Downloadable - About University

450 Alternative decision-support systems
doctors coded biopsies of patients with Hodgkin’s disease and then
made an overall rating of severity. These overall ratings were very poor
predictors of survival time, but the variables the doctors coded made
excellent predictions when utilized in a linear additive model.
These early studies of the statistical modeling of judgment showed the
superiority of bootstrapping models over holistic judgment. However,
more recent studies present contradictory evidence, and it is to these
that we turn next.
Recent research
Many studies, conducted in real-world settings, have looked at bankruptcy
prediction. Libby, 43 in a major study, had experienced loan
officers make predictions for 60 real but disguised companies, half of
which had failed. These predictions were made on the basis of the
limited financial information contained in five financial ratios. Other
information such as absolute amount of income, notes to the accounts,
etc. was excluded from the experimental study. Nevertheless, the mean
predictive accuracy of the loan officers’ judgments was high, at 74%.
However, in this artificially limited study only nine of the 43 judges did
better than the ratio of assets to liabilities (see Dawes 44 for an insightful
discussion of the issues). A recent study investigated a group of venture
capitalists who were considered expert in identifying high-potential new
ventures. 45 Venture capitalist-backed ventures survive at a much higher
rate than those ventures backed by other sources. Nevertheless, 20%
of venture capitalist-backed firms still fail within five years. The study
found that bootstrap models of the venture capitalists outperformed
all but one participant – who achieved the same accuracy rate as the
bootstrap model. However, it is important to note that in this study the
venture capitalists were given a standardized set of data about real (but
disguised) companies – some of which had survived and some of which
had failed.
Whitred and Zimmer 46 point out that, in principle, loan officers may
outperform a linear model by the valid use of non-linear relationships
between ratios and (non-)bankruptcy. However, the robustness of the
models to violations of non-linearity will make this potential advantage
of man over model practically immaterial. For loan officers to systematically
outperform the model they must have access to information
unavailable to the model, information which may be prevalent in real
life rather than in laboratory situations.