Quantitative Methods Forum @ Norm Endler Room (BSB 164)
Speaker: Alyssa Counsell, York University
Department of Psychology
Title: Is Item-Level Non-Invariance Always Important?
Abstract: Differential Item Functioning (DIF) refers to measurement non-invariance across groups. In other words, DIF is present when individuals from two distinct groups with equivalent levels of the latent trait or ability demonstrate different response patterns. The implication is that group membership (instead of the latent trait) accounts for the difference in responding. There are several methods that test for DIF but I will use item response theory (IRT) in the current presentation. Specifically I will discuss DIF results that compare Canadian and German participants’ response patterns on each of the items of the General Self Efficacy Scale (Schwarzer & Jerusalem, 1995). The results demonstrate a practical concern for researchers. When DIF is present in some items, the implications for research are not always clear. In some instances the pattern of DIF may not consistently favour one group, and instead, item-level group differences may appear to cancel each other out if the total test information curve is examined. In psychology where groups are typically compared on test information rather than on an item-to-item basis, DIF may not represent a meaningful or important effect.
Speaker: Victoria Ng and Joo Ann Lee, York University
Department of Psychology
Speaker: Victoria Ng
Title: The Yuen-Welch and Generalized Linear Model Approaches for Analyzing Skewed and Heteroscedastic Data in Psychology
Abstract: Many psychological studies are designed for testing whether there are group mean differences for some continuous outcome variable. However, the assumptions of normality and heteroscedasticity underlying traditional methods (i.e., ANOVA/OLS regression) are often violated. Two alternative methods are discussed: the Yuen-Welch with trimmed means, and the Generalized Linear Model (GLM). Given the many specifications that are possible in the GLM, selected studies on competing estimators from health outcomes literature are touched upon. With the premise that one would ideally choose the method that yields both adequate power and estimates that represent all relevant data (i.e., including distribution tails), I address the motivation for comparing the Yuen-Welch and the GLM by simulation and discuss potential implementations of such a study.
Speaker: Joo Ann Lee
Title: A brief survey of current statistical methods for
meta-analyzing data produced by single-case experimental designs
Abstract: Single-case experimental designs (SCEDs; also known as n-of-1 trials,
small-n designs, single-subject designs, and interrupted time-series
experimental designs, among others) are a set of experimental designs
that employ repeated data collection over time on a single unit of
interest such as an individual, a family, or an institution. SCEDs are
especially beneficial when the research areas studied have high
variability, or a low prevalence rate, because the unit serves as its
own control. More specifically, SCEDs explicitly focus on
within-individual variability. Unfortunately, a single SCED provides
very little, if any, information about between-individual variability.
This disadvantage however, can be remediated by meta-analyzing results
from separate SCEDs. Nonetheless, the meta-analytic methods of SCEDs
are just beginning to be developed. The presentation will begin with a
review of the type of data common to SCEDs, followed by illustrations
of current popular methods to analyze and meta-analyze SCED data, and
conclude with future research in the area.
SORA-TABA workshop will be held on Wednesday, April 30th, 2014
University of Toronto
Health Sciences Building (the auditorium – HS610),
155 College Street, Toronto ON.
Recent Advanced in Deep Learning:
Learning Structured, Robust, and Multimodal Models
Building intelligent systems that are capable of extracting meaningful representations from high-dimensional data lies at the core of solving many Artificial Intelligence tasks, including visual object recognition, information retrieval, speech perception, and language understanding.
In this talk I will first introduce a broad class of hierarchical probabilistic models called Deep Boltzmann Machines (DBMs) and show that DBMs can learn useful hierarchical representations from large volumes of high-dimensional data with applications in information retrieval, object recognition, and speech perception. I will then describe a new class of more complex models that combine Deep Boltzmann Machines with structured hierarchical Bayesian models and show how these models can learn a deep hierarchical structure for sharing knowledge across hundreds of visual categories, which allows accurate learning of novel visual concepts from few examples. Finally, I will introduce deep models that are capable of extracting a unified representation that fuses together multiple data modalities. I will show that on several tasks, including modelling images and text, video and sound, these models significantly improve upon many of the existing techniques.
Ruslan Salakhutdinov
Assistant Professor,
Department of Computer Science and
Department of Statistical Sciences
University of Toronto
SSC Student Conference in Toronto
Date: May 24, 2014
Location: University of Toronto
Link
SSC Annual Meeting in Toronto
Date: May 25 to May 28, 2014
Location: University of Toronto
Link
Title: Using Animal Instincts to Find Efficient Experimental Designs
Speaker: Weng Kee Wong, Dept of Biostatistics, UCLA
ABSTRACT
Using popular models from the biological and pharmaceutical sciences as examples, I demonstrate how particle swarm optimization (PSO) searches for different types of optimal experimental designs in dose response studies, including mini-max types of optimal designs where effective algorithms to find such designs have remained elusive until now.
Topic: Interrupted time series
Presenter: Yan Chen, St. Michael’s Hospital; TARGet Kids project
Abstract: more details to follow
When: Monday October 20th, 2014
6-7 pm seminar
7 pm dinner (optional, but registration required)
Where: Room 696, UoT Health Sciences Building,
155 College Street <----------Note different location
Cost: Seminar - free
Dinner - $15 (students $5)
To attend, please RSVP by or on Friday the 17th of October:
* for seminar only, please email us by clicking here: mailto:taba.exec@gmail.com?subject=SeminarOnly
* for seminar and dinner, please email us by clicking here: mailto:taba.exec@gmail.com?subject=SeminarAndDinner
If you plan to attend the dinner, we accept a mailed cheque, Interac email money transfer or cash at the event. If you have to cancel the dinner we request a 48 hour notice otherwise TABA has to pay for it. For Interac email money transfer (not PayPal): do the transfer via your bank to taba.exec@gmail.com and send a separate email via your own email program to us (taba.exec@gmail.com) giving the answer to the security question you used so that we can accept the transfer.
For cheques: mail cheque, payable to TABA, to
Lorinda Simms
Eli Lilly Canada Inc.
3650 Danforth Ave
Toronto, ON
M1N 2E8
Topic: Maternal and Child Health Gains in Afghanistan in the Post-Taliban Era: A Synthesis of Big Data (Presenter: Nadia Akseer)
You are invited to join our statistics seminar next week on Oct 15 at 2pm (Nross 638). The speaker is Dr. Yinglin Qin from University of Waterloo.
Here is the abstract. Hope to see you there.
Title:� Testing the order of a population spectral distribution for
high-dimensional data
�Abstract:
Yingli Qin
Large covariance matrices play a fundamental role in various high-dimensional statistics. Investigating the limiting behavior of
�the eigenvalues can reveal informative structures of large covariance
�matrices, which is particularly important in high-dimensional
�principal component analysis and covariance matrix estimation. In this
paper, we propose a framework to test the number of distinct
population eigenvalues for large covariance matrices, i.e. the order
of a Population Spectral Distribution. The limiting distribution of
our test statistic for a Population Spectral Distribution of order 2 is
developed along with its (N,p) consistency. We will also report
some simulation results and real data analysis.