Factor Analysis of Humor Items

Methodological Considerations

In general, the success of such studies depends on how representative the sampled humor items are. Ideally, one first defines a universe of humor items and then draws items from it randomly. Such a universe could be, for example, all jokes and cartoons in a culture available at a certain time. Not defining the entire universe, or not drawing randomly, will result in a biased sample and hence not a comprehensive model. Another issue determining the outcome is whether the answer format is unrestricted (e.g., a rating scale) or ipsative (e.g., a Q-sort, where every person has the same mean). This has been used in humor statements (to compensate for social desirability, as pointed out by Kenneth Craik, Martin Lampert, and Arvalea Nelson in 1996), or in ratings of jokes to avoid cases where people like everything or nothing, as Raymond Cattell and Lester Luborsky (1947) point out. Further considerations relate to the analysis: the relation of the components (independent of each other, orthogonal or interrelated, oblique), the hierarchy (general or multiple factors), and the arrangement (e.g., a circumplex).

A factor (e.g., of humor appreciation) will only emerge if at least two markers for the factor are included in the sampled items. Bloated specifics occur when a very narrow and unimportant aspect is oversampled; for example, when many variants of “coping with humor” enter formulations of items, a very potent factor will emerge that gives the appearance of being more important than it actually is. Another issue is that all variables should be roughly at the same level of abstraction; a global concept, such as appreciation of humor, should not be mixed with specific variables, for example liking of a comedian, a particular funny film, or a single punch line. Furthermore, the sample or participants should not be restricted in variance (or be too homogeneous); for instance, in a sample of successful comedians, low scores in humor will be missing altogether, which will lower the intercorrelations among scales and thus distort the factor structure.

...