Samenvatting

The purpose of the study is to test whether Modified Cognitive Interviewing (MCI) is an effective method for detecting deceptive human eyewitness accounts in a computer-mediated communications with limited text space (such as in Twitter). The study is based on a previous study from Morgan, Christian, Rabinowitz, Palin and Kennedy (2015), where MCI has proven to be an effective method, with the use of face-to-face interviews. 
In total 44 college students participated in this study, where they either had to perform, or pretend that they had performed a cognitive task. 15 students had to perform a task, and were instructed to answer completely honest when interviewed (truthful group), 16 students performed the task, and when interviewed they had to deny that they performed it, and instead make up a story (deceptive group), and 13 students read the instructions of the task, and had to claim that they performed the task, without actually having done it (false claim group). 
The interview was held in a Twitter format, where six questions were sent out, and participants answered them one by one, with a maximum of 140 characters per answer. Every interview took place in a supervised space, where participants were able to fully concentrate. The questions (tweets) were the six prompts as known in the MCI method. Further explanation on these six prompts is given in Appendix B. 
After the information gathering, the tweets were first used as input for a survey. The tweets processed in the survey were rated by (former) law enforcement professionals, and people with expertise in lie detection. The experts had to rate the first tweet and the whole Twitter conversation, and had to give their confidence level for every rating. The rater’s ability to correctly distinguish truthful participants from deceptive participants for only the initial prompt in MCI resulted in 48%, while the rating of the whole interviews resulted in 54% of success. The 54% success rate of the whole interviews went along with a confidence level of 3.6 out of 5 (72%), this means that the human raters overestimated their abilities to discriminate truthful from deceptive participants. 
Other than for the surveys, the tweets were also used as input for the computer-processed text analysis. After eliminating all fill words and repeating sentences in the tweets, the analysis determined the Response Length (RL), Unique Word count (UW), and Type Token Ratio (TTR). These so called “speech-content variables” were compared with each other, related to the different groups of participants (truthful, deceptive, and false claim). The outcome is that the computer-processed analysis, by using the speech-content variables RL and UW, is effective in lie detecting through computer-mediated communication with a limited text space, such as in Twitter (RL: 70% and UW: 79%). 
As a comprehensive answer to the problem statement, we can conclude that the use of MCI is an effective method of detecting deception in a limited text based environment, such as Twitter. However, this only applies when using computer-processed analysis, with the speech-content variables RL and UW as the leading factors.