标签：Assignment Rubin results data Report MAST90125 Bayesian using chains
MAST90125留学生作业代写、代做Bayesian课程作业、R程序设计作业调试、R语言作业代写
Assignment 2, Question 2 MAST90125: Bayesian
Statistical Learning
Due: Friday 20 September 2019
There are places in this assignment where R code will be required. Therefore set the random
seed so assignment is reproducible.
set.seed(123456) #Please change random seed to your student id number.
Question Two (20 marks)
In lecture 3, we discussed how a Bayesian framework readily lends itself to combining information from
sequential experiments. To demonstrate, consider the following data extracted from the HealthIron study.
Serum ferritin levels were measured for two samples of women, one of C282Y homozygotes (n = 88) and
the other of women with neither of the key mutations (C282Y and H63D) in the HFE gene, so-called HFE
‘wildtypes’(n = 242). The information available is
• idnum: Participant id.
• homc282y: Indicator whether individual is Homozygote (1) or Wildtype (0).
• time: Time since onset of menopause, measured in years.
• logsf: The natural logarithm of the serum ferritin in µg/L.
The data required to answer this question are Hiron.csv, which can be downloaded from LMS.
a) Fit a standard linear regression,
E(logsf) = β0 + β1time
with responses restricted to those who are homozygote (homc282y = 1). This can be done using the lm
function in R. Report the estimated coefficients βˆ, estimated error variance,
b) Fit a Bayesian regression using a Gibbs sampler to only the wildtype (homc282y=0) data. Use the
output from your answer in a) to define proper priors for β, τ . For help, refer to lecture 13. For the
Gibbs sampler, run two chains for 10,000 iterations. Discard the first 1000 iterations as burn-in and
then remove every second remaining iteration to reduce auto-correlation. When storing results, convert
τ back to σ2
. When running the Gibbs sampler, incorporate posterior predictive checking, using the
test statistic T(y, β) = Pn
, where ei
is the predicted residual for
observation i at simulation j and e
is the replicate residual for observation i at simulation j. Report
posterior means, standard deviations and 95 % central credible intervals for β0, β1, σ2
combining results
for the two chains.
c) Perform convergence checks for the chain obtained in b). Report both graphical summaries and
Gelman-Rubin diagnostic results. For the calculation of Gelman-Rubin diagnostics, you will need to
install the R package coda. An example of processing chains for calculating Gelman-Rubin diagnostics
is given below.
Processing chains for calculation of Gelman-Rubin diagnostics. Imagine you have 4 chains of
a multi-parameter problem, and thinning already completed, called par1,par2,par3,par4
1
Step one: Converting the chains into mcmc lists.
library(coda)
par1<-as.mcmc.list(as.mcmc((par1)))
par2<-as.mcmc.list(as.mcmc((par2)))
par3<-as.mcmc.list(as.mcmc((par3)))
par4<-as.mcmc.list(as.mcmc((par4)))
Step two: Calculating diagnostics
par.all<-c(par1,par2,par3,par4)
gelman.diag(par.all)
d) Fit a standard linear regression,
E(logsf) = β0 + β1time
to all the data using the lm function in R. Report βˆ, and associated 95 % confidence intervals. Comparing
these results to the results from b), do you believe that sequential analysis gave the same results as fitting
the regression on the full data.
e) Report the results of posterior predictive checking requested in b). Do you believe the postulated model
was plausible. If not, what do you think is a potential flaw in the postulated model.
因为专业，所以值得信赖。如有需要，请加QQ：99515681 或邮箱：99515681@qq.com
微信：codehelp
标签：Assignment,Rubin,results,data,Report,MAST90125,Bayesian,using,chains 来源： https://www.cnblogs.com/lpka/p/11519850.html
本站声明： 1. iCode9 技术分享网（下文简称本站）提供的所有内容，仅供技术学习、探讨和分享； 2. 关于本站的所有留言、评论、转载及引用，纯属内容发起人的个人观点，与本站观点和立场无关； 3. 关于本站的所有言论和文字，纯属内容发起人的个人观点，与本站观点和立场无关； 4. 本站文章均是网友提供，不完全保证技术分享内容的完整性、准确性、时效性、风险性和版权归属；如您发现该文章侵犯了您的权益，可联系我们第一时间进行删除； 5. 本站为非盈利性的个人网站，所有内容不会用来进行牟利，也不会利用任何形式的广告来间接获益，纯粹是为了广大技术爱好者提供技术内容和技术思想的分享性交流网站。