Exploring the relationship between motor planning and initiation J. Teman 1 , R. Ivry 1,2 , I. Greenhouse 3 Introduction 1 Department of Psychology, 2 Helen Willis Neuroscience Institute, University of California, Berkeley; 3 Department of Human Physiology, University of Oregon Haith, A. M., Pakpoor, J., & Krakauer, J. W. (2016). Independence of Movement Preparation and Movement Initiation. The Journal of Neuroscience : The Official Journal of the Society for Neuroscience, 36(10), 3007–3015. https://doi.org/10.1523/JNEUROSCI.3245-15.2016ipsum Imposing response deadlines yields faster response times (RTs) at the expense of increased error rates, a speed-accuracy tradeoff (SAT). Sequential sampling models (SSMs) assume that evidence for an action accumulates toward a deci- sion boundary. The SAT would emerge in these models given the assumptions that task instructions modulate the boundary and that a response decision nec- essarily precedes response initiation. Haith et al. 2016 demonstrated an exception to the traditional SAT by showing that, when forced, people can respond faster than their voluntary (Free) RTs without an increase in errors under certain conditions. This violation of the tradi- tional SAT is consistent with the existence of independent motor preparation and motor initiation processes that can unfold in parallel. Here, we set out to replicate this finding using a two-alternative forced choice task framework that allowed us to employ a hierarchical drift-diffusion SSM model, one commonly used to separate out decision and non-decision compo- nents of RT. Free, Pressured, and Forced responses Accuracy (%) 80 90 100 Button (top) & EMG (bottom) RT Free Pressured Forced 0 0.1 0.2 0.3 RT (s) Task Conditions Lorem ipsum random response accurate response Movement planning Movement planning motor initiation movement initiation Stimulus encoding & Neural conduction delays Stimulus encoding & Neural conduction delays Serial Parallel Stimulus onset time Response time movement initiation Δ Δ Δ primary determinant of RT accuracy? Time Model comparisons Acknowledgments Hierarchical drift-diffusion model Pressure but not Force decreases movement times Forcing speeded responses inflates RT variance Motivation in the Pressured task results in a standard speed-accuracy tradeoff favoring speed at the cost of accuracy. Replicating Haith et al. 2016, participants in the Forced condition demonstrated a higher proportion of accurate responses at lower RTs. 16 0.1 0.2 0.3 0.4 0.5 0.6 0.7 RT (s) 0 2 4 6 8 10 12 14 Probability Density 0 0.5 0.1 0.15 0.2 0.25 0.3 0 5 10 15 20 RT (s) Probability Density Haith et al. 2016 Free speed? Independence model Drift rate Non-decision time Boundary separation Posterior of boundary group means Posterior of t group means Posterior of drift-rate means drift rate (v) non decision (t) boundary (a) Model fits Higher drift rate predicts reduced RT without loss of accura- cy and indicates a faster accumulation of evidence. Drift rate was estimated to be higher according the HDDM re- sults for the Forced condition. Decreased boundary separation predicts faster and less accurate responses. This is consistent with what was ob- served for the Pressured condition. Left target not shown. target target onset window timing fixation “go” cue coincidence timing threshold window failure to respond before target disappears RT Effective RT time (ms) RT Free Pressured Forced “Too soon!” “Too late!” -300 -35 0 35 feedback feedback e e e e e e e e e Pressure significantly reduced movement time as deadlines became increasingly stringent (0.8 to 0.35, 0.35 to 0.3, 0.3 to 0.2; t-test, p < 0.05). The mean reduction in movement time was (27 ms, p < 0.05) , while the Forced exhibited no such trend. Pressured movement time also decreased significantly relative to the Forced (~20%, p < 0.05). Measurements of Forced response times in terms of “effective RT” (see Task Conditions) show that savings in accuracy come at the cost of increased RT variability. An inherent challenge with the Forced design is that effective RT measurements are necessarily variable due to the nature of the uniform target presentation; however, the variability within the observed window is higher than theoretical predictions. Pressured responses had significantly higher between-subject variability relative to the Forced (F-test, p < 0.05). Three parameters describe diffusion toward a boundary for a two-alternative choice decision: 1. drift rate (v) 2. non-decision time (t) 3. boundary separation (a). The non-decision time parameter captures a combination of stimulus perception and response execution time. A decrease in non-decision time suggests reduced RT without loss of accuracy. However, the non-decision parameter estimates for the Forced and Pressured conditions are unreasonable given our movement time data. Free Pressured Forced Traditional sequential sampling models assume serial processes determine RT. Alternatively, independent movement planning and movement initiation processes may happen in parallel to permit fast and accurate responses. Three task conditions were used. The Free condition measured voluntary responses in the absence of any imposed timing constraints. The Pressured condition permitted responses only within a limited response interval. The Forced condition required participants to execute a response at a given time regardless of whether information about which response had been provided. In the Forced condition accuracy, effective RT, and movement time analyses were performed in four selected intervals (-0.3 to -0.2, -0.2 to -0.186, -0.186 to -0.1, -0.1 to 0; in seconds) for comparison against the four harshest deadlines in the Pressured condition (saturated below for emphasis). Empirical speed-accuracy tradeoff curves were obtained by calculating the moving average of the probability that a response was successful given a 50 ms RT window. Significance was calculated according to a binomial test (p < 0.00001). Probability of accurate movement over 50ms windows 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 RT (s) 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 Probability of accurate movement Not significant ns * * Discussion Consistent with Haith et al., 2016, RT data support parallel movement planning and initiation processes during the performance of a two-alternative choice task. Movements were executed more quickly when the allowable time for responses decreased (Pres- sured). In contrast, movement time variability increased when a fixed response initiation time was imposed (Forced). This may reflect a three-way tradeoff between movement speed, response accuracy, and movement time variability. HDDM may compensate for parallel response processes with unrealistic non-decision time parameter estimates. • • • Assumptions Reaction time is determined by an initiation pro- cess independent from motor planning The underlying preparation distributions gener- ated by Free, Pressured, and Forced conditions are similar • • Pressured included seven blocks of progressively harsher deadlines in seconds (2.0, 0.8, 0.35, 0.3, 0.2, 0.186, 0.1) followed by an eighth block, like the first, with effectively no deadline (2.0 s), used to separate performance increases due to learning from motivational influence. Forced target presentation followed a uniform distribution (-0.3 s, 0 s) relative to a fixed initiation deadline. In the independence model, T I is a random variable representing the time motor initiation is complete and T p represents the independent random time motor preparation is complete. Above: a plot from Haith et al. 2016 showing T I, obtained directly from their Free RT data, and T p, , which they estimated from their Forced effective RT data according to the probability model above, where P(H | RT) stands for the probability of success on a trial given an RT with the preparation parameters that maximize the summed log likelihood of this function across trials. On the left: T I and estimated T p distributions obtained with our data according to this model. Free Pressured Forced 40 60 80 100 Accuracy (%) Response accuracy Response accuracies were calculated as the proportion of correct responses in each bin. Free 2 0.8 0.35 0.3 0.2 0.186 0.1 2 0.3 0.2 0.186 0.1 (s) Pressured Forced ns ns * 0.1 0.2 0.3 0.4 RT (s) 0.02 0.04 0.06 0.08 0.1 0.12 SD (s) * Response times were calculated as the offset between target presentation and button presses using the right or left index finger. Response times Free 2 0.8 0.35 0.3 0.2 0.186 0.1 2 0.3 0.2 0.186 0.1 (s) Pressured Forced Movement times ns ns ns ns ns 0 0.02 0.04 0.06 Movement time SD (s) 0.06 0.08 0.1 0.12 0.14 0.16 Movement time (s) * * * * * Movement times were calculated as the difference between button press RTs and EMG onset RTs recorded from the FDI of the left and right hands. Free 2 0.8 0.35 0.3 0.2 0.186 0.1 2 0.3 0.2 0.186 0.1 (s) Pressured Forced Model This work was generously funded by NIH grant NS092079. We thank Leah Carrol with the Haas Scholar’s Program for her support and Weixin Liang for assistance with data collection.