Implementation Criteria for the Shotgun Stochastic Parameter Estimation Algorithm
Date
2014-01-09
Authors
Moodley, Devandren
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The use of rank ordered data is widespread and commonplace in fields of
research from marketing to biology and medicine. Complexity and the
mathematical rigour required for the analysis of rank ordered data has resulted
in the use of parametric methods for data analysis. There have been conflicting
views on the validity of the results produced by parametric methods.
Developments in stochastic methods have allowed researchers the opportunity
to bypass the complexity and rigour associated with rank ordered data analysis
by trading accuracy for simplified solutions. The cost of computing has been the
limiting factor for the use of these techniques. With the recent decline in
computing cost and increasing accessibility more researchers are now turning
their attention back to these alternatives.
Stacey (2006) introduced one such method known as the Shotgun Stochastic
Parameter Estimation Algorithm, to determine the mean and variance of rank
ordered data. This study determined the reliability of the algorithm by
developing guidelines that ensured consistency in the results. This was
achieved by observing the output as key variables were changed. Another
important contribution of the algorithm was its ability to analyse partially ranked
data. This research investigated the behaviour of the SSPEA with partially
ranked data and quantified the error introduced including possible solutions to
reduce the error.
The data used for the study was obtained from Grant (2010) for some test
cases and simulated internally for others. Similar to the incorrect application of
parametric methods to rank ordered data; this research proved that the
algorithms true potential is realised by correct choice of simulation software and
efficient implementation. A set of guidelines and key considerations is produced
that will assist researchers utilising this algorithm to prevent potential pitfalls in
the data analysis. The study has demonstrated that the algorithm can be used
with confidence to analyse rank ordered data provided these guidelines are
followed and considerations noted.
Description
MBA thesis
Keywords
Rank ordered data, Stochastic methods