Monthly Seminar: What do the experts know? A model for assessing and aggregating the predictions of forensic handwriting experts

Posted 25 August 2017

Facility: Stats Central

STATS CENTRAL WILL BE HOLDING MONTHLY SEMINARS DISCUSSING STATISTICAL PRINCIPLES THAT ARE RELEVANT TO RESEARCHERS OF ANY DISCIPLINE. THEY WILL BE HELD ON THE SECOND THURSDAY OF THE MONTH FROM 3PM-4PM AND BE FOLLOWED BY AFTERNOON TEA AND AN OPPORTUNITY FOR DISCUSSION.

 

“What do the experts know? A model for assessing and aggregating the predictions of forensic handwriting experts”

   Daniel Navarro, School of Psychology

 

 

Forensic handwriting experts are typically asked to testify in court as to the provenance of handwriting samples (do these samples match?) and as such need to have the ability to notice which features of handwriting are common and which are rare. Until recently, however, no general purpose "ground truth" measures were available that could be used to assess whether the experts had implicitly acquired the relevant knowledge through everyday experience. 

 

In this talk I discuss an data analysis situation that arose when forensic psychologists at UNSW were able to obtain access to the "ground truth" database shortly before that same information became more widely available. This situation allowed a very brief time window in which the researchers could use the data base to design a data collection exercise that measured the implicit knowledge of experts and novices before the experts had the opportunity to explicitly learn from that same data base.

 

As a consequence of the opportunistic data collection (and because of the limited number of experts), the data analyst is "stuck" with a very small sample of experts and a moderate sized sample of novices, and the goal is to work out who knows what about the true frequencies of different handwriting features. I describe a Bayesian approach to solving the problem, one that helps to tease apart the difference between novices and experts, and potentially provides a method for extracting "wisdom from the crowd" by eliciting better calibrated responses from the group than simply averaging everyone's answers.

 

The talk will be about 30 minutes long and will be followed by (free!) afternoon tea and time for discussion.

Please register by Wednesday 13th Septembe, 12:00 pm, so we can cater properly for afternoon tea!

 

 

Date and time:  Thursday, 14th September 2017 - 3:00 pm to 4:00 pm

 

Venue:  Red Centre Room M032

 

 

Please register your interest for catering purposes through the button below.

 

Eventbrite - Understanding how to interpret linear regression models