What the research says

Mixed Examples Teach Recognition: The Paintings Study

People learning painters’ styles did far better when the paintings were mixed than when grouped by artist. Most of them still believed grouping had worked better, even after the test.

By the Giga Quizzes team4 min read
Paintbrushes resting on a wooden palette covered in bright oil paint
Photo: Jennie Razumnaya, Unsplash, Unsplash License, cropped.
In this post
  1. 1The experiment
  2. 2What they found
  3. 3What learners believed
  4. 4The limits
  5. 5Mixing examples in a lesson

How do you learn to recognise a painter’s style, a type of rock, a bird species or a kind of maths problem? By seeing examples. But in what order? The natural answer is to group them: show all of one artist’s paintings together so the common style stands out, then move to the next. Nate Kornell and Robert Bjork at UCLA expected exactly that to work best. Their 2008 experiments found the opposite, and found something just as interesting about what learners believed.

The experiment

Participants were UCLA undergraduates. Their task was to learn the styles of 12 landscape painters, chosen to be unfamiliar, among them Georges Braque, Henri-Edmond Cross, Georges Seurat, Philip Juras and Marilyn Mylrea. They saw six paintings by each artist, 72 in all, for three seconds each, with the artist’s name.

  • Grouped, or massed: six paintings by one artist in a row, then six by the next.
  • Mixed, or spaced: paintings by different artists interleaved, so that consecutive paintings were by different painters.

In the first experiment, 120 participants saw half the artists grouped and half mixed. In a follow-up with 72 new participants, each person saw only one arrangement, to rule out any effect of combining the two.

Then came the real test: new paintings by the same 12 artists, never seen before. For each, participants picked the artist from a list of names. Learning the style was the only way to get it right.

What they found

First block of the testArtists seen mixedArtists seen grouped
Experiment 1a (120 participants)61%35%
Experiment 1b (72 participants)59%36%
Kornell & Bjork (2008), proportion of new paintings correctly attributed

Mixed presentation won by more than 20 points, in both versions. A third experiment changed the test to rule out a different explanation: perhaps mixing only helped people remember the names. This time 80 new participants were shown new paintings and asked only whether each was by a familiar artist or not, with no names needed. Mixed presentation won again, 77% against 67%.

The authors had expected the reverse. Grouping should, in theory, make the shared features of an artist’s work easier to notice. What mixing seems to do instead is highlight the differences between artists: seeing a Seurat right after a Braque draws attention to what separates them, and recognising a style is largely a matter of telling it apart from others.

What learners believed

After the first experiment, participants were told what grouped and mixed meant and asked which had helped them learn more. Most chose grouping.

The numbers make the point sharply. 78% of participants did better on the artists they had seen mixed. Yet 78% said grouping had been as good as or better than mixing. They had just taken the test, and their own results contradicted them, but the feeling of the study phase won.

The same misjudgement turned up again years later, in the online replication described below. Of its participants, 71% said grouped paintings had taught them most, 17% said the two arrangements were equally good, and only 12% picked mixing, the arrangement that had actually worked better on their test.

The researchers checked whether people simply could not remember which artists had been in which condition. Asked to sort artists into grouped and mixed, participants did well above chance for mixed artists but not for grouped ones. Their judgements seemed to come from how the learning had felt. Grouped examples feel smooth and easy, as if the style is being absorbed. Mixed examples feel scattered and hard. The feeling is reliable; the conclusion drawn from it is not.

The limits

The material was paintings, the learners were university students, and the test followed soon after study. A later replication of the recognition experiment, by Peter Verkoeijen and Samantha Bouwmeester with paid online participants, found the advantage for mixing again, of a similar size. The explanation above also implies a limit: if the categories are obviously different, contrasting them has little to teach.

Mixing examples in a lesson

Many subjects involve learning categories: types of rock, kinds of poem, periods of architecture, bacteria under a microscope, chord progressions, maths problem types. In each, the research suggests showing examples of different categories side by side rather than one category at a time.

A recognition quiz built on the study

  • Use a picture question per example: a painting, a rock, a leaf, a graph.
  • Mix the categories, so that each question could be any of them.
  • Include categories that are easy to confuse.
  • After each answer, say what gave it away, which points to the difference that matters.
  • Expect students to find it harder than a grouped set, and tell them why that is useful.

The same principle in maths is called interleaving: see mixing topics in practice works better. For making the most of images in quizzes, see picture questions.

Sources

  1. Kornell, N. & Bjork, R. A. (2008). Learning concepts and categories: Is spacing the “enemy of induction”?. Psychological Science, 19(6), 585–592.
  2. Verkoeijen, P. P. J. L. & Bouwmeester, S. (2014). Is spacing really the “friend of induction”?. Frontiers in Psychology, 5, 259.

More from the blog

All posts →