Cluster analysis reveals at least three, and possibly five distinct handedness groups

Neuropsychologia, Vol. 30. No. 4, pp. 373-380, 1992 Printed in Great Britain.

C028-3932/92 $5.00+0.00 0 1992 Pergamon Press Ltd

CLUSTER ANALYSIS REVEALS AT LEAST THREE, AND POSSIBLY FIVE DISTINCT HANDEDNESS GROUPS

MICHAEL PETERS* and KELLY MURPHY

Department of Psychology, University of Guelph, Guelph, Ontario, Canada NlG 2Wl

(Received 12 July 1991; accepted 6 December 1991)

Abstract-Multivariate techniques have used data from hand preference questionnaires to group hand preference items, but no attempt has been made to date to use multivariate analyses to group indiuiduch in terms of handedness groups. This study analyzed the responses of 645 subjects on the Waterloo 60-item handedness questionnaire with a cluster analysis (BMDP) in order to determine the grouping of individuals in terms of hand preference patterns. Five distinct handedness groups were recognized by this procedure and a Discriminant Function Analysis revealed a very high accuracy of assigning individuals to the five groups. A cluster analysis of a shorter 1Citem questionnaire suggested three distinct handedness groups, and the degree of accuracy of assigning individuals to these groups was also very high. As is the case with all multivariate techniques in neuropsychology, the question ofwhether the clusters form meaningful groupings awaits an answer in terms of their different neuropsychological properties.

THE CLASSIFICATION of handedness groups, as based on questionnaire data, has rested on more or less arbitrary operational definitions. For instance, using questionnaires such as the ones developed by OLDFIELD [6], all persons who express only right hand preferences have been classified as right-handers while the remainder may be labelled “non-right-handers” or “adextrals” [13]. Others [lo] simply take the “0” laterality quotient as the neutral point betw?n handedness groups; right-handers are those with a quotient larger than “0” and left- handers are those with the quotient smaller than “0”. Another set of criteria is used by those who distinguish between left-handers, mixed-handers, and right-handers [ 1). A classification procedure that is similar in general principle distinguishes between consistent left-handers, inconsistent left-handers and right-handers [S, 93. In short, there is no single generally accepted criterion for classifying handedness on the basis of questionnaire data. Often, the classification scheme used is the one that relates handedness to some other neuropsychological variable, rather than the one that has been constructed as a result of a particular view of what handedness is.

We have wondered whether a cluster analysis of the sort used to establish particular groupings of individuals on the basis of common characteristics might be helpful in establishing handedness groups from a different perspective. Factor analytic procedures have been used to establish the grouping of items that make up handedness questionnaires [3, 11, 121 and KRONENBERG and BEUKELAAR [4] have used a clustering procedure to group hand preference items but systematic attempts to group indiuiduals on the basis of handedness are vqry rare. ANNETT [2] classified right-handers into five subgroups, and left-handers into

*To whom all correspondence should be addressed.

313

374 M. PETERS and K. MURPHY

three subgroups, on the basis of combination of information from preference and performance measures. However, we are not aware of any report in the literature where multivariate procedures have been used to distinguish between handedness groups. The results of this approach are described below.

Subjects

METHOD

The 60-item Waterloo questionnaire [ll], and a modified 14-item OLDFIELD [6] questionnaire were administered to 645 undergraduate university students who had volunteered to participate in a handedness study. These students were enrolled in psychology classes at the University of Guelph. Consistent with sample compositions in similar studies [3,11], there were more females (ratio 70:30) than males in the sample. Of the 645 subjects, 118 wrote with the left hand and 527 wrote with the right hand. The original sample contained 11% left-handers. We added left- handed subjects to the sample because we wanted a sufficient number of left-handers to allow clustering of subgroups if such would be found. The general properties of the sample in terms of the questionnaire response patterns were found to be very similar to those of a comparable sample studied by STEENHUIS and BRYDEN [l 11; a principal components analysis of the questionnaire produced item loadings and a factor structure that were almost identical to those observed by STEENHUIS and BRYDEN. The questionnaire asks subjects to answer each item on a 5- point scale “l= always left, 2 = mostly left, 3 = either hand, 4 = mostly right, 5 = always right”. The analysis program used was the BMDP cluster and Discriminant Function Analysis. The BMDP statistical software* is a public domain software, and clear descriptions of the analyses described in this paper, and how to implement them, are provided in the BMDP manuals.

RESULTS AND DISCUSSION

Cluster analysis of items for the 60-item questionnaire (BMDP, 1M)

The clustering procedure can be used to cluster items or individuals into groups. First, we report on the clustering of items (BMDP, 1M). Initially, each item is considered a variable and each variable is initially considered as a separate cluster. The next step is to find the variable (item) that is most closely linked to the first one. In successive iterations, the item or items that are most closely linked to the first two are identified, and so on until the entire 60 items have been included in one cluster.

Depending on how closely linked items are, they join singly, or in clusters that are more related to each other than to joining items. The program used the variable “writing” as focal point around which the clustering procedure revolved, because “writing” had the highest F- ratio. Table 1 shows the outcome of this procedure. There are some general similarities between the outcome of this procedure and factor analytic [ 1 l] procedures. For instance, all of the items listed under Factor 1 in the principal components analysis used by STEENHIJIS

and BRYDEN [ 111 show the closest proximity to “write”in the clustering procedure. The items loading on Factor 2 (“pick up . . . “) cluster together in Table 1, and this is also the case for items that load under Factors 3 and 4. Thus, the clustering procedure on items gives results that are similar in nature to the factor analytic procedures used by STEENHUIS and BRYDEN

[ll] and HEALEY et al. [3]. Such procedures do not, of course, allow comments on the grouping of individuals.

Cluster analysis of individuals (BMDP, KM)

The 604tem questionnaire. The BMDP “K-MEANS” procedure separates subjects into a pre-specified number of cluster groups. The process can be speeded up by specifying an

*We used the BMDP statistical software manual, 1990 release, University of California Press.

HANDEDNESS GROUPS 375

Table 1. Schematic representation of the amalgamation clustering procedure, whereby items join the initial item (writing hand) singly or in clusters in terms of the degree of similarity they share with that item

Cumulative No. of items Items

Cumulative No. of items Items

3

10

13

14

15

16

17

18

19

20

21

22

24

25

26

30

31

32

write draw use pencil eraser

throw dart throw spear throw ball shoot basketball insert pin into material hold needle to sew hold tennis racquet

hammer manipulate tools knife to cut bread

33 tighten screw by hand

34 pick up penny

35 use paperclip

36 erase blackboard

37 hold cloth to dust

38 screw in light-bulb

40 point to distant object wave good-bye

pick up glass of water

dial push button phone

41

42

shoot marble

strike match

strike with fly-swatter

use tweezers

comb hair

hold razor

strongest hand

scissors

extract small object hand for picking up small objects

hand to hit someone

flip coin

45

46

47

48

49

50

51

52

53

57

pick up paperclip, pick up piece of paper, pick up marble, pick up pin

pick up screw

pick up book

58

59

60

hold heavy object pick up heavy suitcase carry heavy suitcase

hold cloth to wash face

unscrew jar lid

catch ball barehanded

pet cat or dog

wind stopwatch

hand used to break fall

turn on faucet

pick up phone receiver

shoulder to rest bat on swing axe over shoulder hand used to bat are you r/l batter

snap fingers

push open swinging doors

hand used to cartwheel

*Items bounded by lines are more similar to each other than to adjacent items outside the line boundaries. The farther away the number of a cluster is from the first cluster, the less similarity with the initital group of items there is.

indicator variable (e.g. “writing”) which is used by the program to identify initial cluster membership. “Writing” was used as indicator variable because of all variables it discriminates most clearly between the groups, and has the highest F-ratio. Use of “writing” as indicator variable is also justified on the grounds that in everyday usage the writing hand serves to label individuals as right- or left-handed. The program begins by establishing two clusters which are defined by the indicator variable “writing” (those who write with the left and those who write with the right hand) and then proceeds through several iterations until


all subjects have been assigned to the specified number of clusters. The procedure allows some flexibility in terms of choosing different indicator variables and numbers of clusters to be formed. In this case, the best solution was found for five clusters and use of “writing” as initial indicator variable. Evaluation of a “best solution” depends to a great extent on the Discriminant Function Analysis (see Table 3) that indicates the accuracy with which individuals can be assigned to the groups. For instance, if four clusters are specified, the Discriminant Function Analysis does not indicate nearly as accurate an assignment of individuals to clusters as is the case with the five-cluster structure. The results are shown in Table 2. In labelling the five clusters, convention was followed in terms oflabelling those who write with the right hand as right-handers and those who write with the left hand as left- handers. On this basis there were two groups of right-handers and three groups of left- handers. The question arises: How accurate was the determination of group membership for individuals? In order to answer this question, a Discriminant Function Analysis (BMDP 7M) was performed. The results are shown in Table 3. It can be seen that the individuals had

Table 2. K-means cluster analysis of individuals (Waterloo Handedness Questionnaire)

Cluster No. Name No. of members Group M Group SD

1 LEFT 40 1.685 0.404 2 WLEFT 42 2.332 0.740 3 ILEFT 36 3.404 0.678 4 WRIGHT 219 3.943 0.518 5 CRIGHT 308 4.414 0.428

Group M= aggregate mean rating on 5 point scale for the 60 items.

Table 3. Discriminant Function Analysis (BMDP 7M)* of K-means cluster results: accuracy of group classification and case assignment for Waterloo questionnaire

No. of cases classified into groups % Correct LEFT WLEFT WRIGHT ILEFT CRIGHT

LEFT 97.1 33 1 0 0 0 WLEFT 87.5 ; 35 0 0 0 ILEFT 87.5 0 4 28 0 WRIGHT 89.1 0 0 171 0 21 RIGHT 92.1 0 0 20 2 258

Total 36 38 195 30 279

*The Discriminant Function Analysis procedure, unlike the clustering procedure, rejects data from any individual for whom any of the items has a missing value.

been assigned to the five handedness clusters with a remarkable degree of accuracy. The analysis shows that, in general, individuals were not misclassified across the right-handed and left-handed categories. The program, of course, does not assign the qualitative importance to “writing hand” that we normally do-even though “writing” is considered the most important item for categorization in the program. In general, however, the questionnaire data allowed a highly accurate assignment of individuals into five distinct handedness groups. The reasons for labelling the three groups of left-handers as LEFT (consistent left choices), WLEFT (weaker left choices), ILEFT (inconsistent left choices) RIGHT (consistent right choices) and WRIGHT (weaker right choices) are given below.


The ,K-means cluster analysis provides F-ratios for each item and this permits a characberization of the different groups in terms of their preference patterns. The 10 most important preference items, in the order of greatest to least impact were: write, draw, use eraser at end of pencil, throw spear, throw dart, hold needle to sew, hold hammer, hold racquet, throw ball, use knife to cut bread. The least important of all 60 items were: “hand used to begin a cartwheel” and “hand used to wind a stop watch” and “hand used to snap fingers”. Table 4 shows the averages for the five groups for the 10 most important items. It can be seen that LEFT and RIGHT individuals have groups means that very strongly favor either the left or right hand. ILEFT individuals have group means which are intermediate; it should be noted that the group means hide the categorical distinctions between groups and that ILEFT do not only differ from the other groups in terms of the intermediate mean but in the con@stent choice of the right arm for throwing items. WLEFT and WRIGHT individuals have group means that are relatively close to LEFT and RIGHT groups and there is a question as to whether these should be grouped into distinct clusters. Table 4 shows the items

Table 4. Mean ratings for selected item groupings for the five handedness groups

Group LEFT WLEFT ILEFT WRIGHT RIGHT

Top ten items* Writing? Throwing1 Baseball4 Pick up item/I

1.2 1.4 3.3 4.1 4.9 1.1 1.2 u 4.8 4.9 1.3 1.4 4s 4.8 4.9 1.8 4.J 4.0 3.J 4.8 2.1 2.6 3.1 3.3 4.0

*List of the 10 items that were most important in classifying groups. ?-Items that relate to writing (write, draw, use eraser at end of pencil). IThrowing (throw baseball, throw dart, throw javelin). §Baseball batting (rest bat on shoulder, bat left/right). 11 Pick up item, average for seven questions (e.g. pick up pin).

on which the LEFT and WLEFT and RIGHT and WRIGHT differed most clearly. The WLEFT showed a strong tendency to answer the questions relating to baseball batting with a right hand preference, and the converse was true for the WRIGHT individuals. The ILEFT clearly favored the right hand for throwing. It should be noted that the averages for the “pick up” items give the most continuous distribution of choices across the five groups.

Our primary goal was to see if multiple handedness groups could be recognized with the help of a comprehensive questionnaire, and the answer was affirmative. But there is a legitimate question as to what would happen if this analysis would also be performed on a short questionnaire, similar in nature to the OLDFIELD [6] questionnaire.

The 14-item questionnaire

When the cluster analysis was performed on a short selection of items that were contained in the 60-item questionnaire (write, hammer, throw, unscrew lid of jar, hold knife to cut bread, use toothbrush, strike match, hold racquet, use scissors, draw, use spoon, hold sewing needle, hold bat, hold hockey stick), three clear clusters emerged (Table 5). There were two groups of left-handers and one group of right-handers. The left-handers divided into a group with quite consistent left hand preferences and a group that can best be described like ILEFT in the five-cluster solution. These individuals prefer the left hand for writing and drawing and


the right hand for throwing, as indicated in Table 6. The results of the Discriminant Function Analysis are shown in Table 7. It can be seen that the accuracy of classification is very high.

Table 5. K-means cluster analysis of cases for short questionnaire selection

Cluster No. Name

3 LEFT 2 ILEFT 1 RIGHT

No. of members Group M Group SD

59 1 SOS 0.385 48 3.018 0.788

537 4.608 0.312

Table 6. Mean ratings for selected item groupings for the three handedness groups established by the short questionnaire

LEFT ILEFT RIGHT

Top ten items* 1.3 2.9 4.7 Writing? 1.1 1.7 5.0 Throwing$ 1.6 4.J 4.7 Bat9 2.3 j.9 4.4

*List of the 10 items (out of 14) that were most important in classifying the three groups.

tWrite, draw. IThrow. $Bat baseball.

Table 7. Discriminant Function Analysis (BMDP 7M) of K-means cluster results: accuracy of group classification and case assignment for short

questionnaire

LEFT ILEFI RIGHT

Total

% Correct

96.4 78.0 99.6

No. of cases classified into groups LEFT ILEFT RIGHT

53 2 - 1 36 4

2 490

54 40 494

DISCUSSION

The recognition of three or five handedness groups does not relieve us of the obligation to provide some independent rationale of whether or not it makes sense to recognize these different clusters as meaningful and separate groupings. We can begin with the matter of naming the clusters. We chose names that are descriptive but this does not mean that the correspondence between what the program defines as a cluster and what we call it is perfect. There would be some agreement that the two groups of individuals classed as LEFT and WLEFT are left-handers and those classed as WRIGHT and RIGHT are right-handers.


Indeed, because the cluster analysis places much weight on the distinctions between these groups on the basis of an activity that is not universal (batting stance), there is a legitimate questian as to whether LEFT and WLEFT and RIGHT and WRIGHT should be considered distinct in a meaningful sense. There is less doubt about the distinctive status of ILEFT individuals. Typically, such individuals themselves are uncertain as to handedness and, asked about their handedness, tend to answer “I write with the left hand but , . .“. Because the ILEFT group emerges in the short and in the long questionnaire, and because this group has been shown to differ categorically from the other groups, with superior throwing performance of the right and superior fine manual performance on the left [8,9], it would make sense to recognize at least these three groups. On the basis of the preference pattern, the most c nvenient way of defining these three groups would be to classify individuals on the basis o P a short questionnaire, such as the OLDFIELD [6] questionnaire, according to whether they show consistent left (LEFT), right (RIGHT) preferences, or whether they prefer the left hand for writing and the right hand for throwing (ILEFT). Whether or not such a grouping is meaningful can only be determined relative to some other neuropsychological variables. Do ILEFT individuals have different language representation patterns than LEFT or RIGHT individuals? Will the pattern of apraxic disturbances after damage in the appropriate cortical regions differ between individuals who use different hands for different skilled activities and individuals who consistently prefer one hand for different skilled activities? These questions are not answered by the clustering methods and answers must await neuropsychological work that allows the examination of deficits of individuals relative to a somewhat richer handedness classification than is normally used.

Finally, an aspect of the data that raises difficulties for current genetic theories of handedness [ 1,5] must be discussed. In our sample, the item “throwing” is the major factor in defining the distinct group of left-handers who write with the left hand and throw with the right hand (ILEFT). Slightly less than half of all of the left-handed writers throw with the right hand; 47% in the present sample. That such individuals exist presents no difficulties to the two major current genetic theories of handedness [l, 51. However, both theories would expect’that for every left-handed writer who throws with the right, these should be at least one right-handed writer who throws with the left. This expectation was not met because there were only nine left-handed throwers among the 572 right-handed writers, an insufficient number for a separate cluster (IRIGHT) to be formed. The models predict that if no cultural pressures are active, the number of right-handed writers who throw with the left would be equal to the number of left-handed writers who throw with the right. In the presence of cultural pressures, some individuals who might have become left-handed writers and left- handed throwers will change to become right-handed writers and left-handed throwers, or right-handed writers and right-handed throwers. For this reason, any calculations of expected proportions of the two groups that are based on the models by Annett and McManus would expect more right-handed writers who throw with the left than left-handed writers who throw with the right. The exact magnitude of the two expected proportions would depend on various assumptions, e.g. whether or not hand choice for writing and throwing is considered independent or not.

The threat to the genetic models posed by the low proportion of right-handed writers who throw with the left hand can be reduced considerably if it is shown that there were some biases~in the questionnaire method that might favor a “right” set in right-handers. Such an argument would suggest that if questionnaires, especially very long ones, are given to subjeats in groups, the very large number of “right” responses given by the majority of right-


handers to most items might lead to a mindless set of simply going down the answer sheets with a “right right right” response bias. This would lead to a slight underestimation of the number of right-handed writers who would perhaps have responded with “left” if tested individually. It is conceivable that a study that had used individual administration of the questionnaire might have identified enough right-handed writers and left-handed throwers to form a sixth handedness group (IRIGHT).

In summary, this first attempt of clustering individuals instead of items on terms of hand questionnaire data raises some serious questions about the common neuropsychological practice of recognizing only two globally defined groups, left-handers and right-handers, when evaluating handedness relative to some other neuropsychological variable. There is no certainty that the different handedness clusters will be meaningful from a neuropsychological perspective but unless that evaluation is undertaken, the possibility cannot be discounted. Indeed, the seemingly unclear relationship especially between left-handedness as conven- tionally defined and various neuropsychological indicators, such as language localization, may eventually be resolved by the recognition of distinct subgroups with different neuropsychological characteristics.

Acknowledgement-This work was supported by Natural Sciences and Engineering Research Council of Canada Grant A 7054.

REFERENCES 1. ANNET~, M. Left, Right, Hand and Brain. Lawrence Erlbaum Associates, London, 1985. 2. ANNE’IT, M. A. A classification of hand preference by association analysis. The tribution of manual asymmetry.

Br. J. Psychol. 61, 30>332, 1970. 3. HEALEY, J. M., LIEDERMAN and GESCHWIND, N. Handedness is not a unidimensional trait. Cortex 22,33-53,

1986. 4. KRONENBERC, L. J. and BEUKELAAR, P. M. Towards a conceptualization of hand preference. Br. J. Psychol. 74,

3345,1983. 5. MCMANUS, I. C. Handedness, language dominance and aphasia: A genetic model. Psychological Medicine

Monograph Supplement 8. Cambridge University Press, Cambridge, 1985. 6. OLDFIELD, R. C. The assessment and analysis of handedness: The Edinburgh Inventory. Neurospychologia 9,

97-113,197l. 7. PETERS, M. Phenotype in normal left-handers: An understanding of phenotype is the basis for understanding

mechanism and inheritance of handedness. In Left-handedness: Behavioral Implications and Anomalies, Advances in Psychology Series, S. COREN (Editor), pp. 167-192. Elsevier, Amsterdam, 199Oa.

8. PETERS, M. Subclassification of left-handers poses problems for theories of handedness. Neuropsychologia 28, 27%289, 1990b.

9. PETERS, M. and SERVOS, P. Performance of subgroups of left-handers, and right-handers. Can. J. Psychol. 43, 341-358,1989.

10. PORAC, C. and COREN, S. Lateral Preferences and Human Behavior. Springer-Verlag, 1981. 11. STEENHUIS, R. E. and BRYDEN, M. P. Different dimensions ofhand preference that relate to skilled and unskilled

activities. Cortex 25, 289-304, 1989. 12. WILLIAMS, S. M. Factor analysis of the Edinburgh handedness inventory. Cortex 22, 325-326, 1986. 13. WITELSON. S. F. The brain connection: The corpus callosum is larger in left-handers. Science 229, 665-668,

1985.

Cluster analysis reveals at least three, and possibly five distinct handedness groups

Documents

Transcript of Cluster analysis reveals at least three, and possibly five distinct handedness groups