Hierarchical levels of behavioral syndromes
Comparison of estimates of correlations in 2 behaviors quantitified on different levels reveals that the phenotypic-level correlation corresponds surprisingly well with the correlation on the underlying, hierarhically-lower levels.

‚ÄčHierarchies in behavioral syndromes

Perhaps the simplest way to partition a phenotypic correlation in behaviors is to divide it in a correlation which occurs on the level of individuals vs the within-individual level. A body of literature with its basis in quantitative genetics tells us that for behaviors, the within-individual correlation will strongly determine the  phenotypic correlation. This is because most of the variation in behaviors arises due to differences we observe when taking consecutive measures of behavior on individuals. Behaviors, of course, are not static traits which are fixed. An individual will vary in, says, its aggressiveness each time we (manage to) measure its aggressiveness. Within-individual correlations arise when, as we gather repeated measures on individuals, behaviors show changes in the same direction. For example, aggression is a bit increased, but also the individual's boldness, and the next time both decrease. These kind of within-individual correlations contrast to what most people tend to think of as personality, which is something more "intrinsic" and stable independent of fluctuations across measurements. This latter association is quantified by the individual-level correlation. If we would have measured each individual many times, the individual-level correlation would be the correlation of the means in their behaviors. 

Phenotypic correlations capture individual-level correlations

Barbara Class and I were interested in seeing to what extent the phenotypic correlation beween behaviors indeed reflected this "intrinsic" individual-level correlation. We therefore compiled literature estimates, which especially in the last years have accumulated. We found 109 estimates from a wide variety of taxa and behaviors. Frankly, I expected that within-individual correlations would be absent, because I believed that fluctuations in multiple behaviors over repeated measures taken on the same individual largely reflect stochastic noise (I argued so in this opinion article). Given I expected that the within-individual correlation would be zero, it seemed logical to me that the individual-level correlation would be much higher than the phenotypic correlation (remember, the latter is a sort of average of both lower hierarchical levels). To my surprise, however, the data proved me wrong! 

We find phenotypic correlations in behaviors resemble quite closely the individual and within-individual level correlations. This pattern appears to be real and not caused by biases arising due to statistical or publication processes. Also, there are several studies and behavioral syndromes for which this generalisation does not hold, but overall and based on what is out there now, it seems clear that scientists have not underestimated the strength of "intrinsic correlations" between behaviors. 

For me, the next puzzle is to understand how this pattern is possible.

-Jon Brommer