Running head: IMPROVING TIMING ABILITY
Improving Timing Ability: Generalization of Explicit Timing Training to Nontrained Intervals and Implicit Timing.
Gusta Marcus, BSc Supervisor: dr. Michael Vliek
Date: 04-‐07-‐2014
Final version of thesis report
University of Amsterdam Research Master Psychology Program Group: Social Psychology
Second assessor: prof. dr. Maurits van der Molen Number of credits: 25 ec
IMPROVING TIMING ABILITY 2
Abstract
This study investigated to what extent training of interval production improves timing ability. The first aim was to examine if training of one specific interval would generalize to improved performances on nontrained intervals in the 1-‐2 seconds range. The second aim was to examine if this training would transfer to improved performances on a
prediction-‐motion task, in which participants had to use temporal information to predict the location of a moving stimulus. Participants in the experimental condition were trained in producing a 1.7-‐s interval. Their performances improved more than
participants in the control condition on the trained interval and on nontrained intervals surrounding the trained interval. Against our expectations, participants in the
experimental condition did not show a different change in performances on the prediction-‐motion task compared the control condition. However, participants who improved more during training also showed more improvement on the prediction-‐ motion task. These findings suggest that learning is not specific for the duration that is trained, but can influence other durations via updating of representations, modifying the decision criteria, or better directing attention, based on the internal clock model.
Furthermore, an indication is found for a relation between the training to another form of timing.
IMPROVING TIMING ABILITY 3
Improving Timing Ability: Generalization of Explicit Timing Training to Nontrained Intervals and Implicit Timing.
Timing is an essential ability in people’s daily lives and it is needed in many activities. For instance, Dutch drivers are urged to use the 2-‐seconds rule to secure safe distances to other vehicles. The rule states that there should be at least 2 seconds between two cars passing a certain point, for example a tree, so the driver has enough time to react on surprising situations. Another example in traffic that requires timing ability is when a pedestrian crosses the road and needs to predict if there is enough time to walk across the street before a moving car has reached him.
These two examples are considered two different forms of timing; the 2-‐seconds rule can be seen as explicit timing and crossing the street can be seen as implicit timing. The organization of various forms of timing into explicit and implicit timing is illustrated in Figure 1 (Coull & Nobre, 2008). Explicit timing is involved in tasks where participants have to give accurate duration estimations. Within explicit timing, a distinction can be made between motor timing, such as producing a time interval, and perceptual timing, such as discriminating between temporal stimuli. Implicit timing is involved in tasks with nontemporal goals, but with temporal characteristics of the stimuli or motor responses, which makes timing a by-‐product. Within implicit timing, a similar
distinction can be made between motor and perceptual timing. Implicit motor timing is called emergent timing and is involved when timing emerges as a result of movements, for example when circles are drawn continuously at the same speed and consequently in the same time. Implicit perceptual timing is called temporal expectations and is involved
IMPROVING TIMING ABILITY 4 when using temporal information to predict the location of a moving stimulus, as was the case in the example of crossing the road.
The cognitive processes behind explicit and implicit timing can be described with the internal clock model. This model represents the process of making judgments about durations and assumes the existence of an internal clock (Gibbon, Church & Meck, 1984). When a specific duration has to be estimated, a pacemaker emits pulses that are send through an attention-‐based switch that controls the passage of pulses to an
accumulator that counts the pulses. For explicit timing, this value of counted pulses is compared with previously stored values for this specific duration in the reference memory. During this comparison, a decision criterion is used to decide if the counted value matches the stored value and an estimation of the duration is given. For implicit timing, the value in the accumulator could be used as a temporal representation to perform the implicit task (Piras & Coull, 2011). For example, when circles must be drawn within two ticks of a metronome, the counted pulses can be saved as a
representation of the time. When the metronome has stopped, that representation can be used to continue circle drawing in the same time and with the same speed.
Relations Between Different Forms of Timing
Researchers have tried to answer the question if a common timing mechanism exists for all different forms of timing (Merchant, Zarco & Prado, 2008). The internal clock would be involved in all behavior that requires accurate timing. However, the various timing tasks are so different from each other that it would be surprising if they would rely on functions of entirely the same mechanism and the same brain system (Lewis & Miall, 2006). This is supported by evidence that a large, distributed neural
IMPROVING TIMING ABILITY 5 system is involved in various forms of timing instead of a single brain area (Buhusi & Meck, 2005; Allman, Teki, Griffiths & Meck, 2014).
If various forms of timing have different mechanisms, the next important question is how they relate to each other. To get more insight in these relations, different approaches are being used. The neuronal approach links specific timing processes to brain areas. Many different brain areas are proposed to be involved in timing, but researchers have not yet reached a consensus about the neural substrates of specific forms of timing and which different forms use the same brain areas (Wittmann, 2009).
The behavioral approach tries to find evidence for connections between different forms of timing by searching for relations in performances of different tasks. One method is to compare patterns of performances of different timing tasks with each other. Another method is to train one form of timing and examine the influence on timing tasks that are similar or different than the trained task.
Before training can influence other forms of timing, it has to influence
performances of the task that is trained. Training in estimating a specific duration can result in improved performance of that duration (Montare, 1988). Feedback on
performance usually increases the accuracy and reduces the variability of estimating the specific duration, called the learning effect (Ryan & Robey, 2002). Proposed cognitive mechanisms for this improvement, based on the internal clock model, are directing more attention to timing, updating representations of durations in reference memory and altering the decision process by modifying the decision criteria (Franssen & Vandierendonck, 2002; Lamotte, Izaute & Droit-‐Volet, 2012).
IMPROVING TIMING ABILITY 6 Relations Within Explicit Timing
Within explicit timing, a relation is found between perceptual and motor timing. Similar patterns of variability were found between perception and production tasks and this could be an indication of partially overlapping neural networks (Merchant, Zarco & Prado, 2008). Additional evidence of this relation is found in training studies in which people who were trained in discriminating a specific interval from other intervals (explicit perceptual timing) also became better in producing that specific interval
(explicit motor timing; Meegan, Aslin, & Jacobs, 2000; Planetta & Servos, 2008). This was found when auditory or somatorsensory stimuli were used to indicate the intervals during training. The researchers suggested that the sensory and motor representations of the durations are connected, so that they can influence each other.
Evidence for the relation between different intervals is less clear. In the two studies about the relation between perceptual and motor timing, performances of the trained interval were also compared with performances of a nontrained interval (Meegan, Aslin, & Jacobs, 2000; Planetta & Servos, 2008). Production of the nontrained interval improved less than production of the trained interval and the authors argue that the discrimination training of one interval does not generalize to motor production of another interval.
These two studies used an indirect method to investigate the influence on a nontrained interval, because they trained perceptual timing and tested motor timing. When a direct method was used and people were both trained and tested in interval production (motor timing), limited evidence was found that training a singe interval can generalize to nontrained intervals. This is called the generalization effect. A study using intervals in the millisecond range (450, 650 and 850 ms) found improved timing on non-‐
IMPROVING TIMING ABILITY 7 trained intervals close to the trained intervals (Bartolo & Merchant, 2009). Another study provided some evidence that training of 10 s generalized to improved
performance in producing an interval of 30 s (Saito & Tayama, 2012).
The two mentioned studies showed somewhat different results. In the first study, the produced intervals were less variable after the training than before the training, while the researchers did not measure the accuracy (Bartolo & Merchant, 2009). In the second study, participants became more accurate, that is, the mean of produced
intervals was closer to the target value for participants who received training than those without training (Saito & Tayama, 2012). However, the study did not find differences in variability between groups.
These differences in generalization between the two studies can arise from the use of different methods. Participants in the first study were trained with 7200 trials
distributed over 8 days, whereas participants in the second study were trained with only 15 trials. Since accuracy improves rapidly and variability improves slowly (Sohn & Lee, 2013), the extensive training caused improvements in variability, whereas the small training caused the lack of improved variability.
An additional explanation for the differences in results between the two studies is the use of different timing processes that can result in different generalization
processes. Intervals in the millisecond range (short intervals) are processed with automatic mechanisms, whereas intervals in the seconds range (long intervals) are processed with cognitive mechanisms (Lewis & Miall, 2003). Generalization of short intervals involves both improved precision of the timing mechanism and sensorimotor learning (Bartolo & Merchant, 2009). Generalization of long intervals involves more cognitive processes, such as counting and small calculations (for example, x times the
IMPROVING TIMING ABILITY 8 trained interval; Saito & Tayama, 2012). If the cognitive processes can improve accuracy on long intervals with a small training and the automatic processes can improve
variability on short intervals with a more extensive training, it would be most advantageous to find a situation where both processes are being used.
Both automatic and cognitive processes are being used in timing of durations between 1 and 2 seconds, since these durations are in the transition range between the automatically and cognitively processed durations (Witmann, 2009). For example, for values upward of 1.2 s, a counting strategy effectively improves performances (Grondin, Meilleur-‐Wells & Lachance, 1999). Also, durations till 2 s are efficiently processed within the motor system (Morillon, Kell & Giraud, 2009). This suggests that processes in the 1-‐2 s range are overlapping. It is possible that this overlap could improve the outcome and generalization of training to nontrained intervals, since participants can either choose the best process or use both processes.
Relations Between Explicit and Implicit Timing
In daily life, people rarely have to estimate exact durations, whereas most real-‐ world timing is implicit. Examples of implicit timing in the 1-‐2 seconds range can be found in sports (hitting a tennis ball), traffic (merging onto the highway), or public transport (closing of metro door after signal). Therefore, training of explicit timing would have more practical utility when it improves performances on implicit timing. This transfer of a learned skill to another task is called the transfer effect (Ryan & Fritz, 2007). Beside, it is theoretically interesting to examine the relation between explicit and implicit timing, because evidence is inconclusive about the possible shared mechanisms. According to the explicit-‐implicit framework, explicit and implicit timing are functionally distinct (Coull & Nobre, 2008) and partially distinct neural substrates are
IMPROVING TIMING ABILITY 9 suggested for explicit and implicit timing mechanisms (Wiener, Turkeltaub & Coslett, 2010). Nevertheless, it seems that these two forms of timing have some shared processes (Zelaznik et al., 2005).
Explicit and implicit timing are suggested to share the same representation of time, but use different timing mechanisms dependent of the particular task requirements. In a study by Piras and Coull (2011), explicit perceptual timing (discrimination between intervals) was compared with implicit perceptual timing (predictable interval between cue and target that facilitated reaction times). Accuracy patterns over intervals were similar in the two tasks. This indicated that explicit and implicit timing processes share the same mechanism for representations of durations; in the internal clock model, this is represented as the process where pulses are emitted and counted. However, variance patterns over intervals were different in the two tasks for longer durations (≥ 600 ms) and this indicated that the representations are used in different ways.
Explicit and implicit timing do not only use the same representations of time, but it even seems that representations of explicit motor timing are needed to perform implicit motor timing. In a study by Zelaznik et al. (2005), explicit motor timing (finger tapping in a specific interval) was compared with implicit motor timing (circle drawing in a specific interval). Similar timing processes were found between the first circle and finger tapping, but no relation was found between the subsequent circles and tapping. This suggests that circle drawing first needs a temporal representation and uses explicit timing to establish the control processes. After the first circle, movement dynamics, such as speed and angle, can be used to keep the goal of drawing circles in the specific time frame and it has become emergent timing. This suggests that the skills underlying explicit motor timing are needed to perform the implicit motor timing task.
IMPROVING TIMING ABILITY 10 The described studies have investigated the relation between explicit perceptual and implicit perceptual timing (Piras & Coull, 2011) and the relation between explicit motor and implicit motor timing (Zelaznik et al., 2005). The relation between explicit motor timing and implicit perceptual timing has to our knowledge not yet been investigated. Based on the notion that explicit and implicit timing share the same
representations of time, we predicted such relation exists. Besides, the described studies only show indirect evidence of a relation between explicit and implicit timing, because they compared patterns of performances of different tasks to infer underlying shared processes. In contrast, a training method provides the possibility to directly manipulate the relation between training and the timing task.
The Current Study
The first aim of the current study was to investigate to what extent training of explicit motor timing could generalize to nontrained intervals. We tried to replicate the generalization effect in an interval range of 1 to 2 seconds. The second aim of this study was to investigate to what extent training of explicit motor timing could improve
implicit perceptual timing.
To study this, half of the participants were trained to produce an interval of 1.7 s (experimental condition), while the other half of the participants performed a filler task (control condition). Before (time 1) and after (time 2) the training/ filler task, all
participants were tested on explicit and implicit timing. Explicit timing was measured with a production task that tested production of the trained interval (i.e. 1.7 s) to examine the learning effect. Furthermore, productions of nontrained intervals surrounding the trained interval (i.e. 1.3, 1.5, 1.9, 2.1 s) were tested to examine the generalization effect.
IMPROVING TIMING ABILITY 11 Implicit perceptual timing was measured with the prediction-‐motion (PM) task to examine the transfer effect. In this task, a moving object disappears and participants have to predict when it will arrive at a target point. This task is often used to measure the prediction of motion when visual information is lacking, called time-‐to-‐contact judgments (Bennnet, Baures, Hecht & Benguigui, 2010). Two different mechanisms are involved to accomplish the task (DeLucia & Liddell, 1998). The first mechanism is creating a cognitive representation of the motion to make an accurate prediction. The second mechanism is using a timing component by estimating and counting the time before the object reaches the target. Recently, variations of the prediction-‐motion task are used to measure implicit perceptual timing (Coull, Vidal, Goulon, Nazarian & Craig, 2008; Sohn & Lee, 2013). The prediction-‐motion task represents implicit perceptual timing, because temporal information is necessary to make the predictions, but the goal is nontemporal (i.e. to make a prediction about the movement). Furthermore, the task is comparable to another implicit perceptual timing task where temporal cues are used to facilitate reaction times, because similar brain areas are activated during both tasks (Coull et al.).
We expected to find three different effects:
1) Learning effect: Performances on the trained interval of the experimental condition improves more between time 1 and time 2 than performances of the control condition.
2) Generalization effect: Performances on the nontrained intervals of the experimental condition improves more between time 1 and time 2 than performances of the control condition
IMPROVING TIMING ABILITY 12 3) Transfer effect: Performances on the prediction-‐motion task of the experimental
condition improves more between time 1 and time 2 than performances of the control condition
Method Participants
A total of 67 participants (51 women, 16 men, Mage = 21.4 years, age range: 18-‐57 years) were recruited through the subject pool website of the University of Amsterdam. Participants received either course credit for participation (46 first-‐year psychology students), received €10 (19 participants, varying from higher-‐years psychology students to non-‐students) or did not receive a reward (2 participants, who were willing to do the experiment for free).
Most of the participants were right-‐handed (55 participants), some were left-‐ handed (9 participants) and a few did not have a preferred hand (3 participants). Participants were excluded if they did not have normal or corrected to normal vision (1 participant) or if they stated they did not participate seriously (no participants). The study was approved by the Ethics Review Board of the Faculty of Social and Behavioral Sciences of the University of Amsterdam.
Procedure
Participants started the experiment with reading the information brochure and signing the informed consent form. They were asked to remove their watch and to turn off their mobile phone. They indicated if they had normal or corrected to normal vision and the time of the day was noted. They took place in front of a computer in a cubicle without any distractions.
IMPROVING TIMING ABILITY 13 Participants were randomly assigned to the experimental or control condition. Participants in the experimental condition received a training to produce an interval of 1.7 s. Instead of training, participants in the control condition performed a filler task. Before (time 1; T1) and after (time 2; T2) the training or filler task, participants made a production task and a prediction-‐motion task to test if performances had
changed. The production task, prediction-‐motion task, production training and filler task all started with a few practice trials, to make the participants familiar with the tasks. Participants filled in a mood questionnaire before and after all the tasks to control for mood effects. At the end of the experiment, participants answered some exit questions. Materials
The experiment was performed using Presentation® software (Version 17.0, www.neurobs.com). The experiment was presented on a computer with a display resolution of 1920 x 1080 pixels and a refresh rate of 60 Hz. The tasks are described in the next sections.
Training task. Training of explicit motor timing was done with a training task, in which participants were trained in the production of 1.7 s. Participants were asked to produce this interval by pressing the spacebar of the keyboard and holding it for the duration they thought would be the target interval. When holding the spacebar, a circle appeared on the screen, so the produced interval could be seen. We thought this
facilitated learning, because discrimination mechanisms could be used to compare the produced intervals.
After each trial, feedback was provided in the form of the exact interval that was produced (e.g. 1.685 s). We chose an interval of 1.7, because a fraction of a second is difficult to produce, so participants needed to rely on the feedback during training. The
IMPROVING TIMING ABILITY 14 training consisted of 150 trials, divided in 5 blocks with a small break between the
blocks.
Production task. Explicit motor timing was measured with a production task, in which participants had to produce five different target intervals (i.e. 1.3, 1.5, 1.7, 1.9 and 2.1 s). Participants were asked to produce the intervals by pressing the spacebar and holding it for the duration they thought would be the target interval. When holding the spacebar, a circle appeared on the screen, similar as in the training task. The intervals were randomly presented and each interval was shown 15 times in total. The task was divided in three blocks of 25 trials and participants could take a small break between the blocks.
Prediction-Motion (PM) task. Implicit perceptual timing was measured with a prediction-‐motion (PM) task. In this task, a box moved from left to right and
disappeared behind an object (i.e. a larger box). The box started to move when
participants pressed the spacebar and they had to hold the spacebar in order to keep the box moving. Participants were asked to predict when the moving box would arrive at the end of the object by releasing the spacebar. The box never appeared behind the object, because that would have served as feedback, which was not the intention of the task. Figure 2 shows a representation of the task with an indication when the spacebar should be pressed.
An interval was counted as the time between the moment the box had disappeared (left side of box was equal to the left side of the object) and the moment the box would be at the end of the object (right side of the box was equal to the right side of the object). The 5 different target intervals were manipulated by changing the length of the object
IMPROVING TIMING ABILITY 15 (speed was held at a constant level) and were equal to the target intervals in the
production task (1.3, 1.5, 1.7, 1.9 and 2.1 s).
The size of the box was 60 x 60 pixels and the starting position was 180 pixels from the object. The box moved 2 pixels per screen refresh (ca. 16.7 ms), so it was visible during 2000 ms. The height of the object box was 200 pixels and the length varied from 216 pixels for the 1.3-‐s interval to 312 pixels for the 2.1-‐s interval.
The intervals were randomly presented and each interval was shown 15 times in total. The task was divided in three blocks of 25 trials and participants could take a small break between the blocks.
Filler task. The filler task was similar in length, difficulty and attention
requirements as the training task and generated reaction times in the 1-‐2 s range. It consisted of a two-‐choice reaction time tasks measuring choice selection and
psychomotor speed (Stout et al., 2011). To make the task as challenging as the training task, four versions of the two-‐choice reaction time task were presented with different rules concerning the responses.
In all versions of the two-‐choice reaction time task, participants saw an empty circle that became colored and they had to respond quickly with the left key (‘S’) or right key (‘L’) in agreement with the specified rule. In version 1, an empty circle at the left and right side of the screen was presented. One of the circles turned blue and participants had to respond to the left or right circle with respectively the left or right key. In version 2 and 3, one circle in the middle of the screen became orange or purple and participants had to respond with the associated left or right key that differed per version (version 2: purple-‐left and orange-‐right; version 3: orange-‐left and purple-‐right). In version 4, an empty circle at each side of the screen was presented, similar to version 1. One of the
IMPROVING TIMING ABILITY 16 circles turned orange or purple and participants had to ignore the color and respond to the left or right circle with respectively the left or right key.
A trial started when participants pressed the spacebar to indicate they were ready and a fixation cross was shown. After a random interval between 1000 and 2000 ms, the stimuli appeared (i.e. a circle became colored) on which they had to respond. After each trial, feedback was presented to indicate if the response was correct or wrong. Each of the four blocks contained one version of the two-‐choice reaction time task with 30 trials and participants could take a small break between the blocks.
Mood. If participants became more irritable or less attentive in one condition compared to the other, then they might have wished to finish soon and underestimated the durations. To take into account this possible effect, mood was measured using the Positive and Negative Affect Schedule (PANAS) scale (Watson, Clark & Tellegen, 1988). The questionnaire consisted of 10 positive mood states, for example ‘interested’ and ‘attentive’, and 10 negative mood states, for example ‘irritable’ and ‘nervous’.
Participants were asked to indicate how they felt at that moment and they could answer on a 5-‐point scale with the labels ‘very slightly’, ‘a little’, ‘moderately’, ‘quite a bit’, and ‘very much’. The PANAS scale has good reliability and validity; the Flemish version that was used has a reported Cronbach’s alpha of .79 and .85 for the PA and NA scales respectively (Engelen, De Peuter, Victoir, Van Diest, & Van den Bergh, 2006).
Exit questions. The exit questionnaire consisted of demographical questions and questions about the experiment. Participants were asked for their gender, age,
handedness, and education level. We asked participants how many hours a week they normally spent on activities that require accurate timing. For the training task,
IMPROVING TIMING ABILITY 17 strategy they used to accomplish the tasks. Furthermore, we asked them about their opinion of the tasks, if they had participated seriously, which hand they used for timing in the tasks, and if they had used coffee, tobacco, alcohol or dugs within 2 hours before the experiment.
Data-Analytic Procedure
Before computing the dependent variables, all produced intervals under 300 ms were excluded, because we expected the participant had made a mistake. Produced intervals were considered outliers and were excluded if they fell outside a specified range, based on the interquartile range (IQR). The range had a minimum value of Q1 – 3(IQR) and a maximum value of Q3 + 3(IQR), based on the method to identify extreme outliers with a boxplot (Tukey, 1993). It was calculated separately for every type of interval of the training, production and prediction-‐motion task.
In the training, production and prediction-‐motion task, the produced intervals were compared with the target intervals. For each participant, two dependent variables were computed per type of interval per task. Accuracy score (acc) was the mean error of responses (absolute difference between produced and target intervals divided by the target interval). Variability score (var) was the coefficient of variation (SD divided by mean of produced intervals, per type of interval).
In computing the dependent variables, the raw variables were divided by interval, because this provided a correction for type of interval. The reason for this correction was to make the intervals more comparable, because otherwise the error and standard deviation increase with longer intervals according to the scalar property (Gibbon, Church, & Meck, 1984).
IMPROVING TIMING ABILITY 18 The dependent variables were indications of the performances on the tasks and were compared with each other over time. Improvement was defined as smaller accuracy scores (i.e. smaller errors) and smaller variability scores (i.e. smaller coefficients of variation).
The two dependent variables were analyzed with multivariate analysis of variance (MANOVA) and the F-‐statistics of Pillai’s trace were reported. Repeated measures were used to analyze the differences in time. The statistical level of
significance was set at an alpha of .05. If the assumption of sphericity was violated, a correction was used (Huynh-‐Feldt correction for estimates of sphericity > 0.75;
Greenhouse-‐Geisser correction for estimates < 0.75; based on recommendation of Field, 2009).
The dependent variables were computed with R (R Development Core Team, 2012). The statistical analyses were performed with SPSS (version 19).
Results
Of the entire data set, 197 trials were excluded because they were less than 300 ms (0.64%) and 59 trials were considered outliers (0.19%). For one participant, 83% of the data points of the PM task had to be excluded, therefore no reliable dependent variables could be calculated for PM T1 and those values were missing.
Due to technical problems1, 12 participants had to make a small part of the experiment again (7 participants made one of the three blocks of PM T1 again, 4 participants made one of the three blocks of PM T2 again and 1 participant made the
1 Most problems arose in the PM task. Participants had to press the spacebar to let the box move.
If they pressed the spacebar before the trial had started (i.e. for the first 14 participants, within 50 ms after they pressed enter to go to the next trial and for the others within 16 ms), the
program could not register the spacebar press and it gave an error. All participants were warned beforehand by the experimenter to wait for the start of the next trial. One participant got an unexplainable error in the control task; the program did not react on anything and it had to be forced to stop.
IMPROVING TIMING ABILITY 19 control task again). This could have led to extra noise in the data. Nevertheless, similar results were obtained without those participants. Therefore, data from those
participants were not excluded.
The data was not normally distributed, but skewed to the right. Although the assumption of normality was violated, the F-‐statistic of Pillai’s trace is robust, because the groups are approximately equal (Stevens, 2009).
Type of Interval Effect
When we computed the dependent variables, we used a correction to control for type of interval effect (i.e. more error and variation for longer intervals in the
uncorrected variables). To check if this type of interval effect was present, the uncorrected variables were examined with repeated measures ANOVAs with type of interval (1.3, 1.5, 1.7, 1.9, 2.1 s) as within-‐subjects variable. Indeed, larger scores were found with longer intervals on the production and PM tasks for all uncorrected
dependent variables (uncorrected accuracy scores: all F’s ≥ 18.8, all p’s < .001; uncorrected variability scores: all F’s ≥ 3.4, all p’s ≤ .01).
We therefore expected that after correction, the dependent variables would have comparable values for each interval. However, an opposite effect was found: accuracy and variability scores decreased with longer intervals. This type of interval effect in the corrected variables was also examined with repeated measures ANOVAs with type of interval (1.3, 1.5, 1.7, 1.9, 2.1 s) as within-‐subjects variable. The effect of smaller scores with longer intervals was found on the production and PM tasks for all dependent variables (accuracy scores: all F’s ≥ 4.0, all p’s ≤ .01, variability scores: all F’s ≥ 6.3, all p’s ≤ .001).
IMPROVING TIMING ABILITY 20 Although the corrected variables seem to be overcorrected, we decided to still use them for analyses, because the differences between intervals with corrected variables were smaller compared to the differences with uncorrected variables. The differences between intervals are not a problem when the intervals are compared over conditions or over time, which is the case in all analyses.
Performance Effect
Participants in the experimental condition received training at producing an interval of 1.7 s. Performances on the 1.7 interval of the production task before training were compared with performances during the 5 blocks of training in repeated measures MANOVAs. Accuracy scores can be seen in Figure 3 and variability scores can be seen in Figure 4. In both figures, the lower line with multiple points represents the average score for the five blocks of training.
Performances were more accurate and less variable during training (MAcc = 0.073,
SD = 0.023; MVar = 0.092, SD = 0.028) than on the 1.7-‐s interval of the production task
before training (MAcc = 0.338, SD = 0.173; MVar = 0.215, SD = 0.094). This was confirmed
with a multivariate test that compared the 1.7-‐s interval of the production task at time 1 with the mean of the training blocks, V = 0.82, F(2, 32) = 71.0, p < .001, ηp2 = .816. The univariate tests showed lower accuracy scores during training than before training, F(1, 33) = 87.33, p < .001, ηp2 = .726, and lower variability scores, F(1, 33) = 66.23, p < .001, ηp2 = .667.
Performances became also more accurate and less variable over blocks of
training (block 1 MAcc = 0.087, SD = 0.028; MVar = 0.112, SD = 0.035; block 5 MAcc = 0.068,
SD = 0.027; MVar = 0.081, SD = 0.030). This was confirmed with the multivariate test
IMPROVING TIMING ABILITY 21 univariate tests showed lower accuracy scores over blocks, F(4, 132) = 7.41, p < .001, ηp2 = .183, and lower variability scores, F(4, 132) = 13.54, p < .001, ηp2 = .291.The
improvements over blocks had linear trends for the accuracy scores, F(1, 33) = 13.62, p = .001, ηp2 = .292, and variability scores, F(1, 33) = 26.42, p < .001, ηp2 = .445.
The improvement during training was an indication that our training worked well. This was a necessary condition for the other effects to reveal themselves. Learning Effect
We examined if the participants in the experimental condition could keep up their improved performances after training. We compared performances on the 1.7-‐s interval of the production task over time and between conditions. This was done with a 2 x 2 MANOVA with time as within-‐subjects factor (T1, T2) and condition as between-‐ subjects factor (experimental, control group).
In Figure 3 and Figure 4, the accuracy and variability scores are shown for the trained interval (circle points) over time with separate lines for conditions. Participants in the experimental condition improved more than those in the control condition. This was confirmed by the interaction between condition and time for the multivariate test, V = 0.18, F(2, 64) = 7.23, p = .001, ηp2 = .184 and the univariate tests for accuracy scores, F(1, 65) = 6.12, p = .016, ηp2 = .086, and variability scores, F(1, 65) = 11.25, p = .001, ηp2 = .148.
The interaction effect was further examined with simple contrasts. At time 1, the control and experimental condition did not differ from each other, as was indicated by the multivariate comparison, V = .04, F(2, 64) = 1.21, p = .304, and simple contrasts on accuracy (control MAcc = 0.363, SD = 0.219; experimental MAcc = 0.338, SD = 0.173, p =
IMPROVING TIMING ABILITY 22 .600), and variability scores (control MVar = 0.188, SD = 0.063; experimental MVar = 0.215,
SD = 0.094, p = .164).
In the control condition, performances did not change over time, as was indicated by the multivariate comparison, V = .08, F(2,64) = 2.79, p = .069. Accuracy scores did not change over time (T1 MAcc = 0.363, SD = 0.219; T2 MAcc = 0.323, SD = 0.180, p = .264).
Although variability scores became slightly smaller over time, it was not significant with a corrected alpha of .025 (T1 MVar = 0.188, SD = 0.063; T2 MVar = 0.155, SD = 0.062, p =
.026). In the experimental condition, performances did improve, as was indicated by the multivariate comparison, V = .49, F(2, 64) = 30.27, p < .001, ηp2 = .486. Participants got smaller accuracy scores over time (T1 MAcc = 0.338, SD = 0.173; T2 MAcc = 0.173, SD =
0.162, p < .001) and smaller variability scores (T1 MVar = 0.215, SD = 0.094; T2 MVar =
0.114, SD = 0.052, p < .001).
At time 2, therefore, performances of the experimental condition were better than those of the control condition, as was indicated by the multivariate comparison, V = .20, F(2, 64) = 8.18, p = .001. Participants in the experimental condition had lower scores than participants in the control condition on accuracy (control MAcc = 0.323, SD = 0.180;
experimental MAcc = 0.173, SD = 0.162, p = .001) and on variability (control MVar = 0.155,
SD = 0.062; experimental MVar = 0.114, SD = 0.052, p = .005).
This confirmed our first hypothesis that training caused improved performances on the 1.7-‐s interval compared to no training; we found a learning effect. Surprisingly, participants in the control condition also became slightly less variable and this trend is against our expectations.
IMPROVING TIMING ABILITY 23 Generalization Effect
We examined if the training also had an effect on the intervals that were not trained. The mean of the 4 non-‐trained intervals of the production task was compared over time and between conditions. This was done with a 2 x 2 MANOVA with time as within-‐subjects factor (T1, T2) and condition as between-‐subjects factor (experimental, control group).
In Figure 3 and Figure 4, the accuracy and variability scores are shown for the mean of the nontrained intervals (square points) over time with separate lines for conditions. Participants in the experimental condition improved more than those in the control condition. This was confirmed by the interaction between condition and time for the multivariate test, V = 0.12, F(2, 64) = 4.22, p = .019, ηp2 = .116, and the univariate tests for accuracy scores, F(1, 65) = 5.21, p = .026, ηp2 = .074, and variability scores, F(1, 65) = 6.08, p = .016, ηp2 = .086.
The interaction effect was further examined with simple contrasts. At time 1, the control and experimental condition did not differ from each other, as was indicated by the multivariate comparison, V = .04, F(2, 64) = 1.46, p = .240, and simple contrasts on accuracy scores (control MAcc = 0.379, SD = 0.212; experimental MAcc = 0.339, SD = 0.158,
p = .383), and variability scores (control MVar = 0.190, SD = 0.070; experimental MVar =
0.213, SD = 0.083, p = .225).
In the control condition, performances did not change over time, as was indicated by the multivariate comparison V = .09, F(2, 64) = 3.07, p = .053. Accuracy scores did not change over time (T1 MAcc = 0.379, SD = 0.212; T2 MAcc = 0.336, SD = 0.178, p = .183), but
variability scores became smaller over time, which was even significant with a corrected alpha of .025 (T1 MVar = 0.190, SD = 0.070; T2 MVar = 0.161, SD = 0.063, p = .018). In the
IMPROVING TIMING ABILITY 24 experimental condition, performances improved over time, as was indicated by the
multivariate comparison, V = .40, F(2, 64) = 21.59, p < .001, ηp2 = .403. Participants got smaller accuracy scores over time (T1 MAcc = 0.339, SD = 0.158; T2 MAcc = 0.194, SD =
0.186, p < .001) and smaller variability scores (T1 MVar = 0.213, SD = 0.083; T2 MVar =
0.142, SD = 0.050, p < .001).
At time 2, therefore, performances of the conditions differed from each other, as was indicated by the multivariate comparison, V = .17, F(2, 64) = 6.71, p = .002, ηp2 = .173. Participants in the experimental condition had lower scores than participants in the control condition on accuracy (control MAcc = 0.336, SD = 0.178; experimental T2
MAcc = 0.194, SD = 0.186, p < .001), but no difference between conditions was found on
variability scores (control MVar = 0.161, SD = 0.063; experimental MVar = 0.142, SD =
0.050, p = .177).
This confirmed our second hypothesis that training of the 1.7-‐s interval caused improved performances on nontrained intervals surrounding the trained interval compared to no training; we found a generalization effect. It is against our expectations that participants in the control condition also became less variable over time.
We examined if improvements would differ across the 5 intervals of the
production task separately. We expected that performance of the trained interval would improve more than performances of the non-‐trained intervals and that performance of the intervals close to the trained interval (i.e. 1.5 and 1.9 s) would improve more than performances of the intervals further away (i.e. 1.3 and 2.1 s). This was analyzed with a 2 x 2 x 5 MANOVA with condition as between-‐subjects factor (control, experimental), time as within-‐subjects factor (T1, T2), and type of interval as within-‐subjects factor (1.3, 1.5, 1.7, 1.9, 2.1 s).
IMPROVING TIMING ABILITY 25 The improvement did not vary across the different intervals when comparing the experimental with the control condition, as was indicated by the interaction between time, interval and condition, V = .15, F(8, 58) = 1.24, p = .294, ηp2 = .146. This was against our expectations and it is remarkable that for the experimental condition, performance of the trained interval is not better than performance of the non-‐trained intervals. It is an indication that training causes similar improvements on the trained as the
nontrained intervals. Transfer Effect
We examined if the training also had an effect on performance of the prediction-‐ motion (PM) task. The mean of the 5 intervals of this task was compared over time and between conditions. This was done with a 2 x 2 MANOVA with time as within-‐subjects factor (T1, T2) and condition as between-‐subjects factor (experimental, control group). In Figure 5, the accuracy and variability scores are shown for the mean of the PM intervals over time with separate lines for condition. The change in performances over time of the experimental condition was not different than the change of the control condition, as was indicated by the non-‐significant interaction between time and condition, V= 0.01, F(2, 63) = .38, p = .687.
However, a main effect of time was found in an unexpected direction:
participants in both conditions showed higher accuracy scores at time 2 compared to time 1, which indicates an impairment (T1 MAcc = 0.221, SD = 0.098; T2 MAcc = 0.267, SD
= 0.137). The main effect of variability followed the expected direction of smaller variability scores over time (T1 MVar = 0.182, SD = 0.049; T2 MVar = 0.168, SD = 0.051).