2. Brief historic overview
Donders (1868/1969) formulated a way to investigate human information processing
by demonstrating that some tasks take longer to initiate than others. He proposed
a stage insertion/deletion method in which a relatively simple task could be made a
bit more difficult by requiring participants to do an extra ‘computation’ before
executing a response. By carefully manipulating this extra computation, Donders
reasoned that the time needed by this computation could be deduced by subtracting
the time to initiate the easier task from the time to start the more difficult task, much
like comparing the latency to generate the outcome of the sum 3þ4 with that of the
sum 3þ4þ5. To implement this general method in more detail, Donders defined
three characteristic tasks that differed in the processes needed to correctly fulfil the
task requirements. Type-a reactions/tasks were simple RTs reflecting only sensory
584 P. M. C. Lemmens et al.
and motor processes. The b-type reactions were choice RTs with two stimulus and
two response alternatives (including discrimination and response choice next to the
sensory and motor processes of the a-type task), whereas the type-c reactions were
similar to a go/no-go paradigm (with discrimination processes but no response
choice). This way the duration of, for instance, the response–choice process could be
estimated by subtracting response times of c-tasks from b-tasks. This method of pure
stage insertion was the first processing stages approach to choice RT (Sanders 1998).
However, difficulties were found creating sufficiently different type-b and type-c
tasks as it appeared that both types of tasks required response choice. To overcome
the methodological as well as statistical criticisms on Donders’ subtraction method,
Sternberg (1969) defined the additive factors method (AFM, sometimes also referred
to as additive factors logic, AFL) that assumed, like the subtraction method of
Donders, that response-time latencies consist of the sum of the intervals that a
sequentially ordered modular set of processing stages take to translate a stimulus
into a response and start executing this response (Sanders 1998). Each of these stages
takes care of processing a certain aspect of the stimulus or response, possibly
influenced by an experimental manipulation. The benefit of the AFM over Donders’
method were the statistical assumptions that Sternberg formulated, creating a solid
statistical framework in which experimental results from factorial designs could be
interpreted in only a single way. Interactions between two (or more) factors signified
that all factors acted upon the same information processing stage whereas additive
effects implied that each involved factor influenced a different stage. Thus, during
the heydays of the AFM around seven different stages were proposed or discovered,
among which the stages of feature extraction, stimulus identification, response
selection and motor programming received most attention and consensus (Gopher
and Sanders 1984, Sanders 1980, 1983, 1990, 1998).
However, these statistical premises, including ones derived from the basic
assumptions, were violated quite easily. For instance, the stage-robustness criterion
(Gopher and Sanders 1984), a derived assumption arguing for constancy of the stage
structure over many variations (including number) of factors over experiments, was
violated using stimuli typically used in experiments employing stimulus–response
compatibility paradigms (Ridderinkhof et al. 1995, Sanders 1998). Dual-route
architectures were proposed (de Jong, Liang and Lauber 1994, Kornblum et al. 1990,
Ridderinkhof et al. 1995) to overcome the limitations that the AFM imposed on
SRC experiments.
3. Stimulus–