In my Advanced Social Psychology class, the presenting student selected Srull & Wyer’s ”The role of category accessibility in the interpretation of information about persons: Some determinants and implications” as his classic empirical paper. That is the “Donald” paper, ancestor of much social priming research.
It also became the paper that we pick apart so that we can better understand and critique research. As it is early in the semester (my first seminar, their second), the picking apart ended up being me asking questions from the presenting student, and then writing it down on the white board. (I have yet to master proper disposition of the board. I managed to fill it up just by the first set of conditions).
The Donald paper is legendary. Well worth actually acquainting oneself with. For example, turns out there were two hostile (and two kind) vignettes. They publish one in full. But, I digress.
We actually focused mostly on method. What exactly did they do?
The idea is that if you prime a trait concept, like being hostile, you are more likely to use this primed concept to interpret described behaviors later. This is something we all know more or less.
To prime, they used the now ubiquitous sentence unscrambling task which evidently was first created by Cotin in 1969. Not as an independent variable it seems (just looking at the title, now how diligent do you think I would be for a mere blog post?) but as a measure of hostility. Each sentence consisted of four words such as” “leg break arm his” or “her found knew I”. Two different 3-word sentences can be created out of each set. Notably, for the hostile one, both will be hostile, and for the neutral ones, both will be neutral.
The sentence task was presented in four different conditions. All conditions contained hostile sentences, but in different proportions. For two conditions, 20% were hostile words, for two 80%. In addition, participants were given either 60 sentences to unscramble, or 30 sentences to unscramble. So, participants unscrambled either 6, 12, 24 or 48 Hostile sentences; the 12 and 48 belonging to much longer tasks. Got that?
The idea here is that the more a concept is primed, the more likely it is to be used when interpreting ambiguous behavior.
But, as they say, that’s not all! Priming is likely to fade over time, in part because other types of concepts get brought to mind replacing the earlier primes. Thus, they also tested 3 types of delay. Participants did either read about, and rate Donald immediately after priming, 1 hour later, or 24 hours later.
That is a 4 x 3 between subjects single measure study. Total number of subjects: 96. Eight participants in each cell. (Given the F stat showing df’s of 1,72, I suspect some data may have been lost).
Eight participants in each cell!
And, I can’t make out if each participant read only one of the vignettes or both. (There were evidently two other tasks, but they were not relevant for the results presented in the paper, as they say).
What were the measures?
Well, first participants were asked to form an impression of Donald, and then rate him along a series of trait dimensions on a scale from 0 to 10 (not at all to extremely). Six of them could be considered related to hostility: hostile, unfriendly, dislikable, kind, considerate and thoughtful. (I leave it to the reader to work out which ones were reverse-scored ;),) six were evaluative, but not related to hostility: boring, selfish, narrow-minded, dependable, interesting, intelligent.
The second measure was to rate a series of individual behaviors on perceived hostility, again on a 0-10 scale. These were behaviors that had been collected and rated earlier, which had also been used as the basis for the vignettes. Five clearly were hostile, ten ambiguous, and five not hostile, as per the pre-test.
This is kind of nice. They do spend a lot of time working out their materials, and then also test what happens after priming.
Finally the participants are asked to do a co-occurrence rating. Per the description participants are given a sentence that reads “ If a person is hostile, how likely is it that he is…..?” the blank is filled in by one of 11 other traits, which I now presume come from the first measure. I also assume that it is always “hostile and…” as rating co-occurrence between all of them would lead to a combinatorial explosion. *
This is about how far we came in analyzing the paper. We did not go into the statistics this time. But, as I look over it, I’m struck by the lack of standard deviations in the reporting.
The results do show interesting orderings: the more hostile sentences the participant unscrambled, the more hostile (and less non-hostile) they thought he was. This also fades over time. Interestingly, the pattern is similar for the evaluative but not-related traits.
The behavioral ratings seem affected in similar ways, although the spread is the largest for the ambiguous behaviors. This is, of course, the famous finding. The spread is quite large – going from a mean rating of about 4, 5 to a rating of 9, depending on the number of hostile sentences in the unscrambling task when you do the test right after the priming.
They are suggesting that there is an interaction between trait type and delay, but none of the interactions are true cross-over interactions. The time-delays are not linear (0, 1 hr, 24 hr), so I don’t think much can be said about that.
They do nothing, that I can see (in my skimming) about the co-occurrence data. At least, nothing is plotted.
(On edit). Now I see. They use the co-occurrence data to verify that what they deemn descriptively related traits (the hostile etc) really are related as they expect. Well, I think methodologically they do a very nice job.
Experiment 2, the priming of kindness, seems to yield weaker data.
It is certainly interesting. Probably quite interesting in 1979 when it was published.
They do ask participants if they had realized that the priming task and the judgment task belonged together. Mostly they did not. In some instances people did, but seemed to guess. But, this is really more of a “connect the two” than an attempt to unconscious/unaware priming (which Tom Stafford beautifully critiques here). Perfectly fine, of course, and they also do not claim anything about things being unconscious, but more about not being aware that there was a possible connection.
I’m still struck by the low number of participants in each cell. When I did my sentence priming task, I had thirty participants in each, and had a hard time finding evidence for priming of the emotion concept.
I was also struck by the fact that hostility was primed in all conditions. I think it is fine titrating the amount of priming in this way, but that was also very different from what I was doing, where we were careful about not having too many priming sentences, in order to not hit them over the head with it. You are surely hit over the head by unscrambling 60 sentences, where, for 48 of them, you are forced to put together sentences involving arm or leg breaking or similar hostile acts.
Now I’m almost getting curious about tracing the tasks further, as the priming literature got subtler and wobblier.
*(12! would lead to 479001600 combinations, and I’m too lazy right now to work out whether it really would be 12! or a subset of it. But, you know, taking far longer time to do than any student would be willing to participate. 2 seconds per item, which would be on the quick side, would take roughly 30 years to do).
Srull, Thomas K. & Wyer, Robert S., (1979) The role of cateogry accessibility in the interpretation of information about persons: Some determinants and implications. Journal of Personality and Social Psychology, 37(10) 1660-1672.