NPR series on lie detection

NPR has just finished an interesting series on lie detection. You can listen to it at

http://www.npr.org/templates/story/story.php?storyId=15670581

This series is all about accuracy, and it makes it clear that lie detection is not very good, and will never be perfectly accurate. That means it will always be necessary to set a threshold, and the threshold, of course, will depend on the costs and benefits of various outcomes. I am not surprised that a series like this says nothing about thresholds. But the people who develop and use these methods should certainly be concerned about thresholds.

It is also interesting that there is a reference to mental "systems" in the second segment of the report, which is about the use of fMRI to determine whether people are telling the truth. Although it is not stated explicitly, we can infer that they are assuming that lying requires analytic cognition, and therefore activates "system 2," which can be detected using fMRI.

Task Properties Interactions

So the UConn trip went well I'd say. For those who couldn't make it, I did a short presentation on my dissertation topic - building towards a theory of task properties. By this, I mean using regression analysis to measure the influence of task properties (i.e. surface {display} & depth {statistical})* on dependent variables such as judgment performance (ra), environmental certainty (Re), knowledge (G), and judgmental consistency (Rs).

Because there are ten proposed task properties, it is important to choose wisely which properties to analyze given the constraints of research design (i.e. I don't want to spend five years getting the 1500 subjects a large-scale test would require). I would like to start a discussion with anyone interested in helping me to find clarity in this jumble.

I have been thinking about analyzing the relationship of three properties - number of cues, cue intercorrelations, and time to complete task. Intuitively, it makes sense to me that the interaction of time and number of cues will be strong. If subjects have to assess a large number of cues in a short time, intuitive cognition should be induced.

Other task properties and their interactions may be of equal or greater importance. Kathlea (UConn) pointed out the importance of the property "task decomposition." No doubt she hit the nail on the head. Her suggestion also made me think of the cues in terms of functional relations to one another (i.e. task decomposition = f(number of cues, display, etc) - thus adding more complexity to this already complex puzzle. If anyone doesn't mind brainstorming, look at the list from the Hammond et al. citation below and throw out some ideas.


* For more information on proposed task properties, check out Hammond et al. (1987; p. 756).

**Note: the dependent variables are the parameters of the Lens Model Equation.

Research subjects

Hey y'all,

I just thought that I would get some conversation going on a topic that has floated around occasionally since we started meeting over the summer. It is about time for many of us to start planning on how many/how to persuade research subjects for our dissertation experiments. I know Tom is encouraging us all to apply for dissertation grants through UAlbany. We should also discuss ideas to make this process efficient for all of us (i.e. Team Stewart as I like to call us) - perhaps a common subjects pool. Anyway, let's start throwing out ideas or putting current ideas into action...

Ryan

JDM journals

Prescriptive area

• Management Science
• Decision Sciences

Descriptive area

• Organizational Behavior and Human Decision Processes
• Behavioral Decision Making

Other

• IEEE Transactions on Systems, Man and Cybernetics
• Human Factors