Sunday, July 27, 2014

RATs! A way to measure creative thinking

One of the new tasks included in PEBL 0.14 is an implementation of the Remote Associates Task (RAT).  A RAT is not a creepy vermin with a tail; rather it is a word puzzle in which you are given three words that are not really related to one another, but they are all related to a third word.  This can be either linguistically-related (phone, match, play are all words that precede 'book'), or more meaning-related (stick, light, cake all relate to candle).  The screenshot shows how the PEBL version is arranged. The red bar is a timer that shrinks; you have 30 seconds to identify and type the answer.



There are a number of good resources for RAT-like problems.   For example, John Kihlstrom at  at Berkeley maintains a really nice page, whose problems I stole for the PEBL test (He transcribed them from Shames and Mednick and Mednick).  THere is also another set of 'compound remote associate Problems" (CRAPs) that are in some way different.  For the moment, I've included a .csv file including this in the RAT directory, but you cannot yet use them in testing unless you are prepared to hand-edit files.

There are quite a few parameters that can be set.  The standard set includes problems of three difficulty levels, and you can choose which ones you want to use (by default, only the easy and medium set).  You can mix them or do them blocked, you can change the time limit.  There is also a parameter that allows the participant to retry any answer until they get it right.  This defaults to off, because for some problems, it is not really clear whether there is a right answer.  I've done some limited testing with these problems on children, and it seems like it is better to not tell them they got it wrong, and just let them provide whatever answer they feel is right.  Adults may be a different story.


There is a pair of parameters called seedrandom and rngseed.  Suppose you want to just test 20 random problems, but you want everyone to get the same 20.  You could set maxprobs to 20, and set seedrandom to 1, and then give rngseed any number.  It will always use the same subset in the same order.  You should take a look at which problems it uses, and if you don't like it, try a new seed.  You could probably achieve something similar by editing the data files too.

Below is a list of all the parameters you can control.

timeout|30|time per problem in s
doeasy|1|Should you do the easy problems,
domed|1|Should you do the medium difficulty problems
dodiff|0|SHolud you do the hard problmes],
mixproblemsets|1|Should problem sets be mixed
blocksize|100|Give a pause after this many trials (>68 means no pause)
maxprobs|68|Do at most only this many problems. They will be mixed between conditions specified by doeasy domed and dodiff
seedrandom|0|Set this to use the same randomization every time
rngseed|1000|Choose a random number to seed order
allowretry|0|should you be able to retry until you get correct (and time up)

If you are using this as an assessment of creativity, I'd suggest looking at the number of correct problems achieved, perhaps weighted by the difficulty, and maybe also the mean time to complete.

No comments: