| January 21, 2004 | |
| Speaker: | Rebecca Nugent, graduate student UW Statistics |
| January 28, 2004 | |
| Speaker: | Fadoua Balabdaoui, graduate student UW Statistics |
| February 4, 2004 | |
| Title: | Marginal Regression Modeling of Longitudinal, Categorical Response Data |
| Speaker: | Jonathan Schildcrout, graduate student UW Statistics |
| Abstract: |
Longitudinal regression analysis is important in a variety of settings when
the goal is to characterize changes that occur over time. The focus of this
talk is on marginal regression models for longitudinal, categorical response
data. I will first discuss a consistency-efficiency tradeoff with
semi-parametric modeling when the goal is to estimate the cross-sectional
relationship between the response and an exposure E[Y(t) | X(t)]. Next, I
will describe the "marginalized" model class which permits likelihood-based
estimation of marginal regression parameters. I will extend this class to
accomodate response dependence that I have seen with long series of response
data (the functional form of response dependence has both serial and
long-range components). Finally, I will discuss prospective inference with
retrospective,
outcome dependent sampling. One situation where such a sampling scheme
might be important is in a study where interest is in estimating the
relationship between a response and a time-varying exposure, the exposure is
expensive to measure, and a number of subjects exhibited no response
variation during the study period (e.g., never had symptoms). With this
sampling design, under certain conditions, we are able to make valid
inference that is efficient when we exclude subjects without response
variation as long as we account for the covariate ascertainment mechanism.
|
| February 11, 2004 | |
| Title: | Two-stage multiple imputation |
| Speaker: | Ofer Harel, postdoctoral fellow UW Biostatistics |
| Abstract: |
Conventional multiple imputation (MI) replaces the missing values in a
dataset by m>1 sets of simulated values. I explore a two-stage
extension of MI in which the missing data are partitioned into two
parts and imputed $N=mn$ times in a nested fashion. Two-stage MI
divides the missing information into two components of variability,
lending insight when the missing values are of two qualitatively
different types. Point estimates and standard errors from the N
complete-data analyses are consolidated by simple rules derived by
analogy to nested analysis of variance. I present simple examples of
two-stage MI and discuss a variety of potential applications. I also
clarify the inferential role of the missingness indicators, extending
Rubin's concept of ignorability to accommodate two types of missing
values.
|
| February 18, 2004 | |
| Title: | Bayesian image analysis of cDNA microarray data |
| Speaker: | Raphael Gottardo, graduate student UW Statistics |
| Abstract: |
DNA microarrays are an increasingly important tool that allow biologists
to gain insight into the function of thousands of genes in a single
experiment. By using an array containing many DNA
samples, scientists can measure---in a single experiment---the expression
levels of hundreds or thousands of genes within a cell by measuring the
amount of labelled cDNA bound to each site on the array. In a typical
two-color microarray experiment, two mRNA samples, from control and
treatment situations, are compared for gene expression.
Both mRNA samples, or targets, are reverse-transcribed into cDNA, labeled
using
different fluorescent dyes (red and green dyes), then mixed and hybridized
with the arrayed DNA sequences. The hybridized arrays are then imaged to
measure the red and green intensities for each spot on the glass slide.
Image analysis is an important aspect of microarray experiments, whose
purpose is to provide estimates of the foreground and background
intensities for both the red and green channels.
In this talk, I will take a Bayesian approach to the problem.
In broad terms, the Bayesian approach treats the recorded raw images as
numerical data, generated by a statistical model, involving both a
stochastic component (to accommodate the effects of noise due to the
environment and imperfect sensing) and a systematic component (to describe
the true scene under view). Using Bayes' theorem, the corresponding
likelihood is combined with a prior distribution on the true scene
description to allow inference about the
scene on the basis of the recorded image.
I will not assume any knowledge on cDNA microarrays. I will give a brief
introduction about the technology and review the main statistical issues
involved in the analysis of the images.
|
| March 3, 2004 | |
| Title: | Optimal Dynamic Treatment Regimes |
| Speaker: | Erica Moodie, graduate student UW Biostatistics |
| Abstract: |
Dynamic treatment regimes offer an ethical and flexible protocol for
studying the effects of treatments which are adjusted over time according
to response to treatment. A dynamic treatment regime is a list of decision
rules - one for each time interval - for how levels of treatment should be
allocated. Consequently, dynamic regimes may reduce non-compliance due to
toxicity or under-treatment.
Until recently, few methods existed to study these regimes. I will examine the traditional approach of dynamic programming to solving these problems as well as recent advances in the area which rely on least squares methods and estimating equations.
|
| March 10, 2004 | |
| Title: | Empirical Evaluation of Data Transformation and Ranking Statistics for Microarray Analysis |
| Speaker: | Lixuan Qin, graduate student UW Biostatistics |
| Abstract: |
Many choices in the analysis of a microarray dataset affect the results,
such as normalization, background adjustment, and test statistics. Some
procedures (e.g., background adjustment) are common practice now, but
whether they truly benefit the analysis has not been fully evaluated. We
used ten spike-in microarray experiments to evaluate the relative
effectiveness of analysis choices in three categories: background
adjustment, normalization, and ranking statistic. Findings support the use
of an intensity-based normalization procedure and also indicate that local
background-adjustment is harmful. We find that t-statistics perform poorly
in identifying differentially expressed genes; more robust statistics are
preferred. During this talk microarray experiments and some commonly-used
terms will be briefly introduced. This work is done with Dr. Katie Kerr.
|