4 options for running a card sort

3 min read Andrew Mayfield

This morning I eavesdropped a conversation between Amy Worley (@worleygirl) and The SemanticWill™ (@semanticwill) on “the twitters“.

Aside from recommending two books by Donna Spencer (@maadonna), I asked Nicole Kaufmann, one of the friendly consultants at Optimal Usability, if she had any advice for Amy about reorganising 404 books into categories that make more sense.

I don’t know Amy’s email address and this is much too long for a tweet. In any case I thought it might be helpful for someone else too so here’s what Nicole had to say:

In general I would recommend having at least three sources of information (e.g. 1x analytics + 1 open card sort + 1 tree test, or 2 card sorts + 1 tree test) in order to come up with a useful and reliable categorisation structure.

Here are four options for how you could consider approaching it (starting with my most preferred to least preferred):

Option A

  • Pick the 20-25 cards you think will be the most difficult and 20-25 cards that you think will be the easiest to sort and test those in one open card sort.
  • Based on the results create one or two sets of categories structures which you can test in a one or two closed card sorts. Consider replacing about half of the tested cards with new ones.
  • Based on the results of those two rounds of card sorting, create a categorisation structure and pick a set of difficult cards which you can turn into tasks which you can test in a tree test.
  • Plus: Categorisation is revised between studies. Relative easy analysis.
  • Minus: Not all cards have been tested. Depending on the number of studies needs about 80-110 participants. Time intensive.

Option B

  • Pick the 20-25 cards you think will be the most difficult and 20-25 cards that you think will be the easiest to sort and test those in one open card sort.
  • Based on the results do a closed card sort(s) excluding the easiest cards and adding some new cards which haven’t been tested before.
  • Plus: Card sort with reasonable number of cards, only 40-60 participants needed, quick to analyse.
  • Minus: Potential bias and misleading results if the wrong cards are picked.

Option C

  • Create your own top level categories (5-8) (could be based on a card sort) and assign cards to these categories, then pick random cards within those categories and set up a card sort for each (5-8).
  • Based on the results create a categorisation structure and a set of task which will be tested in a tree test.
  • Plus: Limited set of card sorts with reasonable number of cards, quick to analyse. Several sorts for comparison.
  • Minus: Potential bias and misleading results if the wrong top categories are picked. Potentially different categorisation schemes/approaches for each card sort, making them hard to combine into one solid categorisation structure.

Option D

  • Approach: Put all 404 cards into 1 open card sort, showing each participant only 40-50 cards.
  • Plus: All cards will have been tested
  • Do a follow up card sort with the most difficult and easiest cards (similar to option B).
  • Minus: You need at least 200-300 completed responses to get reasonable results. Depending on your participant sources it may take ages to get that many participants.
Andrew Mayfield

Andrew Mayfield

Andrew is the CEO of Optimal Workshop.