Usability Testing: What are the maximum number of cards one should use in a card sorting exercise?

Studies in the past have shown that anywhere between 30 to 100 cards give good results. Anything below 30, won't give you enough data and anything above 100 are too many cards for the participants to process. A study by Tullis and Wood (link below) has shown that participants are willing to spend at the most 20-30 minutes performing a card-sorting experiment. So consider conducting a pilot study to make sure you don't go over 30 minutes.   Personally, I like to use around 60 cards which result in 8-10 categories. This pa@per by Tullis and Wood is a little academic, but here is the link if you want to read it. http://www.measuringux.com/CardSorting/Tullis&Wood- CardSorting-2005.pdf Another related question that often comes up is, how many users should you test your card sort on? Again, research by Tullis and Wood shows that any number between 20-30 is ideal. But Jakob Nielsen argues that 15 participants for a card sorting exercise are enough! http://www.useit.com/alertbox/20040719.html

In a web-based environment, I have found that even 15 "cards" are too many. Users get confused quickly. (Note: Cards in this case are really just line items in a list of piroritized items.) For web-based tests, pairwise comparison tests are much easier to execute. http://www.Allourideas.org provides a great way to achieve the same effect as card sorting while each user only has to choose between two options. You will see though that users will get excited and often vote 20 or more times.

Definitely you should not overwhelm people with number of cards. People might get confused and just get a feeling that nothing really goes together. There is a cool article about common mistakes in card sorting tests: http://blog.usabilitytools.com/make-common-card-sorting-mistakes

About fifty cards seems to be a good amount. Although it depends on the complexity of the terms and whether you are testing single people or groups of people. If you're testing in groups, you can use a larger number of cards. I've had good results with up to 100.

I agree with and would recommend not to use more than 60 cards and only do the test with one user a time. It's more time intensive but that allows you to follow one users thought in depth. When recording the session, use the "Thinking Aloud"method which basically means that the user is telling his thoughts behind his actions.

I usually do Card Sorts using an online-tool like https://www.optimalworkshop.com/optimalsortThe main advantage of doing it online is that you don´t have to do the quite cumbersome analysis by hand and that it produces beautiful dendrogram-clusters. As has pointed out, you should use between 30-50 cards, with 100 being the absolute maximum for motivated participants.I would like to note however, that you should use more participants than cards, as otherwise the main quality measure will suffer. This quality measure is NOT a correlation but the http://www-01.ibm.com/support/docview.wss?uid=swg21479963In order to generate usable results, the KMO needs to be above 0.6 and for the scientific minded the KMO should be above 0.8. The main remedy of a low KMO is to add participants with the rule of thumb being#cards > #participants

15 minimum, but I'd say 50-70 is optimal. Read more about it here: http://blog.usabilitytools.com/

The maximum number of cards is determined by how long you can get participants to sort them. The only relation to statistical significance and number of cards is that if you have too few people sort, you'd lack confidence that the target user population would sort/think about the concepts the same way. If you have too many cards to sort, you can run variations of the sort using subsets of the cards with groups of users to get the full sort done. More cards isn't inherently better, as long as you've got enough cards to represent what you want sorted. More participants is better in terms of reliability of the findings, as long as the participants represent the target population of users.

Anyone ever test subsets of cards (each person sees 50 of 80 total cards for example) and then increase the number of people to compensate? Several online tools offer this option... but I'd love some confirmation that it works as expected, without undesirable side-effects... On the one hand, many experts recommend using a subset of the total content anyway... So wouldn't this just be an extension of that approach?