Preferential choices are often explained using models within the evidence accumulation framework: value drives the drift rate at which evidence is accumulated until a threshold is reached and an option is chosen. Although rarely stated explicitly, almost all such models assume that decision makers have knowledge at the onset of the choice of all available attributes and options. In reality however, choice information is viewed piece-by-piece, and is often not completely acquired until late in the choice, if at all. Across four eye-tracking experiments, we show that whether the information was acquired early or late is irrelevant in predicting choice: all that matters is whether or not it was acquired at all. We recorded the eye movements participants made when choosing between a series of pairs of (a) photographs of foods, (b) "posters" from the International Affective Picture System 3, (c and d) lotteries over three monetary outcomes. We recorded the choice participants made, the time taken to make the choice, and the eye movements during the choice. Models with potential alternative assumptions were posited and tested, such as 1) accumulation of instantaneously available information or 2) running estimates as information is acquired. These provided poor fits to the data. We are forced to conclude that participants either are clairvoyant, accumulating using information before they have looked at it, or delay accumulating evidence until very late in the choice, so late that the majority of choice time is not time in which evidence is accumulated. Thus, although the evidence accumulation framework may still be useful in measurement models, it cannot account for the details of the processes involved in decision making.Imagine choosing between an apple and some chocolate. In one instant you will be thinking about the lovely, crunchy apple. The next you are thinking about the smooth, indulgent chocolate. Your attention will dart back and forth between the apple and the chocolate until you make your mind up. This research is about exactly what is happening, at each instant, as you deliberate and choose. We will combine laboratory experiments (detailed in the Case for Support) and translational field experiments with real consumers making real choices (with our industry collaborators, detailed in the Pathways to Impact). Using eye-tracking technology, we will record millisecond-by-millisecond where people are looking as they choose. This will allow us to build an explanation of the link between attention and choice. Psychology, Neuroscience, and Economics have developed a first-generation model based on some key findings. When someone is choosing they take time to deliberate and shift their attention back and forth repeatedly. We now know that people are likely to look more at the option they ultimately choose and that if we intervene to make people look more at one option they are more likely to choose it. This has led to the development of drift diffusion models. Although complicated, these models instantiate the following idea: While an individual is looking at a specific option, they are biased towards thinking about why they should choose that option. For example, if you are looking at the chocolate, you are more likely to be thinking about the chocolate and are edging towards choosing the chocolate. In the language of the model, we say you are drifting towards choosing the chocolate while attending to the chocolate. If the chocolate is really much nicer than the apple, you will chose it quickly. In the language of the model, you have a high drift rate. If the chocolate and the apple are pretty evenly matched, you will take ages to decide. In the language of the model, you have a low drift rate. Now, when you look at the chocolate bar, we can't tell whether you are thinking about the sweetness, the high calories, or the pretty wrapper. So we asked people in our lab to choose between options with spatially separate attributes. For example, we had people choose between lotteries where the cash prize and the chance of winning were written separately, next to one another. We were surprised by the result. We found that looking more at a particular lottery was associated with choosing that lottery-just like with the chocolate. But this result did not hold for the separate attributes. You might expect people who looked more at the high cash prize to be more likely to pick the lottery and people who looked more at the low chance of winning to be less likely to pick the lottery. But, across a whole series of experiments, looking at the bad properties of an alternative was just as strongly associated with choosing that alternative as looking at the good properties. Our results show that while the link between what you are looking at and what you are thinking about is solid at the level of whole alternatives, it is broken at the level of the composite attributes. We need to rethink the model of how we choose. In the laboratory, we will replicate our findings across a wide number of tasks. We will test the effect of manipulating which attributes people look at. We will test for a feedback loop where looking increases drift rate, which in turn increases looking, and so on. The outcome will be new experimental evidence and new ideas about how we build up evidence over time to choose. We will translate this laboratory research into the everyday, with real consumers choosing for real. By working with the Financial Conduct Authority, the regulator for consumer financial products, the UK credit card industry, and the Which? consumer group, we will implement field trials, demonstrating these effects in real choices and measuring their impact.

Aside from the variations noted below, the experiments proceeded as follows. Participants were recruited from the University of Warwick participation pool and were paid for their participation. To limit movement, participants used a chin rest placed approximately 70cm away from the screen. Monocular eye movements were recorded at 500Hz with an EyeLink 1000 Plus (SR Research, Osgoode, ON, Canada) eye-tracker. Fixations were identified by the eye-tracker software using velocity algorithms. Participants initially underwent a 13 point calibration and validation cycle of the eye-tracker, which was repeated throughout the experiments. Stimulus presentation was controlled by MATLAB using Psychtoolbox extensions. Trials were excluded if their reaction times were greater than 1.5 times the interquartile range above the mean reaction time across all trials within the experiment or less than 200ms. Food Useable data was collected from 41 participants. Data was excluded for 4 additional subjects because of poor eye tracking data quality (as indicated by the proportion of time during trials where eye gaze was detected by the tracker being outside of the normal distribution across all participants). The stimuli were pictures of 50 different snack items. These were comprised of five types of snack: crisps, fruit candy, sweet carbonated drinks, health and sports drinks, and chocolate. This experiment was split into two parts. In the first, the participant rated the desirability of 50 snack items on a 1-9 scale. In the second, the 50 stimuli were paired to create 100 binary choice trials. Choice pairs were created such that the rating difference between the two items was 3 or less, and so that an individual snack item was not present in more than 5 trials. At the end of the experiment one trial was randomly selected and the subject was given the item they chose. Posters Useable data was collected from 53 participants. The experiment was displayed on a widescreen monitor (1920 x 1080 resolution, 60Hz refresh). Additional data was collected from 13 participants but 12 were excluded due to a programming error and one because their gaze location could be measured for less than 70% of the time across all trials. The stimuli were chosen from the International Affective Picture System. The pictures were all positive in affect (average ratings between 5=neutral and 7=mildly positive for both males and females) and had differences in value ratings of no more than 1.5 between male and female raters. After visual inspection, a further 7 images were removed for containing sexual images and 32 images were removed because they had a portrait aspect ratio. The 200 stimuli for each participant were randomly sampled without replacement from the 253 pictures that met these criteria. All participants completed binary choice and strength-of-preference tasks in a counterbalanced order. Here, we only analyse the binary choice data. Two landscape pictures (each 514 x 384px) were displayed side by side following a fixation cross. The response scale was presented horizontally centered, below the stimuli. For the binary choice task, two labels (Option A'' andOption B'') were shown underneath the appropriate stimuli. The current choice was signified by a red, square marker (30 x 30px) above the label. The marker was initially centered equidistant between the two images. To respond, the participants had to press the left mouse button. Reaction times were measured from the start of the trial to the mouse click onset. A blank, black screen was displayed for 500ms between each trial. Finally, participants had to rate their overall liking for each picture on a Likert scale, vertically displayed to the left of each image. The eye-tracker was recalibrated at the beginning of each condition and then after every 25 trials. Lottery Useable data was collected from 54 participants. The data of an additional 5 subjects was excluded: 1 because they failed to complete the task within a reasonable time frame and withdrew, and 4 because of poor quality eye tracking data. The stimuli were gambles with three equally likely outcomes. Because the outcomes were equally likely, no probabilities or likelihoods were displayed during the trials themselves. For each trial, two gambles were presented, one on the top and one on the bottom of the screen. The three payouts of a single gamble were presented in horizontal alignment and were displayed as white text within a solid grey circle. This was done to reduce the discriminability of the numbers in peripheral vision, so that subjects had to directly fixate the number to determine its value. Each gamble consisted of three possible outcomes which were always one low value (10-30), one medium value (40-60), and one high value (70-90). The specific values were randomly drawn on each trial. The main task consisted of 100 trials. Trials were presented in a random order, with trials from the different conditions intermixed. Each trial began with a fixation cross displayed in the centre of the screen, and the trial only began once the subject was looking at the fixation cross. Trials were split into 3 different conditions: no change, low to high, and high to low. There were 34 trials in the no change condition, and 33 in each of the other two. Here, we only analyse the no change condition. The eye-tracker was recalibrated at the beginning of the main block and then after every 20 trials. Currency Usable data was collected from 46 participants from the California Institute of Technology participation pool. Additional data had been collected from 14 participants but four were excluded due to incomplete data (they ran over the experiment slot) and 10 because of poor eye tracking accuracy (their gaze location could be measured for less than 68% of the time averaged across all trials). The stimuli were two gambles with three equally likely outcomes, presented at the top and bottom of the screen. The three payouts of each option were presented side-by-side as white text within a solid grey circle. There were two within-subject conditions, counterbalanced for order. In the commensurate condition, there were 42 trials, in a third of trials all attributes were displayed in pounds, yen and ``Q'' respectively. In the incommensurate condition, on every trial, the three attributes of each option were presented in the three difference currencies. Here, we only analyse the commensurate condition. Participants were told the exchange rates for pounds, yen and Q to dollars. Here we analysed the equivalent dollar values. The eye-tracker was calibrated at the beginning and every 21 trials. Participants pressed the up key if they preferred the top lottery, and the down key if they preferred the bottom lottery. Areas of interested were defined horizontal distance of 320 pixels, and a vertical distance of 340 pixels apart on the screen.