Robust Regression Methods For Massively Decayed Intelligence Data

ROBUST REGRESSION METHODS FOR MASSIVELY DECAYED INTELLIGENCEDATA byAKIVA JOACHIM LORENZMay 2014 Advisor: Dr. Barry MarkmanMajor: Evaluation and ResearchDegree: Doctor of Philosophy Homeland Security, sponsored by governmental initiatives, has become a vibrantacademic research field. However, most efforts were placed with the recognition ofthreats (e.g. theory) and response options. Less effort was placed in the analysis of thecollected data through statistical modeling. In a field that collects more than 20terabyte of information per minute though diverse overt and covert means and indexesit for future research, understanding how different statistical models behave when itcomes to massively decayed data is of vital importance.Using Monte Carlo methods, three regression techniques (ordinary least squares,least-trimmed, and maximum likelihood) were tested against different data decaymodels presumed to be found in homeland security research studies in order to testwhether these techniques will preserve the Type I error rate in the t-test ofstandardized beta.The results of these Monte Carlo simulations (sample size n=30,90,120,240,480and 100,000 iterations) showed that the least trimmed squares method should beavoided under any circumstance due to the lack of a defined standard error, while themaximum likelihood technique should be avoided with smaller sample sizes due to theinflated Type I errors. Interestingly, although it is known that the ordinary least squares