Dear Applied Statistics Community,
Please join us for our final meeting tomorrow when Thomas Yee, Department of
Statistics, University of Auckland will present ``Vector generalized linear
and additive models". Thomas provided the following abstract for his talk:
The class of vector generalized linear and additive models (VGLMs/VGAMs) is
very large and contains many statistical models relevant to quantitative
social science, e.g., univariate and multivariate distributions, categorical
data analysis, time series, survival analysis, extreme value analysis,
mixture models, correlated binary data, and nonlinear regression. I'll first
give an overview of the framework and tie it in with practice using my VGAM
package for R. Then we will focus on two sub-topics: reduced-rank VGLMs and
quantile/expectile regression. The former handles the reduced-rank
multinomial logit model (aka stereotype model) and Goodman's row-column
association model; applications of the latter are becoming popular in many
fields. Time allowing, I'll describe several sub-projects I'm currently
working on since arriving at IQSS.
The Applied Statistics workshop meets each Wednesday in room K-354,
CGIS-Knafel (1737 Cambridge St). We start at 12 noon with a light lunch,
with presentations beginning around 1215 and we usually wrap up around 130
pm.
I hope you can make it!
Justin Grimmer
Dear Applied Statistics Community,
Please join us this Wednesday when Alberto Abadie, Professor of Public
Policy, will present ``A General Theory of Matching Estimation", joint work
with Guido Imbens. Alberto provided the following abstract for his talk:
Matching methods provide simple and intuitive tools for adjusting the
distribution of covariates among samples from different populations.
Probably because of their transparency and intuitive appeal, matching
methods are widely used in evaluation research to estimate treatment effects
when all treatment confounders are observed (Rubin, 1973, 1977; Rosenbaum,
2002). In spite of their popularity, the problem of establishing the large
sample distribution of matching estimators remains largely unsolved, with
the exception of some special cases (see Abadie and Imbens, 2006). The
reason is that matching estimators are non-smooth functionals of the data,
which makes their large sample theory particularly challenging. This talk
will describe a new general method to establish the large sample
distribution of matching estimators. As an example of the applicability of
the method, we will describe how to derive the distribution of matching
estimators when matching is carried out without replacement, a result
previously unavailable in the literature. We will also discuss how to adjust
the standard errors for propensity score matching estimators to take into
account first step estimation of the propensity score, a result also
previously unavailable.
The Applied Statistics Workshop meets each Wednesday at 12 noon in K-354
CGIS-Knafel (1737 Cambridge St). The workshop begins with a light lunch and
presentations usually start around 1215 and last until about 130 pm.
Justin Grimmer
Dear Applied Statistics Community,
We are pleased to announce a special presentation that should be of
interest. David Firth, Professor of Statistics at the University of
Warwick, will present on Quasi Variances this *Thursday* from 12-2 pm in
room K-354 in CGIS-Knafel (1737 Cambridge St, the usual meeting place for
the applied statistics workshop). Professor Firth provided the following
abstract for his presentation:
The notion of quasi variances, as a device for both simplifying and
enhancing the presentation of additive categorical-predictor effects in
statistical models, was developed in Firth and de Menezes (Biometrika,
2004, 65-80). The approach generalizes the earlier idea of "floating
absolute risk" (Easton et al., Statistics in Medicine, 1991), which has
become rather controversial in epidemiology. In this talk I will outline and
exemplify the method, and discuss its extension to some other contexts such
as parameters that may be arbitrarily scaled and/or rotated.
Everyone (especially graduate students) is welcome and encouraged to attend.
A bit of background on Professor Firth. He is Professor of Statistics at
the University of Warwick. He specializes in statistical theory and methods,
and has a particular interest in generalized linear models---especially as
applied to the social sciences. He has published extensively in the
discipline's major journals of record, such as JRSS and Biometrika, and has
written several packages for the R language and environment. He has made
several significant contributions to the field, and is well known as the
inventor of bias-reduced logistic regression (also known as 'Firthit').
He is at IQSS as a Distinguished Visiting Fellow (April 7--17), and will be
spending part of his time here working with Arthur Spirling on models of
momentum for contest data.
We hope everyone will be able to attend
Cheers
Justin
Dear Applied Statistics Workshop,
The workshop will meet tomorrow, when Sandra Sequeira, a PhD candidate in
public policy, will present her work on the efficiency cost of corruption,
work that is joint with Simeon Djankov. Sandra provided the following
abstract for her talk:
This paper estimates the efficiency cost of corruption. We generate an
original dataset on bribe payments at ports in Southern Africa that allows
us to take an unusually close look into the black box of corruption,
observing how bureaucrats set bribes and measuring their economic costs on
firms and on the broader economy. We find that bribes are product-specific,
frequent and substantial. Bribes can represent up to a 14\% increase in
total shipping costs for a standard 20ft container and a 600\% increase in
the monthly salary of a port official. Bribes are paid primarily to evade
tariffs, protect cargo on the docks and avoid costly storage. We further
identify three systemic effects associated with this type of corruption: a
``diversion effect" where firms go the long way around to avoid the most
corrupt port; a ``revenue effect" as bribes reduce overall tariff revenue;
and a ``congestion effect" as the re-routing of firms increases congestion
and transport costs by causing imbalanced cargo flows in the transport
network. The evidence supports the theory that bribe payments at ports
represent a significant distortionary tax on trade, as opposed to just a
transfer between shippers and port officials that greases slow-moving
clearing queues.
The Applied Statistics Workshop meets each Wednesday at 12 noon in K-354
CGIS-Knafel (1737 Cambridge St). The workshop begins with a light lunch and
presentations usually start around 1215 and last until about 130 pm.
Apologies for the late notice and I hope you'll be able to attend,
Justin Grimmer