Virtual Child Witness-Effects of single and multiple use on performance with Novice and Expert cohorts in a structured virtual human interview (bibtex)
by Talbot, Thomas B., Lyon, Thomas D., Rizzo, Albert and John, Bruce
Abstract:
Virtual human avatars can be used to train and assess a myriad of complex skills, such as interviewing, interpersonal, and clinical skills, in a safe environment that provides consistency, reduced cost, greater accessibility, and objective feedback. We created a structured virtual human interview which consisted of a conversational avatar that interacts verbally in response to on screen question choices. Our prototype was a forensic interview simulation called Virtual Child Witness (VCW). VCW provides a content-rich interview in response to open-ended questions and is designed to assess user’s interviewing strategy. In a quasi-experimental design, we evaluated 222 subjects to determine if the system could discriminate between Experts (M = .713, SD = .29961) and Novices (M = .373, SD = .27547), t (230) = 9.002, p \textless .001, d = 1.18, indicating easy discrimination between cohorts. A subset of 92 subjects completed the simulation multiple times; multiple use was associated with much higher performance (M = .727, SD = .23669) than single use (M = .489, SD = .33688), t (88.847) = 5.491, p \textless .001, d = .92. Our results demonstrate that multiple use improved performance and that differing skill level is measured by the VCW system. The gamut of question topics and the question-dependent branching architecture of the conversation made a practice effect unlikely, as compared to subjects interpreting feedback data and adapting their interview strategy. Choice-based virtual interviews have the advantage of using inexpensive technology that does not require the language processing artificial intelligence of more complex systems. The system tracked increases in performance with a group of subject Novices that was near equivalent to a single performance from a subject-matter Expert group. This study demonstrates that structured virtual encounters can provide valid assessments and that intentional practice can improve performance. This successful design can serve as a model for virtual interview training in other professional domains such as military interrogation, standardized medical patients, and national security applications.
Reference:
Virtual Child Witness-Effects of single and multiple use on performance with Novice and Expert cohorts in a structured virtual human interview (Talbot, Thomas B., Lyon, Thomas D., Rizzo, Albert and John, Bruce), In Proceedings of the Interservice/Industry Training, Simulation, and Education Conference (I/ITSEC) 2016, 2016.
Bibtex Entry:
@inproceedings{talbot_virtual_2016,
	address = {Orlando, Florida},
	title = {Virtual {Child} {Witness}-{Effects} of single and multiple use on performance with {Novice} and {Expert} cohorts in a structured virtual human interview},
	url = {http://ict.usc.edu/pubs/Virtual%20Child%20Witness%20Effects%20of%20single%20and%20multiple%20use%20on%20performance%20with%20Novice%20and%20Expert%20cohorts%20in%20a%20structured%20virtual%20human%20interview.pdf},
	abstract = {Virtual human avatars can be used to train and assess a myriad of complex skills, such as interviewing, interpersonal, and clinical skills, in a safe environment that provides consistency, reduced cost, greater accessibility, and objective feedback. We created a structured virtual human interview which consisted of a conversational avatar that interacts verbally in response to on screen question choices. Our prototype was a forensic interview simulation called Virtual Child Witness (VCW). VCW provides a content-rich interview in response to open-ended questions and is designed to assess user’s interviewing strategy. In a quasi-experimental design, we evaluated 222 subjects to determine if the system could discriminate between Experts (M = .713, SD = .29961) and Novices (M = .373, SD = .27547), t (230) = 9.002, p {\textless} .001, d = 1.18, indicating easy discrimination between cohorts. A subset of 92 subjects completed the simulation multiple times; multiple use was associated with much higher performance (M = .727, SD = .23669) than single use (M = .489, SD = .33688), t (88.847) = 5.491, p {\textless} .001, d = .92. Our results demonstrate that multiple use improved performance and that differing skill level is measured by the VCW system. The gamut of question topics and the question-dependent branching architecture of the conversation made a practice effect unlikely, as compared to subjects interpreting feedback data and adapting their interview strategy. Choice-based virtual interviews have the advantage of using inexpensive technology that does not require the language processing artificial intelligence of more complex systems. The system tracked increases in performance with a group of subject Novices that was near equivalent to a single performance from a subject-matter Expert group. This study demonstrates that structured virtual encounters can provide valid assessments and that intentional practice can improve performance. This successful design can serve as a model for virtual interview training in other professional domains such as military interrogation, standardized medical patients, and national security applications.},
	booktitle = {Proceedings of the {Interservice}/{Industry} {Training}, {Simulation}, and {Education} {Conference} ({I}/{ITSEC}) 2016},
	author = {Talbot, Thomas B. and Lyon, Thomas D. and Rizzo, Albert and John, Bruce},
	month = dec,
	year = {2016},
	keywords = {MedVR}
}
Powered by bibtexbrowser