On Preferring A to B, while also preferring B to A

By Alex Tabarrok

When people evaluate two or more goods separately versus jointly it’s common to see “preference reversals”. In a random survey, for example, people were asked to value the following dictionaries:

  • Dictionary A: 20,000 entries, torn cover but otherwise like new
  • Dictionary B: 10,000 entries, like new

When asked to value just one dictionary, either A or B, the average value was higher on Dictionary B. But when people were asked to evaluate both dictionaries together the average value was higher on Dictionary A.

What’s going on? Most people have no idea how many words a good dictionary has so telling them that a dictionary has 10K or 20K entries just fades into the background–it’s a dictionary of course it defines a lot of words. On the other hand, we all know that “like new” is better than “torn cover” so dictionary A gets the higher price. When confronted with the pair of dictionaries, however, we see that Dictionary A has twice as many entries as Dictionary B and it’s obvious that more entries makes for a better dictionary and in comparison to more entries, the sine qua non of a dictionary, the torn cover fades into importance.

Cass Sunstein collects a bunch of these examples (these two from List and Lowenstein respectively):

  • Baseball Card Package A: 10 valuable baseball cards, 3 not-so-valuable baseball cards
  • Baseball Card Package B: 10 valuable baseball cards
  • Congressional Candidate A: Would create 5000 jobs; has been convicted of a misdemeanor
  • Congressional Candidate B: Would create 1000 jobs; has no criminal convictions

In each case B tends to have a higher value when evaluated separately but A tends to evaluate higher with joint evaluation. When is separate evaluation better? When is joint evaluation better?

There is a tendency to think that joint evaluation is always better since it is the “full information” condition. Sunstein pushes against this interpretation because he argues that full information doesn’t mean full rationality. Even with full information we may still be biased. The factor that becomes salient when the goods are evaluated jointly, for example, need not be especially relevant. Is a dictionary with 20k entries actually better than one with 10k entries? Maybe 95% of the time it’s worse because it takes longer to find the word you need and the dictionary is less portable. We might let the seemingly irrefutable numerical betterness of A overwhelm what might actually be more relevant, the torn cover.

Sellers could take advantage of the bias of joint evaluation by emphasizing information that consumers might think is important but actually isn’t–our computer screen has 1.073 billion color combinations while our competitors only has 16.7 million–while making less salient 6 hours of battery life versus 8 which may in practice be more important.

Personally, I’d go for full information and trust myself to figure out what is truly important but maybe that is my bias. See the paper for more examples and thought-experiments.