Page 15 - Windhoek Graduation (20 May)(SESSION 2)(PhDs, Masters)

Executive Dean: Prof Martha N Nickanor
Faculty of Agriculture, Engineering & Natural resources
SCHOOL OF SCIENCE
Associate Dean: Prof. Veikko Uahengo

DOCTOR OF PHILOSOPHY IN SCIENCE (APPLIED STATISTICS)

CANDIDATE: CHARAMBA Vonai

CURRICULUM VITAE

Vonai Charamba was born in Masvingo, Zimbabwe. Her qualifications include
MSc Statistics (Distinction with Dean’s Meritorious Award), BSc Special Honours
in Statistics (First Class) and a Diploma in Statistics - all from the University of
Zimbabwe (UZ). She also holds a Diploma in Secondary Education (Distinction
in Mathematics) from Mutare Teachers College (UZ). Vonai did her Advanced and Ordinary Level studies at Lundi
High School. She is currently working as a Biostatistician in the School of Agriculture and Fisheries Sciences of the
University of Namibia. Her past working experience includes working as a Statistics Lecturer at Chinhoyi University
of Technology and as a Statistical Analyst for SNV Netherlands Development Organisation and Infoholdings. She
also worked as a Monitoring and Evaluation Officer for CARE International in Zimbabwe and as a Mathematics
Teacher (HOD Science and Mathematics Department) at Chitowa 1 Secondary School in Murewa. Her research
interests are in the application of Item Response Theory Modelling to Food Security Measurement, Bayesian
Estimation and application of Statistics in Agriculture.

CANDIDATE’S DISSERTATION

A HIERARCHICAL NON-PARAMETRIC BAYESIAN TESTLET MODEL FOR DUAL LOCAL DEPENDENCE

The doctoral study was undertaken and completed under the supervision of Prof Lawrence Kazembe of the
University of Namibia as Main-Supervisor and Prof Ndeyapo M. Nickanor of the University of Namibia as Co-
Supervisor.

The candidate investigated the effect of ignoring person and item clustering in item response theory (IRT)
proficiency measurement. Standrd IRT models assume local person and item independence. However,
respondents are often clustered while test items are grouped according to sub-content measuring the same
trait. Thus study presented a non-parameteric polytomous multilevel testlet model for simultaneously modelling
person and item clustering effects. The person grouping variable was assumed to be unknown and determined
from the data using the Dirichlet process. The model was compared to models ignoring either person or item
clustering effects or ignoring both, in terms of systematic, random and total errors in parameter estimation and
test information and reliability for simualted and real life data. Ignoring person clustering effects resulted in
increased bias and total errors in the estimation of proficiencies while ignorance of item clustering negatively
impacted on item parameter estimation increasingly with sample size, testlet size and number of category
options. In addition, ignoring item dependency resulted in overestimation of test information and reliability.
The application of the model to real life data has shown that it can offer an alternative to food security access
measurement by combining the Household Food Insecurity Access Scale, the Household Dietary Diversity Scale
and the Months of Adequate Household Food Provision into one scale. That indicates that it is worthy of further
investigation.

15