acase@tulane.edu

Replication of dengue virus is mediated by the viral replicase recognizing the terminal regions of the dengue genome. Conserved RNA sequences in the 5' and 3' terminal regions form secondary structures for recognition and initiation of (-) and (+) strand RNA synthesis by the viral replicase. Based on an understanding of dengue replication, the 5' and 3' terminal regions were used as promoters for replication and the inducible expression of a foreign gene, eGFP. A transient expression system used cells infected with Vaccinia virus-T7 and transfected with a T7 promoter driven dengue virus genomic plasmid and T7 promoter driven dengue terminal region constructs containing eGFP. The orientation of the dengue terminal region constructs placed the T7 promoter driving expression of the 3' UTR, eGFP, the 5' TR and finally the hepatitis delta virus ribozyme in the reverse orientation. This generated a negative strand RNA species that did not express eGFP by itself. The presence of dengue viral replicase complex proteins was required to transcribe the negative strand template to a positive strand species with a functional open reading frame for reporter gene expression. This transient expression system was used to determine the requirements for (+) strand RNA synthesis from (-) strand RNA dengue terminal region constructs by the viral replicase. As a result, the requirements for inducing expression of the foreign gene were determined using several variants of this system containing either the 5' UTR or larger fragments of 200 or 250bp of the dengue 5' terminal region / acase@tulane.edu

acase@tulane.edu

Monte Carlo simulation techniques were used to replace missing completely at random data (MCAR) in Likert scaled items with a monotone missing pattern. Seven missing data techniques (MDTs) were implemented: item mean substitution (IMS), person mean substitution (PMS), single expectation-maximization imputation (EM), regression multiple imputation (REG), Markov Chain Monte Carlo multiple imputation (MCMC), propensity scores multiple imputation (PROP), and complete case analysis (CC). In addition to these MDTs, five other factors were manipulated: percentage of missing subjects, percentage of missing items, sample size, number of items per scale, and number of response choices Phase I examined bias and accuracy. Only the scale mean was unbiased by most of the MDTs. PMS was an exception because it underestimated the mean. The scale variance was statistically biased by IMS, PMS, REG, and PROP. The variance was consistently underestimated by IMS, REG, and PROP while overestimated by PMS. The sample size had the least impact on the bias of the variance, and the bias was most affected by an increase in the number of items. Cronbach's alpha was biased by every method except EM. IMS, PROP, MCMC, and CC led to underestimation whereas PMS and REG caused overestimation of alpha Overall, EM provided the most accurate estimates of the scale mean, variance, and alpha, followed by MCMC and REG. The least accurate MDTs for the mean were PMS and CC while IMS and PROP were the least accurate for the variance and alpha Phase II used the best MDTs from phase I (EM, MCMC, and REG) to predict linear regression coefficients in a real data application. All three MDTs provided accurate results. The goal of this dissertation was to establish guidelines for the future analysis of Likert scaled items with missing data. This study found EM, MCMC, and REG to be the best MDTs for monotone MCAR data / acase@tulane.edu

acase@tulane.edu

acase@tulane.edu

acase@tulane.edu

acase@tulane.edu

acase@tulane.edu

acase@tulane.edu