Fundamental Issues, Models, and Theories 295calculate the speed of nerve transmission in humans. This
procedure involved measuring RT as a function of the dis-
tance away from the brain by applying a shock to the skin.
However, Helmholtz concluded that this procedure does not
yield an accurate measure of nerve conduction because the
measurements “suffer from the unfortunate fact that a part
of the measured time depends on mental processes”
(Helmholtz, 1867, p. 228).
The research of Helmholtz and others using RT to esti-
mate the speed of nerve conduction stimulated Donders and
his students to pursue the use of RT as a means for measur-
ing mental processes. De Jagger’s dissertation (1865/1970)
provided the first account of the experiments conducted in
Donders’s lab. The first part of De Jagger’s study continued
Helmholtz’s notion of measuring the speed of nerve conduc-
tion, but the second part focused on measuring the time re-
quired to identify a stimulus and select a motor response. In
one set of experiments, subjects were required to respond to
a red light with the right hand and a white light with the left
hand. The mean RT was 356 ms, which was 172 ms longer
than a simple reaction (executing a single response when a
stimulus is presented) to the same stimuli. De Jagger inter-
preted this time as the duration of the central processes in-
volving stimulus discrimination and response initiation.
Donders (1868/1969) formalized the subtractive method
used by De Jagger, emphasizing specifically that the time for
a particular process could be estimated by adding that process
to a task and taking the difference in RT between the two
tasks. He distinguished three types of reactions: type a (simple
reaction), type b (choice reaction), and type c (go or no-go
reaction; responding to one stimulus but not another). These
types of reactions allowed separate measures of the stimulus
identification and decision processes that were assessed
together by De Jagger. The difference between the type-c
and type-a reactions was presumed to reflect the time for
stimulus identification, and the difference between the type-b
and type-c reactions the time for “expression of the will”
(p. 424).
Reaction time research in general, and the study of action
selection in particular, continued to flourish throughout the
remainder of the nineteenth century (see Jastrow, 1890).
Wundt (1883) criticized Donders for using the type-c reac-
tion as a measure of stimulus identification, reasoning that
subjects must distinguish whether to respond, and suggested
using the type-d reaction instead as a pure measure. The type-
d reaction is measured by presenting subjects with the same
stimuli and having them make the same response every time,
as in the type-a reaction, with the difference being that they
are instructed not to respond until they have identified the
stimulus. However, Wundt’s type-d reaction quickly fell out
of favor because it is subjective and highly variable, and after
practice, the type-d reaction time does not differ from the
type-a reaction time. Criticisms of the subtractive method in
general led to its demise in the early twentieth century.Methodological and Modeling IssuesWith the advent of the information processing approach in
the 1950s and 1960s, the subtractive method was resurrected.
This method, and the stage analysis of RT data on which it is
based, came to be seen as sufficiently important to establish
Donders as a major figure in the history of human perfor-
mance. One influential use of the subtractive method was to
estimate the rate of mental rotation by varying the amount
that one stimulus was rotated relative to another to which
it was to be compared, and measuring the slope of the RT
function (Cooper & Shepard, 1973). Mean RT increased by
approximately 240 ms for each 20° increase in angle of rota-
tion, suggesting a continuous transformation in which each
degree of rotation took about 12 ms.
A major advance in stage analysis of RT data was the de-
velopment of the additive factors method by Sternberg
(1969). Like the subtractive method, the additive factors
method assumes discrete serial processing stages. However,
whereas the subtractive method provides duration estimates
for assumed stages, the additive factors method provides a
way to discover the stages themselves. Sternberg showed
that if two or more factors each influence the durations of
distinct stages, then the effect of one of the factors on total
duration will be invariant across the levels of the other
factors: That is, the effects of the variables on RT will be
additive. If two factors have interactive effects on RT,
then they must influence at least one common stage. Thus,
Sternberg advocated the use of multifactor experiments in
which the presence or absence of interactions among vari-
ables is used to determine the processing stages involved in
task performance.
Numerous limitations of the additive factors method have
been enunciated, including problems of accepting the null
hypothesis for additivity, assuming serial processing stages
with no feedback loops, and assuming constant output from
each stage (see Pachella, 1974). Despite these limitations,
the method has proven to be a useful tool for analyzing
the structure of information processing in a variety of tasks
(see Sternberg, 1998) because, as Sanders (1998) states, “the
method appears to provide a successful summary of a large
amount of experimental data” (p. 65). One criterion for eval-
uating the additive factors method is stage robustness: The
relations between two factors should not change as a function
of levels of other factors. Although there are exceptions,
stage robustness has generally been found to hold (Sanders,
1998).