Вы находитесь на странице: 1из 19

Journal of Experimental Psychology

Monograph
Vol. 79, No. 3, Part 2 March 1969

FOREPERIOD EFFECTS IN SIMPLE REACTION TIME:


ANTICIPATION OR EXPECTANCY? 1

JOAN GAY SNODGRASS 2


New York University

An experimental method for determining the location of S's "true" simple RT


distribution consists of rewarding him for responding consistently by means of
narrow payoff bands, systematically varying the position of these payoff
bands along the time axis, and identifying as the true RT distribution that one
having minimum variability. The location of this true RT distribution was not
significantly affected by increasing the foreperiod range from 0 through 300
msec. These and other results were taken as evidence for an anticipation
model as opposed to an expectancy model for foreperiod effects in simple
RT. The anticipation model proposes an underlying RT distribution of low
variability which is unaffected by foreperiod variability and payoffs, but which
combines with more variable distributions of time estimations to produce the
observed RT distributions. The empirical RT distributions were highly
peaked with high tails and, consistent with previous results, were better de-
scribed by the double monomial (DM) distribution than by either the normal
or exponential. However, the exponential was superior to the DM for
distributions composed primarily of true reactions. Strong sequential effects
over as many as 20 trials were observed, and were greater for no-feedback
than feedback conditions.

Simple reaction time (RT) has been found a linear function of stimulus uncertainty,
to increase with increased uncertainty in when uncertainty is defined as the sum of the
the time of stimulus occurrence, regardless variance of 5"s time estimates of durations
of whether such uncertainty is manipulated equal to the mean FP duration and the vari-
by changes in foreperiod (FP) duration, FP ance of the FP durations imposed experi-
variability (Karlin, 1959; Klemmer, 1956; mentally by E. A related finding is that
Woodrow, 1914), or the proportion of catch within a series of variable FPs, mean RT
trials (Drazin, 1961). Klemmer (1957), varies inversely with duration of the immedi-
analyzing these effects within the context of ate FP presented on the trial (Drazin, 1961;
information theory, found that mean RT was Karlin, 1959; Nickerson, 1965; Woodrow,
1
This paper was based on a dissertation sub- The computer time was provided by National Insti-
mitted to the Faculty of Philosophy, University of tutes of Health Grant FR-1S to the Johnson Re-
Pennsylvania, in partial fulfillment of the require- search Foundation, School of Medicine, University
ments for the PhD degree. The author wishes to of Pennsylvania, and by a grant from the Computer
express her gratitude to R. Duncan Luce, for advice Center of the University of Pennsylvania. A
in the conception and analysis of the experiment, short report of this work was read at the meetings
and to R. Duncan Luce, Francis W. Irwin, Allan J. of the Eastern Psychological Association, Boston,
Nash, and Allen M. Schneider for criticisms of a April 1967.
2
preliminary draft of the paper. Requests for reprints should be sent to Joan Gay
The experimental and theoretical support for this Snodgrass, Department of Psychology, New York
work was provided by National Science Foundation University, University Heights, New York, New
Grant GB 1462 to the University of Pennsylvania. York 10453.
1
1969 by the American Psychological Association, Inc.
JOAN GAY SNODGRASS

1914). This phenomenon has been termed he is less likely to make false reactions un-
"the immediate FP effect." der the former than under the latter condi-
Although there is agreement on the exist- tions. According to the anticipation model,
ence of these empirical phenomena, there is then, mean RT increases with increased FP
disagreement on their interpretation. The duration or variability because spuriously
oldest interpretation, here referred to as the fast RTs due to time estimation drop out,
expectancy model, proposes that RT varies not, as the expectancy model proposes, be-
inversely with level of readiness or attention, cause 5"s level of readiness decreases.
and that readiness in turn decreases as the To test the anticipation model, a method
length or the variability of the set of FPs is needed of observing the true RT distribu-
used within a session increases, or as the tion under different degrees of FP variabil-
length of the FP on a particular trial de- ity. To do this, one must first make some
creases (Karlin, 1959; Woodrow, 1914). assumptions about differences between true
An alternative interpretation, here called reactions and time estimates, and then at-
the anticipation model, was suggested by tempt to take advantage of these differences
Nickerson (1965) to account for the immedi- by means of payoffs.
ate FP effect he observed in his data. He True reactions are assumed to differ from
noted that as the duration of the immediate anticipatory time estimates in two ways:
FP increased, so also did the frequency of first, they are longer; second, they are less
false reactions (responses prior to the reac- variable. The reason why they are longer
tion signal) and the frequency of spuriously is obvious. The reason why they are less
fast RTs (<130 msec.) which he assumed to variable is based on the following reason-
have been initiated prior to the reaction stim- ing: If, as is commonly done, true RT is
ulus. He attributed at least part of the assumed to be the arithmetic sum of stimulus
immediate FP effect to an increased tendency input time, decision time, and motor response
of S to anticipate the signal as time from time, then time estimates must also include
the warning signal, and hence FP duration, each of these three steps plus an additional
increased. componentdelay. The interposed delay is
The anticipation model proposed here is an assumed to add some variability of its own,
elaboration of Nickerson's initial observa- which increases monotonically with the in-
tions. Under the variable FP conditions of terval. In fact, there is good empirical evi-
Nickerson's experiment, 6" is assumed to pro- dence that the variability of time estimates
duce a mixture of anticipations (or time es- increases with the duration of the interval
timates from the warning signal) and true being estimated (Snodgrass, Luce, & Galan-
RTs. Because the proportion of anticipa- ter, 1967; Woodrow, 1930, 1933).
tory time estimates he makes increases as In order to encourage to make true reac-
immediate FP duration increases, and be- tions rather than anticipatory time esti-
cause time estimates result in shorter RTs mates, he should be rewarded for RTs of
than true reactions, 6"'s mean RT decreases intermediate length and low variability. The
with increased FP duration within a vari- band payoff was developed to do just that.
able FP condition. The proposed anticipa- Under the band payoff, 5" is given informa-
tion model thus accounts for the entire im- tion feedback and monetary payoffs for re-
mediate FP effect by assuming differential sponding within a narrow latency band, and
rates of anticipatory time estimates for the is penalized for responding outside the band,
different FP durations. regardless of whether he is too fast or too
The anticipation model can also account slow. Because the location of the true RT
for other FP effects such as the increase in distribution is not known a priori, the band
RT with increased mean FP duration or in- payoff is moved along the time axis from a
creased FP variability. It simply assumes latency band clearly too fast for true RT
that S makes more anticipatory time esti- to one clearly too slow. The true RT dis-
mates under fixed or short FP durations than tribution is identified as that distribution of
under variable or long FP durations because RTs showing least variability, and the band
FOREPERIOD EFFECTS IN SIMPLE RT

location producing this distribution of mini- initiated by the warning signal and are esti-
mum variability is called the optimum band. mates of the FP duration + A t, where A t is
Snodgrass et al. (1967) varied the posi- less than the true RT. These may occur
tion of a 20-msec. wide payoff band from prior to or after the reaction signal. If
85-105 msec, to 185-205 msec, after the on- they occur prior to the signal, they are re-
set of an auditory reaction signal which in corded as false reactions, and if they occur
turn was delivered after a fixed 2-sec. FP. after the signal, they are recorded as spuri-
They found that the location of the optimum ously fast RTs. These time estimates are
band varied from 105-125 msec, to 145-165 here denoted EWs, and differential rates of
msec, over four 5"s. As the band position their occurrence have been postulated pre-
departed from .9's optimum in either direc- viously to account for traditional FP effects
tion, RT variability generally increased in RT.
monotonically for each 6". These results The second kind of time estimates are ini-
were interpreted as indicating that under the tiated by the reaction signal and result in
optimum band, 5" reacted to the reaction sig- RTs which are slower than true reactions.
nal and thereby produced the least variable These are here denoted ERs. They are not
RT distribution, whereas under faster bands assumed to play a role in accounting for FP
he made time estimates of the FP duration effects in traditional RT experiments, but
and under slower bands, time estimates from they are assumed to occur whenever 5s are
the reaction signal of a time interval longer explicitly motivated to delay reactions, as is
than the true reaction time. Under condi- the case in the present study. True reactions
tions in which 5" is rewarded for making re- are denoted Rs.
sponses which are slower than true reactions, The model assumes that the distributions
then, he is assumed to make time estimates of the three types of responses are unimodal
from the reaction signal, rather than from but that, as noted above, their means and
the warning signal. variances differ in consistent and significant
The present experiment was designed to ways. Obviously, /ABW < MB < /*ER when
provide a more stringent test of the anticipa- measured with respect to the onset of the
tion model by observing the effect of different reaction signal; for the present experimental
degrees of FP variability on performance conditions, the relationship among the vari-
under payoff bands positioned at various ances is <72B < <T2BB < <T2EW-
latency bands. Before presenting detailed
predictions of the anticipation model for this PREDICTIONS OF THE MODEL
situation, its major assumptions are sum- 1. Predictions of the anticipation model for
marized and some notation introduced. the joint effect of band position and FP vari-
The major assumption of the anticipation ability on RT variability, or more exactly
model is that for given stimulus conditions on the percentage of responses within the
and a given S, there is an underlying dis- payoff band (Pb), are summarized in Fig. 1.
tribution of true RTs which is unaffected by As can be seen in Fig. 1, PD is an inverted
stimulus uncertainty, at least when manipu- U-shaped function of band position for each
lated by FP variability, and unaffected by FP condition. Consistency is maximum at
motivational aspects of the experiment such the optimum band, decreases for slower-than-
as explicit payoffs. However, FP variability optimum bands because 5" must introduce
and payoffs affect the proportion of time esti- a delay into his reactions (and the bigger the
mates S makes. Time estimates are here de- delay, the more variability is introduced),
fined as voluntary estimates of a temporal and decreases for faster-than-optimum bands
interval which are initiated by a stimulus and because a larger and larger proportion of EW
terminated by a motor response. responses are made as band position de-
Time estimates may be of two kinds, one creases. Under the conditions of the present
of which results in responses that are faster experiment, EWs are more variable than
than true reactions and the other, in re- ERs because the former are estimates of
sponses that are slower. The first kind are longer intervals. Thus the consistency func-
JOAN GAY SNODGRASS

as possible, and they are rewarded for con-


HIGH VAR. ft sistency on the basis of the number of re-
MED. VAR. FP sponses in their most consistent, or modal,
LOW VAR. FP
latency band as determined at the end of
the session. The initial determination of
the modal band necessarily involves no feed-
back, but in a subsequent session S is given
information feedback in addition to payoffs
for responding within the same modal band
determined in the no feedback session. This
OPTIMUM method is called the floating-band method, as
opposed to the previous fixed-band method
* FAST SLOW in which predetermined bands are systematic-
BAND POSITION ally varied in position.
FIG. 1. Predictions of the anticipation model of A final purpose of the present experiment
the effect of band position and FP variability on was to investigate the effect of payoff band
the percentage of responses within the payoff band. width under both the floating- and fixed-band
procedures.
tion is lower on the fast side of the optimum
band than on the slow side. THEORETICAL DISTRIBUTION OF RTs
The basic assumption of the model, that A separate problem is that of the correct
FP variability does not affect true RT, taken theoretical distribution for simple RTs. The
with the additional assumption that under the anticipation model makes no assumption
optimum band 51 makes only true reactions, about the distributions of true reactions or
leads to the prediction that FP variability time estimates other than that their distribu-
will have no effect on either the position of tion is unimodal. Various models of the reac-
the optimum band or on the absolute level tion process predict the ordinary gamma dis-
of consistency within that band. Under tribution (McGill, 1963), the general Erlang
faster-than-optimum bands, higher levels of or generalized gamma (McGill & Gibbon,
FP variability will have a deleterious effect 1965), and the negative binomial (LaBerge,
on consistency because the estimated interval 1962) for the distribution of Rs. Snodgrass
is itself variable, whereas at slower-than- et al. (1967) found that their simple RT
optimum bands, FP variability will not affect distributions collected under 20- or 100-msec.
response consistency because estimates are wide payoff bands were better fitted by the
made from the reaction signal, rather than double monomial (DM) distribution than by
from the warning signal. the displaced gamma (with the starting point
2. The anticipation model predicts that a parameter) by a likelihood comparison,
the immediate FP effect will be present at whereas two-choice RT distributions were
faster-than-optimum bands and absent at better fitted by the displaced gamma. The
optimum and longer bands. This is because DM is composed of two power functions back
the immediate FP effect is assumed to be pro- to back, meeting at the theoretical mode,
duced by EW responses and 5" is assumed and has two characteristics found by Snod-
to make these only under faster-than-opti- grass et al. (1967) to be typical of empiri-
mum bands. cal simple RT distributions under band pay-
The present experiment also investigates off conditions-extreme peakedness and high
an alternative method of inducing -Ss to make tails. It was derived by Luce and Galanter
true reactions. It is based on the assump- (1963) as the form for the response generali-
tion that 5"s "know" that true reactions are zation function in magnitude estimation. In
more consistent than time estimates, thus the present study, RT distributons were
proper motivating conditions should elicit analyzed for their theoretical distribution
their true RT distributions. In this method, by fitting the DM, normal, and exponential
5"s are instructed to respond as consistently distributions to each of the empirical distribu-
FOREPERIOD EFFECTS IN SIMPLE RT

tions;, and comparing the adequacy of fit of Each 5 participated in at least 30 experimental
each. sessions. The first six were practice sessions de-
signed to give 6" experience with each of the pro-
METHOD cedures used in the experiment proper. Data from
these sessions are not reported here. There were
Apparatus.The 6"s sat in an Industrial Acoustics 24 experimental sessions, each consisting of 600
Co. acoustic chamber which attenuated outside noise reactions and lasting between 4 and 14 hr., depend-
by 30-65 db. over the audible frequency range from ing on the number and duration of self-imposed
100 Hz. The warning and reaction signals were breaks 5" took. Sessions were normally run on
short (S-msec.) high-intensity clicks produced by separate days, although occasionally two sessions
discharging a capacitor through an amplifier and were run during the same day.
then through a pair of earphones. The click when The 6"s received a base pay of $1.50 per session.
measured on an oscilloscope had a rise time of They received additional money depending on their
5 jttsec. to a maximum peak of 1.7 v. dc. The clicks consistency of responding as measured by the
were presented to the right ear through Permoflux number of responses falling within the payoff band
PDR-8 earphones. in force. The payoff bands were of two types, float-
The FP durations were controlled by a syn- ing and fixed. The positions of the floating bands
chronized motor-driven cam switch and were ac- were determined by 5's performance during sessions
curate to within 2-5 msec. For the fixed FP con- in which no feedback was given. The 6" was in-
dition (FP 1), the FP duration was always 944 structed during the no-feedback (NFB) sessions
msec.; for the medium-variability condition (FP 2), to respond as consistently as possible, and was told
two FPs, 944 and 980 msec., were presented ran- that he would be paid on the basis of the number
domly; and for the high-variability condition (FP of responses in the band of a prescribed width
3), the randomly presented FPs were 786 and 1,070 that contained the maximum number of responses.
msec. Selection of the FPs was controlled by a Each such band was used in a subsequent feedback
punched tape of probability .5 read by a Western (FB) session. The conditions investigated during
Union Teletype. The intertrial interval was 3 sec. the floating-band conditions were the three degrees
The 5" responded by depressing a microswitch of FP variability and two band widths, 20 and 50
that required a small amount of travel. The clock msec.
was started at the onset of the reaction signal (Si) Fixed bands were determined prior to the experi-
and latencies were timed to the nearest millisecond ment and were the same for all 5s. Five 20-msec.-
by a Hewlett-Packard electronic counter and they, wide fixed bands, 70-89, 90-109, 110-129, 130-149,
along with the FP preceding the reaction signal, and 150-169 msec., were run with each of the
were recorded by an associated printer. False reac- three FP conditions.
tions were not timed, but their occurrence was Three band widths'50, 20, and 5 msec.were
recorded by E. A three-section display panel in each run with the fixed foreperiod condition (FP 1)
front of >? provided feedback after every trial. The and feedback. The locations of these three bands
feedback informed 5" of four possible events. A were determined somewhat differently. The posi-
red light indicated a false reaction, a minus () tions of the 50- and 20-msec.-wide bands were
in the leftmost section of the panel a response after determined by 5"s performance during NFB. These
Si but too fast for the band, a one (1) in the are floating bands, as described above. The posi-
middle section a response within the band, and a tion of the 5-msec. band was determined from S's
plus (+) in the rightmost section a response too performance during fixed-band sessions under FP
slow for the band. The feedback was given manually 1. That 5-msec. band of latencies within the opti-
by E after she observed the RT on the electronic mum payoff band under FP 1 which contained the
counter, and thus it appeared to S at somewhat largest number of responses was selected as the
random intervals after the reaction had been made. payoff band.
A rough timing of these intervals indicated that The amount of money won for each response
feedback was given somewhere between 300 and within the band varied with the band width and
400 msec, after the response. was i$ for 50-msec.-wide bands, 4tf for 20-msec.-
The 5" started and stopped the series of trials by wide bands, and 2^ for 5-msec.-wide bands. The
a toggle switch. Verbal communication between E 5"s received no payoff for responses outside the
and 5" was conducted via an intercom. band or for false reactions.
Subjects and procedure.-Five 5"s participated in The 5 was allowed as many practice trials at
the experiment. Three, CP, ES, and AH, had had the beginning of the session as he wished. These
extensive experience in similar RT experiments. trials were not recorded, and he won no money
The remaining two, BR and MW, were naive 5"s for them. The S"s were encouraged to win as much
recruited through the University of Pennsylvania as they could, and they showed every evidence of
student employment service with the promise of being highly motivated to respond within the band
good hourly pay. CP was female. No 5 reported as much as possible. They were also encouraged
any hearing defect, and all 5s were right-handed to refrain from responding if they felt their atten-
with the exception of ES, who nonetheless used his tion had wandered or they were not quite ready
right hand to respond. when the reaction stimulus occurred. The purpose
JOAN GAY SNODGRASS

distributions were plotted by a Control Data


ISO NO FEEPBMK FEEDBtCK Corporation 160A computer. For each S,
10 distributions were obtained with fixed
FPs and 14 with variable FPs, for a total
of 24 that each included 600 observations.
In addition, the data from the 14 variable FP
conditions were separated into the two dis-
tributions corresponding to each FP dura-
\ tion, thus generating 28 more distributions
of 300 observations each, for an overall total
of 52 distributions per 51.
Effect of Information Feedback
The floating-band conditions provide two
types of information: first, a comparison be-
tween performance under feedback compared
with no feedback, and second, a comparison
FIG. 2. Mean median RT over 6"s as a function of -S"s modal bands under floating-band con-
of immediate FP duration and feedback condition ditions with his optimum bands under fixed-
for the 20-msec. floating-band conditions. (The band conditions. The latter comparison al-
filled circle is for FP 1; unfilled circles connected
by solid lines, for FP 2; and unfilled circles con- lows us to determine whether 5 does in fact
nected by dotted lines, for FP 3. The numbers on produce true reactions when instructed to
the abscissa refer to FP durations, where 1 = 786 respond consistently. This comparison will
msec., 2 = 944 msec., 3 = 980 msec., and 4 = 1,070 be postponed to the section on fixed bands,
msec.) and only the effect of information feedback
will be considered here.
of this instruction was to attempt to eliminate out-
lying RTs caused by factors other than those Figure 2 presents average performance of
inherent in the reaction process. ,Ss under the 20-msec. band for no-feedback
Of the 24 experimental conditions, 8 were run (NFB) and feedback (FB) conditions under
under floating-band conditions, 2 of these under a each FP condition. The NFB plot shows
50-msec. band and FP 1, and 6 under a 20-msec. two of the results commonly reported in the
band, 2 per FP condition. Half of the floating-band
conditions were with feedback and half were with- literaturethat overall RT increases with
out. All >?s received the floating-band conditions FP variability, although the effect here only
first, followed by the 15 fixed-band conditions. The occurs between FP 3 and the other two FP
last condition run was the 5-msec. band under FP 1. conditions, and that RT is longer to the
RESULTS shorter FP in each of the variable FP condi-
tions (the immediate FP effect) indicated by
The data were analyzed by an IBM 7040 a negative slope on the plot. The effect of
computer, and some of the summary statis- feedback was to decrease both the magnitude
tics and all of the empirical and theoretical of the immediate FP effect and the influ-
ence of FP variability on overall RT, and
TABLE 1 to increase median RT under FPs 1 and 2.
COMPARISON OF FEEDBACK (FB) AND No-FEEDBACK Feedback also had the effect of increasing
(NFB) PERFORMANCE UNDER FLOATING BANDS
response consistency, both when measured by
Median RT
(msec.) IQR (msec.) Pi, (Percent-
age)
interquartile range (IQR) and by percent-
Band age of responses within the band (Pi,).
NFB FB NFB FB NFB FB Table 1 presents medians, IQRs, and Pb,
20 1 109 114 36 31 33 38 averaged across 5"s, for the three FP condi-
20 2 108 119 38 29 36 41 tions under the 20-msec. band and for FP 1
20 3 135 133 23 20 49 51
50 1 119 123 46 33 57 66 (constant FPs) under the 50-msec. band.
The data for individual Ss show the de-
FOREPERIOD EFFECTS IN SIMPLE RT

crease in IQR under FB in 16 of 20 distribu- 225


FOREPtOlOO 2 FOREPERIOO 3

tions, and the increase in Pb under FB in 14


li 1 ^
of 20 cases, whereas the increase in RT under
.{""
ISO

FB for FPs 1 and 2 occurred in 13 of 15 """


79 l{ '*
individual cases. The Ss responded most
consistently under the highest FP variabil- 229
ity condition (FP 3) regardless of feedback fifl
1}
condition. ISO
If '*
i;
Feedback, then, reduces RT variability, 79 ' " i"" ,.."""
and it appears to do so by eliminating very
fast responses, inasmuch as the median RTs Ki

are lengthened under FB even though the CP

location of the mode of the distribution, or at J 190

least of the modal band, remains essentially ! 79 .f"""'1 ,f.'"""


invariant. Indeed, it would be surprising if I
the mode of the distribution were affected S 229
by feedback since 5"s are given feedback and
payoffs for responding in exactly the same E ,80
ii 'i
band in which they responded most fre- 79 i f '*
1 ,f ' l! " f i l "
quently under no feedback.
229
Joint Effect of Band Location cs
and FP Variability ISO
li l{ 19 ' li l!
{

The 15 fixed-band conditions provide evi-


78 f " !" * ''
dence on the joint effect of band position and
FP variability when band position is prede- 2 3 4 9
termined by E. Figure 3 shows the effect of BAND

these two variables on median RT and IQR FIG. 3. Median RT and interquartile range for
for each 6" separately. The 5"s are generally each 5 and fixed-band-position. (The vertical line
able to respond so that their median RTs to the left of each set of measures indicates the
range of the band. The bands are numbered in
fall within the band, with the exception of the order from fastest to slowest, where 1 = 70-89 msec.,
fastest bands under FP 3. However, the 2 = 90-109 msec., 3=110-129 msec., 4=130-149
IQR is generally at a minimum at some in- msec., and S = 150-169 msec.)
termediate band location, and increases for
either faster or slower bands. criterion differed from that by a minimum
Figures 4 and 5 show the effect of band IQR criterion are noted by footnotes.
position and FP condition on the two mea- Considering first the location of the op-
sures of variability, Pb and IQR. Particu- timum bands, all but one 5" (MW) show
larly for the Pb measure, response con- identical or almost identical optimum bands
sistency is maximum at some intermediate regardless of FP variability condition, and
band location, designated as the optimum the values of Pb across FP variability condi-
band, and the optimum band location tends to tions are reasonably similar.
be the same for all three FP conditions. Comparing the locations of the optimum
Table 2 presents the locations of optimum and modal bands, although in every case but
bands, using the criterion of highest value of one there is some overlap, if only by a few
Pb, and, for comparison, the location of the milliseconds (the exception is for BR under
modal bands as determined from the floating- FP 3), the modal band locations are never
band conditions. The value of Pb for the exactly the same as the optimum band posi-
modal bands is for the feedback condition. tions. The overall level of consistency is
Cases in which the optimum band by the Pb higher under the optimum than under the.
JOAN GAY SNODGRASS
PERCENT IN BAND INTERQUARTILE RANGE The data for 5 CP show this effect to the
most marked degree, and an analysis of her
FPI
FP2 RT distributions under FP 3 for Bands 1
FP3
and 2 indicates why. Essentially, her per-
formance under these fast bands was almost
indistinguishable from her performance un-
der the optimum band. Whereas she only
IO-LIO- responded within Payoff Band 1 when it
was in effect 3% of the time, and within Pay-
AH % ms off Band 2 when it was in effect 29% of the
50 so
/\
/ A ' V""
time, she responded within her optimum
40 band (Band 3) 52% of the time under Band
I/; 30 30 s 1 payoff and 48% of the time under Band 2
\ =- payoff. Both of these rates compare favor-
',' 20 20 ^<^^'
10' 10
ably with her response rate of 54% within
1 1 1 1 1 ct. ft 1 1 t t 1 Band 3 under Band 3 payoff. Other 5s show
similar but less dramatic tendencies.
Figures 6 and 7 show the combined em-
50 50 pirical RT distributions under each band
40- 40 for FP Cond. 1 and 3, respectively, along
30 30 with the best fitting DMs (continuous solid
20 20
lines). These DM fits will be discussed in a
later section.
10 '10

3
-0-1.0-
I -i- Immediate FP Effect
BAND POSITION
The effect of band location on the immedi-
FIG. 4. Percent responses within the payoff band
(left) and interquartile range (right) as a function ate FP effect is shown in Fig. 8 for each S
of band location for each FP condition, for S"s ES, separately and for the average across 5s. As
AH, and CP. in Fig. 2, a negative slope indicates the

PERCENT IN BAND INTERQUARTILE RANGE


modal bands (average Pb for the optimum
bands is 48% and for the modal bands is
43%), but whether this is due to an inoppor-
tune positioning of the modal band by 5s'
no-feedback performance or to the fact that
the optimum (fixed) band conditions were
run later in the series than the modal (float-
ing) band conditions is difficult to say.
However, it will be argued later that be-
cause of some of the characteristics of modal
band performance, 5s did not, in general,
adopt the optimum strategy (according to
the anticipation model) of making only true
reactions when instructed to respond consist-
ently.
Considering performance under the fixed-
band conditions, in general, the two measures
of response variability, Fb and IQR, show
an inverse correlation, with the notable ex- 5 I
BAND POSITION
ception of fast bands under FP 3. For
several 5s, both Pb and IQR tend to be low. FIG. S. Same as Fig. 4 for 5s MW and BR.
FOREPERIOD EFFECTS IN SIMPLE RT

TABLE 2
COMPARISON OF OPTIMUM AND MODAL BAND PERFORMANCE

Optimum bands Modal bands


5
FP 1 FP2 FP3 FP 1 FP2 FP3
CP
Band 110-129 110-129 110-129 100-119 102-121 122-141
Pb 54 46 54 43 44 48
BR
Band 110-129 110-129 110-129 120-139 100-119 143-162
Pb 42 44 40 36 22 47
AH
Band 90-109 110-129 110-129 102-121 95-114 113-132
Pb 48 41 52 39 43 54
ES
Band 90-1 09 110-129 90-109 91-110 97-116 104-123
Pb 66 64 65 60 66 61
MW
Band 110-129 150-169 130-149' 91-110 141-160 135-154
-Pb 32 33 43 14 28 46

Note.Bands are expressed in milliseconds; Pb in percentages.


The minimum IQR occurred for Band 110-129.

usual immediate FP effect. The immediate gressively diminished and at slow bands is
FP effect is large under fast bands, but as sometimes reversed in direction. Further-
band position is increased the effect is pro- more, differences among the median RTs

BAND I

BAND 2

SAND 3

BAND 4

BANDS

260 6 ioo 260 6 r& 2 0 0 0 1 0 0


.ILL^l
REACTION TIME IN MILLISECONDS

FIG. 6. RT distributions (triangles) and the fitted DMs (continuous solid line) obtained under fixed-
band conditions with FP 1. (The payoff band range is shown by the two vertical lines. Data are plotted
in lO-msec.-wide intervals except for the interval containing U. Each distribution consists of 600 RTs.
Fits of the DM rejected at the .05 level are indicated by asterisks.)
10 JOAN GAY SNODGRASS

.022
8AND I
1 -iO.
.022

f(RT)
I BAND 2

.022
BAND 3

.022
BAND 4

1
.022
BANDS

100 200 0 100 200 0 100 200 0 KX> 2OO 0 100 200
REACTION TIME IN MILLISECONDS

FIG. 7. Same as Fig. 6 for FP 3. (Each distribution consists of 600 RTs,


combined across both FP durations.)

obtained under the three FP conditions are RT to the short FP, and the immediate FP
large under the fastest bands, but tend to dis- effect disappears.
appear as band position is increased. The incidence of false reactions steadily
An inspection of the distributions of RTs decreases with increases in fixed-band posi-
to the separate FPs gives some indication tion and is higher under FP 3 than the other
why the immediate FP effect disappears with FP conditions at every band position, partic-
increased band position. These are shown ularly under the fast bands. Although all SB
in Fig. 9 for 5 AH under FP 3 along with showed these qualitative effects, 5s vary ap-
the combined FP distributions. Here it is preciably in absolute rate of false reactions,
apparent that RTs to the short FP are rela- from a low of .22% for ES to a high of
tively unaffected by band position, except 2.44% for AH.
perhaps for Band 1, whereas RTs to the
long FP are markedly affected. At the fast- Effect of Payoff Band Width
est bands, there are a large number of ex- Three different bands widths5, 20, and
tremely fast RTs to the long FP, which tend 50 msec.were each run under the fixed
to disappear as band position is increased. FP condition, FP 1. There were two float-
The existence of such fast RTs lowers ing bands, 20 and 50 msec, wide, and two
median RT to the long FP, thereby produc- fixed bands, the optimum 20-msec. band un-
ing the immediate FP effect. Such fast RTs der FP 1, and the 5-msec. band within the
would be classified under the anticipation optimum band containing the most responses.
model as spuriously fast RTs (i.e., time Because of the method of locating the float-
estimates of the FP duration which occur ing bands, the 20-msec. band is not required
after the reaction signal). As such spuri- to be located within the 50-msec. band, al-
ously fast RTs disappear with increased band though four of the five 5s responded such
position, the RT to the long FP approaches that this was the case. For those four, the
FOREPERIOD EFFECTS IN SIMPLE RT 11

ES MW AH IS CP AVERAGE

150

\4
S.4
o \4

2! 150-

N
3
0,.. \3
~

no

100

90
\,
eo
4 I 25 4 23 4 25 4
FOREPERIOD

FIG. 8. Median RT as a function of immediate FP duration for each S separately and for the mean
across 5"s. (The filled circle is for FP 1; unfilled circles connected by solid lines, for FP 2; and unfilled
circles connected by dotted lines, for FP 3. The number to the right of each point or set of points is the
payoff band.)

mean IQR is virtually identical (35 msec.) 52 per S. Its distribution function is de-
under both band widths. scribed in Snodgrass et al. (1967). Its three
Because the 5-msec. band was chosen parameters, fo (the theoretical mode), and 8
within the optimum 20-msec. fixed band, and e (the two exponents), were estimated
comparisons between these two band widths by maximum likelihood techniques, as de-
are available for every 5". Every 5" except scribed in Appendix B of Snodgrass et al.
CP responded more consistently within the (1967).
20- than the S-msec. band both by the IQR The DM was fitted to the empirical dis-
and by the P\, measure. The mean IQR and tribution by determining end points contain-
Pb for the optimum 20-msec. band was 22 ing approximately 5% of the DM distribu-
msec, and 49%, and for the 5-msec. band, tion and determining the percentage falling
25 msec, and 44%, where Pb is normalized within the corresponding intervals for the
for the 20-msec. band. It would appear empirical distribution. Such a procedure
from this comparison that instead of making generates 21 intervals because different esti-
the empirical distribution more peaked, the mation techniques are used in the interval
5-msec. band was somewhat disruptive of before and after t0. Because these part inter-
performance. vals are sometimes very narrow, there are
occasional large fluctuations in the empirical
Form of the Theoretical RT Distribution data points around to. Chi-squares calcu-
Double monomial.The DM was fitted to lated between the two distributions indicate
each of the 260 empirical RT distributions, that the fits of the DM to 152 of the 260 em-
12 JOAN GAY SNODGRASS

OVERALL DIST. RT TO SHORT FP RT TO LONG FP bined fits are not rejected at the .05 level).
.044 * Furthermore, the mean chi-square value for
* *
the combined distributions is 1.4 times that
.022 A
A
4 for the separate distributions. The 76 sep-

\
f
BAND I arate distributions whose DM fits could not
-B^U -^
P<<'
W^ ' L . L be rejected are approximately equally divided
.044
* between long and short FP distributions.
Comparison of the DM -with the normal
.022
-
< - ' and exponential.Because it is clear that
BAND 2
t, the DM fits are not satisfactory by a chi-
./
, '
y ~A \
square criterion, it seems desirable to deter-
.044 ^* * mine whether other theoretical distributions
<(RT) might do better, particularly with respect
.022 k [ to the two characteristicspeaked modes and
'J
BAND 5
high tailsthat both the DM and the empiri-
" J
' ^ cal distributions seem to share. To see
.044 ^
whether other distributions might do as well,
the fit of the DM was compared to the fits
.022 of the normal and exponential distributions
; ' ;>
BAND 4
with regard to just these two characteristics.
i

T
The normal was fit to the entire empirical
.044 distribution by equating the empirical and
theoretical medians and interquartile ranges,
.022
,1.
and the exponential was fit to the RT dis-
SAND 9 tributions from the median on by equating
100 200
V. 0
iiiiiiinif
100
Sta.
200 0 100
/ V
200
the empirical 75th percentile and the theo-
RT IN MILLISECONDS
retical median.
Three measures of goodness of fit were
FIG. 9. Distributions of RTs to both FPs com- compared, two for the tail of the distribution
bined (Overall Distribution) and to each individual
FP for S AH under FP 3. (The triangles are the and one for the mode. The two tail measures
empirical data points and the continuous solid line is were (a) the value of t such that F (t) =
the fitted DM. The payoff band range is shown .9995, denoted ta, and (6) the probability
by the two vertical lines. Fits of the DM rejected beyond 250 msec., 1 -F (250), denoted p
at the .05 level are indicated by asterisks.)
(250). The measure of the peakedness of
the mode was (c) the probability within the
pirical distributions are rejected at the .05 payoff band, Pb. In order to calculate PI, for
level (df = 17), whereas the expected num- the exponential, the probability predicted
ber is 13 under the hypothesis that simple from the median to the upper limit of the
RTs have a DM distribution. payoff band was calculated and normalized
The number of rejected DM fits to the for the empirical probability within the en-
combined RT distributions under fixed-band tire payoff band. Distributions for which the
conditions increases with FP variability and median exceeded the upper payoff band
is 16, 19, and 23 out of 25 for FP 1, 2, and 3, limit were excluded. Because this analysis
respectively. As might be expected, the was carried out after the DM had been fitted
DM does a better job of fitting distributions to each distribution, two of the measures, ta
of RTs to separate FPs than it does for the and Pb, were not readily available for every
combined FP distributions. Although there distribution for the DM. The ta measure
are only twice as many separate as combined was only compared for those distributions
RT distributions, the DM fits over five times having short empirical or DM tails, compris-
as many separate as combined distributions ing 183 of the 260 distributions. The Pb
(76 of 140 separate and only 15 of 70 com- comparison was made only for the 15 com-
FOREPERIOD EFFECTS IN SIMPLE RT 13

bined RT distributions under fixed-band con- though the deviation from linearity would not
ditions, for a total of 75 distributions. be very marked if the plot were stopped at
The results of this analysis are as fol- that point. This fact is important because
lows : (a) The number of empirical observa- the usual plotting procedure is to end the
tions with "outliers," or the number having plot at l-F ( 0 = . 1 0 (e.g., Carterette,
a longest observation exceeding ts, exceeds Friedman, & Cosmides, 1965; Greenbaum,
by 94 the number expected by chance if the 1963; McGill, 1963). Furthermore, even
normal were the correct theoretical distribu- the theoretical DM, which has a power func-
tion, is fewer by 7 than would be expected tion rather than an exponential tail, does not
if the exponential were the correct form for exhibit very marked deviations from linearity
the right half of the distribution, and the on the plot. This suggests that this pro-
DM is only off by 1 in predicting the number cedure is not very sensitive to deviations
of distributions with outliers, (b) For the from exponential tails, and that it may often
second tail comparison, p (250), the normal lead to erroneous conclusions about the shape
underestimates the probability beyond 250 of the distribution.
msec, by .0023 per distribution, the expo-
nential overestimates it by .0035, and the DISCUSSION
DM overestimates it by only .0003. (c) For Fit of the Anticipation Model
the peakedness comparison, the normal over- Fixed-band performance.A comparison of
estimates the probability within the payoff the predictions of the model for fixed-band con-
band by .015, the exponential slightly over- ditions (Fig-. 1) with the results presented in
estimates it by .002, and the DM overesti- Fig. 4 and 5 indicates that the data for three S's,
mates it by only .001. BR, ES, and CP, are reasonably well fit by
Because in a previous study (Snodgrass the model, whereas for MW and AH the fit
et al., 1967) the DM was found to fit simple is rather poor. Perhaps the S whose results
RT distributions under band payoff condi- most closely conform to those predicted by
the model is BR, although every 51 exhibits
tions better than the gamma distribution by some of the predicted effects.
a likelihood comparison, and because the The most crucial prediction of the model is
present empirical RT distributions appeared that the location of the optimum band and the
to be very similar to those collected in that degree of consistency under it will be invari-
study, the likelihood comparison of the ant across FP variability conditions for indi-
gamma and DM was not repeated here. In- vidual .S's. With regard to location, .9s CP
stead, the distributions were plotted by a pro- and BR had the same optimum band under all
cedure suggested by McGill (1963) in which three FP conditions, ES and AH had very
the logarithm of the probability that RT is nearly the same optimum bands, and MW had
greater than some time t is plotted against t a different optimum band for each FP condition.
on the abscissa. This procedure has the effect However, the results for MW are generally
of spreading out the tail of the distribution more erratic than those for the other 5s. With
so that it can be determined whether a dis- regard to consistency under the optimum band,
tribution with an exponential tail, such as ES and BR performed with about the same
the gamma, would fit the data. If the em- degree of consistency under all FPs, CP was
pirical distribution when plotted in this way almost the same, and AH and MW differed con-
exhibits a "candy-cane" shape in which the siderably in consistency from one FP condition
tail is a straight line, then the distribution to another.
The model also predicts that RT variability
has an exponential tail. Figure 10 shows
under faster-than-optimum bands will be dif-
some representative plots of empirical and
ferentially affected by FP variability, whereas
DM distributions under fixed-band condi- variability under slower-than-optimum bands
tions for all three FP conditions. As is clear will not be affected by FP condition. The pre-
from Fig. 10, the empirical distributions diction for fast bands is confirmed for all 6"s
show quite marked deviations from linearity, except MW but only between FP 3 and the
particularly beyond 1 F () = .10, al- other two FP conditions, and every 5" without
14 JOAN GAY SNODGRASS

FORgPERIOD I

-2

100 200 0 100 ZOO 0 100 200


TIME t IN MILLISECONDS

FORE PERIOD 2

0
-I
-2

(9 -I
O

-3
100 200 0 100 200 0 100 200
TIME t IN MILLISECONDS
FOREPERIOD 3

100 200 0 100 200 0 100 200


TIME t IN MILLISECONDS
FIG. 10. Log of probability that RT is greater than t against t on abscissa, for representative 5s under
each FP condition for the fixed-band conditions. (Band location is from fastest to slowest from left to
right, and band number is given to the left of each plot. The triangles are the empirical data points plotted
in 10-msec. intervals, and the continuous line is the fitted DM.)

exception performed the worst under FP 3 Under slower-than-optimum bands, two 5s,
under the fastest band. The lack of a difference ES and BR, exhibit the superposition of the
between FP 1 and FP 2 is perhaps not too three FP functions predicted by the model, and
surprising in light of the small objective and any deviation from this pattern is always in the
subjective variability in FP 2 (the short FP direction of more consistent responding under
differed from the long FP by only 36 msec.). FP 3.
In fact, only one 5" became aware it was a Another line of evidence supporting the model
variable condition. is the fact that the incidence of false reactions
FOREPERIOD EFFECTS IN SIMPLE RT 15

steadily decreases with band position. Some- SHOR

what paradoxically, however, in light of the


interpretation of FP 3 responding under fast
bands, false reaction rate is much higher for FP
3 under fast bands than for the other FP con-
ditions. This finding is interpreted as indi-
cating not that more EW responses are made I I I I

under FP 3 but rather that of those made, a


much higher proportion are recorded as false
reactions. In contrast, although the absolute
proportion of EWs is assumed to be higher
under FP 1 and FP 2, many more are recorded
as spuriously fast RTs rather than as false
reactions.
Immediate FP effect.The model says that
the immediate FP effect is obtained whenever
S makes some EW responses to the longer FP
duration, and that under the fixed-band pro-
cedure the immediate FP effect should be ob-
tained for shorter-than-optimum bands and dis-
M -
appear for optimum and longer bands. An
inspection of Fig. 8 indicates that this prediction 40 -
is substantiated by the data. Furthermore, Fig.
9 provides qualitative evidence that the im-
mediate FP effect is obtained at fast bands be-
cause of the occurrence of highly variable and
spuriously fast RTs. This has the effect of 0 70 80 90 100 110 120 130 HO 150 160

lowering median RT to the longer compared MEDIAN RT IN MSEC

to the shorter FP duration; however, it should FIG. 11. Relation between median RT and IQR
also have the effect of increasing the variabil- for RT distributions to individual FPs in the vari-
ity of RTs to the longer FP. The model thus able FP conditions. (Left-pointing arrows indicate
predicts that whenever the immediate FP effect the immediate FP effect.)
is obtained, RT to the longer FP should be
more variable than RT to the shorter FP. Taking into account both the relationship
Figure 11 presents the relation between between IQR and median RT, which indicates
median RT and IQR of the pairs of separate the type of response, and the direction of the
RT distributions to each of the two FP dura- arrows, which indicates the FP preceding the
tions. All 14 of the variable FP conditions are response, the strategies employed by 5s can be
included in the plot, but the data for S MW summarized. When they are fast, all of them
have been omitted as they show no consistent tend to react to the short FP and make antici-
trend. A left-pointing arrow indicates the patory time estimates (EWs) to the long FP.
immediate FP effect and a left-pointing arrow This tendency is indicated by left- and upward-
which also points up indicates the predicted re- pointing arrows in the left-hand side of the
lation. For every .S" except ES, most of the figure, and accounts for the immediate FP effect
left-pointing arrows point up. observed at fast-band positions. When they
Ignoring for a moment the direction of the are slow, three of the four appear to continue
arrows, there is a decided tendency for there to to react to the short FP but to make time esti-
be a U-shaped relation between IQR and mates from the reaction signal (ERs) to the
median RT. This effect has been noted previ- long FP. This tendency is indicated by right-
ously for the combined RT distributions, but is and upward-pointing arrows in the right side
also apparent for the separate RT distributions. of the figure, and results in an inverse im-
The model says that RT distributions having in- mediate FP effect at slow bands. In contrast,
termediate median RTs and low IQRs consist when 6" ES is slow, he appears to react to the
primarily of Rs, those of low medians and high long FP and to make ERs to the short FP.
IQRs, of EWs, and those of high medians and This strategy produces an immediate FP effect
high IQRs, of ERs. at slow bands but does so by a process exactly
16 JOAN GAY SNODGRASS

opposite to the one which produces it at fast even though neither of the component distribu-
bands. tions differed appreciably from, e.g., the normal.
In general then, 5"s under slower-than-opti- To investigate this possibility, the author at-
mum bands adopt the strategy of reacting to a tempted to identify those empirical distributions
signal preceded by one FP duration, and making composed primarily of true reactions so that the
delayed reactions to a signal preceded by the comparisons of fits of the three theoretical
other FP duration. Why they should be unable distributions could be repeated on this subset.
to make delayed reactions to either FP duration The anticipation model suggests that R dis-
is puzzling and at present unexplained. tributions for a given 5" have a high proportion
Floating-band performance.The floating-, or of responses within that S"s optimum band.
modal, band procedure, which was considered as Accordingly, the 10 empirical distributions per
an alternative to the fixed-band method for de- S having the highest proportion of responses
termining the location of the optimum band, within his optimum band were selected for
was not successful in producing a high propor- analysis. The optimum bands for ^s CP, BR,
tion of R responses, to judge from the data. and AH were taken to be 110-129 msec., for
For one thing, the location of the modal bands ES, 90-109 msec., and MW was omitted from
differed substantially from one FP condition the analysis. Of the 40 distributions so selected,
to another, particularly between the two lowest 23 were distributions to the separate FPs (14
FP variability conditions (FP 1 and FP 2) and of these to the short FP), 13 were combined
the high-variability condition (FP 3). Fur- distributions to both FPs, and 4 were to the
thermore, Ss responded more consistently under fixed FP condition.
optimum bands than under modal bands, and the In terms of goodness of fit, the DM did some-
mean immediate FP effect was less under opti- what worse with these R distributions than with
mum than under modal bands (.6 msec. vs. 3.1 the entire set of distributions. The DM fit
msec.). Taken together, these observations sug- only 14 (35%) of the 40 distributions whereas
gest that -S"s made more time estimates than true it fit 108 (42%) of the entire set of 260 dis-
reactions with modal bands than with optimum tributions. In terms of the specific characteris-
bands. Furthermore, it would appear that this tics of peaked modes and high tails, the com-
tendency is particularly strong in the absence parisons among the DM, normal, and expo-
of information feedback since, as was noted nential were repeated for these 40 distributions.
earlier, median RT increased, variability de- The ta comparisons were available for 37
creased, and the extent of the immediate FP distributions, the p (250) comparisons for all
effect decreased under feedback compared with 40 of them, and the peakedness measure for
no feedback. only 9.
The R comparisons present a somewhat dif-
Theoretical Distribution of RTs ferent picture from the overall comparisons.
It seems clear that simple RT distributions The exponential now does a better job than
obtained under band payoff conditions are both the DM in accounting for peakedness (although
more highly peaked and have higher tails than it underestimates it somewhat) and it does
can be accounted for by traditional distribu- about as well as the DM in the tail compari-
tions for RT, such as the normal, exponential,
sons. As in the previous comparisons, the nor-
and gamma distributions. The DM fits the em-
pirical RT distributions better because it de- mal underestimates both tail measures and over-
scribes both the extreme peakedness and the estimates peakedness. However, the exponen-
high tails that are characteristic of these dis- tial and DM also underestimate the probability
tributions. The observed outliers may be true beyond 250 msec., in contrast to their perform-
reaction times, in which case the DM probably ances for the entire set of distributions.
also approximately fits the true reaction time Although the results of this analysis are by no
distributions, or they may be the more variable means conclusive, especially in light of the few
time estimates. If the outliers are time esti- distributions available for the peakedness com-
mates, then it is possible that the observed dis- parison, they do suggest that one reason the DM
tributions are composed of a mixture of two does so well for the empirical distributions as
distributions: true reactions which have a small a group is that it accounts for the characteristics
variance and time estimates which have a large of a mixture of two distributionstrue reac-
variance. This would produce an overall dis- tions which are highly peaked and time esti-
tribution that is both peaked and high-tailed mations which are highly variable.
FOREPERIOD EFFECTS IN SIMPLE RT 17

Sequential Effects Comparisons were made between the paired


A separate question from that of the theo- NFB-FB conditions on two criteria: the mean
absolute amplitude over frequencies and the
retical distribution of RTs is the extent of
degree of preponderance of slow over fast fre-
sequential dependencies among the trial-ordered
quencies. Of the 20 comparisons, NFB had
RTs. Bertelson (1961) has reported significant
sequential dependencies among choice RTs; higher mean absolute amplitude in 15 cases and
a greater preponderance of low frequencies in
these were strongly dependent on the Markovian
properties of the stimulus presentation schedule. 16 cases.
Weiss, Laties, Siegel, and Goldstein (1966) re- Figure 12 shows a typical example of spectra
port both long-term and short-term cyclic effects for NFB and FB sessions. These are based
on data for CP under the 20-msec. floating
in interresponse times in monkeys exposed to
a DRL schedule of reinforcement. They em- band and FP 1. Comparisons between the spec-
ployed the method of spectral analysis to evalu- tra for NFB and FB indicate that both the
ate these effects. mean amplitude of the component frequencies
and the degree of predominance of low over
For the present experiment, the extent of
sequential effects was evaluated for all the data high frequencies is greater for NFB than FB.
Below each spectrum is the series of RTs
by calculating the distribution of differences
between RTs on Trial n and Trial n + 1 (the from which it was calculated plotted in blocks
of various trial lengths and overlaps.
one-trial difference distribution), and for the
In the present experiment, the intertrial inter-
four paired feedback-no-feedback conditions by
the technique of spectral analysis. val was fixed at 3 sec. Thus the lowest de-
The ratio of the variance of the one-trial dif- tectable frequency has a cycle length of 2 min.
(40 observation points) and the highest fre-
ference distribution to that of the RT distribu-
tion was calculated and its mean value across quency, a cycle length of 6 sec. (2 observa-
tions). There may be lower frequency compo-
^s and conditions found to be 1.75; the ex-
pected value on the hypothesis of independent nents than one cycle per 2 min. present in the
trials is 2.00. Of the 120 distributions analyzed, data, but these would not be detected by the
114 have ratios less than 2.00. The average lag present analysis.
one autocorrelation coefficient (i.e., the Pearson Some of the 5"s reported being aware of slow
cyclic changes during the feedback sessions.
r between RTs on Trial and Trial n + 1) is
+.132. However the mean lag one autocorrela- They would note on the basis of feedback in-
formation that their times drifted from too fast
tion for the four NFS conditions is +.231 com-
pared with a mean of +.109 for the four corre- for the payoff band, through the band, to too
slow for the band, although subjectively they felt
sponding FB conditions. Of the 20 paired
FB-NFB comparisons, 17 NFB conditions have themselves responding in exactly the same
higher autocorrelations. way. It may be that slow fluctuations were
Schoeffler (1965) has reported similar re- more pronounced during no feedback because
the 5"s during feedback worked against this
sults in a two-choice RT experiment in which
tendency to drift, or it may be that the infor-
6"s categorized a light flash as being "long" or mation feedback itself disrupted whatever un-
"short." He found that the lag one auto- derlying process accounts for the slow cyclic
correlations between latencies tended to be changes. On the basis of the present data,
positive. Although no means are reported, it there is no way of deciding which of these two
is clear from his figures that the autocorrela- alternatives is more likely.
tions are higher in the conditions without infor-
mation feedback than they are with feedback. CONCLUSIONS
In order to determine whether sequential ef- The present data are consistent with the hy-
fects lasting over more than one trial were pothesis that each 5* has a distribution of true
present, the data for the eight paired FB-NFB reaction times which is highly peaked and may
conditions were analyzed by the method of spec- have a high tail, but that in an RT experiment
tral analysis, following the procedure outlined he does not necessarily exhibit this distribution
inAbelson (1953). directly. The exact nature of the true RT dis-
Of the 40 distributions analyzed by the tribution is undoubtedly affected by various
method of spectral analysis, all but one showed characteristics of the stimulation conditions, in-
significant deviations (p < .05) from a random cluding the reaction stimulus, but it appears to
series, and in all but two cases, slow frequencies be unaffected by either payoffs or FP variabil-
predominated, i.e., the spectrum was decreasing. ity (at least for fixed mean FP).
18 JOAN GAY SNODGRASS

NO FEEDBACK FEEDBACK

to

b-zoo-

d-0
LOCK* OF TRIAL* LOCKS Of TRIAL*

FIG. 12. Spectral analysis (a) and mean RTs over blocks of 600 trials (b-d) for various block sizes
and overlaps for the NFB and FB sessions of CP under the 20-msec. floating band and FP 1. ([a] The
p is frequency, where 0 indicates one cycle per 40 observation points and 20 indicates one cycle per 2
observation points, and Ut is the corresponding power of frequency p. The upper and lower lines indicate
95% confidence limits for the U,. [b] Blocks of 20 trials with a 3-trial overlap, [c] Blocks of 6 trials
with a 3-trial overlap, [d] Blocks of single trials with no overlap. The horizontal lines indicate the payoff
band in force. Mean RTs > 2SO msec, are plotted as 250 msec.)

The present experiment suggests that the true mixtures of true reactions and time estimates.
reaction time distribution may be observed Empirical distributions thought to contain pri-
directly when one uses a fairly high degree of marily true reactions appear to be too highly
FP variability, a fairly narrow optimum band peaked for the DM, too high-tailed for the
payoff (where optimum means that location normal, and well characterized by an expo-
which maximizes the proportion of responses nential fitted to the right side of the distribution.
within the band), and information feedback.
The band width which minimizes response vari- REFERENCES
ability appears to lie somewhere between 5 and ABELSON, R. P. Spectral analysis and the study of
50 msec. Feedback after every response is individual differences in the performance of
desirable because it decreases RT variability routine, repetitive tasks. Princeton, N. J.: Edu-
which it appears to do by decreasing the fre- cational Testing Service, Research Bulletin RB-
quency of voluntary estimations of the FP dura- 53-2, 1953.
tion and by decreasing the amplitude of natur- BERTELSON, P. Sequential redundancy and speed in
ally occurring slow sequential changes. a serial two-choice responding task. Quarterly
There is strong evidence that the immediate Journal of Experimental Psychology, 1961, 13,
FP effect is due to anticipatory time estimates 90-102.
of the longer FP, and that these may be almost CARTERETTE, E. C., FRIEDMAN, M. P., & COSMIDES,
R. Reaction-time distributions in the detection
completely eliminated by using an optimum of weak signals in noise. Journal of the Acous-
payoff band. tical Society of America, 1965, 38, 531-542.
Although the DM fits the entire set of empiri- DRAZIN, D. H. Effects of foreperiod, foreperiod
cal distributions better than the exponential or variability, and probability of stimulus occur-
normal, this superiority of fit appears to be rence on simple reaction time. Journal of Experi-
due to the fact that these distributions are mental Psychology, 1961, 62, 43-50.
FOREPERIOD EFFECTS IN SIMPLE RT 19

GREENBAUM, H. B. Simple reaction time: A case and relative duration of intersignal interval. Per-
study in signal detection. (Doctoral dissertation, ceptual and Motor Skills, 1965, 21, 3-10.
Columbia University) Ann Arbor, Mich.: Uni- SCHOEFFLER, M. S. Stimulus, response, and rein-
versity Microfilms, 1963. No. 63-7421. forcement factors as determinants of sequential
KARLIN, L. Reaction time as a function of fore- effects in psychophysical experiments. Sym-
period duration and variability. Journal of Ex- posium presented at the meeting of the American
perimental Psychology, 1959, 58, 185-191. Psychological Association, Chicago, September
KLEMMER, E. T. Time uncertainty in simple reac- 1965.
tion time. Journal of Experimental Psychology, SNODGRASS, J. G., LUCE, R. D., & GALANTER, E.
1956, SI, 179-184. Some experiments on simple and choice reaction
KLEMMER, E. T. Simple reaction time as a func-
time. Journal of Experimental Psychology, 1967,
tion of time uncertainty. Journal of Experimental 75, 1-17.
Psychology, 1957, 54, 195-200.
LABERGE, D. A recruitment theory of simple be- WEISS, B., LATIES, V. G., SIEGEL, L., & GOLDSTEIN,
havior. Psychometrika, 1962, 27, 375-396. D, A computer analysis of serial interactions in
LUCE, R. D., & GALANTER, E. Psychophysical scal- spaced responding. Journal of the Experimental
ing. In R. D. Luce, R. R. Bush, & E. Galanter Analysis of Behavior, 1966, 9, 619-626.
(Eds.), Handbook of mathematical psychology. WOODROW, H. The measurement of attention.
Vol. 1. New York: Wiley, 1963. Psychological Monographs, 1914, 17(5, Whole
MC&LL, W. J. Stochastic latency mechanisms. No. 76).
In R. D. Luce, R. R. Bush, & E. Galanter (Eds.), WOODROW, H. The reproduction of temporal inter-
Handbook of mathematical psychology. Vol. 1. vals. Journal of Experimental Psychology, 1930,
New York: Wiley, 1963. 13, 473-499.
McGiLL, W. J., & GIBBON, J. The general-gamma WOODROW, H. Individual differences in the repro-
distribution and reaction times. Journal of duction of temporal intervals. American Journal
Mathematical Psychology, 1965, 2, 1-18. of Psychology, 1933, 45, 271-281.
NICKERSON, R. S. Response time to the second of
two successive signals as a function of absolute (Received June 17, 1968)

Вам также может понравиться