The 2014 Deming Student Scholar award recipients are – Xialu Liu from Rutgers University,
Yanping Li from Temple University, Edward Kennedy from University of Pennsylvania, School of
Medicine and Hyunseung Kang from University of Pennsylvania, Wharton School. They will be
presenting poster on their work at the conference.
Abstracts:
Convolutional Autoregressive Model for Functional Time Series and Its Applications
Xialu Liu and Rong Chen
Functional data analysis, taking the place of scalar or vector data analysis, becomes an
increasingly popular topic in statistical research during the last few
decades. Bosq (2000) first studied functional data with the serial
dependence, and proposed the autoregressive functional time series
model. He came up with a method to estimate the parameters in the
model, but failed to provide its asymptotic properties as parameters
come from an infinite-dimensional space. We propose a convolutional
autoregressive model, and use spline functions with sieve method to
estimate the infinite-dimensional parameters. The asymptotic
properties are investigated to offer a sound theoretical support for data
analysis. Several real data examples are presented.
A Decision Theoretic Approach to Multiple Testing of Grouped Hypotheses
Yanping Liu, Sanat K. Sarkar and Zhigen Zhao
In many modern large-scale multiple testing problems, the hypotheses appear in nonoverlapping groups with the associated p-values exhibiting
dependence within but not between groups. Such group
formation is often a natural phenomenon due to the underlying
experimental process or can be created based on other
considerations. In this paper, we take a compound decision
theoretic approach toward developing a multiple testing
procedure for grouped hypotheses subject to controlling the
false discovery rate (FDR). Our procedure works in two stages. At the first stage, hypotheses in
each group are screened for possible rejection subject to a certain constraint on group-specific
FDR. At the second stage, these hypotheses are ultimately rejected if the corresponding groups
are determined to be rejected when controlling the overall or total FDR at the specified level.
We provide numerical evidence of superior performance of the oracle version of our procedure
over its natural competitors, including the one without using the group structure, in certain
scenarios under two different model settings for the within-group pairs of p-value and the truth
or falsity of the associated null hypothesis.
Combining propensity scores, regression, and matching for robust estimation of treatment
effects
Edward Kennedy
The three most popular adjustment methods in causal inference are propensity scores,
regression, and matching. Propensity score and regression approaches
rely on models and are biased under misspecification; matching
estimators are nonparametric but give poor finite-sample performance
with high-dimensional data and complicate inference. In this work we
describe and evaluate the performance of a simple but novel estimator
that combines the three adjustment methods and retains advantages of
each. The method is applicable to any setting involving estimation of a
treatment effect; we illustrate it here using the famous Lalonde study of
the effects of job training programs on earnings.
Instrumental Variables Estimation With Some Invalid Instruments and its Application to
Mendelian Randomization
Hyunseung Kang, Anru Zhang, T. Tony Cai and Dylan S. Small
Instrumental variables have been widely used for estimating the causal effect between exposure and
outcome. Conventional estimation methods require complete knowledge
about all the instruments' validity; a valid instrument must not have a direct
effect on the outcome and not be related to unmeasured confounders. Often,
this is impractical as highlighted by Mendelian randomization studies where
genetic markers are used as instruments and complete knowledge about
instruments' validity is equivalent to complete knowledge about the involved
genes' functions.
In this paper, we propose a method for estimation of causal effects when this
complete knowledge is absent. It is shown that causal effects are identified
and can be estimated as long as less than 50% of instruments are invalid, without knowing which of the
instruments are invalid. We also introduce conditions for identification when the 50% threshold is
violated. A fast penalized L1 estimation method, called sisVIVE, is introduced for estimating the causal
effect without knowing which instruments are valid, with theoretical guarantees on its performance.
The proposed method is demonstrated on simulated data and a real Mendelian randomization study
concerning the effect of body mass index on health-related quality of life index. An R package sisVIVE is
available on CRAN. Supplementary materials for this article are available online.