Content uploaded by Ruud Meulenbroek
Author content
All content in this area was uploaded by Ruud Meulenbroek
Content may be subject to copyright.
Abstract Many recent studies indicate that memory for
final position is superior to memory for movement. There
is ambiguity about what is meant by the term final posi-
tion, however. Is it final spatial location or final posture?
According to a recently proposed theory by Rosenbaum
et al., which maintains that stored postures form the ba-
sis for movement planning, when people try to return to
recently reached positions, they should try to adopt the
postures they just occupied. An alternative view, which
holds that movements are primarily planned with respect
to spatial locations, predicts that subjects should tend to
return to places in external space. We describe an experi-
ment that tested these opposing predictions. The experi-
ment relied on the notion that if people store and use pos-
tures, they should “copy” the posture adopted with one
arm to the other arm when possible. The results support
this hypothesis.
Key words Reaching movements ·Memory for positions ·
Laterality · Posture copying · Human
Introduction
When one reaches to a final position, what does one later
remember –the movement, the final location, the final pos-
ture, or some combination of these elements? Although
information in the literature bears indirectly on this ques-
tion, we have been unable to find a direct answer to it in
previously published research.
In this article, we review previous work that bears on
the question of what is learned when people move repeat-
edly to a given position. Then we present two theoretical
perspectives which make diverging predictions about what
should be learned in repositioning tasks. One perspective
predicts that final positions are remembered as postures;
the other predicts that final positions are remembered as
locations. We describe an experiment designed to distin-
guish between these two predictions. The experiment in-
dicates that final postures are remembered and are “cop-
ied” from one arm to the other when subjects try to reach
repeatedly to the same location in the midsagittal plane
with alternating arms or when subjects try to reach re-
peatedly to the same location anywhere in the workspace
with the same arm. In the last section of the article, we
discuss the implications of our findings.
Previous research on memory for position
A number of studies have shown that memory for final
position is better than memory for movement. In one of
the first such studies (Marteniuk and Roy 1972) subjects
moved the hand to a stop and the experimenter then re-
turned the subject’s hand either to the original or to a dif-
ferent starting position. Next, subjects were asked to re-
produce the distance they just covered. They did very
badly when they had to reproduce the distance from a
new starting position but they did well when they had to
reproduce the distance from the original starting posi-
tion. In another condition subjects again moved the hand
to a stop and had the hand returned either to the same
start position or to a different starting position; this time
they were asked to bring the hand back to the final posi-
tion they just reached. Surprisingly, they did well in this
task, regardless of whether the starting position was the
same as in the beginning. In fact, subjects managed to
reach the final position even if the hand was moved
passively back and forth during the initial movement.
Results like these have been obtained in many studies
(Kelso and Holt 1980; Jaric etal. 1992,1994; Laabs 1973).
See Smyth (1984) for an excellent review.
David A. Rosenbaum (✉)
642 Moore Building, Department of Psychology,
Pennsylvania State University, University Park, PA 16802, USA
e-mail: dar12@cac.psu.edu
Ruud J. Meulenbroek
Nijmegen Institute for Cognition and Information,
University of Nijmegen, PO Box 9104, 6500 HE,
Nijmegen, The Netherlands
Jonathan Vaughan
Department of Psychology, Hamilton College,
Clinton, NY 13323, USA
Exp Brain Res (1999) 124:503–512 © Springer-Verlag 1999
RESEARCH ARTICLE
David A. Rosenbaum · Ruud J. Meulenbroek
Jonathan Vaughan
Remembered positions: stored locations or stored postures?
Received: 24 September 1997 / Accepted: 11 August 1998
The fact that final-position reproduction is better than
distance reproduction has been taken to support the equi-
librium-point hypothesis of motor control (Feldman 1966;
Kelso and Holt 1980; Latash 1993), which emphasizes
the establishment of target states for the neuromuscular
system. One way of redescribing the superiority of final-
position reproduction over distance reproduction in terms
compatible with the equilibrium-point hypothesis is that
actors remember final equilibrium states but do not re-
member means of reaching those equilibrium states. This
redescription is supported by the observation that there is
usually greater movement variability than terminal-posi-
tion variability in targeted movements (Bootsma and Van
Wieringen 1990; Desmurget et al. 1995; Wiesendanger et
al. 1996).
Defining “final position”
What is meant by the term “final position”? The term
could mean the final location in external space to which
a critical point along the limb segment chain (e.g., the tip
of the index finger) is brought, or a final body state – for
example, a vector of joint angles or, considering the dy-
namics as well as the kinematics of movement, a vector
of muscle stiffnesses. As far as we know, no one has at-
tempted to distinguish between the final-location and fi-
nal-body-state interpretations.
Final location
Which class of interpretation is correct? According to a
classical view, originating with Tolman (1948), one would
expect location to be the critical control parameter for po-
sition reproduction. Tolman showed that organisms learn
the spatial layouts in which they move rather than the
movements they make. Thus, a rat will get to the feeding
area of a maze more quickly than in its first exposure to
the maze even if its limbs are suddenly weighed down, if
the maze is suddenly flooded, or if its starting place in the
maze is altered. The rat has evidently learned the spatial
layout of the maze, not the movements needed to tra-
verse it.
The same general conclusion has been reached by oth-
er investigators working on manual control. In the case
of manual reaction time tasks, the classical view of stim-
ulus-response (S-R) compatibility effects is that S-R com-
patibility depends more on the location in external space
where responses are made than on which hand makes the
response (Brebner et al. 1972; Wallace 1971). Thus, press-
ing a key on the left in response to a stimulus on the left
is easier than pressing a key on the left in response to a
stimulus on the right, regardless of whether the key is
pressed with the left hand or right.
A similar conclusion was reached by Wickens (1938),
whose subjects participated in a shock-avoidance study.
Extension of the finger enabled the subject to bring the
finger away from the shock after a warning tone sound-
ed. After training, the subject’s hand was inverted, so flex-
ion rather than extension of the finger allowed the finger
to escape the shock. Subjects immediately flexed the fin-
ger rather than extending it after the inversion, suggest-
ing that they had not just learned movements.
A number of investigators have likewise proposed that
movements are primarily planned in external spatial co-
ordinates. A well-known hypothesis favoring spatial con-
trol of movement is the minimum-jerk hypothesis of Flash
and Hogan (1985), according to which hand movements
are typically designed to follow straight lines in extrinsic
space. Flash and Hogan proposed more specifically that
the kinematics of the end-effector (e.g., the tip of a hand-
held stylus) generally respects a spatiotemporal pattern
that permits minimization of the mean squared jerk (the
third time derivative of position) of the end-effector in
external space. Georgopoulos and his colleagues (e.g.,
Georgopoulos et al. 1981) have likewise suggested that
the primate motor cortex primarily controls the spatial
properties of arm trajectories. Insofar as both of these
views are correct – something that is questionable, as seen
below – they fit with the view that learned final positions
are learned final locations.
Final body states
Because controlling movements entails generation of ap-
propriate muscle torques, lower-level details of move-
ments must ultimately be taken into account for move-
ments to achieve spatial goals. A growing body of evi-
dence supports the view that movement planning is also
based on these lower-level variables. With respect to Flash
and Hogan’s (1985) minimum-jerk hypothesis, it has been
shown that departures from straight-line motions in ex-
ternal space occur when loads are applied to the hand
(Uno et al. 1989). Moreover, changes in movement cur-
vature associated with different spatial directions can be
ascribed to minimization of mean muscle tension change
(Dornay et al. 1996). Although it has been suggested that
directionally dependent departures from linearity may be
due to visual misperception (Wolpert et al. 1994, 1995),
this hypothesis has been rejected (Osu et al. 1997).
With respect to the hypothesis by Georgopoulos et al.
(1981) that the primate motor cortex codes spatial prop-
erties of movement, recent studies have shown that dis-
charge properties of motor cortex neurons depend on start-
ing and ending arm postures for the same spatial dis-
placements of the end-effector (Scott and Kalaska 1997).
Hence, it can no longer be accepted that only spatial fea-
tures of movement are represented in primate motor cor-
tex, although it is possible that spatial features alone are
represented elsewhere and that spatial features alone are
represented in areas of primate motor cortex that have
not yet been studied.
Findings such as these imply that body states are tak-
en into account in motor control. However, they do not
imply that spatial properties of movements are not taken
into account. Everyday experience suggests that spatial
504
properties must be provided for in some tasks (e.g., draw-
ing a straight line). What is needed, then, is interactive
control across different levels. Recent data concerning the
kinematics of joint-angle changes for targeted hand move-
ments (Haggard et al. 1995) indicate that spatial as well
as kinesthetic constraints do indeed affect movement gen-
eration. Moreover, two recent models of the motor con-
trol system have emphasized cooperative consideration
of constraints at many levels. Kawato (1996) developed
a neural network model for bidirectional communication
between higher and lower levels. D.A. Rosenbaum, R.G.
Meulenbroek, C. Jansen, J. Vaughan (1998, unpublished)
developed a cognitive psychological model with a hier-
archy of constraints that can be adjusted according to
task demands. In the theory of Rosenbaum et al. it is as-
sumed that target locations are generally planned at a
higher level than final postures, and that final postures
are generally planned at a higher level than movements;
see also Rosenbaum et al. (1993, 1995, unpublished) and
Vaughan et al. (1997). The theory leaves open the possi-
bility that final locations or final postures can be remem-
bered since both elements are assumed to play a role in
motor planning. However, because the theory assumes that
movements are planned by making use of previously
stored final postures, it would be favorable to the theory
to show that final postures are remembered.
Experiment overview and predictions
In the present experiment, the subject sat at a table raised
to chest height and began each experimental trial with
the arms outstretched, resting on the table, and bent slight-
ly at the elbows, as shown in Fig. 1. The task was to
move either the left or right hand to a target and then to
return the hand to its rest position. In the control condi-
tions, the subject was supposed to move one hand – ei-
ther the left or right – to the same target 20 times in a
row, returning to the rest position between each reach. In
the experimental conditions, the subject was also sup-
posed to reach the same target 20 times in a row but to
do so with alternating hands, again returning to the rest
position between each reach. In both the control and ex-
perimental conditions, the first ten trials were performed
with visual feedback and the final ten trials were per-
formed without visual feedback. The targets were either
in the body midline or to the left or right of the body mid-
line. The targets in the body midline could be reached with
symmetric arm postures; the targets off the body midline
could not.
The main question concerned the way positions were
remembered in those conditions where, a priori, the posi-
tions could be remembered either as postures or as loca-
tions – namely, midline targets reached for with alternat-
ing hands or with just one hand, and peripheral targets
reached for with just one hand.
Four sets of predictions were tested. The first set
concerned distances between successively adopted posi-
tions. If subjects remember postures when it makes sense
to do so, distances between successively adopted posi-
tions should be lower in the conditions where only one
hand is used than in the conditions where alternating
hands are used. By contrast, if subjects only remember
locations, distances between successively adopted posi-
tions should be no different in the one- and two-hand
conditions. Admittedly, the former, posture-based, pre-
diction is not very strong. Smaller within- than between-
505
E
B
F
C
D
A
Fig. 1 Setup for the experiment. Top panel Schematic view of the
subject as seen from above. The subject faced a video monitor,
shown opposite the subject’s side of the table. Not shown is the
video camera, mounted above the center of the table and pointed
straight down. Bottom panel Schematic of the image seen by the
subject on the video monitor. Neither diagram is drawn exactly to
scale. The left hand is shown here in the resting position
hand distances could be obtained for reasons other than
posture storage. For example, there might be a tendency
to play out the same motor commands repeatedly for the
same hand, or there could be similar proprioceptive bi-
ases for the same hand. Nonetheless, it was important
to test this prediction because its disconfirmation would
vitiate the posture-store hypothesis.
A second set of predictions concerned distances be-
tween positions adopted in lag-2 rather than lag-1 (con-
secutive) reaches. Because all lag-2 distances were pro-
duced by one hand, either separated by a response made
with that same hand (in the control condition), or by a re-
sponse made with the other hand (in the experimental con-
dition), different predictions could be made about lag-2
distances based on the location and posture hypotheses.
If only locations are remembered, lag-2 distances should
be the same in the control and experimental conditions –
that is, regardless of whether the intervening response is
made with the other hand or with the same hand. How-
ever, if postures are remembered, lag-2 distances should,
in general, be smaller in the experimental (alternating-
hand) condition than in the control (repeating-hand) con-
dition because errors should accumulate as more and more
responses are made with the same hand (Bock and Eck-
miller 1986). The only way that the lag-2 distances could
be the same in the control and experimental conditions,
according to the posture hypothesis, is if posture copying
from one arm to the other were perfect.
An elaboration of the prediction introduced in the last
paragraph concerns the difference between lag-2 distanc-
es in the control and experimental conditions for midline
versus non-midline targets. If postures adopted by one
arm are “copied” to the other arm when it makes sense to
do so (i.e., for midline targets reached for with alternat-
ing hands), then the difference between control- and ex-
perimental-condition lag-2 distances should be different
for midline and non-midline targets. The basis for this
prediction is that if the arm whose positions contribute to
the lag-2 distance measure is affected (midline targets)
or unaffected (non-midline targets) by the activity of the
other arm between its first and second reaches, the dif-
ference between lag-2 distances in the control and exper-
imental conditions should be different when posture copy-
ing either occurs (midline targets) or does not occur (non-
midline targets).
A third set of predictions concerned correlations be-
tween positions adopted successively by the two hands
when they moved in alternation to the same targets. Ac-
cording to the posture hypothesis, there should be strong-
er positive correlations in the in-out (y) direction for mid-
line targets than for peripheral targets, and stronger neg-
ative correlations in the horizontal (x) direction for mid-
line targets than for peripheral targets. By contrast, if on-
ly locations are stored, positive correlations should be
observed between positions adopted by the two alternat-
ing hands both in the horizontal (x) and in the in-out (y)
direction, and there should be no difference between posi-
tive correlations for midline versus peripheral targets. The
latter, location-based predictions are based on the idea
that if locations are remembered in a single coherent map,
the hand used to get to a location should not matter. The
former, posture-based, predictions are based on the idea
that when subjects can copy postures from one arm to
the other (i.e., for midline targets), one arm should ap-
proximately mirror what the other arm just did.
A fourth set of predictions provides a check on the
third. Correlations between successively adopted positions
might turn out as predicted by the posture hypothesis be-
cause each arm on its own might change where it ends
up over successive positioning trials. Thus, if each arm
exhibits the same cumulative bias to move in or out, the
two arms moving in alternation could yield a positive cor-
relation in the in-out direction, and this positive correla-
tion could, conceivably, be more pronounced for midline
targets than for peripheral targets. Similarly, if each arm
exhibits the same cumulative bias to move eccentrically
or medially, the two arms moving in alternation could
yield a negative correlation in the horizontal direction, and
this negative correlation could, conceivably, be more pro-
nounced for midline targets than for peripheral targets.
To check these possibilities, a further prediction of the
posture-store hypothesis concerns the effects of interleav-
ing target positions adopted by the two hands when only
the left hand or right hand was used. If subjects copy pos-
tures from one arm to the other when they alternate be-
tween the two arms to reach medial targets, the obtained
correlations should be weaker for interleaved hands than
for alternating hands in the case of the medial targets.
For peripheral targets, there should be no such weakening
of interleaved-hands correlations compared to alternat-
ing-hands correlations because posture copying would not
be expected for peripheral targets. Finally, if subjects re-
turn to previously reached locations but not to previously
reached postures, no difference would be expected be-
tween interleaved-hands and alternating-hands correla-
tions for any target.
Materials and methods
Subjects made successive hand movements in the horizontal plane
to each of six targets. Within each block of trials, subjects were sup-
posed to reach to just one target, which was shown at the start of
the block, either with one hand or with two hands in alternation.
Each trial block began with vision (trials 1–10) followed by no vi-
sion (trials 11–20). The movements always began from fixed start-
ing positions: The left hand extended out and to the left so that a
manipulandum held with the left hand rested against an L-shaped
frame mounted to the far left corner of the table top, as shown in
Fig. 1. Similarly, the right hand extended out and to the right so a
manipulandum held with the right hand rested against an L-shaped
frame mounted to the far right corner of the table top. The target
locations lay between the two arms, and were closer to the frontal
plane of the body. In the control condition movements were made
with just one hand (either the left or the right) in the entire series
of 20 reaches for a given target, but in the experimental conditions
the movements were made by alternating between the two hands
(left, right, left, right ... or right, left, right, left, ...) to the same tar-
get 20 times in a row.
The subject sat at a table, which was raised to chest height, and
moved the upper arm and forearm via shoulder and elbow rota-
tions, respectively; the wrist could not bend. The subject’s forearm
506
and hand rested on a light wooden board with felt on its underside
and two bands across the top for securing the forearm. With this
arrangement, the subject could slide the board with very little fric-
tion over the table. The board for each arm was 55 cm long, 9 cm
wide, and 1.5 cm thick and was equipped with a micro-switch rest-
ing beneath the tip of the index finger. When the subject pressed the
switch, it illuminated a penlight mounted in the center of a thin
wooden circular disk, 13 cm in diameter, positioned 2 cm over the
hand, centered above the micro-switch. A small battery resting on
the board sat just beyond the micro-switch and provided the power
source for the light.
Over the table and pointing down toward it was a video camera
which sent images of the workspace to a video monitor located be-
yond the table top, directly in front of the subject. On the video
monitor, the subject could see his or her own movements during
the opening trials of each experimental block. Movements away
from the frontal plane of the body appeared as upward displace-
ments on the video screen, movements toward the frontal plane of
the body appeared as downward displacements, and movements to
the right or left appeared as right or left displacements, respective-
ly. A clear plastic sheet was superimposed on the video screen to
indicate target positions. Six circles appeared on the sheet. Each
circle had a unique letter: A, B, and C for the top left, top middle,
and top right positions, respectively, and D, E, and F for the bot-
tom left, bottom middle, and bottom right positions, respectively.
The letters appeared above the circles. The circles were 10 cm in
diameter. The image of the hand-driven disk was 8 cm in diameter.
The target circles were positioned so the six possible positions of
the disks filled as much of the workspace as possible and so each
hand could reach comfortably to each position. Because each tar-
get circle’s position was defined by just two spatial coordinates and
only the shoulder and elbow could rotate, any given placement of
the hand-driven circle in a spatial target required a unique posture.
At the start of each block of trials the experimenter announced
which circle was the target for that block, whether one or two
hands should be used to bring the moved circle into the target
circle, and, in the event that two hands would be used, which hand
was supposed to go first. The subject’s task in each trial was to
bring the designated hand circle to the designated target circle, hold
the hand circle at rest until the subject judged it to coincide spa-
tially with the circle on the screen, and then to switch on the light
with the index finger of the positioned hand. Between trials, the
subject had to keep both hands at the far corners of the table. L-
shaped wooden frames were mounted on the table’s far corners so
the round disk above the hand could be pushed against the interior
of the frame with the subject’s arm outstretched. The subject could
bring the manipulandum to the rest position by feel alone.
The same target was supposed to be reached 20 times in a row
within each block of trials. The first ten reaches were completed
with visual feedback. These trials allowed the subject to learn the
target position for that block. After the tenth reach, once the hand
was returned to the holding position, the experimenter flipped an
opaque screen, attached to welder’s goggles worn by the subject,
down over the subject’s eyes. The subject then continued the place-
ment task for another ten trials. The experimenter indicated when
the subject could stop the placements. Thus, the subject was not
responsible for keeping track of the number of positioning trials.
While the task was being performed, the experimenter checked
that the subject activated the light when reaching the judged posi-
tion, that the correct hand was used in each trial, and that both
hands were at their resting positions between trials. The entire ses-
sion was videotaped, allowing for a later check that all aspects of
the procedure were followed in all trials.
The experimental design called for two same-hand conditions
and two alternating-hand conditions. One same-hand condition re-
quired the use of the left hand only (condition LL). The other
same-hand condition required the use of the right hand only (con-
dition RR). One alternating-hand condition began with the left
hand (condition LR). The other alternating-hand condition began
with the right hand (condition RL). The order of conditions was
balanced so all four movement conditions were tested equally of-
ten in all serial positions, and such that all six targets were reached
equally often in each movement condition and equally often in
each serial position.
Nine Penn State undergraduates, four male and five female,
served as subjects. All subjects were told to be as accurate as pos-
sible in all conditions and that a monetary bonus of U.S. $20
would be awarded to the subject whose accuracy was best. Each
subject’s session lasted about 50 min.
The data were analyzed using a video-digitizing technique de-
veloped by Barnes et al. (1989), which allows for recording the Car-
tesian coordinates of points of interest in video still frames. As
shown by Fischer et al. (1994), interrater reliability with this meth-
od exceeds 0.98 and the standard error of discrepancies between
mean positions digitized from video frames and mean positions for
the same points measured directly with a ruler is less than 1 mm.
In the present application, the operator saw the superimposed freeze-
frame video and computer image and used the mouse to click on
the position of the light bulb at the first video frame in which it
was illuminated. This was done for all no-vision trials. The coder
watched the initial ten trials of each block to make sure subjects
complied with the instruction to bring the moved circle into the
target circle when visual feedback was available.
Results
Inspection of the videotape revealed that the basic proce-
dure was followed by all subjects in all trials. During the
vision-present trials, the moved circle was always brought
within the target circle and the light was clicked on while
the circle occupied that region. Furthermore, the subjects
always brought both arms up against the L’s at the far
corners of the table between trials.
A summary of the results (Fig. 2) shows the first posi-
tion adopted after the removal of visual feedback (trial 11)
and the position adopted in the last no-vision trial (trial
20) both in the unimanual and bimanual conditions. Be-
cause there was no indication of a difference in perfor-
mance in conditions LR or RL, those conditions have been
combined.
As seen in Fig. 2, there were large shifts in adopted
positions from the first to the last no-vision trials. In both
of the one-hand conditions (LL and RR), adopted posi-
tions for targets farthest from the subject (top points in
the left and right panels of Fig. 2) were generally farther
from the subject in trial 20 than in trial 11, whereas for
targets closer to the subject (bottom points in the left and
right panels of Fig. 2) the adopted positions were gener-
ally closer to the subject in trial 20 than in trial 11 (ex-
cept for the rightmost target in the left-hand case). In the
alternating-hands conditions, the most striking feature of
the results was the overall symmetry of the shifts: The in-
ner (bottom) points were “attracted” toward the body cen-
ter, and the outer (top) points were “attracted” to the cen-
ters of the distal portion of each arm’s workspace.
Distances between successive positions
As stated earlier, the first prediction of the posture-based
model was that distances between positions adopted in
successive trials (herewith, lag-1 distances) would be larg-
er in the between-hand conditions than in the within-hand
507
conditions. As seen in Table 1, this prediction was con-
firmed. For every target, the lag-1 distance was greater
in the between-hand condition than in the within-hand
condition. These data were evaluated with a three-way
analysis of variance (ANOVA) that tested the effects of
the in-out (near-far) position of the target, the horizontal
(left, middle, or right) position of the target, and response
condition (LL, RR, or LR and RLcombined). The effect of
response condition was highly significant, F(2, 16)=65.80,
P<0.001, as was the interaction between the in-out and
horizontal target positions F(2, 16)=5.67, P<0.01. For
outer (farther) targets, lag-1 distances were larger for mid-
dle positions than for lateral positions, whereas for inner
(closer) targets, lag-1 distances were smaller for middle
positions than for lateral positions. No other interactions
or main effects were statistically significant; all associat-
ed P values exceeded 0.10.
Next consider the second set of predictions, which con-
cerned lag-2 distances. The relevant data are summarized
in Table 2 and were analyzed in the same way as the lag-
508
-45
0
In - Out (cm)
Left - Right (cm)
045
80
Fig. 2 Mean positions adopted
in condition LL (top left panel),
condition RR (top right panel),
and conditions LR and RL
(bottom panel). Solid points
represent positions in the first
no-vision trial (trial 11). Empty
points represent positions in the
last no-vision trial (trial 20).
Circles represent right-hand
responses. Squares represent
left-hand responses. Dimen-
sions and lengths indicated for
the bottom panel also apply to
the top panels
Table 1 Lag-1 distances (cm)aCondition Target A Target B Target C Target D Target E Target F
LL 2.32 2.39 2.36 2.49 2.14 2.77
RR 2.10 2.39 2.10 2.94 2.14 2.25
LR or RL 4.08 5.85 4.88 5.72 4.33 5.29
a Data from no-vision trials
only
Condition Target A Target B Target C Target D Target E Target F
LL 2.65 2.87 2.82 3.07 2.32 3.20
RR 2.65 2.85 2.46 3.48 2.39 2.90
LR or RL 2.10 2.27 2.69 3.48 2.12 2.79
a Data from no-vision trials
only
Table 2 Lag-2 distances (cm)a
1 distances. An ANOVA tested the effects of the in-out
(near-far) position of the target, the horizontal position
(left, middle, or right) of the target, and response condi-
tion (LL, RR, or LR and RL combined). The ANOVA re-
vealed a significant effect of response condition, F(2, 16)
=5.643, P=0.014. The mean lag 2 distance was 2.82 cm
in the left-hand-only condition (LL), 2.75 cm in the right-
hand-only condition (RR), and 2.45 cm in the two-hand
condition (LR or RL). A Newman-Keuls test showed
that the lag-2 distance in the two-hand condition was sig-
nificantly smaller (P<0.05) than the lag-2 distance in ei-
ther one-hand condition, and the two one-hand conditions
did not differ significantly (P>0.05). These results are
consistent with the expectation that error would build up
for a hand when that hand was used repeatedly, as pre-
dicted by the posture hypothesis. The fact that lag-2 dis-
tances were smaller in the experimental conditions than
in the control conditions is inconsistent with the location
hypothesis, which predicted no such effect.
With respect to the elaboration of the posture hypoth-
esis as applied to the lag-2 distances, the data provide
mixed support for the posture hypothesis. On the basis of
the posture hypothesis, it was predicted that the differ-
ence between lag-2 distances in the experimental and con-
trol conditions would be unequal for midline and periph-
eral targets. Inspection of Table 2 shows that this predic-
tion was only partially confirmed. The difference between
lag-2 distances in the experimental and control conditions
for target B, the far medial target, was 0.59 cm, the larg-
est for any of the six targets. However, the difference be-
tween lag-2 distances for target E, the near medial target,
was only 0.24 cm. The differences between lag-2 distanc-
es for targets A and D, the two left targets, were 0.55 cm
and –0.20 cm, respectively, and the differences between
the lag-2 distances for targets C and F, the two right tar-
gets, were –0.05 cm and 0.26 cm, respectively.
Correlations between positions
The next set of tests concerned correlations between suc-
cessively adopted positions. As seen in Table 3, correla-
tions between successive left-right positions were nega-
tively correlated in the between-hand conditions but were
positively correlated in the within-hand conditions. An
ANOVA showed that the effect of response condition on
horizontal position correlations was highly significant,
F(2, 16)=109.04, P<0.001, as was the effect of the hori-
zontal position of the target, F(2, 16)=21.47, P<0.001,
and the effect of the in-out position of the target, F(1,
8)=13.60, P<0.01. However, no other main effect or in-
teraction was statistically significant (all P’s>0.20).
Because the posture hypothesis predicted negative cor-
relations between successive left-right positions in the
alternating hands condition (LR/RL), it was important to
test whether the correlations were significantly negative.
To do so, for each of the nine subjects we counted the
number of targets for which a negative correlation was ob-
tained between successive left-right positions in the al-
ternating hands condition (LR/RL). Then we performed
a one-tailed t-test to see whether the obtained sample mean
was significantly greater than the number that would be
expected by chance to have negative correlations (i.e.,
three targets). The obtained sample mean, 5.11, was sig-
nificantly different from 3, t=4.64, P<0.001, df=8. This
result, like the others reported so far in this section, sup-
ports the posture hypothesis. As would be expected if par-
ticipants copied one arm’s posture to the other arm, left-
right errors were mirrored from one arm to the other.
Correlations between successive in-out (y) positions
are summarized in Table 4, where it is seen that all the
correlations, except for two, were positive. The only ex-
ceptions were at the far lateral positions in the between-
hand condition. The correlations between successively
clicked in-out positions were evaluated with an ANOVA
which showed that the y-position correlations were sig-
nificantly affected by response condition (LL vs RR vs
LR/RL), F(2, 16)=21.81, P<0.001. The mean values for
the three response conditions were 0.359 for condition
LL, 0.452 for condition RR, and 0.077 for condition LR/
RL. A Newman-Keuls test showed that condition LR/RL
was significantly different (P<0.01) from conditions LL
and RR, but conditions LL and RR were not significantly
different from each other. No other main effect or inter-
action was statistically significant (all P’s>0.09). Because
the posture hypothesis predicted positive correlations be-
tween successive in-out positions in the alternating hands
condition (LR/RL), it was important to test whether the
correlations were significantly positive. To do so, we tal-
lied the number of targets for each subject that yielded
509
Condition Target A Target B Target C Target D Target E Target F
LL 0.230 0.341 0.328 0.561 0.314 0.533
RR 0.479 0.140 0.469 0.329 0.236 0.529
LR or RL −0.342 −0.692 −0.370 −0.279 −0.568 −0.224
aData from no-vision trials only
Table 3 Correlations between
successive left-right positionsa
Condition Target A Target B Target C Target D Target E Target F
LL 0.409 0.467 0.258 0.351 0.264 0.403
RR 0.307 0.409 0.570 0.413 0.492 0.518
LR or RL −0.031 0.077 −0.163 0.097 0.156 0.330
a Data from no-vision trials
only
Table 4 Correlations between
successive in-out positionsa
positive correlations and compared the obtained sample
mean (3.11) to the mean expected by chance alone (3).
Not surprisingly, the null hypothesis could not be reject-
ed, t=0.286, P>0.39, df=8.
The final set of analyses concerned the fourth set of
predictions, which pertained to the effects of interleaving
positions adopted by each arm when only one arm was
used in a block of trials. Recall that the question was
whether the correlations between successively adopted
positions by the two alternating hands would also be ob-
tained if we interleaved the positions of the two hands
when they performed the targeting tasks by themselves
in separate blocks. The posture hypothesis, or more spe-
cifically the hypothesis that postures are copied from one
arm to the other for midline targets, predicted that the
correlations would be more pronounced for the alternat-
ing-hands correlations than for the interleaved-hands cor-
relations in the case of midline targets, but that the corre-
lations would be no more pronounced for the alternating-
hands correlations than for the interleaved-hands correla-
tions in the case of peripheral targets.
The data concerning interleaved and alternating hands
correlations appear in Table 5. Consider first the left-right
dimension. The main effect of data source (alternating-
hands versus interleaved-hands) was highly significant,
F(1, 8)=119.98, P<0.001, with the correlations from the
alternating-hands source being significantly more pro-
nounced (mean correlation=–0.412) than the correlations
from the interleaved-hands source (mean correlation
=0.076). The effect of target was also significant, F(5,
40)=2.91, P<0.03. Pairwise comparisons of alternating
hands versus interleaved hands for individual targets, us-
ing the Newman-Keuls procedure, showed that only the
two midline targets yielded significant differences be-
tween alternating-hands and interleaved-hands sources;
both midline targets had P values less than 0.01 for the
source (alternating versus interleaved) effect. None of the
peripheral targets had a significant pairwise difference
between the two sources; the P values for each of the
four peripheral targets exceeded 0.05.
The in-out dimension failed to yield any effect of data
source. Regardless of whether data came from alternating
or interleaved hands, the mean correlation in successive
in-out positions was slightly positive (0.077 for alternat-
ing hands vs 0.052 for interleaved hands), and no main
effect or interaction involving data source approached
statistical significance; all associated P values exceeded
0.35.
Discussion
The present experiment was designed to reveal how po-
sitions that can be remembered as locations or as pos-
tures are actually remembered. To address this question,
we had subjects move to initially seen and then merely
remembered positions, performing the task either with
one hand or with two hands in alternation. We generated
four sets of predictions to distinguish the location and
510
Table 5 Alternating-hands and interleaved-hands correlations between successive left-right and in-out positionsa
Dimension Target
ABCDEF
Alternating Interleaved Alternating Interleaved Alternating Interleaved Alternating Interleaved Alternating Interleaved Alternating Interleaved
Left−right −0.342 −0.010 −0.692 0.070 −0.370 0.068 −0.279 0.058 −0.568 0.065 −0.224 0.204
In−out −0.031 0.118 0.077 0.099 −0.163 0.004 0.097 0.021 0.156 0.065 0.330 0.009
a Data from no-vision trials only
posture hypotheses; all of the predictions concerned per-
formance in the final ten trials of each block, when the
positioning task was performed without benefit of vi-
sion. The results corresponding to the four sets of predic-
tions generally supported the posture hypothesis. First,
distances between successively adopted hand placements
(lag-1 distances) were smaller in the one-hand conditions
than in the two-hand conditions, consistent with the view
that final postures, rather than only final locations, influ-
enced position reproduction. This effect was different for
middle and for lateral positions, as would be expected if
the lag-1 distances depended on the utility of copying
postures between arms. The fact that the difference was
reversed for near and far targets is something that we
cannot yet explain. Second, distances between positions
adopted in non-consecutive reaches (lag-2 distances) were
smaller in the alternating-hands and in the one-hand con-
ditions, again as would be expected if aspects of subjects’
postures affected position reproduction. The difference be-
tween lag-2 distances for the alternating-hands and one-
hand conditions was largest for the far midline target,
which was one of the two targets that could be reached
with symmetric postures. Third, correlations between hor-
izontal positions adopted by the two alternating hands
were negative, most strongly so for the midline targets.
This outcome is consistent with the view that subjects
mirrored with one hand what they did with the other hand.
Correlations between in-out positions adopted by the two
alternating hands were often positive, as predicted by the
posture hypothesis, but were not reliably positive over
subjects; what this means will be discussed in the next
paragraph. Fourth and finally, using interleaved horizon-
tal positions from each of the hands when those hands
were used individually yielded less pronounced negative
correlations than using positions adopted by the two hands
when those hands were used in regular alternation. This
result indicates that the negative correlations for horizon-
tal position in the alternating-hands conditions reflected
mirroring of one hand’s position by the other hand rather
than an artifact of each hand’s bias. There was no differ-
ence between interleaved- and alternating-hands correla-
tions in the in-out dimension, however.
All in all, the results supported the hypothesis that po-
sitions that could be stored as postures were stored as
such. The only result which failed to support this hypoth-
esis was that correlations involving lag-1 and lag-2 posi-
tions in the in-out direction were not always reliable. Be-
cause the focus on left-right and in-out dimensions was
somewhat arbitrary, however, being motivated principal-
ly by reliance on Cartesian coordinates for purposes of da-
ta analysis, the absence of reliable in-out correlations need
not be taken as strong evidence against the posture hypoth-
esis. What is more important is that the presence of reli-
able correlations for the left-right dimension is sufficient,
along with the other results, to support the conclusion
that subjects remembered postures and copied postures
from one arm to the other when it made sense to do so.
The finding that there is posture-copying is reminis-
cent of the well-known fact that there are interactions be-
tween the two hands during bimanual performance (Swin-
nen et al. 1994). However, those interactions occur while
the two hands move simultaneously. The new finding here
is that interactions between the hands occur even when
there is a considerable delay (about 1 s) between the two
hands’ motions. A future parametric study could vary the
delay between successive hand movements to determine
how long the delay must be for posture-copying to per-
sist. It is interesting that in cyclic movements of the two
hands, one hand slightly leads the other (Stucci and Viv-
iani 1993; Swinnen et al.1996; Treffner and Turvey 1995),
as if information for one hand is passed immediately to
the other hand. Our data suggest that this transfer occurs
even when the other hand is not required to move imme-
diately. In addition, the transferred information appears
to persist for a relatively long time.
A final comment about our results concerns the impli-
cation of our findings for the classical view of learning
mentioned at the outset of this article. According to this
view, which originated with Tolman (1948), what is
learned in spatial performance are spatial locations, not
body positions. Our results do not show that final loca-
tions cannot be learned; indeed, final locations may have
been learned here when the hands reached in alternation
to peripheral targets. The fact that postures were appar-
ently learned as well fits with the posture-based theory
of motor planning developed by Rosenbaum et al. (1993,
1995, 1998) and helps explain the original results of
Marteniuk and Roy (1972) – that when distances must be
reproduced from original starting positions, performance
is good, whereas when distances must be reproduced from
different starting positions, performance is poor. When
distances must be reproduced from original starting posi-
tions, the same final posture can be adopted in the induc-
tion and testing phases. However, when distances must
be reproduced from different starting positions, the same
final posture cannot be adopted in the induction and test-
ing phases. The fact that performance was better when the
same final posture could be adopted fits with the view
that final postures are stored and relied on in the plan-
ning of movement.
Acknowledgements This study was supported by NSF grant
SBR-94-96290, NIH Research Scientist Development Award KO2
MH00977, and a grant from the Research and Graduate Studies
Office of Penn State University. Jason Bloom and Stephanie Lee
helped with apparatus construction, data acquisition, and data anal-
ysis. Cathleen Moore and two anonymous reviewers provided help-
ful comments.
References
Barnes HJ, Vaughan J, Jorgensen MJ, Rosenbaum DA (1989) A
low-cost method for recording videotaped continuous move-
ments with the Macintosh. Behav Res Meth Instr Computers
21:255–258
Bock O, Eckmiller R (1986) Goal-directed arm movements in ab-
sence of visual guidance: evidence for amplitude rather than
position control. Exp Brain Res 62:451–458
Bootsma RJ, Van Wieringen PC (1990) Timing an attacking fore-
hand drive in table tennis. J Exp Psychol Hum Percept Per-
form 16:21–29
511
Brebner J, Shephard M, Cairney P (1972) Spatial relations and S-
R compatibility. Acta Psychol 36:1–15
Desmurget M, Prablanc C, Rossetti Y, Arzi M, Paulignan Y, Urqu-
izar C, Mignot JC (1995) J Neurophys 2:905–910
Dornay M, Uno Y, Kawato M, Suzuki R (1996) Minimum muscle-
tension change trajectories predicted by using a 17-muscle mod-
el of the monkey’s arm. J Mot Behav 2:83–100
Feldman AG (1966) Functional tuning of the nervous system with
control of movement or maintenance of a steady posture: II.
Controllable parameters of the muscles. Biophysics 11:565–578
Fischer M, Engelbrecht SE, Rosenbaum DA (1994) Evaluation of
a video-based motion analysis method. Five College Cognitive
Science Paper No. 94-1, Amherst, Massachusetts
Flash T, Hogan N (1985) The coordination of arm movements: an
experimentally confirmed mathematical model. J Neurosci 5:
1688–1703
Georgopoulos et al. (1981)
Haggard P, Hutchinson K, Stein J (1995) Patterns of coordinated
multi-joint movement. Exp Brain Res 167:254–266
Jaric S, Corcos DM, Latash ML (1992) Effects of practice on final
position reproduction. Exp Brain Res 91:129–134
Jaric S, Corcos DM, Gottlieb GL, Ilic DB, Latash ML (1994) The
effects of practice on movement distance and final position re-
production: implications for the equilibrium-point control of
movements. Exp Brain Res 100:353–359
Kawato M (1996) Bidirectional theory approach to integration. In:
Inui T, McClelland JL (eds) Attention and performance XVI: in-
formation integration. MIT Press, Cambridge, MA, pp 335–367
Kelso JAS, Holt KG (1980) Exploring a vibratory systems analysis
of human movement production. J Neurophys 43:1183–1196
Laabs GJ (1973) Retention characteristics of different reproduction
cues in motor short-term memory. J Exp Psychol 100:168–177
Latash ML (1993) Control of human movement. Human Kinetics,
Champaign, IL
Marteniuk RG, Roy EA (1972) The codability of kinesthetic loca-
tion and distance information. Acta Psychol 36:471–479
Osu R, Uno Y, Koike Y, Kawato M (1997) Possible explanations
for trajectory curvature in multijoint arm movements. J Exp
Psychol Hum Percept Perform 23:890–913
Rosenbaum DA, Engelbrecht SE, Bushe MM, Loukopoulos LD
(1993) Knowledge model for selecting and producing reaching
movements. J Mot Behav 25:217–227
Rosenbaum DA, Loukopoulos LD, Meulenbroek RG, Vaughan J,
Engelbrecht SE (1995) Planning reaches by evaluating stored
postures. Psychol Rev 102:28–67
Scott SH, Kalaska JF (1997) Reaching movements with similar
hand paths but different arm orientations: I. Activity of indi-
vidual cells in motor cortex. J Neurophys 77:826–852
Smyth MM (1984) Memory for movements. In: Smyth MM, Wing
AM (eds) The psychology of human movement. Academic
Press, London, pp 83–117
Stucci N, Viviani P (1993) Cerebral dominance and asynchrony
between bimanual two-dimensional movements. J Exp Psychol
Hum Percept Perform 19:1200–1220
Swinnen SP, Heuer H, Casaer P (eds) (1994) Interlimb coordina-
tion: neural, dynamical, and cognitive constraints. Academic
Press, San Diego
Swinnen SP, Jardin K, Meulenbroek R (1996) Between-limb asyn-
chronies during bimanual coordination: effects of manual dom-
inance and attentional cueing. Neuropsychology 34:1203–1213
Tolman CE (1948) Cognitive maps in rats and man. Psychol Rev
55:189–208
Treffner P, Turvey MT (1995) Handedness and the asymmetric dy-
namics of bimanual rhythmic coordination. J Exp Psychol Hum
Percept Perform 21:318–333
Uno Y, Kawato M, Suzuki R (1989) Formation and control of op-
timal trajectory in human multijoint arm movement: minimum
torque-change model. Biol Cybern 61:89–101
Vaughan J, Rosenbaum DA, Harp CJ (1997) Modeling planning
and reaching. In: Shafto MG, Langley P (eds) Proceedings of
the Nineteenth Annual Conference of the Cognitive Science
Society. Lawrence Erlbaum, Mahwah, NJ, pp 793–795
Wallace RJ (1971) Stimulus-response compatibility and the idea
of a response code. J Exp Psychol 88:354–360
Wickens D (1938) The transference of conditioned excitation and
conditioned inhibition from one muscle group to the anatago-
nistic group. J Exp Psychol 22:101–123
Wiesendanger M, Kazennikov O, Perrig S, Kaluzny P (1996) Two
hands – one action: the problem of bimanual coordination. In:
Wing AM, Haggard P, Flanagan R (eds) Hand and brain: neu-
rophysiology and psychology of hand movement, Academic
Press, San Diego, pp 283–300
Wolpert DM, Ghahramani Z, Jordan MI (1994) Perceptual distor-
tion contributes to the curvature of human reaching move-
ments. Exp Brain Res 98:153–156
Wolpert DM, Ghahramani Z, Jordan MI (1995) Are arm trajecto-
ries planned in kinematic or dynamic coordinates? An adapta-
tion study. Exp Brain Res 103:460–470
512
