1
The Effects of Low Latency on Pointing and
Steering Tasks
Sebastian Friston, Student Member, IEEE, Per Karlstr
¨
om, and Anthony Steed, Member, IEEE
Abstract—Latency is detrimental to interactive systems, especially pseudo-physical systems that emulate real-world behaviour. It
prevents users from making quick corrections to their movement, and causes their experience to deviate from their expectations. Latency
is a result of the processing and transport delays inherent in current computer systems. As such, while a number of studies have
hypothesized that any latency will have a degrading effect, few have been able to test this for latencies less than
∼
50 ms. In this study we
investigate the effects of latency on pointing and steering tasks. We design an apparatus with a latency lower than typical interactive
systems, using it to perform interaction tasks based on Fitts’s law and the Steering law. We find evidence that latency begins to affect
performance at
∼
16 ms, and that the effect is non-linear. Further, we find latency does not affect the various components of an aiming
motion equally. We propose a three stage characterisation of pointing movements with each stage affected independently by latency. We
suggest that understanding how users execute movement is essential for studying latency at low levels, as high level metrics such as total
movement time may be misleading.
Index Terms—Latency, Indirect Input, HCI, Fitts’s law, Human Factors.
F
1 INTRODUCTION
A
DVANCED graphical interfaces are commonly used to fa-
cilitate intuitive visualisation and manipulation of data as
efficiently as possible. Some do this with abstractions such as
widgets or manipulators. Others, such as pseudo-physical interfaces,
exploit knowledge about natural object behaviour to allow more
intuitive interaction techniques. For example, by constraining
virtual objects by the laws of physics [
1
]. When successful, a
user will take the same approach to tasks in this system as
they would to such a system in the real world. This is achieved
through the formation of the sensorimotor loop - the “continued
correlation between proprioception and sensory data” [
2
]. To form
and maintain this loop, the responses of the system to user input
must meet the expectations the interface creates.
One way in which these responses can deviate from such
expectations is in how fast the user receives them - the latency.
Latency is defined as the time between a user’s action and the
response to this action. Keeping latency low is important to
maintain the perception of a correlation between a user’s action
and the response to it. As a product of the inherent processing and
transport delays within a computer system, latency will never reach
zero [
3
]. Previous studies examined the effects of latency on a
number of sensory modalities, from latency detection in immersive
virtual environments, to its effects on indirect physical interaction.
This latter modality has received considerable attention due to
its ubiquity and importance, with many previous studies using
motion primitives such as pointing tasks to investigate the effects
of latency [
4
], [
5
], [
6
], [
7
], [
8
]. Only recently though has it become
practical to build apparatus with latencies low enough that the
limits of its effects may be found [9].
• Sebastian Friston is with University College London.
E-mail: sebastian.friston.12@ucl.ac.uk
• Per Karlstr
¨
om is with Maxeler Technologies Ltd.
E-mail: pkarlstrom@maxeler.com
• Anthony Steed is with University College London.
E-mail: a.steed@cs.ucl.ac.uk
Manuscript received January 08, 2015; revised June 03, 2015.
The human motor system has been modelled as a control loop,
with inherent delays that place natural limitations on performance;
movement cannot be coordinated on time-scales smaller than the
inherent delay [
10
]. We therefore hypothesize that there may be
a non-zero external latency which has no perceptible effect on
the sensorimotor loop. Latency cannot be removed given current
technology, but we can compensate for it. By understanding how
latency affects the different modalities that create an effective
user interface, we can distribute resources of computer systems to
minimize negative effects and create a better user experience.
In this study we investigate the modality of indirect physical
interaction, using familiar desktop based pointing and steering tasks.
We create an apparatus similar to previous studies [
4
], [
7
], [
8
] but
capable of much lower latencies. Our results, taken with those from
studies on other modalities, will help guide the requirements for
future interactive systems and better estimations of the effectiveness
of existing ones. Further, our results have implications for future
studies using physical tasks to investigate latency, as we show
considering only total movement time and not its constituent parts
may result in inconclusive measurements which hide the effects of
latency.
2 PREVIOUS WORKS
An interface affording natural interaction can have a number of
advantages. Exploiting the user’s intuition may reduce learning
time or improve performance. For example, Smith et al. applied real
world physical constraints to the objects in a 3D editor, decreasing
the degrees of freedom of the objects in a familiar way. Users
showed improved performance when interacting with the editor
using 2D interaction techniques [
11
]. Where interfaces emulate
physical systems, users will likely interact with their motor system.
Even when abstractions are present, actions are still predominantly
basic motion primitives such as reaching and pointing [
1
]. To
encourage natural interaction, an interface must exercise the same
functionalities of the user that are exercised day to day by the
2
real world. To do this the system must provide stimuli from which
users can form percepts and react to them as if they were real [
2
].
Limitations of current technology prevent perfect emulation of
these stimuli however. One of these limitations is latency, which
is an unavoidable result of the inherent processing and transport
delays in the computer system itself [3].
2.1 Models of the Motor System
A number of authors have constructed theoretical models to
explain the operation of the visuomotor system. One such model
is that of Botzer & Karniel [
12
]. The authors derived their model
from observations of delay compensation behaviours. Participants
performed Fitts’s law style tests [
13
]. They were allowed to adapt
to different latency conditions, and then the visual feedback was
removed and at the same time the latency changed. By observing
how user motion changed under this new condition, the authors
tested where in the hypothesized control loop delay compensation
was performed. Whether in the feedforward model, which plans
the trajectory, or the feedback loop, where correction commands
are issued based on visual feedback. Overshoot and undershoot
were present in reaching tasks in unexpected delay conditions. This
demonstrates dominance of an adapted visual feedback stage over
the feedforward planning stage. They also found that while discrete
reaching movements returned to baseline conditions (that is, the
users no longer overshoot or undershoot), rhythmic ones do not.
This suggests there is adaption in the forward model, but it is
dependent on movement type, leading to their model incorporating
multiple pathways.
Beamish et al. considered the motor system as a Vector
Integration to Endpoint (
VITE
) circuit. In the
VITE
circuit a
continuous outflow of commands to the muscles are a result
of the motor system attempting to reduce the difference vector
between the intended target position and the present position. The
commands are generated by the neuron population calculating the
difference vector, based on the present position estimation from
a population which integrates all previous movement commands.
They note the
VITE
circuit as one of the earliest models to suggest
how the movement characteristics described by Fitts’s law are
a result of underlying neurobiological mechanisms. The authors
introduce time delay between the two populations into this model.
By drawing comparisons with a servomechanism model, they show
that for a system to be stable the gain (magnitude) of the movement
commands must be below a value which is a function of delay. It
should be noted that the model described above does not take into
account visual feedback - or indeed any external delays. That is,
even considering a system based only on proprioceptive cues the
authors demonstrate a hard upper limit on performance [10].
Beamish et al. pursued this model, using it to estimate the
inherent effective feedback delays in the motor system based
on the results of previous Fitts’s law style experiments. They
expressed the performance of the
VITE
circuit (movement time)
in terms of difference vector neuron population time constant, and
feedback delay. They could then relate these parameters to the
observed Fitts’s law constants
a
&
b
. Using the measurements
available from over 25 previous Fitts’s law style studies, they found
feedback delays between 0-112 ms, generally below 60 ms. They
also found that the nature of the
VITE
circuit imposes a limit on
the performance of unidirectional movement [
14
]. When this limit
is expressed as a Fitts’s law Index of Difficulty (
ID
), it happens to
be the typical range employed by previous experimenters.
2.2 Measuring the Effects of Latency
An advantage of assessing an interface that emulates a physical
system is that there is a clear baseline to compare it to: the
real system. Considering latency and physical interaction, we can
measure a user’s performance to see how this degrades from the
‘real world standard’ as latency increases. A number of studies
have done this using typical motion primitives, such as pointing
and reaching tasks. Performance is defined in terms of completion
time and error rate.
For example, Jay et al. used task completion time and error
rate to measure the performance of users in collaborative physical
manipulation tasks while experiencing delays in haptic feedback.
25-400 ms of latency was added during the experiments. The
authors estimated a base latency of 14 ms (7 ms from the projector
and 7 ms from the network). They found a strong interaction
between latency and both error rate and task completion time. The
authors also documented users employing the impact-perceive-
adapt model of latency compensation. This states that latency
begins to degrade performance, before the user is aware of it. Once
the latency becomes large enough to cause a “breakdown of the
perception of immediate causality” (the sensorimotor loop), the
user adopts a ‘wait and see’ pattern. They act based entirely on
predictions of the result of their motion, wait for a response, and
then make corrections using the same technique. At this point real
time correction of ceases, and the user’s response time consists
almost entirely of the system delay [15].
2.3 Fitts’s law
Most studies on physical interaction, such as that of Jay et al.,
use Fitts’s law style tests. A good review of Fitts’s law is by
Seow [
16
]. Fitts’s law is an emergent property rather than a
description of the motor system operation. This is discussed by
Bootsma et al. [
17
] and Huys et al [
18
]. Both sets of authors
demonstrate that by observing the patterns of motion directly under
different conditions, Fitts’s law is a good summary of complex
motor processes. However there is increased asymmetry in the
amount of time spent in the acceleration stage vs. the deceleration
stage as latency increases. The pattern of movement is significantly
different between rhythmic and non-rhythmic movement, and as
ID
increases rhythmic pattern becomes more like the discrete
pattern. This suggests multiple functionalities acting in parallel,
such as in the model proposed by Botzer & Karniel [
12
]. Botzer &
Karniel referred to Rythmic/Non-Rythmic as Slicing and Reaching
respectively.
As a characterisation of the motor performance, Fitts’s law has
been observed a number of times under a range of conditions. Its
repeatability and invariance make it valuable for testing the effects
of various factors on user interaction. For example, Adam et al.
measured the difference between egocentric guided movement and
allocentric guided movement [
19
]. This was expanded on by Blinch
et al., who found that the most significant effects occurred between
the presence of allocentric markers and the preparation stage of
movement [
20
]. Perrault et al. tested the scale effect using Fitts’s
law [
21
]. Jax et al. tested the effects of obstacles in the movement
path [22].
2.3.1 Fitts’s law and Latency
For the same reasons described above, Fitts’s law has been used
extensively to investigate the effects of latency. MacKenzie & Ware
did one of the first studies in this area, reformulating Fitts’s law to
3
account for additional movement time delay [
4
]. They estimated
the base latency at 8.3 ms, and between 16 and 225 ms of latency
was added. Pavlovych & Stuerzlinger suggest that the base latency
could actually have been
∼
60 ms though [
8
]. Performance began to
decrease significantly at the 75 ms condition. Ware & Balakrishnan
used 3D reaching tasks in order to compare the effects of hand
tracking delay with head tracking delay in an immersive Virtual
Environment. They tested latencies between 87 and 337 ms. Teather
et al. measured the effect of latency and jitter on performance in
Fitts’s style 2D tasks, and 3D object movement tasks, while looking
for an effect of the type of tracker used. They measured the latency
of their system at 73 ms and found that the performance degradation
was equal for the tracker devices [
7
]. Pavlovych & Stuerzlinger
performed a Fitts’s law style test to determine the effects of jitter
and latency. They found a strong interaction with latency and jitter.
Further, with low jitter the effects of latency were dominant, but
the jitter degraded performance at a higher rate than latency. The
authors measured the base latency of their system at 33 ms, and
added up to 100 ms [
8
]. Chung & So considered that latency may
affect the stages of movement differently. They studied the effects
of latency in Fitts’s law style tests but on target width and distance
separately. There was strong interaction between latency and target
width, but not target distance [6].
2.4 The Steering law
There is evidence ([
12
], [
10
]) that the motor control system consists
of multiple complex elements, some acting in parallel, and that
the effects of latency on these is not equivalent. Thus in our
experiment, aside from a Fitts’s law-style task, we introduce a
second task based on the Steering law. It is designed to exercise
the real-time correction functionalities predominantly and force the
user to continually change goals as they move.
The Steering law was introduced by Accot & Zhai. It was
originally derived from Fitts’s law, considering a path as a sequence
of goal crossing tasks. The completion time was the measure
of performance, and was estimated to be the sum of the time
to complete the individual goal crossing tasks, that make up a
path [
23
]. It was extended by Kulikov et al., who used the concept
of effective width to demonstrate that the Steering law was even
more accurate than originally shown [24].
Like Fitts’s law the Steering law has been used to investigate
the effect of specific factors on user performance. Liu et al.
investigated which path properties affected user performance. The
path properties considered were curvature and width [
25
]. Liu
& Liere continued to investigate the effect of these properties
changing within a path. In their test the path was presented as a
tube. Participants were encouraged to remain within it by pushing
a ball through it with the cursor. We model our implementation of
the Steering law task on theirs. On examining the user movements,
they assert that the behaviour does not resemble a goal crossing
task, as much as a set of small ballistic movements [26].
Pavlovych & Stuerzlinger investigated the impact of latency
on tracking tasks. While this task is analogous to the Steering law
task, the authors point out that the Steering law itself does not
apply. This is because there are no boundaries to movement outside
of the target area, and the user is required to correct velocity as
well as direction. The experimental setup had a base latency of 20
ms, and an additional latency of 30-150 ms. The authors observed
a significant effect of latency on tracking accuracy, and that it
was not symmetric: users had a smaller error perpendicular to the
target, than tangential. The latencies that could be tolerated before
a significant interaction was visible were higher than in previous
studies (50 ms for latency and 40 ms for jitter). Another interesting
observation was that performance decreased for the condition with
the lowest additional latency (20 ms), improved between 20-50
ms, then for latencies above 50 ms degraded again but at a slow
rate [27].
2.5 Investigation of very low latencies
The closest study to ours is that of Jota et al [
9
]. They studied
the effects of latency on direct interaction surfaces, with their
HPT (High Performance Touch prototype) - a touchscreen with
a latency of less than 1 ms. A number of previous studies have
investigated the effects of latency on direct touch interaction, but
none at such low levels. Participants performed Fitts’s law style
tests. Of particular interest in this study, is that the user received
visual feedback from both their non-latent hand and the latent
cursor simultaneously. How the potentially conflicting stimuli affect
performance is not clear. Participants showed a range of behaviours
in response to the latent cursor, from ignoring it completely, to
leading it, to slowing their movement so that it remained under
their finger at all times. The additional latencies were between 1-50
ms. The authors reported no observable difference in performance
between latencies of 1 ms and 10 ms. A linear regression fit
suggested the performance floor may not exist. By segmenting
the movement into stages, the authors demonstrate the effects of
increasing latency on these are not symmetric, as Chung & So and
Bootsma et al. showed for increasing ID [6], [17].
3 EXPERIMENT
A number of studies have used performance in motor tasks to detect
the effects of latency. Few though have investigated latencies at very
low levels. Jota et al. found a potential floor for direct interaction
tasks [
9
]. Indirect interaction techniques however remain important
for both 2D and 3D interfaces. They can exceed direct interaction
in both efficiency and precision [
1
]. We therefore continue the
investigation into indirect interaction.
3.1 Apparatus
To conduct the investigation an interface with very low controllable
latency was required. The indirect input Fitts’s law and Steering
law tests require a 2D interface. The participants interacted through
a cursor, which had to respond to the user within the shortest
amount of time possible. As described by Mine, latency consists
of tracker delay, processing delay, rendering & display scan-out
delay, and transport delays between those stages [
28
]. By probing
and optimising the latency between different parts of our system
we constructed a system with a latency of
∼
6 ms using mostly
off-the-shelf components (Figure 1).
3.1.1 Tracker
The tracker was a Kingston Mouse-in-a-Box optical mouse, with
the Control-Display gain set to 1. Many newer mice, such as this
one, can be sampled at 1kHz. The mouse device, connected via
USB, was polled directly by our application, avoiding the event
system of the operating system.
4
3.1.2 Rendering
To drive the display we implemented our own display controller on
a Maxeler Dataflow Engine (
DFE
) [
29
].
DFE
s are processing cards
which execute dataflow computations. Algorithms are described
as dataflow graphs, which are implemented as pipelines of single-
purpose cores executing in parallel in space, rather than sets of
operations executed by a small number of multipurpose cores such
as on CPUs. This spatial parallelism provides high performance,
and a deterministic latency at levels lower than that achievable by
conventional GPUs. We designed an algorithm to render 2D sprites,
driven by an application running on the CPU, and described it as
a dataflow graph using MaxCompiler, Maxeler’s toolchain. Our
algorithm is deterministic. Knowing exactly how long it takes to
compute one pixel, we can begin computation of a pixel using
the latest tracking data, that much time before it is required for
transmission. At no point in the system is a frame buffered, on each
clock tick a new pixel is completed and transmitted to the display.
We used the parallel port of the host computer and an output
from the
DFE
to probe the latency of the rendering stage of our
system. The
DFE
illuminated an LED on receipt of a specific
input. High speed video monitored the input device, and the LED.
This arrangement was chosen as it allowed us to monitor both
the input device and the scan-out of the display, with no further
instrumentation. The latency between the input and the LED was
below the temporal resolution of the video (1 ms). In the best case
scenario the user begins just prior to the cursor is drawn. In this
case the latency is between 1-2 ms - predominantly the mouse
sampling time. In the worst case the user moves immediately after.
In this case the latency is 8-9 ms. This is the mouse sampling
time (1-2 ms), the rendering time (
<
1 ms) and the period of one
frame on our display (6.9 ms). We expect the latency to be
∼
5
ms on average. We measured the total end-to-end latency of our
apparatus using the cross-correlation variant of Steed’s Method.
Correct operation of the apparatus was confirmed by measuring
the latency throughout the investigation, between each participant.
The baseline latency was measured at 6 ms, with the tolerances
described for the measurement method [30].
In our renderer, a number of sprites and a background map are
composited to make a frame. The content and transformations of
these sprites make up the renderer state. The renderer maintains
the state, which is updated asynchronously by the CPU. With
Maxeler’s assistance, a modification was made to MaxCompiler,
which allowed DVI compliant display data to be output directly
from the
DFE
. We also constructed an electrical interface that
would allow the DFE to drive any DVI receiver.
At
∼
1 ms, the latency of our system from input to video
signal output is much lower than previous apparatus. We are
limited by display technology however. The display scan-out time
increases our end-to-end latency to 6 ms. Beyond this, persistence
of the image on the monitor can cause the perception that latency
is greater than the average frame period. This is because the
stimuli at any time is a blur between the current stimuli and the
previous one. We selected a highly responsive, high frame rate
monitor (an ASUS VG248QE), minimizing perceived latency due
to both scan-out delay and persistence. The limitations in available
display technology are shared by previous authors. Out of the
aforementioned studies only Jota et al. secured a better performing
display than the VG248QE. They did this by building a custom
display based on a Digital Micro-mirror Device driven in a very
low chromatic range [9].
3.1.3 Processing and Transport
Our system was based around an Intel Core i7 PC running CentOS
6. The tests were implemented in a thread running with real-
time priority, controlled by a non-realtime manager application.
The renderer was accessed using Maxeler’s low-latency API for
communicating with the
DFE
via the PCIe bus. Like the mouse
access, this makes use of polling, rather than events. The real-time
thread communicated with the managing application via flags in
memory. We profiled the thread to ensure that we only used calls
which would not cause it to yield unintentionally. The thread was
given the highest priority. The result was that the thread was never
pre-empted, and latency due to time-slicing of the CPU was not
introduced.
Fig. 1. Experimental apparatus that the participants interacted with.
3.2 Participants
30 participants (19 M/11 F) with an average age of 27 (Standard
Deviation: 4 years) from within University College London were
recruited for the study. Participants were paid £5 for taking part.
3.3 Procedure
Participants were seated
∼
0.6m in front of the display, their right
hand being obscured by a black cloth. They were invited to move
the chair, display and mouse. Once comfortable, they were shown
the two tasks and allowed to practice each for as long as they
wished. All participants were instructed to move as fast as possible.
The participants spent 20-30 minutes completing the actual tests.
The time to complete the whole experiment was 30-50 minutes.
Our experimental design is very similar to the one-directional
tapping task described in ISO9241-9 [
31
]. We deviated by asking
users to make discrete movements, rather than repeated rhythmic
movements. This is because the motor system behaves differently
during these two types of motion [
18
], [
12
]. Further, the seminal
works using Fitts’s law to investigate latency, such as that of
MacKenzie & Ware [4], use discrete tasks.
3.4 Tasks
3.4.1 Fitts’s law
For the Fitts’s law style tests, participants saw a box on the screen
∼
2cm x 2cm, which remained throughout all the tests (the staging
area). Clicking on this box would start the test, and a target would
become visible to the right. Participants were instructed to click
on the target as fast as possible, then in their own time move back
to the staging area. Clicking the staging area a second time would
begin the second test, and they were to repeat this until all tests
were complete.
5
3.4.2 Steering law
For the Steering law tests, users were presented immediately with
a 2D path, and at the start of the path, a green ball. They were
instructed to push the ball through the path, by placing the cursor
behind the ball and moving it forward through the path. Users were
again told to maximise speed, and were told that keeping the cursor
within the path would be the fastest way to complete the tests.
Examples of the stimuli seen by the users are in Figure 2.
Fig. 2. Images of the stimuli the participants were exposed to.
3.5 Design
The experiments had three independent variables: Latency, Width
and Distance (Fitts’s)/Curvature (Steering). For both Fitts’s law
and the Steering law there were four conditions of spatial difficulty,
summarised in Table 1. For each condition there were six additional
latencies (0, 10, 20, 30, 50, 80) for a total of 2x2x6 (24) unique
conditions. Unique Fitts’s law conditions were repeated 8 times,
and Steering law conditions 5 times. The repetitions were averaged
for each participant, resulting in 720 data points for the Fitts’s
law tests and 720 for the Steering law tests. The tasks had low
entertainment value, and fatigue was a concern. Since we expected
the effect to be small, we optimised for a high number of latencies
and repeats at the expense of spatial difficulty range. The widths
and distances were informed by pre-trials. The range of IDs
found by these matched those of MacKenzie & Ware, and those
estimated by Beamish et al. [
4
], [
10
]. The IDs were calculated
using MacKenzie’s method [32].
The Steering law paths were manually created, with one
designed to emphasize sharper higher rate turns (predominantly
exercising the wrist) and other sweeping turns to exercise the upper
arm (classed as curvatures 2 & 1 respectively). The IDs were
calculated with Accot & Zhai’s method [
23
]. The curves are not
produced from any predicable function. This was deliberate, to
prevent any unanticipated motor process (such as that used for
reciprocal movement) from hiding the effect of latency on the on-
line correction processes. Conditions were distributed to maximize
the difference between sequential latencies. Within this constraint
the widths and distances/curvatures were distributed randomly. All
participants received the same conditions in the same order.
4 RESULTS & DISCUSSION
4.1 Pointing Tasks
We measure Movement Time (
MT
) to be from the time the user
clicks the staging area, to the time they click the target. Figure 3
shows
MT
for each of the latencies. As expected
MT
increases
with ID, with a jagged appearance due to the small number of
spatial (Width & Distance) conditions that do not have overlapping
IDs [
8
]. We clearly see an increase in
MT
with high latencies, but
not for low latencies. This is better illustrated in Figure 4 which
shows how MT changes with latency for each condition.
TABLE 1
Spatial Difficulty Conditions for both types of task.
Condition Index of Difficulty
Fitts’s law
Width (cm) Distance (cm)
1 0.25 4 4.09
2 0.25 11 5.49
3 0.9 4 2.44
4 0.9 11 3.72
Steering law
Width (cm) Curvature
1 0.4 1 45.37
2 0.4 2 50.63
3 0.7 1 25.92
4 0.7 2 28.92
2 2.5 3 3.5 4 4.5 5 5.5
600
800
1000
1200
1400
1600
Index of Difficulty
Movement Time (ms)
6ms
16ms
26ms
36ms
56ms
86ms
Fig. 3. Movement Time against Index of Difficulty, for all latency conditions.
Error bars indicate confidence intervals at 95%.
6 26 46 66 86
600
800
1000
1200
1400
1600
Latency (ms)
Movement Time (ms)
Fitts’s 1
Fitts’s 2
Fitts’s 3
Fitts’s 4
Fig. 4. Movement times for each condition against latency. Error bars
indicate confidence intervals at 95%.
4.1.1 Comparison with Previous Works
Studies conducting experiments most comparable with ours in-
clude [
4
], [
7
], [
8
]. All studies included Fitts’s law style tests using
mice, with latency as the independent variable.
•
MacKenzie & Ware [
4
] investigated latencies estimated to be
between 68 - 315 ms [8].
•
Teather, et al. [
7
] investigated latencies measured at 35 - 255
ms.
•
Pavlovych & Stuerzlinger [
8
] investigated latencies measured
between 33 - 133 ms.
For [
7
] and [
8
] the latencies measured are the total end-to-end
system delay, the same as measured by us. We first consider only
the higher latency conditions (36, 56, 86 ms) which are directly
comparable with the previous studies above.
We fit a model using multiple linear regression and show
a significant interaction with width
(β = −499.81, t(356) =
![Fig. 5. Movement time for the 36, 56 & 86 ms conditions of the current study, compared with the overlapping conditions from Pavlovych & Stuerzlinger (PS) [8], MacKenzie & Ware (MW) [4] and Teather, et al. (T) [7].](/figures/fig-5-movement-time-for-the-36-56-86-ms-conditions-of-the-3up8edyr.webp)
