We used the custom-made experimental apparatus described in Levy-Tzedek et al. (2010, 2011a,b,c), Ben-Tov et al. (2012), and Levy-Tzedek (2017). It is comprised of an arm rest, connected to an encoder that records the flexion/extension angle of the elbow with an accuracy of 0.002° per count at 200 Hz. The arm rest was designed to be as light as possible, to minimize its effect on the natural behavior of the limb. The arm rest was free to rotate around its axis, with no imposed limits on the range of motion. Participants placed their arm on the arm rest, positioned parallel to the table on which it was mounted, and moved their forearm toward and away from their body in a movement similar to that of a windshield wiper. An opaque cover was placed on the table, above the experimental apparatus, such that direct visual information about the arm’s position was precluded. Participants received real-time feedback on their arm’s location via a cursor that appeared on a computer screen that was placed on the table in front of them (see Figure 1C).

Participants’ arm was placed in a wrist brace, to prevent movements of the wrist. The forearm flexion/extension movements were one-dimensional, in the horizontal plane. No explicit timing cues were given.

The computer display presented a phase plane, such that position was displayed on the X-axis, and velocity on the Y-axis (see Figure 1D). On that phase plane, a pair of black concentric ellipses created an enclosed doughnut-shaped area, that defined the lower and the upper bounds of the amplitude (on the X-axis) and speed (on the Y-axis) allowed on the task. The participants controlled a cursor on the phase plane with their arm movements. They were asked to maintain the cursor within the doughnut-shaped area at all times (see yellow trace in Figure 1D). Thus, the amplitude, speed and frequency of the movement were prescribed.

The experimental protocol sequence was as follows: the participants started each trial at a neutral position, from which they could comfortably flex and extend their forearm, and this neutral position defined the center of the concentric ellipses (i.e., it is the point of zero position and zero velocity, prior to initiation of the movement).

The angular displacement of the forearm about the elbow joint, was filtered using a first order Butterworth filter with a cutoff frequency of 20 Hz. Trend was removed from the position data to reduce the effects of drift. This was achieved by removing the best straight-line fit from the angular position data. Data were analyzed with MATLAB^® software (Mathworks, Natick, MA, United States, v.9.1). To quantitatively determine the movement type performed by participants at each time point, we used the measure of harmonicity (Guiard, 1993, 1997; Buchanan et al., 2006; Hogan and Sternad, 2007; Levy-Tzedek et al., 2010, 2011b). The harmonicity index is a unitless number that is calculated for each movement half-cycle between two zero-crossings in the position record (see Figure 3). It is used in order to classify a movement as either highly rhythmic (when the harmonicity value is equal to 1, or close to it) or as containing discrete features. When a single peak in acceleration occurs in the half-cycle, the harmonicity value is set to one; when an inflection occurs in the half-cycle acceleration trace, the harmonicity is computed as the ratio of the minimum to the maximum absolute values of the acceleration within the given half-cycle; finally, if the acceleration trace within the half-cycle changes its sign, the harmonicity value is set to zero (Guiard, 1993, 1997). For instance, the harmonicity value calculated for the velocity trace shown in Figure 1A is 1, and the harmonicity value for the trace shown in Figure 1B is 0. The harmonicity index has previously been demonstrated to be a robust indicator of movement type (see, for example Levy-Tzedek et al., 2010).

Peaks in acceleration were found as follows: the jerk (the third derivative of the position) of the movement was calculated and was scanned for changes in sign. Each change in the sign of the jerk indicated a peak in acceleration occurred at that time point.

Harmonicity values were used to determine the frequency (Fs) at which a switch occurs between movement types (Levy-Tzedek et al., 2011b). In trials where the frequency of the movement continuously increased, the harmonicity values were scanned until the first instance of four consecutive half cycles had a harmonicity value above 0.5. That point was considered as the switch point between the two movement types. The cutoff value of harmonicity = 0.5 was chosen to comply with convention (Guiard, 1997; Buchanan et al., 2003, 2004, 2006; Levy-Tzedek et al., 2011b). Multiple movement cycles were examined to ensure that a switch point – rather than a momentary change in the nature of the movement – was indeed encountered.

If, in a given INC trial, there were no four consecutive half-cycles with a harmonicity value greater than 0.5, the trial was considered a no-switch (NS) trial.

The same method was applied for the trials in which the frequency was decreasing, but in this case checking for harmonicity smaller than 0.5. Similarly, if, in a given DEC trial, there were no four consecutive half-cycles with a harmonicity value smaller than 0.5, the trial was considered a NS trial.

This method for detecting the frequency at which the switch between movement types occurs has previously been validated by using an alternative method (a sigmoid function fit) for switch-point detection (Levy-Tzedek et al., 2011b).

The Wilcoxon signed rank test for paired observations was used to compare the frequencies at which participants switched between the two movement types on INC vs. DEC trials. Data from participants who did not have a switch point at all in all four DEC trials or all four INC trials, or both, were not included in this analysis. Furthermore, a conservative approach was taken, and in cases where there was a single INC or a DEC trial with a switch point, out of the four trials, such that an average between two or more readings could not be taken, this participant’s data were not included in the statistical analysis, to avoid bias due to sampling error. A non-parametric test was chosen to eliminate the need for assumptions regarding population distributions required in parametric tests.

Both age groups performed on average about five practice trials during the training phase.

The velocity profile from two trials of a young participant – one INC trial and one DEC trial – are shown in Figure 4.

Participants performed on average about four practice trials during the training phase.

An implicit task is often defined as one where the ability to explicitly express “what” procedures are performed is diminished or non-existent (such as how to ride a bicycle) (Vidoni and Boyd, 2007), and thus a concurrent cognitive demand should not interfere with its performance. It may seem paradoxical to claim that a cognitive load, or cognitive decline, can affect the performance of an implicit task. However, a closer look at the literature reveals that many motor tasks contain both implicit and explicit components (several examples are reviewed in Reschechtko et al. (2016). Reschechtko et al. (2016) even showed that the same group of muscles, performing a single task, can have different output variabilities on instructed (explicit) vs. non-instructed (implicit) elements of the task. As a result, cognitive state can play a part in the outcome of motor tasks that contain both implicit and explicit components. Indeed, adaptive motor learning, which was once believed to depend solely on implicit memory, has been shown to decline with age as a function of explicit memory performance (Trewartha et al., 2014). Another example of this seemingly paradoxical combination is the effect of cognitive load on APAs. As discussed in the Introduction section, the studies on APAs showed that control of explicit elements (such as bending) is done separately from the control of implicit elements within the same movement (such as maintaining balance during the weight shift), supporting the dual-process hypothesis (Slijper et al., 2002). Indeed, studies on the effects of cognitive load on postural adjustments reveal degraded performance in both in young and in old participants (Shumway-Cook et al., 1997; Melzer et al., 2001). Thus, the current finding, that a cognitive load (and potentially a cognitive decline) affected performance on this task can be explained by recognizing that the task has both implicit and explicit components to it. These are, respectively: switching between motor tasks (rhythmic and discrete), and cyclically moving their forearm to control a cursor on the screen at changing speeds.

The hysteresis phenomenon is commonly found in nature, and can be described as a limited tendency to stay at a given state, despite a changing environment. It has the benefit of avoiding rapid oscillations between states around the switch point between these states, which may have an associated cost (Alexander, 1989). A cost of switching between tasks has been demonstrated in a wide variety of experiments involving both motor and cognitive tasks (e.g., Monsell, 2003; Hirsch et al., 2016).

Most examples found in nature are of classical hysteresis, while reverse hysteresis is less common. A plausible explanation for the reverse hysteresis pattern found in the current, as well as a previous study (Levy-Tzedek et al., 2011b), is that predictive control guides the early switching between movement types, in anticipation of the future. Predictive control is a form of control that incorporates a prediction of the future behavior of the system. The ability to identify a change in the environment and adapt to it in feed-forward manner, rather than rely on the inherently delayed feedback-based response is important. It can serve as a means to avoid injury, or even life-threatening situations (Porr and Worgotter, 2003; Gilbert and Wilson, 2007).

In both Experiments 1 and 2, we found that participants performed a mix of classical-hysteresis (positive hysteresis) and reverse-hysteresis switching patterns. This mix is found for young individuals when studying the walk-to-run (WR) and run-to-walk (RW) transitions in the lower limb. Experimental evidence showed either a classical hysteresis (e.g., Hreljac, 1995), revealed no difference between the WR and RW transitions speed (Prilutsky and Gregor, 2001), or showed reverse hysteresis (Getchell and Whitall, 2004). Diedrich and Warren (1995, 1998) reported individual behavior ranged from a significant classical hysteresis to a significant reverse hysteresis. Evidence for reverse hysteresis is found in Thorstensson and Roberthson (1987) as well, but significance values are not reported. Getchell and Whitall (2004) suggested that the reverse hysteresis phenomenon is the result of the effect of participants’ intention on the coordination dynamics, lending support to our predictive control hypothesis. Researchers have recently theorized that a classical hysteresis pattern emerges in gait patterns when participants are actively engaged in the task, and a reverse hysteresis pattern results when they are asked to evaluate how they would perform the task, without actually performing it (Abdolvahab and Carello, 2015). This theory is not supported by our upper-limb data. A study on arm movements, using an explicit timing cue, reported weak evidence of a classical hysteresis pattern in total movement time, manifest as a greater tendency to maintain a higher-frequency movement, even as the pace of the metronome slows down (Sternad et al., 2013). However, that study used different parameters for describing the arm movement at different frequencies, which preclude direct comparison with the current data.

Thus, data from both the upper and the lower limb are consistent with an interpretation that people often engage in predictive control of movement, and change their control strategy in preparation for a change in the environment, rather than perform this change only in feedback form, after the environmental change has taken place. This behavior can be advantageous, as it bypasses the inherent delays in feedback-based reactions to a changed environmental context.

Recent evidence from behavioral studies further supports this interpretation. Rosenbaum et al. (2014) have demonstrated in a series of experiments the propensity of participants to start a task early, even at the expense of extra physical effort. They termed this unexpected behavior “pre-crastination.” This can be envisaged as yet another form of “pre-acting.” That is, making a change to one’s motor plan in anticipation of future demands.

The results from Experiment 2 suggest that there is a cognitive component to the predictive control of movement. This result is supported by research showing that predictive control of saccadic eye movements is affected by a cognitive load (Lehtonen et al., 2012). The finding that both age and a cognitive load had a similar effect on the predictive control of movement might have suggested that potentially a decline in cognitive abilities underlies the decreased ability to predictively control movements with older age. While this is not reflected in the Mini-Mental scores of the participants in the current experiment, it would potentially be manifest in finer tests that examine executive control directly. The hypothesis that cognitive decline may impair predictive control should be examined in a separate study.

There are several changes that underlie aging, including changes in cognition and in motor-output control (Beauchet et al., 2007; Kilby et al., 2014; Sternäng et al., 2015) that affect activities of daily living, such as driving (Lodha et al., 2016). Can we identify the specific difficulties that lead to reduced performance? In one study, older adults have been found to not take advantage of color cues to predictively control their grip force when lifting an object (Cole and Rotella, 2002). Other work suggests that for older adults, the cost of switching between certain tasks is higher than for younger adults (Hirsch et al., 2016), which would suggest an inertial tendency (Rosenbaum et al., 2007) to maintain a motor plan without switching it, if possible. In other words, older adults appear to employ a more conservative switching strategy. This view is in line with other research suggesting that aging is accompanied by an increased reliance on slower feedback-based control of movement, as opposed to rapid feed-forward control (Bagesteiro and Sainburg, 2009).

Our findings further support these suggestions, in two ways: (1) more participants in the older group showed a positive-hysteresis pattern than in the younger group; and (2) some participants in the older group (and not in the young group, Experiment 1) failed to switch between movement types altogether, despite this being a requirement of the task. It suggests that, for the older adults, a high switching cost between motor plans (Monsell, 2003) leads to perseverance in performing the same type of movement (rhythmic or discrete) over longer extents of time.

And yet, we found that for 40% of the participants in the older age group were successful in predictively controlling their movements, and they displayed a reverse-hysteresis switching pattern. These results are in accordance with recently published data on APAs, showing that aging does not impair these preparatory adjustments (Plate et al., 2016), and with several studies on anticipatory control of grip force in older adults, showing relatively intact anticipatory control during simple grasp manipulations (for a review, see Diermayr et al., 2011). Together, these findings suggest that implicit predictive control of movement is not uniformly affected by age. Future studies should examine a potential correlation between cognitive function and the capacity for predictively controlling movement.

As noted in the introduction section, we specifically chose a task that was previously unfamiliar to the participants, so that any training effects from past experience would not be a factor in task performance. Moreover, we employed a task that calls for an implicit, unaware switch between motor plans, to eliminate the process of a conscious decision making, and get at the basic pattern of predictively switching between motor plans. The task being different from standard activities of daily living, we can only speculate as to whether the results from the current study generalize to familiar, explicitly performed task switching in everyday life. The agreement between previous studies (e.g., Melzer et al., 2001; Diermayr et al., 2011; Trewartha et al., 2014; Hirsch et al., 2016; Plate et al., 2016) and the current results suggest that that would be the case.