The effect of context and individual differences in human-generated randomness

Abstract

Many psychological studies have shown that human-generated sequences deviate from the mathematical notion of randomness. Therefore, the inability to generate perfectly random data is currently considered a well-established fact. What remains an open problem is the degree to which this (in)ability varies between different people and can be affected by contextual factors. In this paper we investigate this problem. We focus on between-subjects variability concerning the level of randomness of generated sequences under different task descriptions. In two studies, we used a modern, robust measure of randomness based on algorithmic information theory to assess human-generated series. We tested hypotheses regarding human generated randomness vis-a-vis effects of context, mathematical experience, fatigue, and the tendency to engage in challenging tasks. Our results show that the activation of the ability to produce random-like series depends on the relevance of contextual cues which rather help to avoid production of trivially non-random sequences than increase the rate of production of highly complex ones. We also show that people tend to get tired very quickly and after first few attempts to generate highly random-like series their performance decrease significantly and they start to produce more markedly patterned, sequences. Based on our results we propose a model that considers two main areas (intellectual and cognitive) of possible psychological factors related to the variability of ability to produce random-like series.

Embodied nonlinear dynamics of cognitive performance.

“…Two simple one-way ANOVAs across the four instruction sets, within each ITI condition failed to reveal significant differences in bias across instruction sets. We interpreted this as an indication that relatively minor variations in verbal characterizations of randomness had little influence on the participants’ performances (Biesaga et al, 2021). This outcome does not preclude relationships between instructions and performance, it simply respects the outcome at hand.…”

“…Recurrence analyses revealed how a relatively simple parameterization of deGuzman and Kelso’s (1991) discrete bimanual coordination model was sufficient to simultaneously approximate all three grain-scales of the participants’ sequences; the fine-grained permutation counts of the 1- to 5-trial sequences, the coarse grained (multiscale) expression of scaling noise, and the deterministic dynamics on intermediate grain scales. The model’s iterative, phase-attractive nature obviated the need for explicit trial-lagged, probabilistic look-back mechanisms that are often used to implement bias in randomness production models (e.g., Biesaga et al, 2021; Warren et al, 2018).…”

“…If it is difficult to unambiguously communicate "randomness" to research participants, then it is unsurprising that participants produce variations on nonrandom themes (Nickerson, 2002). By contrast, some cognitive processing explanations emphasize a variety of limitations in experience, attention, information processing, or working memory (e.g., Biesaga et al, 2021;Hahn & Warren, 2009;Warren et al, 2018). However, other cognitive processing explanations are framed in opposition to the self-same resource-based limitations: because randomness, itself, is memoryless, cognitive faculties such as attention and working memory are thought to introduce bias, from memory, to the production of random sequences (e.g., Baddeley, 1966;Tversky & Kahneman, 1974).…”