Skip to main content

Workshop Series


The Statistical Consulting Studio 
holds several workshops each semester.  The topics are chosen to benefit a broad range of researchers with various backgrounds.  While these classes are free for all faculty and students in the Emma Eccles Jones College of Education and Human Services, space is limited.  Please register early to ensure a seat.




All sessions 1-3 pm in EDUC 454 unless otherwise stated.

Date

Topic

Friday

February 24

 

GEE: Generalized Estimating Equations

Classic linear regression relies on observations being independent or uncorrelated.  Data collected on observations nested within a hierarchical population structure (e.g. students within classes) and longitudinal studies (repeated measurements on each subject) should be analyzed via a method that takes into account the lack of independence between observations. Violating or ignoring this assumption can lead to spurious results such as an inflated Type I error rate. Directly including a correlation structure as part of the model essentially corrects the under-estimated standard errors, resulting in more appropriate and replicable results. This workshop will provide an overview of GEE as a method for handling longitudinal, nested, and non-independent data, and discuss difference with multi-level (mixed-effects) statistical models.

Friday

March 17

Random Forests: Classification & Regression Trees

random forest iconIt is common in statistical modeling to have far more predictor variables available than will be included in a final model (especially when including combinations of variables as interaction effects). Often, predictor variables are correlated with one another and have differing patterns of missingness. In order to arrive at the most parsimonious statistical model that accounts for as much of the variance in the outcome as possible, a systematic approach to variable selection is needed, which also allows for interactions, correlations, missingness, and unbalancedness across groups. This workshop will demonstrate the use of Random Forests as a method to arrive at the strongest and most parsimonious set of predictor variables. Although a limitation of Random Forests is interpretability of effects, methods for using the results of Random Forests as a guide in other modeling approaches will be presented.

Friday

March 31

RSVP by

March 29

1-2pm
Held in Huntsman Hall Room 160

Participant Recruitment and Data CollectionUsing Amazon's Mechanical Turk

mturk logoAmazon’s Mechanical Turk (MTurk) is a recruitment platform that researchers can use to collect survey data (called Human Intelligence Task or HITs) from participants by paying a monetary fee. MTurk is unique in that it allows researchers to quickly collect data from a diverse set of participants, making the process efficient and cost-effective. Although early research found that MTurk yielded lower quality data (Buhrmester, Kwang, & Gosling, 2011), recent advances and developments have led to several useful strategies for ensuring quality data (see Mason & Suri, 2012), such that the reliability and validity of data obtained using MTurk is comparable to that obtained from other methods of data collection (Rouse, 2015). As such, MTurk can be an efficient tool for collecting quality data.

Participants in this workshop will:

  • Be introduced to MTurk as a survey hosting platform
  • Be able to identify and cite relevant literature that has investigated the pros and cons of MTurk
  • Identify strategies and methods for ensuring quality data collection
  • Understand IRB rules and regulations as pertaining to an MTurk study
  • Learn how to set up an MTurk Requester account and HIT
  • Link the MTurk hosting platform with data capture tools such as Qualtrics and REDcap
  • Learn how to validate and pay participants

Friday

April 14

RSVP by

April 7

Ridge & LASSO: Regularized Regression

ridge and lasso logoRidge and LASSO regression are alternative approaches in situations where a large number of predictor variables are available to choose from, and where a high degree of correlation among those predictor variables is possible. These methods guide selection of the most important predictor variables (even from among hundreds) without having to ‘fish’ for significance. The resulting models are parsimonious, with easy to interpret estimates, that also avoid problems with collinearity, biased estimates, and overfitting.

A new system to electronically collect and enter data is now available at USU.
Although similar to Qualtrics, it offers many more advanced features.

Learn more about REDCap it's features and getting started.

Each session will include

  • overview of main features
  • sample demonstration
  • questions and answers

Monthly sessions held 1-2pm in EDUC 454

Flyer for printing and posting.

Researve Your Seat (space is limited)

If you would like to receive periodic emails announcing new sessions and links to register, Subscribe to submit your email address to our mailing list. 

Materials for past workshops may be accessed below.  Clicking on a date below to be directed to a webpage with class notes and files.

 

 

Topic

Date

 

Introduction:
R, For Absolute Beginners

(starting from scratch)

Oct 2015

 

Effect Size & Power Analysis:
Using the FREE ‘G*Power’ Software

Nov 2015

 

Data Visualization:

Easily Exploring Your Data in R using ggplot2

(assumes some familiarity with R)

 

Feb 2016

 

Powerful Publishable Plots

(no software needed)

Apr 2016