Randomness vs. Balance

Randomness and balance are important experimental issues.
  • Show some pictures with a subliminal cue, and some without, and count how many are correctly recognized.

It is tempting to just randomize things. But if you have twenty trials of two conditions (cued and uncued) and you "just randomize" them, you will not necessarily then end up with ten trials in each condition.

Because having even numbers in your various conditions is often important for statistical testing, you should "balance" your design rather than "randomizing" it. This means that you should make sure that all the permutations work out evenly:

  • Each subject should see half cued and half uncued pictures.
  • Each picture should be cued for half the subjects and uncued for the other half.
next: factors
Experimental Design for Wearable Computing -- Minimizing the Effects of Uninteresting Factors

Minimizing the Effects of Uninteresting Factors

Balanced designs help minimize the effects of uninteresting factors.

We are presumably not interested in discovering that:

Even with a balanced design, though, you may find that you need quite a few trials and quite a few subjects to get good data. Be prepared for the potential scope of your experiment!

next: graph your data