Choice perseveration in value-based decision making: The impact of inter-trial interval and mood
Introduction
Value-based and perceptual decision making in humans is often characterized by choice biases (Erev, Plonsky, Cohen, & Cohen, 2017; Hunt, 2014). Such choice biases can be split into two broad categories. One class subsumes biases that take effect due to properties of the actual decision such as framing, loss aversion, over-weighting of rare events (Erev et al., 2017; Kahneman & Tversky, 1979), and previous pay-off information (Noorbaloochi, Sharon, & McClelland, 2015). The second class subsumes biases that operate due to properties of the decision environment such as sunk costs (Bogdanov, Ruff, & Schwabe, 2015; Haller & Schwabe, 2014), a status quo (Nicolle, Fleming, Bach, Driver, & Dolan, 2011; Samuelson & Zeckhauser, 1988), and sequential effects caused by the responses in previous decisions (Bonaiuto, De Berker, & Bestmann, 2016; Hämmerer, Bonaiuto, Klein-Flügge, Bikson, & Bestmann, 2016; Scherbaum et al., 2016; Soetens, Boer, & Hueting, 1985; Soetens, Melis, & Notebaert, 2004).
In this paper, we put the focus on choice perseveration, that is, the tendency to stick to the present choice in future decisions. Choice perseveration and more complex sequential effects have been extensively studied in the context of perceptual decision making (Berlemont & Nadal, 2019; Cho et al., 2002; Fründ, Wichmann, & Macke, 2014; Gao, Wong-Lin, Holmes, Simen, & Cohen, 2009; Soetens et al., 1985). However, this important aspect is often neglected in empirical and modeling work studying value-based decision making, where trials of interest are analyzed in isolation. But why should sequential effects only occur in perceptual and not in value-based decisions? Both in real life and in the laboratory, people usually make many subsequent decisions. Thus, looking at the process dynamics of how one decision affects the next one can complement research on value-based decision making and inform future model developments. Here, we focus on choice perseveration as one example of first-order sequential effects (for an overview of other sequential effects and their mechanisms, see Gao et al., 2009).
We previously described such a dynamic process model that explains how choice perseveration emerges from sustained residual activity from the previous decision (Scherbaum et al., 2016). With the present study, not only aim to replicate this finding but we derive new predictions on choice perseveration and test them empirically. Before we turn to the new predictions, in the next paragraph, we will briefly describe the model as well as how it predicts and explains choice perseveration (for details, see Scherbaum et al., 2016, and the Appendix A).
The model is inspired by a value-based decision task in which people choose between rewards that are small in value and only a short distance away (small/near or SN option), and rewards that are large in value, but a larger distance away (large/far or LF option). For modeling purposes, we assume that people generate subjective values for each option. Taking this as a given, we will focus is on the process dynamics of how the decisions unfold over time (Spivey and Dale, 2004, Spivey and Dale, 2006; van Rooij, Favela, Malone, & Richardson, 2013).
In the model, options are represented as self-sustainable active patterns that constitute attractors in the neural system's state space (Reilly, 2006; Rolls, 2010; Scherbaum et al., 2016; Scherbaum, Dshemuchadse, Ruge, & Goschke, 2012). The dynamics of such models are often illustrated by conceiving of the system's state space in analogy to a potential landscape, where the stability of a state depends on the depth of the attractor (Rolls, 2010; Scherbaum, Dshemuchadse, & Kalis, 2008). In our model, the depth of the attractors is determined by the relative attractiveness of the SN and LF options, i.e. their relative subjective value. Fig. 1 depicts three kinds of possible attractor layouts. The left and right panel of Fig. 1 reflect almost exclusive activation of one option's representation (left panel: SN clearly more attractive; right panel: LF clearly more attractive), and hence illustrate configurations of the system with a preference towards one option. The middle panel of Fig. 1 represents a decision in which both options receive an identical input and are thus equally attractive; here, random noise would tip the system towards one or the other decision state (please see the Appendix A for the formal computational implementation of such a neural system).
A major advantage of such an attractor model is that it predicts dynamics of decision making on different time scales. In our model, the decision is not only determined by the current attractor layout, which is in turn determined by the currently offered options, but also by the history of the system's previous decisions (see Fig. 2) (Scherbaum et al., 2008; Townsend & Busemeyer, 1989). Due to the inertia of the model, the system temporally reclines in the area where it ended up previously—in the vicinity of the vanished attractor representing the recent choice—and relaxes only slowly to the neutral start state under no input. This process explains how choice perseveration emerges. For example, if the model chose the SN option in a first decision trial, it would remain near the SN option's attractor in the inter-trial interval. In a second decision trial, it would hence start the decision with a bias to the previously chosen SN option, even if this trial comprises the LF option being more attractive (see Fig. 2). This choice perseveration bias can be so strong that, in a series of choices where the initially unchosen option becomes increasingly more attractive, people switch to this now more attractive option much later than they would if their choices were unbiased by the previous decision. This phenomenon is called hysteresis or path-dependence (Rączaszek, Tuller, Shapiro, Case, & Kelso, 1999; Tuller, Case, Ding, & Kelso, 1994).
In our previous study, we demonstrated that the model indeed successfully predicts choice perseveration in human participants. Now, we identify potential modulators of choice perseveration and compare model predictions to human decision making. We focus on two modulators: the inter-trial interval (ITI) and mood, for which we derive the following hypotheses.
First, we expect to replicate our previous finding (Scherbaum et al., 2016) that people show choice perseveration in our value-based decision making task (H1).
Second, we expect the ITI to influence choice perseveration. The ITI determines how much residual activity from the previous decision is present at the start of the new decision. If the system had as much time as it needs to fully relax to the neutral start state between two options, no residual activity should be present at the time of the next decision, and choice perseveration should vanish. In contrast, if the system only had a short time span to relax before the start of the next trial, then the residual activity would be maximal, leading to strong perseveration effects. Thus, choice perseveration should decrease with increasing ITI (H2).
Third, we expect influences of mood.2 A manipulation of mood should modulate choice perseveration because mood affects the overall flexibility of behavior. Specifically, positive mood has been linked to increased flexibility and decreased perseveration (Baumann & Kuhl, 2005; Dreisbach & Goschke, 2004; Nadler, Rabi, & Minda, 2010; Wang, Chen, & Yue, 2017; Zwosta, Hommel, Goschke, & Fischer, 2013). Goschke and Bolte (2014) suggest that this increased flexibility might be modulated by the dopaminergic system, for example through changes in the gain parameter (Servan-Schreiber, Printz, & Cohen, 1990), as both mood and gain have been associated with changes in the dopaminergic system (mood and dopamine: Ashby, Isen, & Turken, 1999; Dreisbach & Goschke, 2004; gain and dopamine: Herd et al., 2014; Li, Lindenberger, & Sikström, 2001). In the context of the attractor model, the gain parameter determines how strongly a system responds to input and how stable—and thus how hard to overcome—a certain state is (Herd et al., 2014; Servan-Schreiber et al., 1990). Thus, similarly to positive mood, a lower gain has been associated with decreased perseveration and increased flexibility (Eldar, Cohen, & Niv, 2013; Eldar, Niv, & Cohen, 2016; Herd et al., 2014). Thus, we expect positive mood to lead to decreased choice perseveration, possibly due to a decreased neural gain (H3).
Additionally, through the manipulation of people's mood, decision making in general might be modulated irrespective of perseveration (Phelps, Lempert, & Sokol-Hessner, 2014). Negative mood has been linked to more localized or focused, and hence more analytic processing (for a review see Mitchell & Phillips, 2007), which should increase participants' ability to detect optimal (i.e., best ratio of costs and benefits) choices in our value-based decision making task (H4a). Following from that, we further expect that participants in the negative mood group make more optimal choices than participants in the positive mood group (H4b).
We have simulated both the ITI and the gain manipulation in our neural attractor model (see Appendix A) to validate our verbally derived hypotheses for choice perseveration. As expected, the model simulations show that choice perseveration decreases with increasing ITI and with decreasing gain.
Section snippets
Participants
Fifty-nine female students (mean age = 21.32 years, SD = 4.34 years) of the Technische Universität Dresden took part in the experiment that lasted approximately 75 min. They received class credit for their participation. All participants had normal or corrected to normal vision. Prior to the study, participants gave informed consent and were randomly assigned to one of the two conditions. One participant did not complete the PANAS. As this participant completed all other tasks and
Results
Participants completed 601.81 trials on average (SD = 116.45 trials). This allowed them to repeat each possible condition (2 [ITI] × 2 [Direction] × 8 [Interval] = 32 trials) approximately 18–19 times on average.
We analyzed participants' choice behavior as operationalized by their indifference points (see Section 2.5 for details). We then used those indifference points as the dependent variable in a four-factorial repeated measurements ANOVA with the within factors Interval, Direction, and ITI,
Discussion
In this study, we aimed to replicate previous findings (viz. Scherbaum et al., 2016) as well as test new predictions of our neural attractor model on choice perseveration in a value-based decision task. We created trial sequences in which one of two options—a small but near (SN) or a large but far (LF) option—was initially more attractive but became less attractive throughout the sequence by manipulating the distance between both options. This sequential manipulation allowed us to measure
Declaration of Competing Interest
The authors declare no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Acknowledgments
We thank Annemarie Pielenz.
Funding
This research was partly supported by the German Research Foundation (DFG grant SFB 940/2 to Stefan Scherbaum). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
References (75)
- et al.
Emotional modulation of control dilemmas: The role of positive affect, reward, and dopamine in cognitive stability and flexibility
Neuropsychologia
(2014) - et al.
Sunk costs in the human brain
NeuroImage
(2014) - et al.
A neural network model of individual differences in task switching abilities
Neuropsychologia
(2014) What are the neural origins of choice variability?
Trends in Cognitive Sciences
(2014)- et al.
Neuroeconomics of emotion and decision making
- et al.
The malleability of intertemporal choice
Trends in Cognitive Sciences
(2016) - et al.
Aging cognition: From neuromodulation to representation
Trends in Cognitive Sciences
(2001) - et al.
The psychological, neurochemical and functional neuroanatomical mediators of the effects of positive and negative mood on executive functions
Neuropsychologia
(2007) - et al.
Dynamic goal states: Adjusting cognitive control without conflict monitoring
NeuroImage
(2012) - et al.
On the continuity of mind: Toward a dynamical account of cognition
Decision making in recurrent neuronal circuits
Neuron
Positive emotion facilitates cognitive flexibility: An fMRI study
Frontiers in Psychology
Inertia and decision making
Frontiers in Psychology
A neuropsychological theory of positive affect and its influence on cognition
Psychological Review
Positive affect and flexibility: Overcoming the precedence of global over local processing of visual information
Motivation and Emotion
Perceptual decision making: Biases in post-error reaction times explained by attractor network dynamics
The Journal of Neuroscience
Extending a biologically inspired model of choice: Multi-alternatives, nonlinearity, and value-based multidimensional choice
Philosophical Transactions of the Royal Society B
Transcranial stimulation over the dorsolateral prefrontal cortex increases the impact of past expenses on decision-making
Cerebral Cortex
Response repetition biases in human perceptual decisions are explained by activity decay in competitive attractor models
ELife
The psychophysics toolbox
Spatial Vision
Mechanisms underlying dependencies of performance on stimulus history in a two-alternative forced-choice task
Cognitive, Affective, & Behavioral Neuroscience
A probabilistic, dynamic, and attribute-wise model of intertemporal choice
Journal of Experimental Psychology: General
The role of affect and reward in the conflict-triggered adjustment of cognitive control
Frontiers in Human Neuroscience
How positive affect modulates cognitive control: Reduced perseveration at the cost of increased distractibility
Journal of Experimental Psychology. Learning, Memory, and Cognition
Dimension reduction and dynamics of a spiking neural network model for decision making under neuromodulation
SIAM Journal on Applied Dynamical Systems
The effects of neural gain on attention and learning
Nature Neuroscience
Do you see the forest or the tree? Neural gain and breadth versus focus in perceptual processing
Psychological Science
From anomalies to forecasts: Toward a descriptive model of decisions under risk, under ambiguity, and from experience
Psychological Review
Dynamic field theory of movement preparation
Psychological Review
Quantifying the effect of intertrial dependence on perceptual decisions
Journal of Vision
Sequential effects in two-choice reaction time tasks: Decomposition and synthesis of mechanisms
Neural Computation
Experimental inductions of emotional states and their effectiveness: A review
British Journal of Psychology
Modelling option and strategy choices with connectionist networks: Towards an intergrative model of automatic and deliberate decision making
Judgment and Decision making
Signal detection theory and psychophysics
Selective alteration of human value decisions with medial frontal tDCS is predicted by changes in attractor dynamics
Scientific Reports
The dynamical foundations of motion pattern formation: Stability, selective adaptation, and perceptual continuity
Perception & Psychophysics
The affective control of thought: Malleable, not fixed
Psychological Review
Cited by (10)
An integrated modelling approach examining the influence of goals, habit and learning on choice using visual attention data
2020, Journal of Business ResearchCitation Excerpt :The ‘drift diffusion model’ in psychology (Krajbich & Rangel, 2011) describes the accumulation of information over time in favour of a particular alternative, until evidence in favour of that alternative exceeds a threshold. Similarly, the ‘choice perseveration model’ (Senftleben et al., 2019) posits that previous choices of an alternative cumulatively bias a respondent towards that alternative. One way to capture the learning effect is to regress the exogenous variables of the previous time periods (e.g., experimental constraints such as previous price levels) on the current choice (see Erdem et al., 1999 for further discussion).
Active reinforcement learning versus action bias and hysteresis: control with a mixture of experts and nonexperts
2024, PLoS Computational BiologyValue-Based Decision Making – Simple Analytic Theory
2022, Studies in Computational Intelligence
- 1
Ulrike Senftleben and Martin Schoemann contributed equally to the article and hence, share first authorship.