Look at the faces above.
Suppose that each of these somehow represented the overall financial state of a company,
in which ones would you invest? Suppose each one represented the global data representing the
credit-worthiness of an individual who had applied for a loan, to which ones would you grant the loan?
The Empathic Visualisation Algorithm (EVA) supports this kind of analysis.
In fact the faces above were generated from the financial data of companies (apart from the last two
in the list which illustrate how the faces are displayed).
The EVA requires two inputs – first the actual data, which is any data matrix of n rows on k variables.
For example, this could be n companies measured on k financial indicators. Typically both n and k would
be large (though with n greater than k). Second, the data is always examined for a purpose – let’s call
the person or people who make decisions on the basis of the data the ‘users’. The user has a set of
goals in mind when examining the data. The goals are typically prioritised (though they do not have to
be). What the EVA does is to provide a mapping so that the most important goals are transformed into
the most important emotional facial expressions. For example, suppose the most important goal when
examining a group of companies is that profitability must be above a certain percentage of turnover.
Let’s suppose that this particular goal is mapped into ‘happiness’. Then the higher the profit in
relation to turnover, the more happy the face will look. On the other hand, suppose that staff turnover
is mapped to anxiety. Then the greater the staff turnover, the greater the anxiety shown on the face.
It is then possible to get a face that in its overall aspect, looks happy, but with a trace of anxiety
showing through. Since humans are experts (without training!) of picking up emotional expression in faces,
a single glance will show that such a company is overall doing very well in profitability, but at the
same time there is cause for concern in the higher than wanted staff turnover. Of course this is a simple
example to get the idea across – but this can be generalised to a larger combinations of variables (and many