Embodied decision biases: individually stable across different tasks?

In everyday life, action and decision-making often run in parallel. Action-based models argue that action and decision-making strongly interact and, more specifically, that action can bias decision-making. This embodied decision bias is thought to originate from changes in motor costs and/or cognitive crosstalk. Recent research confirmed embodied decision biases for different tasks including walking and manual movements. Yet, whether such biases generalize within individuals across different tasks remains to be determined. To test this, we used two different decision-making tasks that have independently been shown to reliably produce embodied decision biases. In a within-participant design, participants performed two tasks in a counterbalanced fashion: (i) a walking paradigm for which it is known that motor costs systematically influence reward decisions, and (ii) a manual movement task in which motor costs and cognitive crosstalk have been shown to impact reward decisions. In both tasks, we successfully replicated the predicted embodied decision biases. However, there was no evidence that the strength of the biases correlated between tasks. Hence, our findings do not confirm that embodied decision biases transfer between tasks. Future research is needed to examine whether this lack of transfer may be due to different causes underlying the impact of motor costs on decisions and the impact of cognitive crosstalk or task-specific differences. Supplementary Information The online version contains supplementary material available at 10.1007/s00221-023-06591-z.

as fast as possible between both ends of the room for five trials before experimentation. To achieve an average distance of four steps before reaching the central zone, the starting line was shifted so that the distance between the starting line and the midpoint of the central zone was 0.22 m smaller than the mean walking distance of the first four steps in the calibration trials, but maximally 3.41 m because of the length of the room. The time constraint was calculated as the average time to make four steps in the calibration time plus 2.1 s (Grießbach et al., 2022). For calibration trials, we defined the trial start as the first time where one of the lateral malleoli markers exceeded a horizontal velocity of 0.1 m/s for 0.125 s consecutively, based on the difference (derivative) between the position of consecutive frames.
In the following familiarization trials, to familiarize with the time constraint the required time to finish was projected on the screen for the first six trials, and no auditory feedback was given regarding whether participants were too late or too early. In the following six trials and for the rest of the experiment, auditory feedback was provided to indicate whether participants were in time.

Online analysis
Identical to Grießbach et al. (2022), the following conditions had to be met for 180 frames (1.5 s) to start a trial: 1. Six markers had to be in an area around the starting line (-0.3 to 2.1 m horizontal, -0.3 to 0.3 lateral, and under 0.25 m height). If six markers were in the area, these markers were identified by the assumed starting position (left foot positioned left to the right foot, toe positioned more to the front than the lateral malleolus, lateral malleolus more to the front than the heel).
2. The most forward marker had to be close to the starting line (horizontal ± 0.05 m, lateral ± 0.3 m).
3. The predetermined foot had to be in front.
4. The malleolus marker stood still, i.e., was not displaced between consecutive frames for more than 0.004 m.
To present rewards at the third step, the touch-down (first contact of the foot with the ground) of every step was estimated kinematically (Banks et al., 2015). A touch-down was defined when the horizontal distance of the heel marker of the swing leg and the lateral malleolus marker of the stance leg reached a maximum, i.e., the horizontal difference of the position between two frames inverted from positive to negative. To ensure one maximum per touchdown, the analysis of touchdowns was paused for 0.125 after finding a touchdown. To check whether participants stepped into the central zone, we compared the position of the lateral malleolus marker of all touch-downs to the area of the central zone. If the participant did not step into the central zone a warning message appeared centrally on the projected display ("Markierung beachten" in German, freely translated as "Note marking"). A trial ended if more than four markers were in the target area (more than one foot. The time from trial start to the end was measured with MATLABs intern stopwatch timer ("tic", "toc") and used as comparison and feedback to check whether participants finished in time.

Offline analysis
To identify the correct foot stepping into the zone, we checked the kinematic data visually for trials in which: 1. the foot in the mark was not defined by the online analysis, 2. a second self-written velocity-based algorithm for estimating the touch-down did not agree with the distance-based algorithm from Banks et al. (2015), 3. the toe of the foot first passing the beginning of the zone did not agree with the foot touchdown in the zone, 4. rewards were presented too late when the foot was already in the central zone, 5. participants switched sides rapidly (0.25 s) before a trial finished, as an indication of ignoring the obstacle.
6. a trial was not finished, or there was no kinematic data available, Here, a switch to the upper lane is depicted (i.e., a lower-cost switch). This is comparable to participants choosing a lateral step in the TWWT (see fig. S1). The reward on the respective lane was collected automatically. For original video footage, the reader is referred to the online repository of this study.

Determination of the movement threshold in the MLTT
Given the mouse dpi setting of 1200, a constant weight for the position of the bird avatar of 0.36, and a mouse cursor sensitivity setting of

Data Analysis
Moreover, we used weakly informative priors for all parameters. To determine these priors, we relied on prior predictive checks on the probability scale, aiming for a roughly uniform distribution for decisions. For the intercept we used a normal distribution (µ = 0, σ = 1).
For all other regression coefficients, we used a slightly narrower normal distribution (µ = 0, σ = 0.5). We used an exponential distribution for the standard deviation of all random effects ( = 1).
This results in the following formula for the model to be fitted to the data (McElreath, 2019): indicates the normal distribution with parameters µ (mean) and (standard deviation), Exp the exponential distribution with rate parameter , and LKJ indicates the Lewandowski-Kurowicka-Joe distribution with shape parameter . For fitting this model, we used the following notation as input for the formula to brms in our R script: