ECONOMISTS ARE ALWAYS trying to predict how people will react to new polices or choices by looking at decisions they’ve already made.
However, researchers from CalTech and Stanford have taken an early step towards predicting choices by looking at people’s brains when they’re looking at various prospects but not yet making a decision.
Most prior research has focused on what the brain looks like at the moment of choice. Far more interesting to economists are the sort of things that can help map out choices in advance.
In research described in a new National Bureau of Economic Research paper, Alec Smith, Colin Camerer, and Antonio Rangel of CalTech and B. Douglas Bernheim of Stanford had subjects look at pictures of 100 snacks while undergoing fMRI brain scans. After that, they asked them to choose between 50 pairs of snacks, also while being scanned. Finally, they were asked to rate how much they liked each option.
68.2% success rate
To figure out what sort of neural responses might be predictive, the researchers built a model using data from 48 choices that tried to predict the choice in the two excluded.
For more than half of individual participants, the model worked significantly better at predicting choices than an uninformed choice. Though some of the predictions weren’t particularly confident, the overall success rate was 68.2%.
Here’s the chart showing the success rate based on how many voxels (data points) were retained:
Success rates from the experiment. If you’re having trouble viewing the image, click here. (Image Credit: NBER)
If, as the research implies, there are similar neural indicators across people, more data researchers should be able to create one predictive model and use it on new people, rather than having to recalibrate it each time. That would allow for much larger scale research in areas in which economists traditionally have a lot of trouble.
The researchers argue that this would be particularly useful in the sorts of situations where similar situations or good data aren’t available, or people might be biased in how they make hypothetical choices. An example given by the authors of the former is how much people value pristine coastlines or biodiversity.
As for hypothetical questions, people systematically overstate their willingness to pay for things, and their preference for choices that cast them in a better personal light.
It’s important to note that this is extremely early work on a small sample. It looks at a relatively simple choice, a good amount of data had to be removed from the analysis, and it isn’t compared to alternatives for prediction. It’s expensive, time consuming, and doesn’t appear to work for everybody.
But it’s a step with huge potential towards both getting a better understanding of why and how people make choices, and getting the sort of useful, predictive information that economists always want but rarely get.
Max Nisen
To embed this post, copy the code below on your site
have your say