(This is a guest post by Sarah Fisher. 2000 words; 8 minute read.)
We could all do with imagining ourselves into a different situation right now. For me, it would probably be a sunny café, with a coffee and a delicious pastry in front of me––bliss. Here’s another scenario that seems ever more improbable as time goes by (remember when we played and watched sports…?!):
Imagine that you are a recruiter for a college basketball team. Your job is to search for promising high school basketball players and try to recruit them to your college. You are looking through files for players from local high schools, and you are especially interested in players who can score many points.
So, what do you think? Whereabouts would you rate the player on the scale below?
In a recent psychology experiment, this exact scenario was presented to a group of students at the University of California, San Diego. On average, they rated the player about here:
The experimenters gave the same scenario to another group of students, except this time they changed one detail. Instead of describing the player as having made 40% of his shots last season, they described him as having missed 60%.
Seems like no big deal. After all, missing 60% is exactly the same as making 40%, isn’t it? These are just two ways of saying the same thing. Presumably, then, it didn’t make a difference to how the students responded. Right?
In fact, the average rating dropped to about here:
Not a dramatic difference, but a big enough one that it's unlikely to have happened by chance. It looks as though changing the wording from ‘made’ to ‘missed’ genuinely affects how people judge the player.
This is an example of a framing effect. (In particular, it is known as an ‘attribute framing effect’, according to the typology developed here.) Psychologists have been interested in framing effects since Amos Tversky and Daniel Kahneman first brought them to light about forty years ago. Many experiments conducted since then show that most of us are susceptible to such effects most of the time (see here for a survey of the first twenty years of framing research).
So, it turns out that framing the same information in different words can lead people to draw different conclusions. Huh.
Now, there’s a sense in which this is not surprising at all. We’re pretty well-attuned to these kinds of effects. In fact, you probably sensed straight away that switching from talking about shots made to shots missed would inevitably make the player sound worse. (As you can imagine, it’s useful to know about this effect when you’re trying to be persuasive – if you keep an eye out, you’ll notice how advertisers and politicians, among others, deliberately frame their messages in particular ways).
But why do framing effects happen? How can it make a difference which words are used, if they convey exactly the same information? And, if framing effects seem obvious and familiar in some sense, why don’t we see past them? Are we being irrational?
Many have taken framing effects to be a paradigmatic example of human irrationality. Kahneman has a chapter on them in his classic book about irrational decision-making, Thinking Fast and Slow.
I’m going to outline one way of challenging that view (and I hope to cover some others in a later blog post). On the alternative view, framing effects are a symptom of our sensitivity to subtle linguistic cues, and of our use of these cues to draw very reasonable conclusions.
Before getting to that, let’s first consider why framing effects are so commonly thought to reveal irrationality. Think again about the case of the basketball player. Remember that he is judged more valuable when the focus is on the proportion of shots he made than when the focus is on the corresponding proportion he missed. It seems as though our judgements are not tracking how well or badly he actually performed––the actual proportions he made and––instead, our judgements are merely tracking language. We are swayed by the particular words used to convey the information. But those words, presumably, are merely superficial, surface phenomena, which make no difference to the underlying information they are used to convey. To invoke an analogy, it’s a bit like deciding how much you like a gift purely on the basis of whether the wrapping paper is red or blue.
Going back to our earlier example, it seems clear, on reflection, that the basketball player’s performance is the same whichever way it is described. Presumably, then, if we were perfectly rational, we should rate him similarly under each description.
This is why many psychologists have treated our susceptibility to framing effects as a kind of reasoning error. They assume that the differences in language are irrelevant. And, therefore, we are irrational if we allow these to affect our judgements. Summing up his chapter on ‘Frames and Reality’, Kahneman writes:
As we have seen, again and again, an important choice is controlled by an utterly inconsequential feature of the situation. This is embarrassing – it is not how we would wish to make important decisions.
Why might we make such an error? Kahneman and others have suggested that positive or negative language (as when we hear that a basketball player ‘made’ or ‘missed’ shots) can colour our judgement. This is thought to happen purely by association, perhaps as an emotional response that bypasses our reasoning capacities.
This interpretation implies that we process information fairly shallowly, just responding to surface features (the particular words used) and not thinking through the underlying facts (what’s actually being said with these words).
Recently, though, a different view of framing effects has begun to emerge. Several theorists have argued that we are right to respond differently to alternative frames. The advocates of this view aim to provide a rationalising explanation of framing effects.
The idea I’ll focus on here is that frames ‘leak’ information beyond what is explicitly conveyed – a proposal developed here. The proponents agree that the basketball player made 40% of his shots and missed 60%, regardless of which frame is used. But, it is argued, by choosing to describe the player as having ‘made 40%’, some extra information is implicitly conveyed.
What is this extra information?
Well, think about how the frame affects your expectations about typical high school players. Do you expect them to make more or less than 40%? How does that change if I tell you the player ‘only made 40%’?
According to the reference point hypothesis, describing the player as having ‘made 40%’ conveys that the player made relatively many shots – compared, say, to the average high school player. On the flipside, saying that the player ‘missed 60%’ of his shots conveys the opposite information – that the player missed relatively many shots.
As you might have spotted, it’s possible to switch these expectations with ‘only’. When we hear that the player ‘only made 40%’ we expect that he should have made more. But when we hear that he ‘only missed 60%’ it sounds like he’s a relatively high scorer.
Now, if this theory is right, it would explain why the player is judged to be better under the ‘40% made’ frame than under the ‘60% missed’ frame. Under the ‘40% made’ frame, he is assumed to be a high performer, making more shots than average. In contrast, under the ‘60% missed’ frame, the player is assumed to be a poor performer, making fewer shots than average.
Importantly, it turns out that we’re right to make these assumptions. The extra inferences we make are actually tracking how speakers choose their words (as shown by studies reported here, here, here, here, and here). For example, when a basketball player made more shots than average, a speaker is more likely to talk about how many shots he ‘made’ rather than how many he ‘missed’. In other words, the speaker’s choice about whether to use ‘made’ or ‘missed’ is a pretty reliable guide to how well the player performed. And it’s perfectly reasonable for our evaluation of the basketball player to be sensitive to this. On the ‘information leakage’ account, then, the framing effect is perfectly rational. Going back to gift-wrapping analogy, if people tend to wrap better gifts in red paper, then you should pay attention to the colour of the paper!
It’s worth emphasising that, according to the ‘information leakage’ account, we fully process the information about how many shots the basketball player made and missed. It is just that we also make an extra inference about whether that represents relatively good or bad performance. In this way, framing effects are seen as evidence of deeper, not shallower, processing.
Suppose this is the right analysis. That raises a question: what is the status of the additional information ‘leaked’ by frames. Is it part of the meaning of our words, or something we work out using non-linguistic information?
In this draft paper, I argue that the information should be captured under the broad umbrella of pragmatics, as ‘conversational implicatures’. These are all kinds of extra bits of information, which we glean by ‘reading between the lines’ of what people explicitly say. The concept of an implicature was first introduced in a classic philosophy paper by Grice on ‘Logic and Conversation’ (published in a collection of his work here).
Here’s an example: imagine you ask me whether I read your article. I report to you: ‘I did put the article in front of me and I moved my eyes from left to right along each line’. You’ll probably infer that I didn’t actually read it – perhaps I was bored or distracted. But why do you draw that conclusion? After all, reading (in English) usually involves precisely the process I described.
The answer is that I should simply have said ‘yes’ if that’s what I meant. By replying in such a convoluted way, I implicitly convey that, although I made some attempt to read the article, I didn’t manage to do so. This information is known as an ‘implicature’ of my utterance.
I suggest that framing works a bit like this. If I tell you that a basketball player ‘missed 60%’ of his shots, you’ll probably infer that he’s a poor performer. Why? Because I could have said he ‘made 40%’ but I chose not to do so. Again, what I chose not to say speaks volumes.
If my analysis is right, it suggests that a rational explanation of framing effects was always available, we just needed to appreciate Grice’s insight that the information we communicate to each other can far exceed what we say explicitly. In particular, a speaker’s choice of frame carries implicit information that justifies our sensitivity to what initially seems to be an entirely inconsequential feature of the utterance.
If the case of the basketball player is anything to go by, we shouldn’t automatically jump to the conclusion that framing effects are irrational. We have good reasons to judge the player differently under each frame. Having said that, the research on framing is complex and wide-ranging (sometimes overwhelmingly so!). Various other studies look at the framing of choices, questions, policies, and goals. Some of these investigate not just how information is phrased, but also the order in which it is presented, whether it is in spoken or written form, and all sorts of other possibilities. Different lines of research explore whether someone’s susceptibility to framing effects depends on their age, gender, cognitive ability, or even their political affiliation. It remains an open question whether all the findings in this vast literature can be given a rational explanation. I don’t know the answer to that. But, for now, I’m keeping the glass half full!
If you want to read more about rational interpretations of framing effects, check out Sarah's website, this 2013 paper by David Mandel, and stay tuned for future posts.