Downloads & Free Reading Options - Results

This paper focuses on investigating whether analogy learners outperform explicit learners on motor performance in complex situations (e.g., stress and dual-task situations). Learners learn motor skills through different learning styles, such as analogy and explicit learning. Learners in the practice phase, where both learning approaches promote motor performance, do not differ significantly in motor performance. Afterwards, researchers will test the motor performance stability in the testing phase by different methods, such as inducing stress or completing a second task. Through previous studies, it was found that some researchers believe that analogy learning has better motor performance than explicit learning in complex situations; however, other researchers believe that there is no difference between analogy and explicit learning in motor performance in complex situations. Therefore, this study focuses on examining whether analogy learning outperforms explicit learning on motor performance in complex situations through a systematic review and meta-analysis.

The improvement of athletic performance can be achieved in several ways and performance does not necessarily mean faster or stronger but coordinated and controlled. Within this study, performance enhancement will be discussed in terms of injury prevention, motor control, learning, and development, and feedback techniques to augment skill learning for improved neuromuscular function. There is a substantial body of research put forward dedicated for musculoskeletal training, thus continuously increasing the research gap for cognitive learning and performance during isolated single leg movements. The relatively new tool, QASLS, is frequently used as an outcome measure for patient performance to monitor either improvements or regressions, yet the potential for this tool to be used as a training parameter for cognitive learning has not been utilised. This research proposes the integration of cognition and skill learning for single leg loading tasks (monitored through QASLS) and the use of external visual feedback due to the role in enhancing motor control performance. This approach seeks to bridge the gap in cognition and skill learning by measuring intrasession motor control performance and the influence of visual stimuli. This also has the potential to provide an alternative evidence-based approach to provide practitioners with greater understanding of cognitive learning in functional movement training. Potentially leading to a more effective training regimen and improved patient outcomes.

The improvement of athletic performance can be achieved in several ways and performance does not necessarily mean faster or stronger but coordinated and controlled. Within this study, performance enhancement will be discussed in terms of injury prevention, motor control, learning, and development, and feedback techniques to augment skill learning for improved neuromuscular function. There is a substantial body of research put forward dedicated for musculoskeletal training, thus continuously increasing the research gap for cognitive learning and performance during isolated single leg movements. The relatively new tool, QASLS, is frequently used as an outcome measure for patient performance to monitor either improvements or regressions, yet the potential for this tool to be used as a training parameter for cognitive learning has not been utilised. This research proposes the integration of cognition and skill learning for single leg loading tasks (monitored through QASLS) and the use of external visual feedback due to the role in enhancing motor control performance. This approach seeks to bridge the gap in cognition and skill learning by measuring intrasession motor control performance and the influence of visual stimuli. This also has the potential to provide an alternative evidence-based approach to provide practitioners with greater understanding of cognitive learning in functional movement training. Potentially leading to a more effective training regimen and improved patient outcomes.

The improvement of athletic performance can be achieved in several ways and performance does not necessarily mean faster or stronger but coordinated and controlled. Within this study, performance enhancement will be discussed in terms of injury prevention, motor control, learning, and development, and feedback techniques to augment skill learning for improved neuromuscular function. There is a substantial body of research put forward dedicated for musculoskeletal training, thus continuously increasing the research gap for cognitive learning and performance during isolated single leg movements. The relatively new tool, QASLS, is frequently used as an outcome measure for patient performance to monitor either improvements or regressions, yet the potential for this tool to be used as a training parameter for cognitive learning has not been utilised. This research proposes the integration of cognition and skill learning for single leg loading tasks (monitored through QASLS) and the use of external visual feedback due to the role in enhancing motor control performance. This approach seeks to bridge the gap in cognition and skill learning by measuring intrasession motor control performance and the influence of visual stimuli. This also has the potential to provide an alternative evidence-based approach to provide practitioners with greater understanding of cognitive learning in functional movement training. Potentially leading to a more effective training regimen and improved patient outcomes.

This review aims to synthesize the evidence on the effects of errorless motor learning on the performance of movement tasks among children, young adults, and older adults.

This review aims to synthesize the evidence on the effects of errorless motor learning on the performance of movement tasks among children, young adults, and older adults.

This review aims to synthesize the evidence on the effects of errorless motor learning on the performance of movement tasks among children, young adults, and older adults.

Is motor skill learning modifiable? Previous studies have attempted to use transcranial direct current stimulation (tDCS) to enhance learning, but with mixed results (Buch et al., 2017). The overwhelming majority of these studies targeted the motor cortex or the cerebellum. This project, however, innovates in testing whether anodal tDCS applied to the right parietal lobe (P4) will modulate visuospatial ability and motor skill acquisition. This aim is stemmed from our previous findings that the extent of motor skill learning is related to visuospatial ability such that better visuospatial scores correlate with more retention (Lingo VanGilder, Hengge, Duff, & Schaefer, 2018; Lingo VanGilder, Lohse, Duff, Wang, & Schaefer, 2021; Wang, Infurna, & Schaefer, 2019), as well as neuroimaging findings that common frontoparietal neural structure underlie both motor learning and visuospatial processes (Brandes-Aitken et al., 2019; Steele, Scholz, Douaud, Johansen-Berg, & Penhune, 2012). Right parietal lobe is selected as the stimulation cite because neuropsychological findings have shown that many visuospatial processes are specialized to the right parietal cortex (Corbetta, Kincade, Ollinger, McAvoy, & Shulman, 2000; Foxe, McCourt, & Javitt, 2003). Mixed-results in tDCS research could also be a result of high response variability to tDCS (Vannorsdall et al., 2016). However, explanations to such variabilities have largely ignored differences in the strength of placebo effect experienced by individuals. The expectation of (and susceptibility to) perceptions of tDCS could induce placebo effects comparable to true treatment effects (Moseley et al., 2002). New evidence suggests that expectations can alter treatment outcomes of tDCS (Rabipour, Wu, Davidson, & Iacoboni, 2018; Ray et al., 2019). Moreover, motor performance has been suggested to be manipulated through expectations and verbal suggestions alone (Fiorio, 2018). These rationales prompted us to add a no-tDCS control group to the study design, to account for any confounding placebo effects (L. Colloca & Barsky, 2020). Thus, the primary objective of this project is to test whether anodal tDCS applied to the right parietal lobe (P4 cite, International 10-20 system) will modulate visuospatial ability and motor skill acquisition, while accounting for potential confounds from placebo effects. The secondary objective is to examine how tDCS expectations associated with parietal tDCS affect motor learning and visuospatial performance.

Is motor skill learning modifiable? Previous studies have attempted to use transcranial direct current stimulation (tDCS) to enhance learning, but with mixed results (Buch et al., 2017). The overwhelming majority of these studies targeted the motor cortex or the cerebellum. This project, however, innovates in testing whether anodal tDCS applied to the right parietal lobe (P4) will modulate visuospatial ability and motor skill acquisition. This aim is stemmed from our previous findings that the extent of motor skill learning is related to visuospatial ability such that better visuospatial scores correlate with more retention (Lingo VanGilder, Hengge, Duff, & Schaefer, 2018; Lingo VanGilder, Lohse, Duff, Wang, & Schaefer, 2021; Wang, Infurna, & Schaefer, 2019), as well as neuroimaging findings that common frontoparietal neural structure underlie both motor learning and visuospatial processes (Brandes-Aitken et al., 2019; Steele, Scholz, Douaud, Johansen-Berg, & Penhune, 2012). Right parietal lobe is selected as the stimulation cite because neuropsychological findings have shown that many visuospatial processes are specialized to the right parietal cortex (Corbetta, Kincade, Ollinger, McAvoy, & Shulman, 2000; Foxe, McCourt, & Javitt, 2003). Mixed-results in tDCS research could also be a result of high response variability to tDCS (Vannorsdall et al., 2016). However, explanations to such variabilities have largely ignored differences in the strength of placebo effect experienced by individuals. The expectation of (and susceptibility to) perceptions of tDCS could induce placebo effects comparable to true treatment effects (Moseley et al., 2002). New evidence suggests that expectations can alter treatment outcomes of tDCS (Rabipour, Wu, Davidson, & Iacoboni, 2018; Ray et al., 2019). Moreover, motor performance has been suggested to be manipulated through expectations and verbal suggestions alone (Fiorio, 2018). These rationales prompted us to add a no-tDCS control group to the study design, to account for any confounding placebo effects (L. Colloca & Barsky, 2020). Thus, the primary objective of this project is to test whether anodal tDCS applied to the right parietal lobe (P4 cite, International 10-20 system) will modulate visuospatial ability and motor skill acquisition, while accounting for potential confounds from placebo effects. The secondary objective is to examine how tDCS expectations associated with parietal tDCS affect motor learning and visuospatial performance.

1. Working Title The effect of reducing terminal feedback frequency on the learning and performance of motor skills. 2. Anticipated Start Date & Completion Date Start: August, 2020 End: January, 2021 3. Stage of Review at Completion of Pre-Registration Preliminary reading (COMPLETE) Preliminary scoping & pilot search for studies (COMPLETE) Study Inclusion/Exclusion Criteria and Design (COMPLETE) Systematic Search & Screening (in progress) Data Extraction (not yet started) Quality Assessment of Data (not yet started) Analysis of Data (not yet started) Project Write-Up (not yet started) 4. Research Objectives The motivation for this meta-analysis was to follow up on a seminal review conducted by Kantak and Winstein (2012) exploring the so-called learning-performance distinction in motor learning. In their review, Kantak and Winstein investigated the effect of feedback and practice schedule manipulations on immediate and delayed retention tests and concluded that effects often differed between tests. The present meta-analysis will endeavour to build on this research by accomplishing the objectives described below. This analysis will focus on the effect of a reduced terminal feedback frequency in order to plausibly estimate the true average effect of a specific type of intervention, in contrast to the original study which examined diverse intervention types in a qualitative manner. Feedback frequency was chosen for two reasons: Statistical power and theoretical basis. Studies examining reduced feedback schedules were the most prevalent sub-type of experiment analyzed in the K & W review. Further, the guidance hypothesis (Salmoni, Schmidt, & Walter, 1984) provides a theoretical basis for expecting the effect of a reduced feedback schedule to differ as a function of when retention is measured, and thus for the hypothetical ‘learning-performance distinction” proposed by K & W. The present meta-analysis will investigate the effect of reducing terminal feedback frequency at immediate retention and delayed retention, and will investigate whether this effect significantly changes as a function of test time. Thus, the learning-performance distinction hypothesis will be investigated meta-analytically within the reduced feedback frequency literature. Further, this meta-analysis will re-examine the experiments included in the original K & W review and explore the extent that publication bias may have influenced the results of that study. Primary Objectives: 1) Estimate the effect of providing a reduced frequency of terminal feedback on delayed retention of motor skills in a healthy population. 2) Estimate the same effect on the immediate retention of motor skills. 3) Estimate the same effect for the acquisition period. 4) Investigate whether the effect of reducing terminal feedback frequency changes from immediate to delayed retention. 5) Investigate whether the effect changes from acquisition to delayed retention. 6) Investigate the influence of publication bias on the primary meta-analysis. 7) Investigate the influence of publication bias on the results of Kantak & Winstein (2012). Hypotheses to be tested: 1) Based on the guidance hypothesis (Salmoni, Schmidt, & Walter, 1984): A reduced frequency of feedback during acquisition will result in superior performance on a delayed 24-hour retention test. 2) Based on the guidance hypothesis: A reduced frequency of feedback during acquisition will result in superior performance on an immediate, no feedback retention test. 3) Based on the guidance hypothesis: A 100% frequency of feedback will result in superior performance during acquisition. 4) 4) Based on the guidance hypothesis: The effect of feedback frequency will change from acquisition to delayed retention, such that 100% feedback is more effective for acquisition performance but less effective for delayed retention performance 5) Based on the Kantak & Winstein motor memory paradigm: The effect of feedback frequency will change from immediate retention to delayed retention, such that the benefit of reduced feedback frequency will increase from immediate to delayed retention. 6) Based on our assessment of the motor learning literature: There will be evidence of significant selection effects around p = .025 (one-tailed), such that studies reporting statistically significant results will be overrepresented in the sample. 7) Based on our assessment of the motor learning literature: There will be evidence of significant selection effects around p = .025 (one-tailed) among the studies included in the Kantak & Winstein review, such that studies reporting statistically significant results at delayed retention will be overrepresented in the sample. Secondary Objectives: Test the following potential moderators of the effect of reduced terminal feedback frequency on motor learning: Age group (Children, Adult, Older adult), Skill level (novice, experienced, expert), Task classification (based on Gentile’s 2 X 2 framework), Number of acquisition trials, Number of acquisition days, Frequency of terminal feedback, Publication status, Bandwidth provisioning, Faded feedback schedule. Hypotheses to be tested: 1) Based on the Challenge-Point Framework (Guadagnoli & Lee, 2001): Children and older adults will perform more effectively on delayed retention tests after having practiced with 100% feedback during acquisition, while younger adults will perform more effectively after having received a reduced frequency of feedback. 2) Based on the Challenge-Point Framework: Higher skill performers will benefit from less feedback in acquisition than lower skilled learners when performance is measured on a delayed retention test. 3) Based on our assessment of the motor learning literature: Published articles will report significantly larger benefits of reduced feedback frequency when assessed at delayed retention. 4) Based on research comparing bandwidth feedback protocols to yoked groups: Providing feedback according to a bandwidth will have a larger benefit for delayed retention performance than reduced feedback frequency. 5) Based on the guidance hypothesis: A faded schedule of feedback during acquisition will be more effective than a static reduced schedule of feedback for delayed retention performance. Tertiary Objective: Examine the sensitivity of the primary analyses to model specification and other analysis decisions. 5. Inclusion criteria Experiments will be included based on the following inclusion criteria: 1. Published article or masters/doctorate thesis. 2. Experiment that employed random assignment to experimental conditions. 3. Included healthy, non-clinical participants. 4. Included immediate OR delayed retention test. 5. Included a feedback frequency manipulation and included a reference group receiving feedback after 100% of acquisition trials. 6. Written in English. 7. Studied a motor skill. 8. Included an objective measure of performance 6. Search Strategy Data-bases: PubMed, PsycInfo. Search query: “feedback” AND “motor learning” PubMed and PsycInfo databases were queried on August 4, 2020. Results from both searches were imported into Covidence for further screening. Covidence automatically identifies and removes duplicate entries and tracks these data on a PRIMSA-style flow chart. A total of 1990 studies were identified, of which 433 were duplicates. The resulting 1557 studies were screened based on title and abstract. Two researchers screened each article independently and decided if the article was potentially relevant to the present meta-analysis or not. Of the 1557 studies, there were 62 conflicts in which the researchers reached different decisions. Each conflict was resolved by a third researcher who was not involved in the first screening decision. Following title and abstract screening, 75 papers were identified for full text review. Again, each article will be reviewed by two researchers independently and conflicts in inclusion/exclusion decisions will be resolved by a third researcher. Additional articles will be sought by performing forward and backward reference searching. The reference sections of each included article will be searched for possible articles of interest. Further, each article will be found on Google Scholar and the “cited by” link will be searched for other potentially relevant articles. The final search strategy employed will be a targeted author search on Google Scholar. Any author with two or more included articles will be searched. 7. Potential Moderators a) Age group: Adult, children (under 18-years old), older adult (over 49-years-old) b) Skill level: Novice, experienced, expert (based on description in article). c) Task classification (based on Gentile’s 2 X 2 framework): Regulatory conditions (in motion, stable), Intertrial variability (with, without). d) Number of acquisition trials e) Number of acquisition days f) Frequency of terminal feedback in acquisition as a percentage of total trials. g) Publication status: published, unpublished. h) Bandwidth provision (yes, no) i) Faded feedback schedule (yes, no) 8. Primary Outcomes Measured (DVs) – Required. Objectively measured motor performance outcome on a delayed 24-hour retention test or immediate retention test. Only one outcome per time point per experiment will be included in the meta-analysis. The prioritization of motor performance outcomes for inclusion is based on the content of the feedback provided. If feedback content maps directly onto an outcome measure, then that measure shall be chosen as the outcome to be included. If feedback does not directly map onto an outcome measure, then prioritization will be as follows: 1) Absolute error, 2) RMSE, 3) Absolute constant error, 4) Absolute timing error, 5) Relative timing error, 6) Variable error, 7), Movement time, 8) Movement form – expert raters, 9) Otherwise unspecified objective performance measure reported first in research report. 9. Data Extraction The data of interest are the standardized mean differences (Hedges’ g) between 100% feedback and reduced feedback frequency groups at a given time-point. To facilitate reproduction, effect sizes will be calculated based on the following priority list: group means and standard deviations, means and standard errors, F values and degrees of freedom, p-values and degrees of freedom. For experiments that included prognostic covariates in the analysis, F-values, degrees of freedom, number of covariates, and correlation between covariates and dependant variables will be used. If the article containing a given experiment does not provide adequate data, an email will be sent to the corresponding author requesting the necessary data and a two-week waiting period will be applied. If data cannot be produced to accurately calculate the effect size for a given experiment that experiment will be left out of further analysis. If full data are produced by the authors and an analysis of covariance can be conducted including pre-test performance as a prognostic covariate, but the original experiment did not perform and analysis of covariance, then both unadjusted and adjusted estimates will be calculated. The unadjusted estimate will be included in the primary meta-analysis and while the adjusted estimates will be included in a subsequent sensitivity analysis. If experiments include multiple groups with reduced feedback frequency, then the group receiving the least frequent (non-zero) feedback will be selected for inclusion in the primary analysis. If reduced frequency groups differ with respect to fading of schedule, the group with the least frequent feedback schedule that is consistent (i.e., not faded) will be included in the main analysis. To be included in the primary meta-analyses, the comparison group cannot receive feedback according to a bandwidth. Additional reduced frequency groups will be included in the moderator analyses of feedback frequency, faded scheduling, and bandwidth scheduling. If experiments include multiple 100% feedback frequency groups, then those groups will be collapsed using the following formula (https://handbook-5-1.cochrane.org/chapter_7/table_7_7_a_formulae_for_combining_groups.htm). In factorial experiments with a significant interaction, the simple effects of the most relevant condition will be included. For example, an experiment may cross neural stimulation with feedback frequency – if there is a significant interaction only the data from the sham conditions will be included. If both levels of the secondary IV are equally relevant but not included as a potential moderator in the meta-analysis, then the main effect will be included even if there is a significant interaction. If the second IV in a factorial experiment is included as a moderator in this meta-analysis (for example, age of participants), then one effect from each level of the IV will be included (for example, one for child, one for adult). If a factorial experiment crosses two IVs that both manipulate feedback frequency (for example, 100% vs.50% KR about absolute timing, 100% vs. 50% KR about relative timing), then the main effect of each IV will be included for the outcome corresponding to the feedback provided (in this example, absolute and relative timing, respectively). Two researchers will independently extract the required data for all time points of interest. Conflicts will be resolved by a third researcher. The compute.es package in R will be used to calculate all effect sizes. 10. Quality Assessment Each article will be assessed using the Cochrane Risk of Bias checklist in the Covidence Review Software. 11. Screening data for influential cases The meta-analysis package ‘metafor’ in R will be used to screen and analyze the data. In order to identify outlier values and exclude them from the final random effects model, the following influence statistics will be calculated for both immediate and delayed retention data: externally standardized residuals, DFFITS values, Cook's distances, covariance ratios, DFBETAS values, the estimates of t2 squared when each study is removed in turn, the test statistics for (residual) heterogeneity when each study is removed in turn, the diagonal elements of the hat matrix, and the weights (in %) given to the observed outcomes during the model fitting. If any study is identified as extremely influential by any three of the influence metrics it will be removed from the primary analysis as an outlier. 12. Data Analysis Alpha will be set at p = .05 for all analyses. To facilitate a decision about equivalence, an effect size of g = .1 will be considered the minimum effect size of interest. A two-one-sided-test (TOST) of equivalence will be conducted on any non-significant result. Publication and other selection biases are expected in this literature. Therefore, the adjusted estimate produced by a weight-function model with a .025 (one-sided) cut-point will be considered the primary outcome in this meta-analysis for all single time point analyses. Since there are currently no bias correction methods derived to account for selection effects on a Time X Group interaction, mixed model analyses will be considered the primary outcomes for these analyses. However, interpretation of these results will depend heavily on the extent of publication bias revealed in the univariate weight-function models. If selection effects are estimated to be substantial and influential for any time point of interest, then conclusions regarding Time X Group interactions will be withheld. The following script will be applied to the dataset of effects sizes with any outliers already removed. Additional rows may need to be removed from the dataset because they are intended for moderator analyses only. This information will be added to the analysis script before final publication. Any deviations from this script will be noted in the final publication and a justification will be provided. # import datasets #terminal feedback frequency data Data # Kantak & Winstein (2012) review data KWdat # Data including bandwidth protocols BandData # load the necessary libraries library(metafor) library(meta) library(weightr) # in order to estimate the average effect of reduced feedback at delayed retention, a naïve random #effects model will be fit to only the experiments that included a delayed retention test #select only delayed retention results delay <- data[!(data$time== “3”),] delayed <- rma(ret_g, ret_v, data = delay) #select only immediate retention results imd <- data[!(data$time== “2”),] # estimate the effect at immediate retention immediate <- rma(ret_g, ret_v, data = imd) #select only acquisition results acq <- data[!(data$time== “1”),] # estimate the effect at immediate retention acquisition <- rma(ret_g, ret_v, data = acq) # test the time (immediate, delayed) X feedback frequency interaction with mixed effects model # dummy code the immediate and delayed levels of the time variable data$t.immediate <- ifelse(data$time == "Immediate", 1, 0) data$t.delay <- ifelse(data$time == "Delayed", 1, 0) Interaction <- rma.mv(ret_g, ret_v, mods = cbind(t.immediate, t.delay), random = cbind(t.immediate, t.delay) | as.factor(study),struct = "CS", data = data) # estimate the effect at immediate and delayed retention while accounting for repeated measures Estimates <- Interaction <- rma.mv(ret_g, ret_v, mods = cbind(t.immediate, t.delay) + intercept = FALSE, random = cbind(t.immediate, t.delay) | as.factor(study),struct = "CS", data = data) # test the time (acquisition, delayed) X feedback frequency interaction with mixed effects model # dummy code the acquisition level of the time variable data$t.acquisition <- ifelse(data$time == "Acquisition", 1, 0) AcqInteraction <- rma.mv(ret_g, ret_v, mods = cbind(t.acquisition, t.delay), random = cbind(t.acquisition, t.delay) | as.factor(study),struct = "CS", data = data) # estimate the effect at acqusition and delayed retention while accounting for repeated measures AcqEstimates <- Interaction <- rma.mv(ret_g, ret_v, mods = cbind(t.acquisition, t.delay) + intercept = FALSE, random = cbind(t.acquisition, t.delay) | as.factor(study),struct = "CS", data = data) # inspect results of each analysis Summary(delayed) Summary(immediate) Summary(acquisition) Summary(Interaction) Summary(Estimates) Summary(AcqInteraction) Aummary(AcqEstimates) # evaluate the influence of publication bias on the delayed retention results x <- delay$ret_g v <- delay$ret_v weightfunct(x,v) # evaluate the influence of publication bias on immediate retention results g <- imd$ret_g s <- imd$ret_v weightfunct(g,s) # evaluate the influence of publication bias on the results reviewed by Kantak & Winstein (2012) #remove immediate retention results KWdelay <- KWdat[!(KWdat$time== “1”),] # fit weight-function model k <- KWdelay$ret_g w <- KWdelay$ret_v weightfunct(k,w) #remove delayed retention results KWimd <- KWdat[!(KWdat$time== “2”),] #fit weight-function model d <- KWimd$ret_g r <- KWimd$ret_v Weightfunct(d,r) # exploratory tests of categorical moderators for delayed retention results # any significant moderator will be followed up with analysis to estimate effect at each level using following code where X = factor heading rma(ret_g,ret_v, mods = ~factor(X)-1, data= delay) # age group rma(ret_g,ret_v, mods = ~factor(age), data= delay) # skill level rma(ret_g,ret_v, mods = ~factor(skill), data= delay) # task class rma(ret_g,ret_v, mods = ~factor(task), data= delay) # publication status rma(ret_g,ret_v, mods = ~factor(pub), data= delay) # faded schedule rma(ret_g,ret_v, mods = ~factor(faded), data= delay) # bandwidth protocols Banddelay <- BandData[!(BandData$time== “3”),] rma(ret_g,ret_v, mods = ~factor(bandwidth), data= Banddelay) # exploratory tests of continuous moderators for delayed retention results # number of acquisition trials rma(ret_g,ret_v, mods = ~ trials, data= delay) # number of acquisition days rma(ret_g,ret_v, mods = ~ days, data= delay) # frequency of terminal feedback rma(ret_g,ret_v, mods = ~ frequency, data= delay) # exploratory tests of categorical moderators for immediate retention results # age group rma(ret_g,ret_v, mods = ~factor(age), data= imd) # skill level rma(ret_g,ret_v, mods = ~factor(skill), data= imd) # task class rma(ret_g,ret_v, mods = ~factor(task), data= imd) # publication status rma(ret_g,ret_v, mods = ~factor(pub), data= imd) # exploratory tests of continuous moderators for immediate retention results # number of acquisition trials rma(ret_g,ret_v, mods = ~ trials, data= imd) # number of acquisition days rma(ret_g,ret_v, mods = ~ days, data= imd) # frequency of terminal feedback rma(ret_g,ret_v, mods = ~ frequency, data= imd) 12. Sensitivity analyses Sensitivity analyses will be conducted on the Time X Feedback frequency interaction analyses as well as on the publication bias analysis. A) Standardized mean change will be calculated wherever data allow to test the Time X Feedback frequency interaction. Data required: Immediate retention mean and standard deviation, delayed retention mean and standard deviation, sample size, and correlation between immediate and delayed retention. The correlation between time-points is expected to be unknown in the majority of cases. Therefore, any known correlations will be used to estimate a range of plausible values. Sensitivity analyses will be conducted using the Standardized Mean Change with a correlation value estimated from each quartile of the overall range of correlation estimates. Therefore, there will be three runs of this analysis. The analysis will be coded and run according to the methods described here: http://www.metafor-project.org/doku.php/analyses:morris2008 B) Publication bias correction methods are sensitive to unknown underlying conditions. Therefore, on the basis of the first run of analyses, performance checks under a range of plausible conditions will be run using the following application: http://www.shinyapps.org/apps/metaExplorer/ On the basis of these performance checks, additional bias correction methods will be applied in an attempt to achieve the most possible coverage of plausible conditions. One additional need for sensitivity analysis may emerge following our quality assessment of included articles. Analyses may be re-run with articles judged to have a high risk of bias excluded in order to evaluate the extent that the overall analysis is sensitive to the inclusion of high risk studies.

1. Working Title The effect of reducing terminal feedback frequency on the learning and performance of motor skills. 2. Anticipated Start Date & Completion Date Start: August, 2020 End: January, 2021 3. Stage of Review at Completion of Pre-Registration Preliminary reading (COMPLETE) Preliminary scoping & pilot search for studies (COMPLETE) Study Inclusion/Exclusion Criteria and Design (COMPLETE) Systematic Search & Screening (in progress) Data Extraction (not yet started) Quality Assessment of Data (not yet started) Analysis of Data (not yet started) Project Write-Up (not yet started) 4. Research Objectives The motivation for this meta-analysis was to follow up on a seminal review conducted by Kantak and Winstein (2012) exploring the so-called learning-performance distinction in motor learning. In their review, Kantak and Winstein investigated the effect of feedback and practice schedule manipulations on immediate and delayed retention tests and concluded that effects often differed between tests. The present meta-analysis will endeavour to build on this research by accomplishing the objectives described below. This analysis will focus on the effect of a reduced terminal feedback frequency in order to plausibly estimate the true average effect of a specific type of intervention, in contrast to the original study which examined diverse intervention types in a qualitative manner. Feedback frequency was chosen for two reasons: Statistical power and theoretical basis. Studies examining reduced feedback schedules were the most prevalent sub-type of experiment analyzed in the K & W review. Further, the guidance hypothesis (Salmoni, Schmidt, & Walter, 1984) provides a theoretical basis for expecting the effect of a reduced feedback schedule to differ as a function of when retention is measured, and thus for the hypothetical ‘learning-performance distinction” proposed by K & W. The present meta-analysis will investigate the effect of reducing terminal feedback frequency at immediate retention and delayed retention, and will investigate whether this effect significantly changes as a function of test time. Thus, the learning-performance distinction hypothesis will be investigated meta-analytically within the reduced feedback frequency literature. Further, this meta-analysis will re-examine the experiments included in the original K & W review and explore the extent that publication bias may have influenced the results of that study. Primary Objectives: 1) Estimate the effect of providing a reduced frequency of terminal feedback on delayed retention of motor skills in a healthy population. 2) Estimate the same effect on the immediate retention of motor skills. 3) Estimate the same effect for the acquisition period. 4) Investigate whether the effect of reducing terminal feedback frequency changes from immediate to delayed retention. 5) Investigate whether the effect changes from acquisition to delayed retention. 6) Investigate the influence of publication bias on the primary meta-analysis. 7) Investigate the influence of publication bias on the results of Kantak & Winstein (2012). Hypotheses to be tested: 1) Based on the guidance hypothesis (Salmoni, Schmidt, & Walter, 1984): A reduced frequency of feedback during acquisition will result in superior performance on a delayed 24-hour retention test. 2) Based on the guidance hypothesis: A reduced frequency of feedback during acquisition will result in superior performance on an immediate, no feedback retention test. 3) Based on the guidance hypothesis: A 100% frequency of feedback will result in superior performance during acquisition. 4) 4) Based on the guidance hypothesis: The effect of feedback frequency will change from acquisition to delayed retention, such that 100% feedback is more effective for acquisition performance but less effective for delayed retention performance 5) Based on the Kantak & Winstein motor memory paradigm: The effect of feedback frequency will change from immediate retention to delayed retention, such that the benefit of reduced feedback frequency will increase from immediate to delayed retention. 6) Based on our assessment of the motor learning literature: There will be evidence of significant selection effects around p = .025 (one-tailed), such that studies reporting statistically significant results will be overrepresented in the sample. 7) Based on our assessment of the motor learning literature: There will be evidence of significant selection effects around p = .025 (one-tailed) among the studies included in the Kantak & Winstein review, such that studies reporting statistically significant results at delayed retention will be overrepresented in the sample. Secondary Objectives: Test the following potential moderators of the effect of reduced terminal feedback frequency on motor learning: Age group (Children, Adult, Older adult), Skill level (novice, experienced, expert), Task classification (based on Gentile’s 2 X 2 framework), Number of acquisition trials, Number of acquisition days, Frequency of terminal feedback, Publication status, Bandwidth provisioning, Faded feedback schedule. Hypotheses to be tested: 1) Based on the Challenge-Point Framework (Guadagnoli & Lee, 2001): Children and older adults will perform more effectively on delayed retention tests after having practiced with 100% feedback during acquisition, while younger adults will perform more effectively after having received a reduced frequency of feedback. 2) Based on the Challenge-Point Framework: Higher skill performers will benefit from less feedback in acquisition than lower skilled learners when performance is measured on a delayed retention test. 3) Based on our assessment of the motor learning literature: Published articles will report significantly larger benefits of reduced feedback frequency when assessed at delayed retention. 4) Based on research comparing bandwidth feedback protocols to yoked groups: Providing feedback according to a bandwidth will have a larger benefit for delayed retention performance than reduced feedback frequency. 5) Based on the guidance hypothesis: A faded schedule of feedback during acquisition will be more effective than a static reduced schedule of feedback for delayed retention performance. Tertiary Objective: Examine the sensitivity of the primary analyses to model specification and other analysis decisions. 5. Inclusion criteria Experiments will be included based on the following inclusion criteria: 1. Published article or masters/doctorate thesis. 2. Experiment that employed random assignment to experimental conditions. 3. Included healthy, non-clinical participants. 4. Included immediate OR delayed retention test. 5. Included a feedback frequency manipulation and included a reference group receiving feedback after 100% of acquisition trials. 6. Written in English. 7. Studied a motor skill. 8. Included an objective measure of performance 6. Search Strategy Data-bases: PubMed, PsycInfo. Search query: “feedback” AND “motor learning” PubMed and PsycInfo databases were queried on August 4, 2020. Results from both searches were imported into Covidence for further screening. Covidence automatically identifies and removes duplicate entries and tracks these data on a PRIMSA-style flow chart. A total of 1990 studies were identified, of which 433 were duplicates. The resulting 1557 studies were screened based on title and abstract. Two researchers screened each article independently and decided if the article was potentially relevant to the present meta-analysis or not. Of the 1557 studies, there were 62 conflicts in which the researchers reached different decisions. Each conflict was resolved by a third researcher who was not involved in the first screening decision. Following title and abstract screening, 75 papers were identified for full text review. Again, each article will be reviewed by two researchers independently and conflicts in inclusion/exclusion decisions will be resolved by a third researcher. Additional articles will be sought by performing forward and backward reference searching. The reference sections of each included article will be searched for possible articles of interest. Further, each article will be found on Google Scholar and the “cited by” link will be searched for other potentially relevant articles. The final search strategy employed will be a targeted author search on Google Scholar. Any author with two or more included articles will be searched. 7. Potential Moderators a) Age group: Adult, children (under 18-years old), older adult (over 49-years-old) b) Skill level: Novice, experienced, expert (based on description in article). c) Task classification (based on Gentile’s 2 X 2 framework): Regulatory conditions (in motion, stable), Intertrial variability (with, without). d) Number of acquisition trials e) Number of acquisition days f) Frequency of terminal feedback in acquisition as a percentage of total trials. g) Publication status: published, unpublished. h) Bandwidth provision (yes, no) i) Faded feedback schedule (yes, no) 8. Primary Outcomes Measured (DVs) – Required. Objectively measured motor performance outcome on a delayed 24-hour retention test or immediate retention test. Only one outcome per time point per experiment will be included in the meta-analysis. The prioritization of motor performance outcomes for inclusion is based on the content of the feedback provided. If feedback content maps directly onto an outcome measure, then that measure shall be chosen as the outcome to be included. If feedback does not directly map onto an outcome measure, then prioritization will be as follows: 1) Absolute error, 2) RMSE, 3) Absolute constant error, 4) Absolute timing error, 5) Relative timing error, 6) Variable error, 7), Movement time, 8) Movement form – expert raters, 9) Otherwise unspecified objective performance measure reported first in research report. 9. Data Extraction The data of interest are the standardized mean differences (Hedges’ g) between 100% feedback and reduced feedback frequency groups at a given time-point. To facilitate reproduction, effect sizes will be calculated based on the following priority list: group means and standard deviations, means and standard errors, F values and degrees of freedom, p-values and degrees of freedom. For experiments that included prognostic covariates in the analysis, F-values, degrees of freedom, number of covariates, and correlation between covariates and dependant variables will be used. If the article containing a given experiment does not provide adequate data, an email will be sent to the corresponding author requesting the necessary data and a two-week waiting period will be applied. If data cannot be produced to accurately calculate the effect size for a given experiment that experiment will be left out of further analysis. If full data are produced by the authors and an analysis of covariance can be conducted including pre-test performance as a prognostic covariate, but the original experiment did not perform and analysis of covariance, then both unadjusted and adjusted estimates will be calculated. The unadjusted estimate will be included in the primary meta-analysis and while the adjusted estimates will be included in a subsequent sensitivity analysis. If experiments include multiple groups with reduced feedback frequency, then the group receiving the least frequent (non-zero) feedback will be selected for inclusion in the primary analysis. If reduced frequency groups differ with respect to fading of schedule, the group with the least frequent feedback schedule that is consistent (i.e., not faded) will be included in the main analysis. To be included in the primary meta-analyses, the comparison group cannot receive feedback according to a bandwidth. Additional reduced frequency groups will be included in the moderator analyses of feedback frequency, faded scheduling, and bandwidth scheduling. If experiments include multiple 100% feedback frequency groups, then those groups will be collapsed using the following formula (https://handbook-5-1.cochrane.org/chapter_7/table_7_7_a_formulae_for_combining_groups.htm). In factorial experiments with a significant interaction, the simple effects of the most relevant condition will be included. For example, an experiment may cross neural stimulation with feedback frequency – if there is a significant interaction only the data from the sham conditions will be included. If both levels of the secondary IV are equally relevant but not included as a potential moderator in the meta-analysis, then the main effect will be included even if there is a significant interaction. If the second IV in a factorial experiment is included as a moderator in this meta-analysis (for example, age of participants), then one effect from each level of the IV will be included (for example, one for child, one for adult). If a factorial experiment crosses two IVs that both manipulate feedback frequency (for example, 100% vs.50% KR about absolute timing, 100% vs. 50% KR about relative timing), then the main effect of each IV will be included for the outcome corresponding to the feedback provided (in this example, absolute and relative timing, respectively). Two researchers will independently extract the required data for all time points of interest. Conflicts will be resolved by a third researcher. The compute.es package in R will be used to calculate all effect sizes. 10. Quality Assessment Each article will be assessed using the Cochrane Risk of Bias checklist in the Covidence Review Software. 11. Screening data for influential cases The meta-analysis package ‘metafor’ in R will be used to screen and analyze the data. In order to identify outlier values and exclude them from the final random effects model, the following influence statistics will be calculated for both immediate and delayed retention data: externally standardized residuals, DFFITS values, Cook's distances, covariance ratios, DFBETAS values, the estimates of t2 squared when each study is removed in turn, the test statistics for (residual) heterogeneity when each study is removed in turn, the diagonal elements of the hat matrix, and the weights (in %) given to the observed outcomes during the model fitting. If any study is identified as extremely influential by any three of the influence metrics it will be removed from the primary analysis as an outlier. 12. Data Analysis Alpha will be set at p = .05 for all analyses. To facilitate a decision about equivalence, an effect size of g = .1 will be considered the minimum effect size of interest. A two-one-sided-test (TOST) of equivalence will be conducted on any non-significant result. Publication and other selection biases are expected in this literature. Therefore, the adjusted estimate produced by a weight-function model with a .025 (one-sided) cut-point will be considered the primary outcome in this meta-analysis for all single time point analyses. Since there are currently no bias correction methods derived to account for selection effects on a Time X Group interaction, mixed model analyses will be considered the primary outcomes for these analyses. However, interpretation of these results will depend heavily on the extent of publication bias revealed in the univariate weight-function models. If selection effects are estimated to be substantial and influential for any time point of interest, then conclusions regarding Time X Group interactions will be withheld. The following script will be applied to the dataset of effects sizes with any outliers already removed. Additional rows may need to be removed from the dataset because they are intended for moderator analyses only. This information will be added to the analysis script before final publication. Any deviations from this script will be noted in the final publication and a justification will be provided. # import datasets #terminal feedback frequency data Data # Kantak & Winstein (2012) review data KWdat # Data including bandwidth protocols BandData # load the necessary libraries library(metafor) library(meta) library(weightr) # in order to estimate the average effect of reduced feedback at delayed retention, a naïve random #effects model will be fit to only the experiments that included a delayed retention test #select only delayed retention results delay <- data[!(data$time== “3”),] delayed <- rma(ret_g, ret_v, data = delay) #select only immediate retention results imd <- data[!(data$time== “2”),] # estimate the effect at immediate retention immediate <- rma(ret_g, ret_v, data = imd) #select only acquisition results acq <- data[!(data$time== “1”),] # estimate the effect at immediate retention acquisition <- rma(ret_g, ret_v, data = acq) # test the time (immediate, delayed) X feedback frequency interaction with mixed effects model # dummy code the immediate and delayed levels of the time variable data$t.immediate <- ifelse(data$time == "Immediate", 1, 0) data$t.delay <- ifelse(data$time == "Delayed", 1, 0) Interaction <- rma.mv(ret_g, ret_v, mods = cbind(t.immediate, t.delay), random = cbind(t.immediate, t.delay) | as.factor(study),struct = "CS", data = data) # estimate the effect at immediate and delayed retention while accounting for repeated measures Estimates <- Interaction <- rma.mv(ret_g, ret_v, mods = cbind(t.immediate, t.delay) + intercept = FALSE, random = cbind(t.immediate, t.delay) | as.factor(study),struct = "CS", data = data) # test the time (acquisition, delayed) X feedback frequency interaction with mixed effects model # dummy code the acquisition level of the time variable data$t.acquisition <- ifelse(data$time == "Acquisition", 1, 0) AcqInteraction <- rma.mv(ret_g, ret_v, mods = cbind(t.acquisition, t.delay), random = cbind(t.acquisition, t.delay) | as.factor(study),struct = "CS", data = data) # estimate the effect at acqusition and delayed retention while accounting for repeated measures AcqEstimates <- Interaction <- rma.mv(ret_g, ret_v, mods = cbind(t.acquisition, t.delay) + intercept = FALSE, random = cbind(t.acquisition, t.delay) | as.factor(study),struct = "CS", data = data) # inspect results of each analysis Summary(delayed) Summary(immediate) Summary(acquisition) Summary(Interaction) Summary(Estimates) Summary(AcqInteraction) Aummary(AcqEstimates) # evaluate the influence of publication bias on the delayed retention results x <- delay$ret_g v <- delay$ret_v weightfunct(x,v) # evaluate the influence of publication bias on immediate retention results g <- imd$ret_g s <- imd$ret_v weightfunct(g,s) # evaluate the influence of publication bias on the results reviewed by Kantak & Winstein (2012) #remove immediate retention results KWdelay <- KWdat[!(KWdat$time== “1”),] # fit weight-function model k <- KWdelay$ret_g w <- KWdelay$ret_v weightfunct(k,w) #remove delayed retention results KWimd <- KWdat[!(KWdat$time== “2”),] #fit weight-function model d <- KWimd$ret_g r <- KWimd$ret_v Weightfunct(d,r) # exploratory tests of categorical moderators for delayed retention results # any significant moderator will be followed up with analysis to estimate effect at each level using following code where X = factor heading rma(ret_g,ret_v, mods = ~factor(X)-1, data= delay) # age group rma(ret_g,ret_v, mods = ~factor(age), data= delay) # skill level rma(ret_g,ret_v, mods = ~factor(skill), data= delay) # task class rma(ret_g,ret_v, mods = ~factor(task), data= delay) # publication status rma(ret_g,ret_v, mods = ~factor(pub), data= delay) # faded schedule rma(ret_g,ret_v, mods = ~factor(faded), data= delay) # bandwidth protocols Banddelay <- BandData[!(BandData$time== “3”),] rma(ret_g,ret_v, mods = ~factor(bandwidth), data= Banddelay) # exploratory tests of continuous moderators for delayed retention results # number of acquisition trials rma(ret_g,ret_v, mods = ~ trials, data= delay) # number of acquisition days rma(ret_g,ret_v, mods = ~ days, data= delay) # frequency of terminal feedback rma(ret_g,ret_v, mods = ~ frequency, data= delay) # exploratory tests of categorical moderators for immediate retention results # age group rma(ret_g,ret_v, mods = ~factor(age), data= imd) # skill level rma(ret_g,ret_v, mods = ~factor(skill), data= imd) # task class rma(ret_g,ret_v, mods = ~factor(task), data= imd) # publication status rma(ret_g,ret_v, mods = ~factor(pub), data= imd) # exploratory tests of continuous moderators for immediate retention results # number of acquisition trials rma(ret_g,ret_v, mods = ~ trials, data= imd) # number of acquisition days rma(ret_g,ret_v, mods = ~ days, data= imd) # frequency of terminal feedback rma(ret_g,ret_v, mods = ~ frequency, data= imd) 12. Sensitivity analyses Sensitivity analyses will be conducted on the Time X Feedback frequency interaction analyses as well as on the publication bias analysis. A) Standardized mean change will be calculated wherever data allow to test the Time X Feedback frequency interaction. Data required: Immediate retention mean and standard deviation, delayed retention mean and standard deviation, sample size, and correlation between immediate and delayed retention. The correlation between time-points is expected to be unknown in the majority of cases. Therefore, any known correlations will be used to estimate a range of plausible values. Sensitivity analyses will be conducted using the Standardized Mean Change with a correlation value estimated from each quartile of the overall range of correlation estimates. Therefore, there will be three runs of this analysis. The analysis will be coded and run according to the methods described here: http://www.metafor-project.org/doku.php/analyses:morris2008 B) Publication bias correction methods are sensitive to unknown underlying conditions. Therefore, on the basis of the first run of analyses, performance checks under a range of plausible conditions will be run using the following application: http://www.shinyapps.org/apps/metaExplorer/ On the basis of these performance checks, additional bias correction methods will be applied in an attempt to achieve the most possible coverage of plausible conditions. One additional need for sensitivity analysis may emerge following our quality assessment of included articles. Analyses may be re-run with articles judged to have a high risk of bias excluded in order to evaluate the extent that the overall analysis is sensitive to the inclusion of high risk studies.

1. Working Title The effect of reducing terminal feedback frequency on the learning and performance of motor skills. 2. Anticipated Start Date & Completion Date Start: August, 2020 End: January, 2021 3. Stage of Review at Completion of Pre-Registration Preliminary reading (COMPLETE) Preliminary scoping & pilot search for studies (COMPLETE) Study Inclusion/Exclusion Criteria and Design (COMPLETE) Systematic Search & Screening (in progress) Data Extraction (not yet started) Quality Assessment of Data (not yet started) Analysis of Data (not yet started) Project Write-Up (not yet started) 4. Research Objectives The motivation for this meta-analysis was to follow up on a seminal review conducted by Kantak and Winstein (2012) exploring the so-called learning-performance distinction in motor learning. In their review, Kantak and Winstein investigated the effect of feedback and practice schedule manipulations on immediate and delayed retention tests and concluded that effects often differed between tests. The present meta-analysis will endeavour to build on this research by accomplishing the objectives described below. This analysis will focus on the effect of a reduced terminal feedback frequency in order to plausibly estimate the true average effect of a specific type of intervention, in contrast to the original study which examined diverse intervention types in a qualitative manner. Feedback frequency was chosen for two reasons: Statistical power and theoretical basis. Studies examining reduced feedback schedules were the most prevalent sub-type of experiment analyzed in the K & W review. Further, the guidance hypothesis (Salmoni, Schmidt, & Walter, 1984) provides a theoretical basis for expecting the effect of a reduced feedback schedule to differ as a function of when retention is measured, and thus for the hypothetical ‘learning-performance distinction” proposed by K & W. The present meta-analysis will investigate the effect of reducing terminal feedback frequency at immediate retention and delayed retention, and will investigate whether this effect significantly changes as a function of test time. Thus, the learning-performance distinction hypothesis will be investigated meta-analytically within the reduced feedback frequency literature. Further, this meta-analysis will re-examine the experiments included in the original K & W review and explore the extent that publication bias may have influenced the results of that study. Primary Objectives: 1) Estimate the effect of providing a reduced frequency of terminal feedback on delayed retention of motor skills in a healthy population. 2) Estimate the same effect on the immediate retention of motor skills. 3) Estimate the same effect for the acquisition period. 4) Investigate whether the effect of reducing terminal feedback frequency changes from immediate to delayed retention. 5) Investigate whether the effect changes from acquisition to delayed retention. 6) Investigate the influence of publication bias on the primary meta-analysis. 7) Investigate the influence of publication bias on the results of Kantak & Winstein (2012). Hypotheses to be tested: 1) Based on the guidance hypothesis (Salmoni, Schmidt, & Walter, 1984): A reduced frequency of feedback during acquisition will result in superior performance on a delayed 24-hour retention test. 2) Based on the guidance hypothesis: A reduced frequency of feedback during acquisition will result in superior performance on an immediate, no feedback retention test. 3) Based on the guidance hypothesis: A 100% frequency of feedback will result in superior performance during acquisition. 4) 4) Based on the guidance hypothesis: The effect of feedback frequency will change from acquisition to delayed retention, such that 100% feedback is more effective for acquisition performance but less effective for delayed retention performance 5) Based on the Kantak & Winstein motor memory paradigm: The effect of feedback frequency will change from immediate retention to delayed retention, such that the benefit of reduced feedback frequency will increase from immediate to delayed retention. 6) Based on our assessment of the motor learning literature: There will be evidence of significant selection effects around p = .025 (one-tailed), such that studies reporting statistically significant results will be overrepresented in the sample. 7) Based on our assessment of the motor learning literature: There will be evidence of significant selection effects around p = .025 (one-tailed) among the studies included in the Kantak & Winstein review, such that studies reporting statistically significant results at delayed retention will be overrepresented in the sample. Secondary Objectives: Test the following potential moderators of the effect of reduced terminal feedback frequency on motor learning: Age group (Children, Adult, Older adult), Skill level (novice, experienced, expert), Task classification (based on Gentile’s 2 X 2 framework), Number of acquisition trials, Number of acquisition days, Frequency of terminal feedback, Publication status, Bandwidth provisioning, Faded feedback schedule. Hypotheses to be tested: 1) Based on the Challenge-Point Framework (Guadagnoli & Lee, 2001): Children and older adults will perform more effectively on delayed retention tests after having practiced with 100% feedback during acquisition, while younger adults will perform more effectively after having received a reduced frequency of feedback. 2) Based on the Challenge-Point Framework: Higher skill performers will benefit from less feedback in acquisition than lower skilled learners when performance is measured on a delayed retention test. 3) Based on our assessment of the motor learning literature: Published articles will report significantly larger benefits of reduced feedback frequency when assessed at delayed retention. 4) Based on research comparing bandwidth feedback protocols to yoked groups: Providing feedback according to a bandwidth will have a larger benefit for delayed retention performance than reduced feedback frequency. 5) Based on the guidance hypothesis: A faded schedule of feedback during acquisition will be more effective than a static reduced schedule of feedback for delayed retention performance. Tertiary Objective: Examine the sensitivity of the primary analyses to model specification and other analysis decisions. 5. Inclusion criteria Experiments will be included based on the following inclusion criteria: 1. Published article or masters/doctorate thesis. 2. Experiment that employed random assignment to experimental conditions. 3. Included healthy, non-clinical participants. 4. Included immediate OR delayed retention test. 5. Included a feedback frequency manipulation and included a reference group receiving feedback after 100% of acquisition trials. 6. Written in English. 7. Studied a motor skill. 8. Included an objective measure of performance 6. Search Strategy Data-bases: PubMed, PsycInfo. Search query: “feedback” AND “motor learning” PubMed and PsycInfo databases were queried on August 4, 2020. Results from both searches were imported into Covidence for further screening. Covidence automatically identifies and removes duplicate entries and tracks these data on a PRIMSA-style flow chart. A total of 1990 studies were identified, of which 433 were duplicates. The resulting 1557 studies were screened based on title and abstract. Two researchers screened each article independently and decided if the article was potentially relevant to the present meta-analysis or not. Of the 1557 studies, there were 62 conflicts in which the researchers reached different decisions. Each conflict was resolved by a third researcher who was not involved in the first screening decision. Following title and abstract screening, 75 papers were identified for full text review. Again, each article will be reviewed by two researchers independently and conflicts in inclusion/exclusion decisions will be resolved by a third researcher. Additional articles will be sought by performing forward and backward reference searching. The reference sections of each included article will be searched for possible articles of interest. Further, each article will be found on Google Scholar and the “cited by” link will be searched for other potentially relevant articles. The final search strategy employed will be a targeted author search on Google Scholar. Any author with two or more included articles will be searched. 7. Potential Moderators a) Age group: Adult, children (under 18-years old), older adult (over 49-years-old) b) Skill level: Novice, experienced, expert (based on description in article). c) Task classification (based on Gentile’s 2 X 2 framework): Regulatory conditions (in motion, stable), Intertrial variability (with, without). d) Number of acquisition trials e) Number of acquisition days f) Frequency of terminal feedback in acquisition as a percentage of total trials. g) Publication status: published, unpublished. h) Bandwidth provision (yes, no) i) Faded feedback schedule (yes, no) 8. Primary Outcomes Measured (DVs) – Required. Objectively measured motor performance outcome on a delayed 24-hour retention test or immediate retention test. Only one outcome per time point per experiment will be included in the meta-analysis. The prioritization of motor performance outcomes for inclusion is based on the content of the feedback provided. If feedback content maps directly onto an outcome measure, then that measure shall be chosen as the outcome to be included. If feedback does not directly map onto an outcome measure, then prioritization will be as follows: 1) Absolute error, 2) RMSE, 3) Absolute constant error, 4) Absolute timing error, 5) Relative timing error, 6) Variable error, 7), Movement time, 8) Movement form – expert raters, 9) Otherwise unspecified objective performance measure reported first in research report. 9. Data Extraction The data of interest are the standardized mean differences (Hedges’ g) between 100% feedback and reduced feedback frequency groups at a given time-point. To facilitate reproduction, effect sizes will be calculated based on the following priority list: group means and standard deviations, means and standard errors, F values and degrees of freedom, p-values and degrees of freedom. For experiments that included prognostic covariates in the analysis, F-values, degrees of freedom, number of covariates, and correlation between covariates and dependant variables will be used. If the article containing a given experiment does not provide adequate data, an email will be sent to the corresponding author requesting the necessary data and a two-week waiting period will be applied. If data cannot be produced to accurately calculate the effect size for a given experiment that experiment will be left out of further analysis. If full data are produced by the authors and an analysis of covariance can be conducted including pre-test performance as a prognostic covariate, but the original experiment did not perform and analysis of covariance, then both unadjusted and adjusted estimates will be calculated. The unadjusted estimate will be included in the primary meta-analysis and while the adjusted estimates will be included in a subsequent sensitivity analysis. If experiments include multiple groups with reduced feedback frequency, then the group receiving the least frequent (non-zero) feedback will be selected for inclusion in the primary analysis. If reduced frequency groups differ with respect to fading of schedule, the group with the least frequent feedback schedule that is consistent (i.e., not faded) will be included in the main analysis. To be included in the primary meta-analyses, the comparison group cannot receive feedback according to a bandwidth. Additional reduced frequency groups will be included in the moderator analyses of feedback frequency, faded scheduling, and bandwidth scheduling. If experiments include multiple 100% feedback frequency groups, then those groups will be collapsed using the following formula (https://handbook-5-1.cochrane.org/chapter_7/table_7_7_a_formulae_for_combining_groups.htm). In factorial experiments with a significant interaction, the simple effects of the most relevant condition will be included. For example, an experiment may cross neural stimulation with feedback frequency – if there is a significant interaction only the data from the sham conditions will be included. If both levels of the secondary IV are equally relevant but not included as a potential moderator in the meta-analysis, then the main effect will be included even if there is a significant interaction. If the second IV in a factorial experiment is included as a moderator in this meta-analysis (for example, age of participants), then one effect from each level of the IV will be included (for example, one for child, one for adult). If a factorial experiment crosses two IVs that both manipulate feedback frequency (for example, 100% vs.50% KR about absolute timing, 100% vs. 50% KR about relative timing), then the main effect of each IV will be included for the outcome corresponding to the feedback provided (in this example, absolute and relative timing, respectively). Two researchers will independently extract the required data for all time points of interest. Conflicts will be resolved by a third researcher. The compute.es package in R will be used to calculate all effect sizes. 10. Quality Assessment Each article will be assessed using the Cochrane Risk of Bias checklist in the Covidence Review Software. 11. Screening data for influential cases The meta-analysis package ‘metafor’ in R will be used to screen and analyze the data. In order to identify outlier values and exclude them from the final random effects model, the following influence statistics will be calculated for both immediate and delayed retention data: externally standardized residuals, DFFITS values, Cook's distances, covariance ratios, DFBETAS values, the estimates of t2 squared when each study is removed in turn, the test statistics for (residual) heterogeneity when each study is removed in turn, the diagonal elements of the hat matrix, and the weights (in %) given to the observed outcomes during the model fitting. If any study is identified as extremely influential by any three of the influence metrics it will be removed from the primary analysis as an outlier. 12. Data Analysis Alpha will be set at p = .05 for all analyses. To facilitate a decision about equivalence, an effect size of g = .1 will be considered the minimum effect size of interest. A two-one-sided-test (TOST) of equivalence will be conducted on any non-significant result. Publication and other selection biases are expected in this literature. Therefore, the adjusted estimate produced by a weight-function model with a .025 (one-sided) cut-point will be considered the primary outcome in this meta-analysis for all single time point analyses. Since there are currently no bias correction methods derived to account for selection effects on a Time X Group interaction, mixed model analyses will be considered the primary outcomes for these analyses. However, interpretation of these results will depend heavily on the extent of publication bias revealed in the univariate weight-function models. If selection effects are estimated to be substantial and influential for any time point of interest, then conclusions regarding Time X Group interactions will be withheld. The following script will be applied to the dataset of effects sizes with any outliers already removed. Additional rows may need to be removed from the dataset because they are intended for moderator analyses only. This information will be added to the analysis script before final publication. Any deviations from this script will be noted in the final publication and a justification will be provided. # import datasets #terminal feedback frequency data Data # Kantak & Winstein (2012) review data KWdat # Data including bandwidth protocols BandData # load the necessary libraries library(metafor) library(meta) library(weightr) # in order to estimate the average effect of reduced feedback at delayed retention, a naïve random #effects model will be fit to only the experiments that included a delayed retention test #select only delayed retention results delay <- data[!(data$time== “3”),] delayed <- rma(ret_g, ret_v, data = delay) #select only immediate retention results imd <- data[!(data$time== “2”),] # estimate the effect at immediate retention immediate <- rma(ret_g, ret_v, data = imd) #select only acquisition results acq <- data[!(data$time== “1”),] # estimate the effect at immediate retention acquisition <- rma(ret_g, ret_v, data = acq) # test the time (immediate, delayed) X feedback frequency interaction with mixed effects model # dummy code the immediate and delayed levels of the time variable data$t.immediate <- ifelse(data$time == "Immediate", 1, 0) data$t.delay <- ifelse(data$time == "Delayed", 1, 0) Interaction <- rma.mv(ret_g, ret_v, mods = cbind(t.immediate, t.delay), random = cbind(t.immediate, t.delay) | as.factor(study),struct = "CS", data = data) # estimate the effect at immediate and delayed retention while accounting for repeated measures Estimates <- Interaction <- rma.mv(ret_g, ret_v, mods = cbind(t.immediate, t.delay) + intercept = FALSE, random = cbind(t.immediate, t.delay) | as.factor(study),struct = "CS", data = data) # test the time (acquisition, delayed) X feedback frequency interaction with mixed effects model # dummy code the acquisition level of the time variable data$t.acquisition <- ifelse(data$time == "Acquisition", 1, 0) AcqInteraction <- rma.mv(ret_g, ret_v, mods = cbind(t.acquisition, t.delay), random = cbind(t.acquisition, t.delay) | as.factor(study),struct = "CS", data = data) # estimate the effect at acqusition and delayed retention while accounting for repeated measures AcqEstimates <- Interaction <- rma.mv(ret_g, ret_v, mods = cbind(t.acquisition, t.delay) + intercept = FALSE, random = cbind(t.acquisition, t.delay) | as.factor(study),struct = "CS", data = data) # inspect results of each analysis Summary(delayed) Summary(immediate) Summary(acquisition) Summary(Interaction) Summary(Estimates) Summary(AcqInteraction) Aummary(AcqEstimates) # evaluate the influence of publication bias on the delayed retention results x <- delay$ret_g v <- delay$ret_v weightfunct(x,v) # evaluate the influence of publication bias on immediate retention results g <- imd$ret_g s <- imd$ret_v weightfunct(g,s) # evaluate the influence of publication bias on the results reviewed by Kantak & Winstein (2012) #remove immediate retention results KWdelay <- KWdat[!(KWdat$time== “1”),] # fit weight-function model k <- KWdelay$ret_g w <- KWdelay$ret_v weightfunct(k,w) #remove delayed retention results KWimd <- KWdat[!(KWdat$time== “2”),] #fit weight-function model d <- KWimd$ret_g r <- KWimd$ret_v Weightfunct(d,r) # exploratory tests of categorical moderators for delayed retention results # any significant moderator will be followed up with analysis to estimate effect at each level using following code where X = factor heading rma(ret_g,ret_v, mods = ~factor(X)-1, data= delay) # age group rma(ret_g,ret_v, mods = ~factor(age), data= delay) # skill level rma(ret_g,ret_v, mods = ~factor(skill), data= delay) # task class rma(ret_g,ret_v, mods = ~factor(task), data= delay) # publication status rma(ret_g,ret_v, mods = ~factor(pub), data= delay) # faded schedule rma(ret_g,ret_v, mods = ~factor(faded), data= delay) # bandwidth protocols Banddelay <- BandData[!(BandData$time== “3”),] rma(ret_g,ret_v, mods = ~factor(bandwidth), data= Banddelay) # exploratory tests of continuous moderators for delayed retention results # number of acquisition trials rma(ret_g,ret_v, mods = ~ trials, data= delay) # number of acquisition days rma(ret_g,ret_v, mods = ~ days, data= delay) # frequency of terminal feedback rma(ret_g,ret_v, mods = ~ frequency, data= delay) # exploratory tests of categorical moderators for immediate retention results # age group rma(ret_g,ret_v, mods = ~factor(age), data= imd) # skill level rma(ret_g,ret_v, mods = ~factor(skill), data= imd) # task class rma(ret_g,ret_v, mods = ~factor(task), data= imd) # publication status rma(ret_g,ret_v, mods = ~factor(pub), data= imd) # exploratory tests of continuous moderators for immediate retention results # number of acquisition trials rma(ret_g,ret_v, mods = ~ trials, data= imd) # number of acquisition days rma(ret_g,ret_v, mods = ~ days, data= imd) # frequency of terminal feedback rma(ret_g,ret_v, mods = ~ frequency, data= imd) 12. Sensitivity analyses Sensitivity analyses will be conducted on the Time X Feedback frequency interaction analyses as well as on the publication bias analysis. A) Standardized mean change will be calculated wherever data allow to test the Time X Feedback frequency interaction. Data required: Immediate retention mean and standard deviation, delayed retention mean and standard deviation, sample size, and correlation between immediate and delayed retention. The correlation between time-points is expected to be unknown in the majority of cases. Therefore, any known correlations will be used to estimate a range of plausible values. Sensitivity analyses will be conducted using the Standardized Mean Change with a correlation value estimated from each quartile of the overall range of correlation estimates. Therefore, there will be three runs of this analysis. The analysis will be coded and run according to the methods described here: http://www.metafor-project.org/doku.php/analyses:morris2008 B) Publication bias correction methods are sensitive to unknown underlying conditions. Therefore, on the basis of the first run of analyses, performance checks under a range of plausible conditions will be run using the following application: http://www.shinyapps.org/apps/metaExplorer/ On the basis of these performance checks, additional bias correction methods will be applied in an attempt to achieve the most possible coverage of plausible conditions. One additional need for sensitivity analysis may emerge following our quality assessment of included articles. Analyses may be re-run with articles judged to have a high risk of bias excluded in order to evaluate the extent that the overall analysis is sensitive to the inclusion of high risk studies.

1. Working Title The effect of reducing terminal feedback frequency on the learning and performance of motor skills. 2. Anticipated Start Date & Completion Date Start: August, 2020 End: January, 2021 3. Stage of Review at Completion of Pre-Registration Preliminary reading (COMPLETE) Preliminary scoping & pilot search for studies (COMPLETE) Study Inclusion/Exclusion Criteria and Design (COMPLETE) Systematic Search & Screening (in progress) Data Extraction (not yet started) Quality Assessment of Data (not yet started) Analysis of Data (not yet started) Project Write-Up (not yet started) 4. Research Objectives The motivation for this meta-analysis was to follow up on a seminal review conducted by Kantak and Winstein (2012) exploring the so-called learning-performance distinction in motor learning. In their review, Kantak and Winstein investigated the effect of feedback and practice schedule manipulations on immediate and delayed retention tests and concluded that effects often differed between tests. The present meta-analysis will endeavour to build on this research by accomplishing the objectives described below. This analysis will focus on the effect of a reduced terminal feedback frequency in order to plausibly estimate the true average effect of a specific type of intervention, in contrast to the original study which examined diverse intervention types in a qualitative manner. Feedback frequency was chosen for two reasons: Statistical power and theoretical basis. Studies examining reduced feedback schedules were the most prevalent sub-type of experiment analyzed in the K & W review. Further, the guidance hypothesis (Salmoni, Schmidt, & Walter, 1984) provides a theoretical basis for expecting the effect of a reduced feedback schedule to differ as a function of when retention is measured, and thus for the hypothetical ‘learning-performance distinction” proposed by K & W. The present meta-analysis will investigate the effect of reducing terminal feedback frequency at immediate retention and delayed retention, and will investigate whether this effect significantly changes as a function of test time. Thus, the learning-performance distinction hypothesis will be investigated meta-analytically within the reduced feedback frequency literature. Further, this meta-analysis will re-examine the experiments included in the original K & W review and explore the extent that publication bias may have influenced the results of that study. Primary Objectives: 1) Estimate the effect of providing a reduced frequency of terminal feedback on delayed retention of motor skills in a healthy population. 2) Estimate the same effect on the immediate retention of motor skills. 3) Estimate the same effect for the acquisition period. 4) Investigate whether the effect of reducing terminal feedback frequency changes from immediate to delayed retention. 5) Investigate whether the effect changes from acquisition to delayed retention. 6) Investigate the influence of publication bias on the primary meta-analysis. 7) Investigate the influence of publication bias on the results of Kantak & Winstein (2012). Hypotheses to be tested: 1) Based on the guidance hypothesis (Salmoni, Schmidt, & Walter, 1984): A reduced frequency of feedback during acquisition will result in superior performance on a delayed 24-hour retention test. 2) Based on the guidance hypothesis: A reduced frequency of feedback during acquisition will result in superior performance on an immediate, no feedback retention test. 3) Based on the guidance hypothesis: A 100% frequency of feedback will result in superior performance during acquisition. 4) 4) Based on the guidance hypothesis: The effect of feedback frequency will change from acquisition to delayed retention, such that 100% feedback is more effective for acquisition performance but less effective for delayed retention performance 5) Based on the Kantak & Winstein motor memory paradigm: The effect of feedback frequency will change from immediate retention to delayed retention, such that the benefit of reduced feedback frequency will increase from immediate to delayed retention. 6) Based on our assessment of the motor learning literature: There will be evidence of significant selection effects around p = .025 (one-tailed), such that studies reporting statistically significant results will be overrepresented in the sample. 7) Based on our assessment of the motor learning literature: There will be evidence of significant selection effects around p = .025 (one-tailed) among the studies included in the Kantak & Winstein review, such that studies reporting statistically significant results at delayed retention will be overrepresented in the sample. Secondary Objectives: Test the following potential moderators of the effect of reduced terminal feedback frequency on motor learning: Age group (Children, Adult, Older adult), Skill level (novice, experienced, expert), Task classification (based on Gentile’s 2 X 2 framework), Number of acquisition trials, Number of acquisition days, Frequency of terminal feedback, Publication status, Bandwidth provisioning, Faded feedback schedule. Hypotheses to be tested: 1) Based on the Challenge-Point Framework (Guadagnoli & Lee, 2001): Children and older adults will perform more effectively on delayed retention tests after having practiced with 100% feedback during acquisition, while younger adults will perform more effectively after having received a reduced frequency of feedback. 2) Based on the Challenge-Point Framework: Higher skill performers will benefit from less feedback in acquisition than lower skilled learners when performance is measured on a delayed retention test. 3) Based on our assessment of the motor learning literature: Published articles will report significantly larger benefits of reduced feedback frequency when assessed at delayed retention. 4) Based on research comparing bandwidth feedback protocols to yoked groups: Providing feedback according to a bandwidth will have a larger benefit for delayed retention performance than reduced feedback frequency. 5) Based on the guidance hypothesis: A faded schedule of feedback during acquisition will be more effective than a static reduced schedule of feedback for delayed retention performance. Tertiary Objective: Examine the sensitivity of the primary analyses to model specification and other analysis decisions. 5. Inclusion criteria Experiments will be included based on the following inclusion criteria: 1. Published article or masters/doctorate thesis. 2. Experiment that employed random assignment to experimental conditions. 3. Included healthy, non-clinical participants. 4. Included immediate OR delayed retention test. 5. Included a feedback frequency manipulation and included a reference group receiving feedback after 100% of acquisition trials. 6. Written in English. 7. Studied a motor skill. 8. Included an objective measure of performance 6. Search Strategy Data-bases: PubMed, PsycInfo. Search query: “feedback” AND “motor learning” PubMed and PsycInfo databases were queried on August 4, 2020. Results from both searches were imported into Covidence for further screening. Covidence automatically identifies and removes duplicate entries and tracks these data on a PRIMSA-style flow chart. A total of 1990 studies were identified, of which 433 were duplicates. The resulting 1557 studies were screened based on title and abstract. Two researchers screened each article independently and decided if the article was potentially relevant to the present meta-analysis or not. Of the 1557 studies, there were 62 conflicts in which the researchers reached different decisions. Each conflict was resolved by a third researcher who was not involved in the first screening decision. Following title and abstract screening, 75 papers were identified for full text review. Again, each article will be reviewed by two researchers independently and conflicts in inclusion/exclusion decisions will be resolved by a third researcher. Additional articles will be sought by performing forward and backward reference searching. The reference sections of each included article will be searched for possible articles of interest. Further, each article will be found on Google Scholar and the “cited by” link will be searched for other potentially relevant articles. The final search strategy employed will be a targeted author search on Google Scholar. Any author with two or more included articles will be searched. 7. Potential Moderators a) Age group: Adult, children (under 18-years old), older adult (over 49-years-old) b) Skill level: Novice, experienced, expert (based on description in article). c) Task classification (based on Gentile’s 2 X 2 framework): Regulatory conditions (in motion, stable), Intertrial variability (with, without). d) Number of acquisition trials e) Number of acquisition days f) Frequency of terminal feedback in acquisition as a percentage of total trials. g) Publication status: published, unpublished. h) Bandwidth provision (yes, no) i) Faded feedback schedule (yes, no) 8. Primary Outcomes Measured (DVs) – Required. Objectively measured motor performance outcome on a delayed 24-hour retention test or immediate retention test. Only one outcome per time point per experiment will be included in the meta-analysis. The prioritization of motor performance outcomes for inclusion is based on the content of the feedback provided. If feedback content maps directly onto an outcome measure, then that measure shall be chosen as the outcome to be included. If feedback does not directly map onto an outcome measure, then prioritization will be as follows: 1) Absolute error, 2) RMSE, 3) Absolute constant error, 4) Absolute timing error, 5) Relative timing error, 6) Variable error, 7), Movement time, 8) Movement form – expert raters, 9) Otherwise unspecified objective performance measure reported first in research report. 9. Data Extraction The data of interest are the standardized mean differences (Hedges’ g) between 100% feedback and reduced feedback frequency groups at a given time-point. To facilitate reproduction, effect sizes will be calculated based on the following priority list: group means and standard deviations, means and standard errors, F values and degrees of freedom, p-values and degrees of freedom. For experiments that included prognostic covariates in the analysis, F-values, degrees of freedom, number of covariates, and correlation between covariates and dependant variables will be used. If the article containing a given experiment does not provide adequate data, an email will be sent to the corresponding author requesting the necessary data and a two-week waiting period will be applied. If data cannot be produced to accurately calculate the effect size for a given experiment that experiment will be left out of further analysis. If full data are produced by the authors and an analysis of covariance can be conducted including pre-test performance as a prognostic covariate, but the original experiment did not perform and analysis of covariance, then both unadjusted and adjusted estimates will be calculated. The unadjusted estimate will be included in the primary meta-analysis and while the adjusted estimates will be included in a subsequent sensitivity analysis. If experiments include multiple groups with reduced feedback frequency, then the group receiving the least frequent (non-zero) feedback will be selected for inclusion in the primary analysis. If reduced frequency groups differ with respect to fading of schedule, the group with the least frequent feedback schedule that is consistent (i.e., not faded) will be included in the main analysis. To be included in the primary meta-analyses, the comparison group cannot receive feedback according to a bandwidth. Additional reduced frequency groups will be included in the moderator analyses of feedback frequency, faded scheduling, and bandwidth scheduling. If experiments include multiple 100% feedback frequency groups, then those groups will be collapsed using the following formula (https://handbook-5-1.cochrane.org/chapter_7/table_7_7_a_formulae_for_combining_groups.htm). In factorial experiments with a significant interaction, the simple effects of the most relevant condition will be included. For example, an experiment may cross neural stimulation with feedback frequency – if there is a significant interaction only the data from the sham conditions will be included. If both levels of the secondary IV are equally relevant but not included as a potential moderator in the meta-analysis, then the main effect will be included even if there is a significant interaction. If the second IV in a factorial experiment is included as a moderator in this meta-analysis (for example, age of participants), then one effect from each level of the IV will be included (for example, one for child, one for adult). If a factorial experiment crosses two IVs that both manipulate feedback frequency (for example, 100% vs.50% KR about absolute timing, 100% vs. 50% KR about relative timing), then the main effect of each IV will be included for the outcome corresponding to the feedback provided (in this example, absolute and relative timing, respectively). Two researchers will independently extract the required data for all time points of interest. Conflicts will be resolved by a third researcher. The compute.es package in R will be used to calculate all effect sizes. 10. Quality Assessment Each article will be assessed using the Cochrane Risk of Bias checklist in the Covidence Review Software. 11. Screening data for influential cases The meta-analysis package ‘metafor’ in R will be used to screen and analyze the data. In order to identify outlier values and exclude them from the final random effects model, the following influence statistics will be calculated for both immediate and delayed retention data: externally standardized residuals, DFFITS values, Cook's distances, covariance ratios, DFBETAS values, the estimates of t2 squared when each study is removed in turn, the test statistics for (residual) heterogeneity when each study is removed in turn, the diagonal elements of the hat matrix, and the weights (in %) given to the observed outcomes during the model fitting. If any study is identified as extremely influential by any three of the influence metrics it will be removed from the primary analysis as an outlier. 12. Data Analysis Alpha will be set at p = .05 for all analyses. To facilitate a decision about equivalence, an effect size of g = .1 will be considered the minimum effect size of interest. A two-one-sided-test (TOST) of equivalence will be conducted on any non-significant result. Publication and other selection biases are expected in this literature. Therefore, the adjusted estimate produced by a weight-function model with a .025 (one-sided) cut-point will be considered the primary outcome in this meta-analysis for all single time point analyses. Since there are currently no bias correction methods derived to account for selection effects on a Time X Group interaction, mixed model analyses will be considered the primary outcomes for these analyses. However, interpretation of these results will depend heavily on the extent of publication bias revealed in the univariate weight-function models. If selection effects are estimated to be substantial and influential for any time point of interest, then conclusions regarding Time X Group interactions will be withheld. The following script will be applied to the dataset of effects sizes with any outliers already removed. Additional rows may need to be removed from the dataset because they are intended for moderator analyses only. This information will be added to the analysis script before final publication. Any deviations from this script will be noted in the final publication and a justification will be provided. # import datasets #terminal feedback frequency data Data # Kantak & Winstein (2012) review data KWdat # Data including bandwidth protocols BandData # load the necessary libraries library(metafor) library(meta) library(weightr) # in order to estimate the average effect of reduced feedback at delayed retention, a naïve random #effects model will be fit to only the experiments that included a delayed retention test #select only delayed retention results delay <- data[!(data$time== “3”),] delayed <- rma(ret_g, ret_v, data = delay) #select only immediate retention results imd <- data[!(data$time== “2”),] # estimate the effect at immediate retention immediate <- rma(ret_g, ret_v, data = imd) #select only acquisition results acq <- data[!(data$time== “1”),] # estimate the effect at immediate retention acquisition <- rma(ret_g, ret_v, data = acq) # test the time (immediate, delayed) X feedback frequency interaction with mixed effects model # dummy code the immediate and delayed levels of the time variable data$t.immediate <- ifelse(data$time == "Immediate", 1, 0) data$t.delay <- ifelse(data$time == "Delayed", 1, 0) Interaction <- rma.mv(ret_g, ret_v, mods = cbind(t.immediate, t.delay), random = cbind(t.immediate, t.delay) | as.factor(study),struct = "CS", data = data) # estimate the effect at immediate and delayed retention while accounting for repeated measures Estimates <- Interaction <- rma.mv(ret_g, ret_v, mods = cbind(t.immediate, t.delay) + intercept = FALSE, random = cbind(t.immediate, t.delay) | as.factor(study),struct = "CS", data = data) # test the time (acquisition, delayed) X feedback frequency interaction with mixed effects model # dummy code the acquisition level of the time variable data$t.acquisition <- ifelse(data$time == "Acquisition", 1, 0) AcqInteraction <- rma.mv(ret_g, ret_v, mods = cbind(t.acquisition, t.delay), random = cbind(t.acquisition, t.delay) | as.factor(study),struct = "CS", data = data) # estimate the effect at acqusition and delayed retention while accounting for repeated measures AcqEstimates <- Interaction <- rma.mv(ret_g, ret_v, mods = cbind(t.acquisition, t.delay) + intercept = FALSE, random = cbind(t.acquisition, t.delay) | as.factor(study),struct = "CS", data = data) # inspect results of each analysis Summary(delayed) Summary(immediate) Summary(acquisition) Summary(Interaction) Summary(Estimates) Summary(AcqInteraction) Aummary(AcqEstimates) # evaluate the influence of publication bias on the delayed retention results x <- delay$ret_g v <- delay$ret_v weightfunct(x,v) # evaluate the influence of publication bias on immediate retention results g <- imd$ret_g s <- imd$ret_v weightfunct(g,s) # evaluate the influence of publication bias on the results reviewed by Kantak & Winstein (2012) #remove immediate retention results KWdelay <- KWdat[!(KWdat$time== “1”),] # fit weight-function model k <- KWdelay$ret_g w <- KWdelay$ret_v weightfunct(k,w) #remove delayed retention results KWimd <- KWdat[!(KWdat$time== “2”),] #fit weight-function model d <- KWimd$ret_g r <- KWimd$ret_v Weightfunct(d,r) # exploratory tests of categorical moderators for delayed retention results # any significant moderator will be followed up with analysis to estimate effect at each level using following code where X = factor heading rma(ret_g,ret_v, mods = ~factor(X)-1, data= delay) # age group rma(ret_g,ret_v, mods = ~factor(age), data= delay) # skill level rma(ret_g,ret_v, mods = ~factor(skill), data= delay) # task class rma(ret_g,ret_v, mods = ~factor(task), data= delay) # publication status rma(ret_g,ret_v, mods = ~factor(pub), data= delay) # faded schedule rma(ret_g,ret_v, mods = ~factor(faded), data= delay) # bandwidth protocols Banddelay <- BandData[!(BandData$time== “3”),] rma(ret_g,ret_v, mods = ~factor(bandwidth), data= Banddelay) # exploratory tests of continuous moderators for delayed retention results # number of acquisition trials rma(ret_g,ret_v, mods = ~ trials, data= delay) # number of acquisition days rma(ret_g,ret_v, mods = ~ days, data= delay) # frequency of terminal feedback rma(ret_g,ret_v, mods = ~ frequency, data= delay) # exploratory tests of categorical moderators for immediate retention results # age group rma(ret_g,ret_v, mods = ~factor(age), data= imd) # skill level rma(ret_g,ret_v, mods = ~factor(skill), data= imd) # task class rma(ret_g,ret_v, mods = ~factor(task), data= imd) # publication status rma(ret_g,ret_v, mods = ~factor(pub), data= imd) # exploratory tests of continuous moderators for immediate retention results # number of acquisition trials rma(ret_g,ret_v, mods = ~ trials, data= imd) # number of acquisition days rma(ret_g,ret_v, mods = ~ days, data= imd) # frequency of terminal feedback rma(ret_g,ret_v, mods = ~ frequency, data= imd) 12. Sensitivity analyses Sensitivity analyses will be conducted on the Time X Feedback frequency interaction analyses as well as on the publication bias analysis. A) Standardized mean change will be calculated wherever data allow to test the Time X Feedback frequency interaction. Data required: Immediate retention mean and standard deviation, delayed retention mean and standard deviation, sample size, and correlation between immediate and delayed retention. The correlation between time-points is expected to be unknown in the majority of cases. Therefore, any known correlations will be used to estimate a range of plausible values. Sensitivity analyses will be conducted using the Standardized Mean Change with a correlation value estimated from each quartile of the overall range of correlation estimates. Therefore, there will be three runs of this analysis. The analysis will be coded and run according to the methods described here: http://www.metafor-project.org/doku.php/analyses:morris2008 B) Publication bias correction methods are sensitive to unknown underlying conditions. Therefore, on the basis of the first run of analyses, performance checks under a range of plausible conditions will be run using the following application: http://www.shinyapps.org/apps/metaExplorer/ On the basis of these performance checks, additional bias correction methods will be applied in an attempt to achieve the most possible coverage of plausible conditions. One additional need for sensitivity analysis may emerge following our quality assessment of included articles. Analyses may be re-run with articles judged to have a high risk of bias excluded in order to evaluate the extent that the overall analysis is sensitive to the inclusion of high risk studies.

This article is from Frontiers in Human Neuroscience , volume 7 . Abstract Skilled performers such as athletes or musicians can improve their performance by imagining the actions or sensory outcomes associated with their skill. Performers vary widely in their auditory and motor imagery abilities, and these individual differences influence sensorimotor learning. It is unknown whether imagery abilities influence both memory encoding and retrieval. We examined how auditory and motor imagery abilities influence musicians' encoding (during Learning, as they practiced novel melodies), and retrieval (during Recall of those melodies). Pianists learned melodies by listening without performing (auditory learning) or performing without sound (motor learning); following Learning, pianists performed the melodies from memory with auditory feedback (Recall). During either Learning (Experiment 1) or Recall (Experiment 2), pianists experienced either auditory interference, motor interference, or no interference. Pitch accuracy (percentage of correct pitches produced) and temporal regularity (variability of quarter-note interonset intervals) were measured at Recall. Independent tests measured auditory and motor imagery skills. Pianists' pitch accuracy was higher following auditory learning than following motor learning and lower in motor interference conditions (Experiments 1 and 2). Both auditory and motor imagery skills improved pitch accuracy overall. Auditory imagery skills modulated pitch accuracy encoding (Experiment 1): Higher auditory imagery skill corresponded to higher pitch accuracy following auditory learning with auditory or motor interference, and following motor learning with motor or no interference. These findings suggest that auditory imagery abilities decrease vulnerability to interference and compensate for missing auditory feedback at encoding. Auditory imagery skills also influenced temporal regularity at retrieval (Experiment 2): Higher auditory imagery skill predicted greater temporal regularity during Recall in the presence of auditory interference. Motor imagery aided pitch accuracy overall when interference conditions were manipulated at encoding (Experiment 1) but not at retrieval (Experiment 2). Thus, skilled performers' imagery abilities had distinct influences on encoding and retrieval of musical sequences.

The original study concerns motor learning and consolidation in adults with ADHD in comparison to healthy controls. Theoretical accounts of ADHD, such as the state regulation model (van der Meere, 2005) and dual-process models (Halperin & Schulz, 2006; Johnson et al., 2007) propose that deficits in cognitive performance in ADHD may reflect problems in regulating arousal. A low intensity task-irrelevant vibratory stimulation (VtSt) was afforded during training on a sequence of finger opposition movements to young adults with normal (typical controls) and decreased (ADHD) arousal levels. Under VtSt, typical individuals had reduced overnight, consolidation phase, gains; performance partly recovering one week later. In contrast, participants with ADHD benefitted from VtSt both during the acquisition (online) and the overnight skill consolidation (offline) phases. Thus, environmental noise that is detrimental to memory consolidation in typical adults is beneficial for peers with reduced baseline arousal levels (adults with ADHD). Learning deficits in ADHD may be fully compensated in specific bio-behavioural conditions.

The original study concerns motor learning and consolidation in adults with ADHD in comparison to healthy controls. Theoretical accounts of ADHD, such as the state regulation model (van der Meere, 2005) and dual-process models (Halperin & Schulz, 2006; Johnson et al., 2007) propose that deficits in cognitive performance in ADHD may reflect problems in regulating arousal. A low intensity task-irrelevant vibratory stimulation (VtSt) was afforded during training on a sequence of finger opposition movements to young adults with normal (typical controls) and decreased (ADHD) arousal levels. Under VtSt, typical individuals had reduced overnight, consolidation phase, gains; performance partly recovering one week later. In contrast, participants with ADHD benefitted from VtSt both during the acquisition (online) and the overnight skill consolidation (offline) phases. Thus, environmental noise that is detrimental to memory consolidation in typical adults is beneficial for peers with reduced baseline arousal levels (adults with ADHD). Learning deficits in ADHD may be fully compensated in specific bio-behavioural conditions.

The purpose of the present review was to evaluate the scientific evidence concerning the effects of attentional focus strategies on motor learning and the performance of racket sports.

This study evaluates the relationships that exist between three types of visual and perceptual-motor tasks (coincidence-anticipation, tracking with rotary pursuit, and a unique two-dimensional discrete motor task) and investigates the nature of learning demonstrated by the subjects on each of the three tasks. Thirty male students were given 20 trials on each of the tasks. A correlation matrix of task variables was computed and learning curves plotted. Results indicated that an individual's ability to perform one visual cue perceptual-motor task is relatively unrelated to performance on another task. Positive learning was demonstrated on the two-dimensional tasks (pursuit rotor and light panel), but the slight learning progress demonstrated during the early trials of the coincidence-anticipation task was almost negated during the final five trials. (Author)

During adolescence, athletes navigate a sensitive phase marked by high divergence in their developmental trajectories (Mudrak et al., 2020; Mudrak & Zabrodska, 2015). As they enter the specialising years of their athletic development (Baker & Côté, 2006; Côté, 1999; Côté et al., 2003), the importance of structured, focused, and effortful engagement grows to sustain continuous successful progression. This critical step towards sports competence appears to be challenging for many youth athletes as adolescents generally reduce their participation in physical activity (Dumith et al., 2011) and a large portion of adolescent athletes drop out of sports during this developmental period (Fraser-Thomas et al., 2008; Mudrak, 2010; Wylleman et al., 2004). Coach and the coach-created motivational climate appear to be crucial factors facilitating the successful developmental transitions of youth athletes (Alvarez et al., 2012; Curran et al., 2015; Duda, 2013; Duda & Balaguer, 2007). It appears that the motivational climate created by the coach affects the ways in which athletes experience their sports participation and, in this way, their decision to maintain their sports involvement or drop out from sports (Alvarez et al., 2012; Mudrak, 2010). The feedback that athletes receive from their proximal environment has also been related to various aspects of their motor performance and motor learning, including balance, accuracy, maximum force production, speed and endurance, or movement form (Chua et al., 2018; Chua et al., 2021; Wulf et al., 2018; Wulf & Lewthwaite, 2016). A majority of the reviewed research on the effects of coach-created motivational climate based their assumptions on two major theoretical frameworks – achievement goal theory (AGT) and self-determination theory (SDT) (Alvarez et al., 2012; Appleton & Duda, 2016; Duda, 2013; Duda & Balaguer, 2007). The key assumptions within these two frameworks are that the types of goals coaches endorse, and the conditions coaches create for satisfying basic human needs of autonomy, relatedness, and competence significantly shape the athletes’ experience of sports participation and, subsequently, athletes’ subjective and objective outcomes. In this way, the coach-created motivational climate depends 1) on the degree to which coaches endorse goals aimed at winning and comparison with others (performance×ego) and/or goals aimed at self-improvement and comparison with previous performance (mastery×task) (AGT) and 2) on the level of autonomy and support or control coaches provide to the athletes (SDT) (Duda, 2013). Recently, Duda and her colleagues (Appleton & Duda, 2016; Duda, 2013; Duda et al., 2017; Duda & Appleton, 2016) proposed a concept of “Empowering/disempowering coaching” (EDC) that merges assumptions of both AGT and SDT frameworks, proposing that a coach-created climate should be approached as including five key dimensions, three positive (task-involving, autonomy-supportive, socially-supportive), and two negative (ego-involving, and controlling coaching). Another integrative attempt to merge SDT and other socio-cognitive approaches explaining the development of sports competence has been provided by the OPTIMAL theory (Optimising Performance through Intrinsic Motivation and Attention for Learning) (Wulf & Lewthwaite, 2016). Compared to the concept of EDC, which rather focuses on the more pervasive motivational climate, as well as the subjective motivational and emotional outcomes in EDC, the OPTIMAL framework investigates the effectiveness of motivational factors on objective measures such as performance or learning outcomes of motor skills across different ranges of motor skills such as balance, accuracy, maximum force production, speed and endurance, or movement form (Chua et al., 2018, 2020; Wulf et al., 2018; Wulf & Lewthwaite, 2009, 2016). From this perspective, coaches should consider general precursors of motor learning and performance to maximise the effectiveness of their approach to athletes: enhanced expectancies (EE), autonomy support (AS), and an external focus of attention (EF). Research has shown that each of these factors enhances motor performance in diverse types of motor skills (for a review, see Wulf & Lewthwaite, 2016). Moreover, researchers demonstrated that combining these components has remarkable benefits for motor performance of vertical jump height (N = 36) (Chua et al., 2018). This study’s results indicate a medium effect size for the Group main effect (η_p^2 = 0.12); the Condition main effect (AS, EE, EF) had a small effect size (η_p^2 = 0.04); the Group × Block interaction effect was medium (η_p^2 = 0.09); in the Post-hoc tests, medium to large effects were observed for Block 3 (η_p^2 = 0.012) and 4 (η_p^2 = 0.16). Any association between two of these factors, such as EE and AS (Wulf et al., 2014), led to additional improvement in motor performance (overhand throw, N = 64). In the retention test, AS and EE both had a small effect size η_p^2 = 0.62; in the transfer test, AS had a medium effect size η_p^2 = 0.105, while EE had a small to medium η_p^2 = 0.081 (Wulf et al., 2014) (the authors of referred articles did not provide further/detailed measures of effect sizes). Both the EDC (Appleton & Duda, 2016; Duda, 2013) and OPTIMAL (Wulf & Lewthwaite, 2016) frameworks seem useful in both research and practical applications concerning adolescent athlete development. However, several conceptual, methodological, and practical limitations within each framework limit our current insights into their effectiveness and applicability. Both frameworks involve a limited number of dimensions, exploring relatively simple relationships between single variables while focusing on different types of athletes’ outcomes (subjective × objective). Evidence for the EDC primarily relies on cross-sectional questionnaire studies, with particular emphasis on subjective motivational and emotional outcomes (Duda & Appleton, 2016; Duda & Balaguer, 2007). In contrast, the OPTIMAL framework predominantly includes experimental studies, focusing on objective athlete outcomes such as motor performance and learning (Wulf & Lewthwaite, 2016). It can be argued that both approaches focus on various elements within a broader process, where coaches' environmental and situational conditions impact the development of athletes' sports competence. We suggest that the coach-created climate and feedback affect the athletes’ development through athletes’ sense of agency, i.e., “the sense that I am the one who is causing or generating an action” (David et al., 2008, p. 524) as adolescents need to become “agents of their own learning, not just recipients of information” (Bandura, 2006, p. 10) to ensure further successful development of competence through and beyond adolescence (Mudrak & Zabrodska, 2015). To address the shortcomings of both previous frameworks, we propose a new theoretical framework, which we label “Coaching for Agency” (CfA). The CfA integrates the main assumptions of the EDC and OPTIMAL frameworks regarding the effects of coach-created climate and feedback on the development of sport competence in adolescent athletes (Laroëre et. al., in prep). We believe that such a complex framework is needed to provide a more comprehensive understanding of the interactions between proximal social environment, individual athletes acting as agents of their own development, and multiple outcomes which can be seen as products but also as drivers of ongoing development towards sport competence (Mudrak et al., 2020). We introduce the main assumptions of the CfA model in Figures 1-7 (see OSF Laroëre et al., 2024). We chose aesthetic sports as a model type of sport for testing our proposed framework. Gymnastics and other aesthetic sports have a complex and particular training process, with high levels of both demands (physical and psychological) and stress (Bobo-Arce & Méndez-Rial, 2013) and to our best knowledge, there is only a limited number of studies investigating the current psychological environment and the effects of different types of coach-created feedback on these aesthetic sports athletes. The larger project, that this study is part of, aims to explore the relationships between the coach-created psychosocial environment and the outcomes of adolescent athletes, including both subjective (e.g., athletes’ engagement and burnout) and objective (e.g., motor performance) measures. Previous work focused on subjective outcomes through an online longitudinal survey investigating athletes’ engagement and burnout over time (see OSF Laroëre et al., 2024). In contrast, the current study - a randomised controlled trial (RCT) - focuses on objective outcomes, specifically examining motor performance and motor learning. This RCT builds on our earlier research and connects with it by further exploring the impact of the coach-created psychosocial environment on athletes' performance and well-being, utilising the CfA model and the OPTIMAL theory to provide a comprehensive understanding of these dynamics. The current study will consist of a series of randomised controlled trials conducted within the methodological framework of the OPTIMAL theory of motor learning extended by the assumptions of CfA. We will compare OPTIMAL-based feedback in AS vs control conditions and EE vs control conditions. In this way, we will assess the causal effects of different types of coach-created feedback based on the motivational components of OPTIMAL theory (i.e., AS and EE) on athletes’ motor performance and learning. We predict that participants under autonomy-supportive and expectancies-enhancing conditions will improve their task-related motor performance and learning compared to control conditions (see Hypotheses below).

During adolescence, athletes navigate a sensitive phase marked by high divergence in their developmental trajectories (Mudrak et al., 2020; Mudrak & Zabrodska, 2015). As they enter the specialising years of their athletic development (Baker & Côté, 2006; Côté, 1999; Côté et al., 2003), the importance of structured, focused, and effortful engagement grows to sustain continuous successful progression. This critical step towards sports competence appears to be challenging for many youth athletes as adolescents generally reduce their participation in physical activity (Dumith et al., 2011) and a large portion of adolescent athletes drop out of sports during this developmental period (Fraser-Thomas et al., 2008; Mudrak, 2010; Wylleman et al., 2004). Coach and the coach-created motivational climate appear to be crucial factors facilitating the successful developmental transitions of youth athletes (Alvarez et al., 2012; Curran et al., 2015; Duda, 2013; Duda & Balaguer, 2007). It appears that the motivational climate created by the coach affects the ways in which athletes experience their sports participation and, in this way, their decision to maintain their sports involvement or drop out from sports (Alvarez et al., 2012; Mudrak, 2010). The feedback that athletes receive from their proximal environment has also been related to various aspects of their motor performance and motor learning, including balance, accuracy, maximum force production, speed and endurance, or movement form (Chua et al., 2018; Chua et al., 2021; Wulf et al., 2018; Wulf & Lewthwaite, 2016). A majority of the reviewed research on the effects of coach-created motivational climate based their assumptions on two major theoretical frameworks – achievement goal theory (AGT) and self-determination theory (SDT) (Alvarez et al., 2012; Appleton & Duda, 2016; Duda, 2013; Duda & Balaguer, 2007). The key assumptions within these two frameworks are that the types of goals coaches endorse, and the conditions coaches create for satisfying basic human needs of autonomy, relatedness, and competence significantly shape the athletes’ experience of sports participation and, subsequently, athletes’ subjective and objective outcomes. In this way, the coach-created motivational climate depends 1) on the degree to which coaches endorse goals aimed at winning and comparison with others (performance×ego) and/or goals aimed at self-improvement and comparison with previous performance (mastery×task) (AGT) and 2) on the level of autonomy and support or control coaches provide to the athletes (SDT) (Duda, 2013). Recently, Duda and her colleagues (Appleton & Duda, 2016; Duda, 2013; Duda et al., 2017; Duda & Appleton, 2016) proposed a concept of “Empowering/disempowering coaching” (EDC) that merges assumptions of both AGT and SDT frameworks, proposing that a coach-created climate should be approached as including five key dimensions, three positive (task-involving, autonomy-supportive, socially-supportive), and two negative (ego-involving, and controlling coaching). Another integrative attempt to merge SDT and other socio-cognitive approaches explaining the development of sports competence has been provided by the OPTIMAL theory (Optimising Performance through Intrinsic Motivation and Attention for Learning) (Wulf & Lewthwaite, 2016). Compared to the concept of EDC, which rather focuses on the more pervasive motivational climate, as well as the subjective motivational and emotional outcomes in EDC, the OPTIMAL framework investigates the effectiveness of motivational factors on objective measures such as performance or learning outcomes of motor skills across different ranges of motor skills such as balance, accuracy, maximum force production, speed and endurance, or movement form (Chua et al., 2018, 2020; Wulf et al., 2018; Wulf & Lewthwaite, 2009, 2016). From this perspective, coaches should consider general precursors of motor learning and performance to maximise the effectiveness of their approach to athletes: enhanced expectancies (EE), autonomy support (AS), and an external focus of attention (EF). Research has shown that each of these factors enhances motor performance in diverse types of motor skills (for a review, see Wulf & Lewthwaite, 2016). Moreover, researchers demonstrated that combining these components has remarkable benefits for motor performance of vertical jump height (N = 36) (Chua et al., 2018). This study’s results indicate a medium effect size for the Group main effect (η_p^2 = 0.12); the Condition main effect (AS, EE, EF) had a small effect size (η_p^2 = 0.04); the Group × Block interaction effect was medium (η_p^2 = 0.09); in the Post-hoc tests, medium to large effects were observed for Block 3 (η_p^2 = 0.012) and 4 (η_p^2 = 0.16). Any association between two of these factors, such as EE and AS (Wulf et al., 2014), led to additional improvement in motor performance (overhand throw, N = 64). In the retention test, AS and EE both had a small effect size η_p^2 = 0.62; in the transfer test, AS had a medium effect size η_p^2 = 0.105, while EE had a small to medium η_p^2 = 0.081 (Wulf et al., 2014) (the authors of referred articles did not provide further/detailed measures of effect sizes). Both the EDC (Appleton & Duda, 2016; Duda, 2013) and OPTIMAL (Wulf & Lewthwaite, 2016) frameworks seem useful in both research and practical applications concerning adolescent athlete development. However, several conceptual, methodological, and practical limitations within each framework limit our current insights into their effectiveness and applicability. Both frameworks involve a limited number of dimensions, exploring relatively simple relationships between single variables while focusing on different types of athletes’ outcomes (subjective × objective). Evidence for the EDC primarily relies on cross-sectional questionnaire studies, with particular emphasis on subjective motivational and emotional outcomes (Duda & Appleton, 2016; Duda & Balaguer, 2007). In contrast, the OPTIMAL framework predominantly includes experimental studies, focusing on objective athlete outcomes such as motor performance and learning (Wulf & Lewthwaite, 2016). It can be argued that both approaches focus on various elements within a broader process, where coaches' environmental and situational conditions impact the development of athletes' sports competence. We suggest that the coach-created climate and feedback affect the athletes’ development through athletes’ sense of agency, i.e., “the sense that I am the one who is causing or generating an action” (David et al., 2008, p. 524) as adolescents need to become “agents of their own learning, not just recipients of information” (Bandura, 2006, p. 10) to ensure further successful development of competence through and beyond adolescence (Mudrak & Zabrodska, 2015). To address the shortcomings of both previous frameworks, we propose a new theoretical framework, which we label “Coaching for Agency” (CfA). The CfA integrates the main assumptions of the EDC and OPTIMAL frameworks regarding the effects of coach-created climate and feedback on the development of sport competence in adolescent athletes (Laroëre et. al., in prep). We believe that such a complex framework is needed to provide a more comprehensive understanding of the interactions between proximal social environment, individual athletes acting as agents of their own development, and multiple outcomes which can be seen as products but also as drivers of ongoing development towards sport competence (Mudrak et al., 2020). We introduce the main assumptions of the CfA model in Figures 1-7 (see OSF Laroëre et al., 2024). We chose aesthetic sports as a model type of sport for testing our proposed framework. Gymnastics and other aesthetic sports have a complex and particular training process, with high levels of both demands (physical and psychological) and stress (Bobo-Arce & Méndez-Rial, 2013) and to our best knowledge, there is only a limited number of studies investigating the current psychological environment and the effects of different types of coach-created feedback on these aesthetic sports athletes. The larger project, that this study is part of, aims to explore the relationships between the coach-created psychosocial environment and the outcomes of adolescent athletes, including both subjective (e.g., athletes’ engagement and burnout) and objective (e.g., motor performance) measures. Previous work focused on subjective outcomes through an online longitudinal survey investigating athletes’ engagement and burnout over time (see OSF Laroëre et al., 2024). In contrast, the current study - a randomised controlled trial (RCT) - focuses on objective outcomes, specifically examining motor performance and motor learning. This RCT builds on our earlier research and connects with it by further exploring the impact of the coach-created psychosocial environment on athletes' performance and well-being, utilising the CfA model and the OPTIMAL theory to provide a comprehensive understanding of these dynamics. The current study will consist of a series of randomised controlled trials conducted within the methodological framework of the OPTIMAL theory of motor learning extended by the assumptions of CfA. We will compare OPTIMAL-based feedback in AS vs control conditions and EE vs control conditions. In this way, we will assess the causal effects of different types of coach-created feedback based on the motivational components of OPTIMAL theory (i.e., AS and EE) on athletes’ motor performance and learning. We predict that participants under autonomy-supportive and expectancies-enhancing conditions will improve their task-related motor performance and learning compared to control conditions (see Hypotheses below).

Law enforcement professionals undergo diverse training and practice conditions to enhance performance, maximize effectiveness, reduce lethality, and ensure self-protection. In addition to physical conditioning (strength, cardiovascular endurance, and power), police training is designed to develop Police Arrest and Self-Defense Skills (ASDS) and refine police-specific motor skills, such as shooting in various scenarios with different firearms or handling conducted energy weapons, such as Tasers. In general, training occurs in multiple environments, with scenario-based training as a key strategy to simulate professional demands and enhance operational readiness. Globally, many law enforcement agencies design operational training based primarily on empirical approaches, relying on the practical experiences of officers. However, as seen in fields such as sports science and rehabilitation, research in Motor Learning can provide an evidence-based framework for optimizing police training. By understanding the underlying mechanisms of skilled motor behavior in law enforcement officers and systematically investigating key influencing factors (e.g., instruction, practice, feedback, and motivation), we can enhance police effectiveness across various operational contexts and under challenging conditions such as stress, fatigue, and sleep deprivation. Therefore, this scoping review will examine existing knowledge on motor performance and learning principles that could improve police training.

Law enforcement professionals undergo diverse training and practice conditions to enhance performance, maximize effectiveness, reduce lethality, and ensure self-protection. In addition to physical conditioning (strength, cardiovascular endurance, and power), police training is designed to develop Police Arrest and Self-Defense Skills (ASDS) and refine police-specific motor skills, such as shooting in various scenarios with different firearms or handling conducted energy weapons, such as Tasers. In general, training occurs in multiple environments, with scenario-based training as a key strategy to simulate professional demands and enhance operational readiness. Globally, many law enforcement agencies design operational training based primarily on empirical approaches, relying on the practical experiences of officers. However, as seen in fields such as sports science and rehabilitation, research in Motor Learning can provide an evidence-based framework for optimizing police training. By understanding the underlying mechanisms of skilled motor behavior in law enforcement officers and systematically investigating key influencing factors (e.g., instruction, practice, feedback, and motivation), we can enhance police effectiveness across various operational contexts and under challenging conditions such as stress, fatigue, and sleep deprivation. Therefore, this scoping review will examine existing knowledge on motor performance and learning principles that could improve police training.

The current experiment is considered a “Proof-of-Concept” in which we aim to establish the feasibility and rationale for an integrated theory that has been developed as part of the German Research Foundation (DFG) project “Motor heuristics and movement analogies in performance and health” (PIs: Markus Raab and Laura Voigt, RA 940/27-1 Ι VO 2789/1-1). For successful sports performance, athletes must choose between actions (movement selection) and then execute the selected action skillfully (movement execution; Raab, 2017; Voigt et al., 2022). A table tennis player who can execute a backhand stroke effectively but does not know when to use the backhand stroke will not succeed in the sport. Equally, a player who knows when to employ the backhand stroke but cannot execute it effectively is unlikely to succeed. Movement selection and execution often need to be performed when time is limited, stress is present, more than one task needs to be processed, or when fatigue or unfamiliar situations occur. Given limited ability to process multiple channels of information, performers need to find ways to manage the wealth of available information economically to make appropriate decisions and execute the chosen movement skillfully in their discipline. The idea that economic management of information is central for successful performance is supported by a common understanding of expertise in sports: During the development of expertise, the nature of the knowledge structures that support motor performance gradually changes over time, with an increasing degree of implicit (unconscious) processing and a decreasing level of explicit (conscious) processing (Fitts & Posner, 1967). Explicit processes involve the retrieval of consciously accessible (declarative) knowledge and depend on working memory. Thus, highly explicit sport performance has often been shown to be disrupted by demands resulting from performance pressure or multiple task requirements (Masters & Maxwell, 2008; Nieuwenhuys & Oudejans, 2012, 2017). In contrast, implicit processes are faster and involve sophisticated complexes of procedural knowledge that can be applied without conscious thought, with greater automaticity and fewer demands for attentional resources (e.g., Anderson, 1983; Lewicki et al., 1992; Kal et al., 2018; Masters & Maxwell, 2004; Shiffrin & Schneider, 1977; Willingham, 1998). Implicit processes are therefore less dependent on working memory, which allows the expert to economically manage multiple streams of information for movement selection and movement execution while being taxed with other demands (for a review of the theoretical architecture and function of working memory, see Baddeley, 2003). But how exactly are these streams of information for movement selection and movement execution managed? In the current study, we aim to deliver a “Proof-of-Concept” for the integration of two theories – motor heuristics (Raab, 2017) and motor analogies (Masters, 2000, 2012). We aim to investigate whether implicit learning helps athletes to manage information during movement selection and movement execution and protects performance under demanding circumstances. Motor heuristics have their theoretical basis in the simple heuristics used for decision making (Gigerenzer & Goldstein, 1996) and advise “what” movement to choose (i.e., movement selection). Motor heuristics are fast and frugal decision-making strategies (i.e., rules of thumb) that exploit information in the environment by ordering pieces of information (i.e., cues) by their validity. Decision makers judge the cues’ validity (i.e., how often in the past a particular piece of information was helpful in making a choice), and then choose the option that is favored by the cue with the highest validity. In sports, a large body of empirical findings demonstrates that superior decision making by experts is characterized by focusing on fewer (task-relevant) options and higher-quality options and decisions (Basevitch et al., 2020; Belling et al., 2015; Laborde & Raab, 2013; Musculus, 2018; Musculus et al., 2021; Raab & Johnson, 2007), especially in demanding circumstances, such as time pressure, opponent pressure, and stress (Musculus et al., 2021). For heuristics to be successful, their use needs to be matched to the environmental structures. Leuker et al. (2018, 2019) showed that statistical regularities (i.e., risk-reward structures) can be learned without explicit instructions via incidental, unsupervised learning and then utilized as heuristics for decision making under uncertainty. Implicit learning processes can thus be an important aspect in building a representation of the environment that, in turn, guides choice behavior (Hertwig et al., 2022). Importantly, the so-called description-experience gap further suggests that learning environmental structures by experience is more beneficial for performance than receiving explicit information about the environmental structures (e.g., Hertwig et al., 2004; Armstrong & Spaniol, 2017). Motor analogies have their theoretical basis in the theory of implicit motor learning (Masters, 1992, 2000) and advise ‘how’ to move (i.e., movement execution). Motor analogies leverage a concept that is already well known, such as “strike the ball while bringing the bat up the hypotenuse of a triangle” in order to convey the complex structure of the motor skill (e.g., a table tennis topspin forehand; Liao & Masters, 2001). It has been proposed that they promote economic management of information for movement control by chunking fundamental technical information (i.e., relevant pieces of information) into one well-known concept – many small “bits” of information are collapsed to into fewer larger chunks (Poolton & Masters, 2014). Although fewer chunks are processed, they contain the relevant information, meaning that information can be processed with relatively less cognitive effort and processing becomes more efficient. At the same time, motor analogies minimize accrual of conscious knowledge of the underlying rules governing the mechanics of movements (e.g., Liao & Masters, 2001). Learning movements implicitly through analogies has been shown to result in more robust motor performance under demanding conditions than learning movements by explicit step-by-step instructions (e.g., Koedijker et al., 2007, Lam et al., 2009; Liao & Masters, 2001; Schlapkohl et al., 2012; for a meta-analysis see Cabral et al., 2022), which has been attributed to the placement of fewer demands on cognitive resources than explicit motor learning. To date, motor heuristics and analogies have only been researched independently from each other. Notably, it has been shown that implicit motor learning through motor analogies did not only improve motor performance, but also decision making, potentially because implicit processing of movement execution left sufficient cognitive resources available for decision making (Lola & Tzetzis, 2021; Masters et al., 2008). Likewise, the use of heuristics (instead of a complex step-by-step planning strategy) showed combined selection and execution advantages, when participants were required to select one of two computer cursors that displace in different directions (i.e., selection) and subsequently navigate from a starting position to a goal as efficiently as possible (i.e., execution; Dundon et al., 2023). We aim to extend these findings by testing performance in demanding circumstances after promoting the economic management of information for both movement selection and execution via implicit learning of motor heuristics and motor analogies.

The current experiment is considered a “Proof-of-Concept” in which we aim to establish the feasibility and rationale for an integrated theory that has been developed as part of the German Research Foundation (DFG) project “Motor heuristics and movement analogies in performance and health” (PIs: Markus Raab and Laura Voigt, RA 940/27-1 Ι VO 2789/1-1). For successful sports performance, athletes must choose between actions (movement selection) and then execute the selected action skillfully (movement execution; Raab, 2017; Voigt et al., 2022). A table tennis player who can execute a backhand stroke effectively but does not know when to use the backhand stroke will not succeed in the sport. Equally, a player who knows when to employ the backhand stroke but cannot execute it effectively is unlikely to succeed. Movement selection and execution often need to be performed when time is limited, stress is present, more than one task needs to be processed, or when fatigue or unfamiliar situations occur. Given limited ability to process multiple channels of information, performers need to find ways to manage the wealth of available information economically to make appropriate decisions and execute the chosen movement skillfully in their discipline. The idea that economic management of information is central for successful performance is supported by a common understanding of expertise in sports: During the development of expertise, the nature of the knowledge structures that support motor performance gradually changes over time, with an increasing degree of implicit (unconscious) processing and a decreasing level of explicit (conscious) processing (Fitts & Posner, 1967). Explicit processes involve the retrieval of consciously accessible (declarative) knowledge and depend on working memory. Thus, highly explicit sport performance has often been shown to be disrupted by demands resulting from performance pressure or multiple task requirements (Masters & Maxwell, 2008; Nieuwenhuys & Oudejans, 2012, 2017). In contrast, implicit processes are faster and involve sophisticated complexes of procedural knowledge that can be applied without conscious thought, with greater automaticity and fewer demands for attentional resources (e.g., Anderson, 1983; Lewicki et al., 1992; Kal et al., 2018; Masters & Maxwell, 2004; Shiffrin & Schneider, 1977; Willingham, 1998). Implicit processes are therefore less dependent on working memory, which allows the expert to economically manage multiple streams of information for movement selection and movement execution while being taxed with other demands (for a review of the theoretical architecture and function of working memory, see Baddeley, 2003). But how exactly are these streams of information for movement selection and movement execution managed? In the current study, we aim to deliver a “Proof-of-Concept” for the integration of two theories – motor heuristics (Raab, 2017) and motor analogies (Masters, 2000, 2012). We aim to investigate whether implicit learning helps athletes to manage information during movement selection and movement execution and protects performance under demanding circumstances. Motor heuristics have their theoretical basis in the simple heuristics used for decision making (Gigerenzer & Goldstein, 1996) and advise “what” movement to choose (i.e., movement selection). Motor heuristics are fast and frugal decision-making strategies (i.e., rules of thumb) that exploit information in the environment by ordering pieces of information (i.e., cues) by their validity. Decision makers judge the cues’ validity (i.e., how often in the past a particular piece of information was helpful in making a choice), and then choose the option that is favored by the cue with the highest validity. In sports, a large body of empirical findings demonstrates that superior decision making by experts is characterized by focusing on fewer (task-relevant) options and higher-quality options and decisions (Basevitch et al., 2020; Belling et al., 2015; Laborde & Raab, 2013; Musculus, 2018; Musculus et al., 2021; Raab & Johnson, 2007), especially in demanding circumstances, such as time pressure, opponent pressure, and stress (Musculus et al., 2021). For heuristics to be successful, their use needs to be matched to the environmental structures. Leuker et al. (2018, 2019) showed that statistical regularities (i.e., risk-reward structures) can be learned without explicit instructions via incidental, unsupervised learning and then utilized as heuristics for decision making under uncertainty. Implicit learning processes can thus be an important aspect in building a representation of the environment that, in turn, guides choice behavior (Hertwig et al., 2022). Importantly, the so-called description-experience gap further suggests that learning environmental structures by experience is more beneficial for performance than receiving explicit information about the environmental structures (e.g., Hertwig et al., 2004; Armstrong & Spaniol, 2017). Motor analogies have their theoretical basis in the theory of implicit motor learning (Masters, 1992, 2000) and advise ‘how’ to move (i.e., movement execution). Motor analogies leverage a concept that is already well known, such as “strike the ball while bringing the bat up the hypotenuse of a triangle” in order to convey the complex structure of the motor skill (e.g., a table tennis topspin forehand; Liao & Masters, 2001). It has been proposed that they promote economic management of information for movement control by chunking fundamental technical information (i.e., relevant pieces of information) into one well-known concept – many small “bits” of information are collapsed to into fewer larger chunks (Poolton & Masters, 2014). Although fewer chunks are processed, they contain the relevant information, meaning that information can be processed with relatively less cognitive effort and processing becomes more efficient. At the same time, motor analogies minimize accrual of conscious knowledge of the underlying rules governing the mechanics of movements (e.g., Liao & Masters, 2001). Learning movements implicitly through analogies has been shown to result in more robust motor performance under demanding conditions than learning movements by explicit step-by-step instructions (e.g., Koedijker et al., 2007, Lam et al., 2009; Liao & Masters, 2001; Schlapkohl et al., 2012; for a meta-analysis see Cabral et al., 2022), which has been attributed to the placement of fewer demands on cognitive resources than explicit motor learning. To date, motor heuristics and analogies have only been researched independently from each other. Notably, it has been shown that implicit motor learning through motor analogies did not only improve motor performance, but also decision making, potentially because implicit processing of movement execution left sufficient cognitive resources available for decision making (Lola & Tzetzis, 2021; Masters et al., 2008). Likewise, the use of heuristics (instead of a complex step-by-step planning strategy) showed combined selection and execution advantages, when participants were required to select one of two computer cursors that displace in different directions (i.e., selection) and subsequently navigate from a starting position to a goal as efficiently as possible (i.e., execution; Dundon et al., 2023). We aim to extend these findings by testing performance in demanding circumstances after promoting the economic management of information for both movement selection and execution via implicit learning of motor heuristics and motor analogies.

The current experiment is considered a “Proof-of-Concept” in which we aim to establish the feasibility and rationale for an integrated theory that has been developed as part of the German Research Foundation (DFG) project “Motor heuristics and movement analogies in performance and health” (PIs: Markus Raab and Laura Voigt, RA 940/27-1 Ι VO 2789/1-1). For successful sports performance, athletes must choose between actions (movement selection) and then execute the selected action skillfully (movement execution; Raab, 2017; Voigt et al., 2022). A table tennis player who can execute a backhand stroke effectively but does not know when to use the backhand stroke will not succeed in the sport. Equally, a player who knows when to employ the backhand stroke but cannot execute it effectively is unlikely to succeed. Movement selection and execution often need to be performed when time is limited, stress is present, more than one task needs to be processed, or when fatigue or unfamiliar situations occur. Given limited ability to process multiple channels of information, performers need to find ways to manage the wealth of available information economically to make appropriate decisions and execute the chosen movement skillfully in their discipline. The idea that economic management of information is central for successful performance is supported by a common understanding of expertise in sports: During the development of expertise, the nature of the knowledge structures that support motor performance gradually changes over time, with an increasing degree of implicit (unconscious) processing and a decreasing level of explicit (conscious) processing (Fitts & Posner, 1967). Explicit processes involve the retrieval of consciously accessible (declarative) knowledge and depend on working memory. Thus, highly explicit sport performance has often been shown to be disrupted by demands resulting from performance pressure or multiple task requirements (Masters & Maxwell, 2008; Nieuwenhuys & Oudejans, 2012, 2017). In contrast, implicit processes are faster and involve sophisticated complexes of procedural knowledge that can be applied without conscious thought, with greater automaticity and fewer demands for attentional resources (e.g., Anderson, 1983; Lewicki et al., 1992; Kal et al., 2018; Masters & Maxwell, 2004; Shiffrin & Schneider, 1977; Willingham, 1998). Implicit processes are therefore less dependent on working memory, which allows the expert to economically manage multiple streams of information for movement selection and movement execution while being taxed with other demands (for a review of the theoretical architecture and function of working memory, see Baddeley, 2003). But how exactly are these streams of information for movement selection and movement execution managed? In the current study, we aim to deliver a “Proof-of-Concept” for the integration of two theories – motor heuristics (Raab, 2017) and motor analogies (Masters, 2000, 2012). We aim to investigate whether implicit learning helps athletes to manage information during movement selection and movement execution and protects performance under demanding circumstances. Motor heuristics have their theoretical basis in the simple heuristics used for decision making (Gigerenzer & Goldstein, 1996) and advise “what” movement to choose (i.e., movement selection). Motor heuristics are fast and frugal decision-making strategies (i.e., rules of thumb) that exploit information in the environment by ordering pieces of information (i.e., cues) by their validity. Decision makers judge the cues’ validity (i.e., how often in the past a particular piece of information was helpful in making a choice), and then choose the option that is favored by the cue with the highest validity. In sports, a large body of empirical findings demonstrates that superior decision making by experts is characterized by focusing on fewer (task-relevant) options and higher-quality options and decisions (Basevitch et al., 2020; Belling et al., 2015; Laborde & Raab, 2013; Musculus, 2018; Musculus et al., 2021; Raab & Johnson, 2007), especially in demanding circumstances, such as time pressure, opponent pressure, and stress (Musculus et al., 2021). For heuristics to be successful, their use needs to be matched to the environmental structures. Leuker et al. (2018, 2019) showed that statistical regularities (i.e., risk-reward structures) can be learned without explicit instructions via incidental, unsupervised learning and then utilized as heuristics for decision making under uncertainty. Implicit learning processes can thus be an important aspect in building a representation of the environment that, in turn, guides choice behavior (Hertwig et al., 2022). Importantly, the so-called description-experience gap further suggests that learning environmental structures by experience is more beneficial for performance than receiving explicit information about the environmental structures (e.g., Hertwig et al., 2004; Armstrong & Spaniol, 2017). Motor analogies have their theoretical basis in the theory of implicit motor learning (Masters, 1992, 2000) and advise ‘how’ to move (i.e., movement execution). Motor analogies leverage a concept that is already well known, such as “strike the ball while bringing the bat up the hypotenuse of a triangle” in order to convey the complex structure of the motor skill (e.g., a table tennis topspin forehand; Liao & Masters, 2001). It has been proposed that they promote economic management of information for movement control by chunking fundamental technical information (i.e., relevant pieces of information) into one well-known concept – many small “bits” of information are collapsed to into fewer larger chunks (Poolton & Masters, 2014). Although fewer chunks are processed, they contain the relevant information, meaning that information can be processed with relatively less cognitive effort and processing becomes more efficient. At the same time, motor analogies minimize accrual of conscious knowledge of the underlying rules governing the mechanics of movements (e.g., Liao & Masters, 2001). Learning movements implicitly through analogies has been shown to result in more robust motor performance under demanding conditions than learning movements by explicit step-by-step instructions (e.g., Koedijker et al., 2007, Lam et al., 2009; Liao & Masters, 2001; Schlapkohl et al., 2012; for a meta-analysis see Cabral et al., 2022), which has been attributed to the placement of fewer demands on cognitive resources than explicit motor learning. To date, motor heuristics and analogies have only been researched independently from each other. Notably, it has been shown that implicit motor learning through motor analogies did not only improve motor performance, but also decision making, potentially because implicit processing of movement execution left sufficient cognitive resources available for decision making (Lola & Tzetzis, 2021; Masters et al., 2008). Likewise, the use of heuristics (instead of a complex step-by-step planning strategy) showed combined selection and execution advantages, when participants were required to select one of two computer cursors that displace in different directions (i.e., selection) and subsequently navigate from a starting position to a goal as efficiently as possible (i.e., execution; Dundon et al., 2023). We aim to extend these findings by testing performance in demanding circumstances after promoting the economic management of information for both movement selection and execution via implicit learning of motor heuristics and motor analogies.

This research study determined the relative importance of various basic motor ability traits possessed by the learner in the process of acquiring neuromuscular skills. Fifty-two university students practiced two novel skills (ball toss and fli-back paddle ball) 5 days per week for 2 weeks. Prior to beginning practice, each subject was measured on 26 basic motor ability items. Factor analysis was employed to isolate basic factors underlying each practice sequence and the gross motor performance traits. Ten stable factors were isolated, including two that described practice for the novel skills and eight that involved gross motor performance. Comparisons of the factor loadings among practice stages for each novel skill across the eight motor performance factors suggested the following conclusions: a) no relationship existed between gross motor factors and performance on the fli-back paddle ball skill; b) gross motor factors were related to performance on the ball toss skill during early, but not late, stages of practice; c) different gross motor factors related to performance on the ball toss skill as practice continued; and d) performance on the ball toss skill should be assessed only after several practice periods to prevent score comparisons. (Author/JA)