 先按照disease status（假设只有两个或几个数量有限的
categories）分组，然后每组内用现成的package做PSM
st
variable |
|
s********a
发帖数: 1100 | 45 试试先用这个variable做1:n的match，然后在结果里再用其他variable match一次 |
|
p********a
发帖数: 5352 | 46 难道不是stratification后取高端STARTA再MATCH？另外，add interaction/higher
order terms.
Exact match什么意思啊？都是分布而已，不可能一模一样吧 |
|
c****u
发帖数: 243 | 47 lz 如果是指 hard match，比如male only matches with male，
应该任何一个包都可以，设置caliper = 0 就可以了 |
|
|
m*********k
发帖数: 10521 | 49 [ebiz] NewEgg Jun 1 ● 节日快乐~发糖果乐
10*20=200
成功奖励 10 伪币的用户: coolyun, BrazilRocks, wiiwii, anywhere, kokosky, dennykarsh, stoneincrazy, yizhitumao, mengxj, UAA836, liuxing3322, propensity, tranquilness, bjjasmine, shaya, VividMay, Cathy2011, jerryIII, hunzaimeiguo, jjmm123
扣除版面:(ebiz)200伪币成功 |
|
b*******n
发帖数: 12321 | 50 难道ebiz的包子每次只发前20个？
dennykarsh, stoneincrazy, yizhitumao, mengxj, UAA836, liuxing3322,
propensity, tranquilness, bjjasmine, shaya, VividMay, Cathy2011, jerryIII,
hunzaimeiguo, jjmm123 |
|
