Dear all,
Please join us for the Applied Statistics Workshop (Gov 3009) this
Wednesday, November 2 from 12.00 - 1.30 pm in CGIS Knafel Room 354. Amber
Brown, Senior Research Scientist at Disney Research, and Joe Marks, Vice
President and Fellow of Disney Research, will give a talk entitled
"Empirical Social Science at Disney Research". As always, a light lunch
will be provided.
The abstract for the presentation is:
*At Disney Research we mostly work on technologies that are relevant to our
various businesses: computer graphics, computer vision, robotics,
human-computer interaction, materials, displays, etc. But we also have
projects in the social sciences, with a heavy emphasis on rigorous
empirical testing. We will describe four recent projects:*
- *Novel pay-what-you-want pricing mechanisms.*
- *Load balancing of park guests via pushed incentives on mobile devices.
*
- *Guest participation in environmental programs.*
- *Introduction of a cinema culture to the developing world.*
An up-to-date schedule for the workshop is available at
http://www.iq.harvard.edu/events/node/1208.
Best,
Konstantin
--
Konstantin Kashin
Ph.D. Student in Government
Harvard University
Mobile: 978-844-0538
E-mail: kkashin(a)fas.harvard.edu
Site: http://people.fas.harvard.edu/~kkashin/
Dear all,
Please join us for the Applied Statistics Workshop (Gov 3009) this
Wednesday, October 26 from 12.00 - 1.30 pm in CGIS Knafel Room 354. Rich
Nielsen, a Ph.D. candidate in the Department of Government at Harvard
University, will give a presentation entitled "Comparative Effectiveness of
Matching Methods for Causal Inference". As always, a light lunch will be
provided.
The abstract for the paper is below and a copy of the paper is available
here: http://gking.harvard.edu/files/psparadox.pdf. This is joint work with
Gary King, Carter Coberley, James E. Pope, and Aaron Wells.
*
*
*
*
Abstract:
*Matching is an increasingly popular method of causal inference in
observational data, but following methodological best practices has proven
difficult for applied researchers. We address this problem by providing a
simple graphical approach for choosing among the numerous possible matching
solutions generated by three methods: the venerable "Mahalanobis Distance
Matching" (MDM), the commonly used "Propensity Score Matching" (PSM), and a
newer approach called "Coarsened Exact Matching" (CEM). In the process of
using our approach, we also discover that PSM often approximates random
matching, both in many real applications and in data simulated by the
processes that fit PSM theory. Moreover, contrary to conventional wisdom,
random matching is not benign: it (and thus PSM) can often degrade
inferences relative to not matching at all. We find that MDM and CEM do not
have this problem, and in practice CEM usually outperforms the other two
approaches. However, with our comparative graphical approach and
easy-to-follow procedures, focus can be on choosing a matching solution for
a particular application, which is what may improve inferences, rather than
the particular method used to generate it.*
An up-to-date schedule for the workshop is available at
http://www.iq.harvard.edu/events/node/1208.
Best,
Konstantin
--
Konstantin Kashin
Ph.D. Student in Government
Harvard University
Mobile: 978-844-0538
E-mail: kkashin(a)fas.harvard.edu
Site: http://people.fas.harvard.edu/~kkashin/
Dear all,
Please join us for the Applied Statistics Workshop (Gov 3009) this
Wednesday, October 19 from 12.00 - 1.30 pm in CGIS Knafel Room 354. Weihua
An, a Lecturer in the Department of Sociology at Harvard University, will
give a presentation entitled "Peer Effects on Adolescent Smoking and Social
Network-Based Interventions". As always, a light lunch will be provided.
The abstract for the presentation is:
*This study addresses a fundamental question in social network analysis:
whether and to what extent peers affect a person's wellbeing. More
specifically, it attempts to identify and quantify peer effects on smoking
among adolescents. *
*
*
*Based on the causal inference terminology, a systematic framework to study
causal peer effects was developed to distinguish several types of peer
effects, including peer effects under control, peer effects under treatment,
etc. To overcome the difficulties in identifying peer effects with
observational data, a novel field experiment was conducted with a partial
treatment group design specifically tuned to estimate peer effects. *
*
*
*More specifically, a smoking prevention intervention composed of
distributing smoking prevention brochures and hosting health education
workshops was assigned to partial randomly chosen members in a number of
classes in six middle schools in China where the experiment was fielded. The
goal was to study how the information contained in the intervention was
spread across students and how it affected their information, knowledge,
intention, and behavior regarding smoking. To accelerate or reinforce the
diffusion, central students or students with their close friends as
identified based on their social network information were also chosen
respectively to receive the intervention in different treated classes. *
*
*
*Descriptive analysis provided strong support for peer effects on the
initiation and maintenance of adolescent smoking. Further statistical
analysis showed that compared with students in the control classes, students
whose classmates were randomly chosen to receive the intervention but who
did not receive the intervention themselves were more likely to exchange
information about the intervention with other students and to remain non-
smokers or change to non-smokers overtime. It was also found that the social
network- based interventions did not consistently bring significant added
value in all the outcomes of interest and their benefits mainly concentrated
on lowering students' intention to smoke and decreasing smokers' popularity.
*
*
*
*Special attention will be paid in the presentation to elaborating how to
choose central students and student groups in a social network. *
An up-to-date schedule for the workshop is available at
http://www.iq.harvard.edu/events/node/1208.
Best,
Konstantin
--
Konstantin Kashin
Ph.D. Student in Government
Harvard University
Mobile: 978-844-0538
E-mail: kkashin(a)fas.harvard.edu
Site: http://people.fas.harvard.edu/~kkashin/
Dear all,
Please join us for the Applied Statistics Workshop (Gov 3009) this
Wednesday, October 12 from 12.00 - 1.30 pm in CGIS Knafel Room 354. Michael
Weissman, a Professor Emeritus from the Physics Department at the University
of Illinois, will give a presentation entitled "From Fourier to Forensics".
As always, a light lunch will be provided.
The abstract for the presentation is:
*Although the statistical and systematic problems of public opinion polls
are fairly widely recognized, we tend to assume that published polling
results reflect some sort of actual poll. In 2009 a prominent blog suggested
that the pollster Strategic Vision might be fabricating data, based in part
on surprising deviations from uniformity of the distribution of trailing
digits of the results.(
http://www.fivethirtyeight.com/search/label/strategic%20vision) Objections
were raised to the assumed uniform distribution, but we were able to use
Fourier analysis together with known polling statistics to show that the
results were weird even if that assumption were dropped.
http://query.nytimes.com/gst/fullpage.html?res=9C03E1DA123AF930A25751C1A96F…
*
*
*
*In 2010 we were contacted by a political consultant who had noticed
anomalies in Research2000 poll reports. Using a variety of elementary
statistical techniques, we showed that those results could not have
accurately represented real polls. (
http://en.wikipedia.org/wiki/Research_2000) Unfortunately, we do not know if
there are other bogus pollsters, disguising results via a random binary
generator (cost $0.01).*
An up-to-date schedule for the workshop is available at
http://www.iq.harvard.edu/events/node/1208.
Best,
Konstantin
--
Konstantin Kashin
Ph.D. Student in Government
Harvard University
Mobile: 978-844-0538
E-mail: kkashin(a)fas.harvard.edu
Site: http://people.fas.harvard.edu/~kkashin/
Dear all,
Please join us for the Applied Statistics Workshop (Gov 3009) this
Wednesday, October 5 from 12.00 - 1.30 pm in CGIS Knafel Room 354. Victoria
Liublinska, a Ph.D. candidate from the Statistics Department at Harvard
University, will present a paper entitled "Addressing missing data issues in
a study with rare binary outcomes constrained by a small sample size". As
always, a light lunch will be provided.
The abstract for the presentation is:
*We (re)analyze the data obtained in a recent study conducted to evaluate
safety and efficacy of a new device designed for vertebroplasty. The
following are just a few issues that had to be addressed: missing data in
some covariates, incorrect analysis applied initially to the primary
endpoint, missing data in secondary endpoints. The latter involved
additional challenges such as panel data (responses were collected twice
over time with a non monotone missingness pattern), secondary endpoints were
rare binary events. The analysis was complicated by a relatively small
sample size. Our work demonstrates how a complex missing data issue can be
broken down into a set of small tasks that are solved individually. Some
tasks involved multivariate missing data imputation using chained equations
(van Buuren and Oudshoorn 2000; Raghunathan et al. 2001) with carefully
chosen conditional models. Other tasks called for new state-of-the-art
solutions, such as z-transformation procedure for combining repeated
p-values (D. Rubin et al. 2011 (to be submitted), C. Licht 2009 Ph.D.
thesis) or enhanced tipping-point graphs that assess sensitivity to various
deviations from assumptions made about the missing data mechanism (Yan et
al. 2009, Campbell et al. 2011).*
This is joint work with D. Rubin and R. Gutman.
An up-to-date schedule for the workshop is available at
http://www.iq.harvard.edu/events/node/1208.
Best,
Konstantin
--
Konstantin Kashin
Ph.D. Student in Government
Harvard University
Mobile: 978-844-0538
E-mail: kkashin(a)fas.harvard.edu
Site: http://people.fas.harvard.edu/~kkashin/