Ml. Levin et S. Mitra, RESEARCH METHODS - VISUALIZATION IN THE NUMERIC SOLUTION OF COMPLEX NONLINEAR EQUATIONS IN MATHEMATICAL DEMOGRAPHY, Social science computer review, 12(4), 1994, pp. 625-640
Citations number
22
Categorie Soggetti
Social, Sciences, Interdisciplinary","Computer Sciences, Special Topics","Computer Science Interdisciplinary Applications
Searching for numeric solutions to complex nonlinear equations is a re
source-consuming, risky business. For a given equation, there can be a
n infinite number of often dramatically quite different solutions meet
ing a specified goodness-of-fit criterion. Although the leading statis
tical packages have nonlinear regression procedures, these procedures
all require ''seeding'' with initial estimates of equation parameters.
Often one has only the vaguest notion as to what might be realistic e
stimates of these parameters. Yet one selection might lead to a failur
e to find any solution, whereas another might lead to a less-than-opti
mal, or even less-than-close-to-optimal, solution. The source of this
problem is, of course, that nonlinear equations can generate a series
of local minma. Consequently, for any specific investigation involving
nonlinear equations, barring some theoretical, empirical, or mathemat
ical basis for estimating initial parameter values, one is reduced to
trial-and-error methods, exceptional good luck, or a combination of bo
th. With the advent of powerful, distributed computing and the diffusi
on of desktop graphical capability, visualization techniques hold pote
ntial for short-circuiting the trial-and-error process. In this paper
we report on one such effort using an off-the-shelf Windows-based, cur
ve-fitting program as the visualization engine for an eight-parameter,
nonlinear equation derived by Heligman and Pollard to model mortality
. First, we review the theory and mathematics of the so-called law of
mortality. Second, we develop the working equation for solution and de
scribe the data sources. Third, we attempt unsuccessfully to use stand
ard techniques to solve the equations. Finally, we develop and illustr
ate a visualization technique designed to generate initial parameter e
stimates. We then apply the visualization technique and successfully s
olve the equations for 22 sets of mortality data.