Agriculture Reference
In-Depth Information
“For a forecasting system to be successful, it must be adopted and implemented by
growers. There must be the
perception
that the grower can realize specific, tangible
benefits from using the forecasting system
that could not be realized in its absence
.”
(Campbell and Madden, 1990, p 424; our emphasis).
The quotation from Campbell and Madden (1990) highlights the important
relationship that any forecasting system (IT-based or not) must have with the
existing knowledge and decision making system of its intended users. The basic
point underlying Campbell and Madden's (1990) perceptive statement is that in
order to be useful a forecasting system must either: (1) provide new information to
the user, or (2) extract previously unrecognized information from already available
knowledge and data. Tools developed with these aims in mind might be better
described as judgement-assistance or risk assessment tools than DSSs. The
objectives for such tools would prioritise user attitudes and preferences, and
therefore increase likelihood of not only use, but also feedback, leading to further
improvement through active participation of the users in development. Such tools
guide growers to a solution, which takes account of the conclusions from the
research but which also makes use of growers' experience and their perceptions of
risk. It is worth noting the obvious (but sometimes overlooked) point that the
combination of objectives (1) and (2) above, constitutes an exhaustive summary of
what it is possible for any forecasting system, constructed from empirical data and
mathematical, statistical and logical relationships, to do; no such system can contain
(or deliver) more information than the content of its components (Doucet and Sloep,
1992). Medawar (1972) made the same point more generally:
“No process of
reasoning whatsoever can,
with logical certainty
, enlarge the empirical content of
the statements out of which it issues
” (our emphasis).
This epistemic limitation of forecasters, tools, or DSSs holds, no matter how
sophisticated the IT infrastructure is that is used in their development or
deployment. Medawar's comment is particularly worth bearing in mind when the
role of the IT-based forecaster or DSS in disease management is considered in more
detail. In addition to the obvious first point (concerning the empirical limitation to
the knowledge content of forecasters) it also makes a more subtle point about
forecasting. Forecasting entails the loss of logical certainty because forecasts always
seek to make statements beyond the empirical content of the information on which
they are based. It is thus logically unavoidable that predictions are uncertain, even if
we construct deterministic predictors of disease outbreaks that ignore that
uncertainty. In the terminology of logical analysis we can say that forecasters are
tools for
inductive
reasoning.
However, acceptance of the inductive nature of forecasts and forecasters offers
the possibility to evaluate them within a coherent analytical framework using
Bayesian analysis within an information theoretical framework. This approach leads
to a simple and quantitative assessment of the conditions under which any given
forecaster or DSS is likely to meet Campbell and Madden's (1990) '
success
'
criterion. (i.e., the extent to which a grower is likely to perceive it as offering
something tangibly informative). Evaluation of forecasters in this context is likely
to result in more efficient use of IT. It could prevent pointless spending on the
development of IT tools which will be useful only to a small proportion of the
