At the start of the project, we focused on developing tools for precisely tracking information sampling during complex choice tasks. These explorations led to our first MEG experiment (N=20, two 5-hour sessions per participant), where participants made decisions among three alternatives differing on a single perceptual feature. We examined the "distractor" effect, which suggests that as the value of the worst alternative increases, it becomes harder to distinguish between the two high-value options. Although we failed to detect this effect at the level of choice probabilities (Fig.1) we found that higher values of the worst alternative slowed decisions. This finding replicated in a behavioural study (N=30).
Decoding attention from MEG signals revealed that participants first focused on the worst alternative to eliminate it (Fig. 2), then used it as a reference for comparing the remaining options. This "anchoring to the worst" strategy enhances the perceived value of the best alternative, counteracting the negative distractor effect anticipated by divisive normalisation. These results have been drafted in a manuscript. In independent analyses we focused on the micromechanisms of information sampling discovering striking rhythmicity (at 11 HZ) in attentional fluctuations and distinct neural signatures that drive the sampling of unexplored information (Siems et al., 2023). We also reanalysed open-access data and conducted online experiments, finding that some reported distractor effects in reward learning could be explained by sampling asymmetries. Overall, we conclude that information sampling plays a crucial role in decision-making, overriding the effects of hardwired distortions like divisive normalisation. These results have been presented in international conferences (e.g. SfN, SBDM meetings) and in invited talks.
In a second pharmacological MEG study (N= 24, three 2-hour sessions, involving the administration of placebo, lorazepam and donepezil) we focused on classical multiattribute preference reversals. This cohort was complemented by N=12 participants performing the three pharmacological sessions without MEG. We devised a novel experimental paradigm (Fig. 3), which we refined using an online cohort of participants, and replicated classical preference reversals (Fig. 4). Lorazepam but not donepezil decreased the magnitude of a specific reversal(the attraction effect). Decoding the locus of attention, we found that participants alternated their sampling across attributes, and within-attribute, followed an “eliminate and anchor” strategy like our first study. The 11 HZ attentional rhythmicity was also detectable in the placebo session, with the rhythm slowing down in the lorazepam session. Due to pandemic delays, data collection concluded towards the end of the funding period. The dataset is still being analysed by the PI, and results are being disseminated at conferences and drafted in manuscripts.
