There were four main findings in this study. First, handedness indices as assessed by both bouts and frequency of hand use were highly correlated and comparable in their sensitivity to individual and population variability in hand preference. Second, population-level right-handedness was evident for the TUBE task, irrespective of the level of analysis and manner in which hand preferences were characterized. Third, hand preferences, as assessed by bouts or frequencies in hand use, were consistent across test sessions. Finally, hand preferences were consistent across a 6-year test-retest time period.
One aim of this study was to evaluate and compare the sensitivity of using bouts contrasted with using frequencies of hand use in the assessment of hand preference. The findings clearly indicated that in the context of the TUBE task, measuring bouts contrasted with frequencies made absolutely no difference in the characterization of hand preferences of the subjects. Support for this conclusion came from the facts that the correlation between the measures was very high (>.90) and the handedness indices derived for each set of measures were nearly identical. Furthermore, analyses for population-level handedness using the SUM-BHI, SUM-FHI, MEAN-BHI, and MEAN-FHI all produced identical results (i.e., population-level right-handedness). Taken together, these results do not support the suggestion by McGrew and Marchant (1997)
that bouts are the best measures of hand preference. The cumulative results also do not support the practice by McGrew and Marchant (1997)
of excluding Studies in their meta-analyses that do not meet their criteria of using bouts as the level of analysis of hand use.
The most relevant difference between the use of bouts and frequencies as the level of analysis for hand use was in the evaluation of individual and population-level hand preferences using z
scores. Using frequencies, there were significantly more lateralized than nonlateralized subjects, and there were significantly more right-handed than there were ambiguously-handed or left-handed subjects. In contrast, there were significantly more nonlateralized than lateralized subjects and there was no difference between the number of left- and right-handed subjects when using bouts as the level of analysis. This finding is not consistent with the handedness indices based on frequencies and bouts, nor is it consistent with the findings based on the classification scheme that was derived based on the degree of variation in handedness values. This could be interpreted as support for the arguments in favor of the use of bouts put forward by McGrew and Marchant (1997)
; however, there are several problems with that interpretation. First, z
scores are sensitive to sample size, and there may not have been enough observations of bout use to reach statistical significance with this measure compared with the case of the raw frequencies. However, it is likely that had we continued to test the subjects, eventually, enough bout data would have been collected, and they would have approximated the frequency data (based on the strong, significant, positive correlation in handedness index measures). The second issue of interpretation has to do with the use of z
scores in the evaluation of individual hand preferences. There is no strong rationale for their use, and arguably, they are unnecessary and a less sensitive measure of hand preference compared with handedness indices. The practice of using z
scores to classify subjects as left- or right-handed reduces statistical power because the data are being transformed from a continuous (such as handedness indices) to a nominal scale of measurement, a type that is less sensitive. The dichotomy of these two approaches is most evident when comparing the SUM-BHI and SUM-FHI scores with the results using the classification scheme based on z
scores. The SUM-BHI (M
= .112) and SUM-FHI (M
= .113) scores are nearly identical, whereas comparing the results of the hand preference classification data is not nearly as consistent. Moreover, using a different classification scheme that was derived based on the degree of variation observed in the sample yielded a significant population bias. These findings suggest that there is much more validity to the use of handedness indices or other classification schemes based on either bouts or frequencies than to the use of classification schemes based on individual z
scores. As Hopkins (1999b)
has articulated, z
scores are useful when the number of observations is large and equal between subjects, but there are often circumstances in Studies with nonhuman primates that prohibit the collection of equal sample sizes. We emphasize that there are other alternatives that are both practical and perfectly valid from a methodological and statistical standpoint.
The operational definition of a bout in hand use adopted in the study was created to address the specific criticisms that others have articulated concerning the measurement of handedness for the TUBE task (see McGrew & Marchant, 1997
). Of course, the issue of bouts can be extended outside of this context and could be considered at other levels of analysis. For example, a bout of hand use could be characterized across days or weeks or months and does not have to be restricted to hand use within a given test session. This is a valid point, but we believe that it highlights the problem with the use of bouts. Bouts are arbitrary units of measurement imposed upon a sequence of ongoing behavior. In the context of hand use for the TUBE task (and probably many other measures), we could have adopted any number of criteria for distinguishing between events that would have constituted bouts or nonbouts of hand use. In our view, the simplest and most parsimonious level of analysis for hand use is to record frequency. This is a perfectly valid measure so long as there are not procedural factors that bias the frequency of hand use, other than those imposed by the subject.
In terms of recording frequency of hand use, it has been argued that hand use can be perseverative, that individual events are not independent, and that this inflates sample sizes and increases the likelihood of finding significant hand preferences (Byrne & Byrne, 1991
; Lehman, 1993
; McGrew & Marchant, 1997
). All of these are valid issues, but there is no reason to assume that the results would yield an asymmetrically distributed population. In other words, when recording frequency of hand use, the probability of having a skewed distribution to the left or right is equal and should result in a bimodal distribution of hand use, all things being equal.
The findings reported in and depicted in indicate that there was consistency in hand preference across test sessions using both bouts and frequencies as the level of analysis. These results are by no means trivial because the design of this study required that the tube be taken by the subjects with either their left or right hand. Thus, we intentionally created the potential for biased and bidirectional hand use for this task. If the chimpanzees simply removed the peanut butter with the hand opposite that used to take the tube, then the correlations should have been significant but negative, rather than positive. This suggests that the preferences expressed by the chimpanzees were not situationally determined, as some have proposed might be the case for those Studies in which population-level hand preference has been reported (McGrew & Marchant, 1997
; but see Hopkins & Carriba, 2000
). These findings further suggest that the hand preferences are expressed because of endogenous factors, perhaps related to a specialization of the left hemisphere for motor skill. This hypothesis warrants further research, perhaps using brain imaging technologies that are now being used with nonhuman primates, including great apes (e.g., Hopkins & Marino, 2000
; Hopkins, Marino, Rilling, & MacGregor, 1998
; Pilcher, Hammock, & Hopkins, 2000
; Zilles et al., 1996
Finally, the correlation between the data originally reported by Hopkins (1995)
and those reported in this article was significant and positive. These results cannot be attributed to learning associated with reinforcement history because the subjects were not tested on the TUBE task during the intervening 6-year period. Thus, these results suggest that the TUBE task is a stable measure of hand preference.
In conclusion, the cumulative results of this study suggest that bouts rather than frequencies are not necessarily a better level of analysis in the evaluation of hand preference in chimpanzees (and probably other nonhuman primates). This is not to suggest that perseveration in responding or a lack of independence of hand use cannot influence the assessment or interpretation of hand preference. There very well may be situations or tasks in which this issue becomes important, but based on these findings, they are not relevant issues for the TUBE task. This may be due, in part, to the fact that the TUBE task is a relatively stable measure of hand preference and does not appear to be influenced by situational factors. Thus, the point at which perseveration of responses or a lack of independence of data points becomes relevant in the assessment of hand preference is likely associated with the degree to which the measure is reliably expressed by the subject and influenced by situational factors.