Reviewer #2: This study addresses a question that is important both theoretically and practically. However, the authors need to rule out the following, less interesting alternative. Namely, the results could be due to task practice effect, as follows.
Since there was no pre-training test, and no practice trials (as far as I can tell), and since the task was an online motor task that participants could not rely on their prior motor experience, trying to launch the ball to the target could only be done via trial and error. For the varied training group, they got to practice at two distances. Therefore, they had a better “calibration” in terms of the relationship between launching speed and target distance. This was likely beneficial both in Exp.1 when both transfer distances were interpolations from the two trained distances, and in Exp.2 when two transfer distances were interpolations and two were extrapolations but the latter two were immediately next to the training distances.
In comparison, since the constant group trained at only a single distance, any transfer distance (or at least the first transfer distance tested) was extrapolation even if this transfer distance was shorter than the trained, because the participants did not know beforehand how to shoot the ball to the shortest distance due to the existence of the barrier. If the transfer distance was longer, for sure that was extrapolation.
Regardless, the above analysis suggests that the constant group would always be a step behind the varied group. The number of trials at each transfer distance may not be sufficient for them to catch up the varied group either (whether there was learning during testing should be checked). If such disadvantage for the constant group is indeed due to the lack of tryout opportunities, then the authors should verify whether the same results still hold if all groups were provided opportunities to practice, or if pre-training tests across all distances were offered.
exponential learning models fit to individual subjects
Reviewer 3 Absolute versus relative distance: From a methodological standpoint, I understand the need to differentiate these two types of distance. However, from a theoretical perspective there may be some issue in differentiating these two concepts. Schema theory relies on relative (or invariant) information to inform the motor program. However, both distances would be important to an instance or exemplar representation. You may want to consider commenting on this issue.
Reviewer 2 For the same reason, the plots showing improvement during training could be due to participants learning the task, rather than fine motor skills. Although task learning and motor learning are impossible to separate cleanly, the common practice in the field is indeed to offer practice trials to reduce the task learning aspects. The authors should address this.
In addition to absolute errors (which is related to variance), the authors should also provide other measures of performance, e.g., the mean of the signed errors, so that readers have a better idea whether there was any meaningful over- or undershooting.
=========================================================================
Condition 610 760 835 910
-------------------------------------------------------------------------
Constant Training 25.28(158.98) 50.82(217.48) 73.14(250.93) 50.76(313.77)
Varied Training 13.85(116.87) 50.59(169.59) 50.52(217.39) 49.94(237.71)
-------------------------------------------------------------------------
Experiment 2 Testing - relative distances
Experimenet 1 - intermittent testing
Code
intTest.half <-readRDS(here::here("data/e1_intTest.rds"))intTest.half %>%ggplot(aes(x=positionX,y=MeanTargetDistance))+geom_bar(aes(group=trainHalf,fill=trainHalf),stat="summary",fun=mean,position=dodge)+facet_wrap(~conditType,ncol=2)+stat_summary(aes(x=positionX,group=trainHalf),fun.data=mean_se,geom="errorbar",position=dodge,width=.8)+ylab("Mean Distance From Center Of Target")+xlab("Intermittent Testing Throw Location")+theme(plot.title =element_text(hjust =0.5))+guides(fill=guide_legend(title="Training Stage"))+theme(legend.title.align=.25)
# possible that scaling required loading special package from devtoolsexp1Train <- e1 %>%filter(stage!="Transfer",mode==1) %>%group_by(Group,sbjCode) %>%mutate(scaleDev=scale_this(AbsDistFromCenter)) %>%ungroup() %>%group_by(Group,sbjCode,stage,conditType)exp1Train = exp1Train %>%summarise(MeanTargetDistance=mean(AbsDistFromCenter),scaledDist=mean(scaleDev,trim=.05))exp1Train$stage <-factor(exp1Train$stage, levels =c("Beginning", "Middle", "End")) #in case the levels get out of orderexp1TrainTrials <- e1 %>%filter(stage!="Transfer",mode==1,trialType!=44) %>%group_by(Group,sbjCode,positionX) %>%mutate(scaleDev=scale_this(AbsDistFromCenter),ind=1,trainIndex=cumsum(ind)) %>%ungroup() %>%group_by(Group,sbjCode,stage,conditType)# manuscript plot - originalggplot(data = exp1Train, aes(x=stage, y=MeanTargetDistance)) +geom_boxplot(aes(fill=conditType),position=position_dodge(1))+stat_summary(fun="mean",aes(group=conditType),position=position_dodge(1))+ylab("Mean Distance From Center Of Target") +xlab("Training Stage")+theme(plot.title =element_text(hjust =0.5))+guides(fill=guide_legend(title="Training Condition"))+theme(legend.title.align=.5)+theme_classic()
Code
lineplot.CI(data=exp1Train,x.factor=stage,group=conditType,response=scaledDist,xlab="Training Stage",x.leg=2,legend=TRUE,ylab="Distance from Target (scaled)",main="Training Performance - Experiment 1",col=c("red","black"))
Code
lineplot.CI(data=exp1Train,x.factor=stage,group=conditType,response=MeanTargetDistance,xlab="Training Stage",x.leg=2,legend=TRUE,ylab="Distance From Target",main="Training Performance - Experiment 1",col=c("red","blue"))
Not in manuscript
fit to testing performance averaged across positions