Biomedical Engineering Reference
In-Depth Information
reaction to such fi ndings, enhancing the likelihood of
program success, will not invalidate the evaluation effort,
even though an initial judgment predicted program
failure. Only a misunderstanding of the nature of evaluative
research could foster the view that the training intervention
is fi xed.
Assume that an evaluation of a training program is
underway. The program, in essence, takes a set of inputs and
given conditions,
Z
, and by means of some process,
G
,
transforms the inputs into an output described by the
dependent variable,
x
. We assume
x
to be a behavior. The
dependent variable may be a business measure, such as
number of reworked batches, or an index, such as
Occupational Safety and Health Administration (OSHA)
recordables. The evaluator is randomly assigning employees
to control or treatment groups, manipulating variables,
recording and communicating results, etc.
Thus behavior
x
is a function
G
of a complex state of
affairs
z
, given by:
x
=
G
(
z
)
[12.1]
This says
G
and an index
z
of the set of independent
variables
Z
are suffi cient for the prediction of the dependent
variable
x
, in the absence of dissemination of
G
or
z
. This
can be represented by a two-dimensional diagram
(Figure 12.1).
We note that for a given interval [z(o), z(i)], x will have a
range of [x(o), x(i)]. Thus z might be an index of
prior training history, on-the-job experience, etc. and x, a
measure of productivity such as unit output, impounded
batches, or quantity reworked. The set Z would include such
items as appear in the employee's training history, etc.
We assume throughout that the interval of x is continuous
and closed.
