370 
Psyche 
[Vol. 85 
To sum, no significant differences in relative importance of 
behaviors were observed over the three colonies. Histograms of 
behavior frequency (Fig. 1 ), rankings of ethogram frequency (Table 
3) and relative time budget frequencies (Table 4) were not signifi- 
cantly different. Therefore, in subsequent analyses of within-species 
social organization below, data were pooled over three colonies. 1 
will return to consideration of between-colony variation below. 
Division of labor among workers was investigated by considering 
the matrix of transition frequencies among behaviors listed in the 
ethogram. Within this 46 x 46 matrix, elements indicate how often 
each behavior followed and preceded every other behavior. To sim- 
plify presentation of the results, transitions among behaviors listed 
in the ethogram are synopsized in Table 5. Diagonal elements 
represent the frequencies by which behaviors in the same categories 
followed each other whereas off-diagonal elements represent transi- 
tion frequencies between behaviors in different groups (Herbers and 
Cunningham 1983). Division of labor is implicated if nonzero tran- 
sition frequencies are clustered in diagonal blocks of the matrix and 
zeroes occur in off-diagonal blocks. 
Examination of Table 5 shows that transitions from (column 1) 
and into (row 1) Personal Behavior commonly occurred. This is no 
surprise, since virtually all workers displayed a form of personal 
behavior. However, among social behaviors, the overall distribution 
of nonzero transitions deviated strongly from random expectation 
(G = 55.24, 1 6 df, P<.00 1 ), in a pattern consistent with organization 
of behaviors into roles: behaviors within the group Brood Care were 
positively correlated in time, as were those within the groups Social 
Interactions, Physical Maintenance, and Colony Provisioning. 
Between these groups, there were significantly fewer transitions than 
random expectation (Table 5). The pattern of overabundance of 
nonzero transitions in diagonal blocks, and under-representation in 
off-diagonal blocks was absolutely consistent with expectation. 
Overall, nonzero transitions clustered in diagonal blocks, thereby 
providing statistical evidence of polyethism. 
Worker behavior can be provisionally categorized into four roles: 
brood care, social interactions, physical nest maintenance, and pro- 
visioning, since transitions among behaviors within a role occurred 
more often than random expectation whereas links between roles 
occurred less often than by chance. Information from the single-step 
