1976] 
Aspey — Response Strategies of Schizocosa 
97 
structed in which the preceding acts are listed as the horizontal 
rows and the following acts are listed as the vertical columns 
(Table I). 
Such a matrix indicates how frequently given behaviors immedi- 
ately follow or precede other specific behaviors, and similar tech- 
niques have been employed by a variety of investigators (Andrew, 
1956; Altmann, 1965; Hazlett and Bossert, 1965; Delius, 1969; 
Wilson and Kleiman, 1974). The construction of an inter-indi- 
vidual transition probability matrix to determine which behaviors 
were exhibited by one spider in response to another’s behavior is 
discussed in Aspey (1976b). Although procedures for constructing 
the intra-individual transition probability matrix of the present 
study are fundamentally the same as for the inter-individual transi- 
tion probability matrix (Aspey, 1976b), no attention was paid in this 
analysis to acts performed by other spiders and their possible in- 
fluence. To obtain the matrix, the complete sequence of agonistic 
behaviors was broken down into a series of two-act sequences. To 
illustrate, the four-act sequence Jerky Tapping — Following 
Walk — Oblique Extend — Vibrate-Thrust, provided three two- 
act sequences: Jerky Tapping — Following Walk, Following 
Walk — Oblique Extend, and Oblique Extend — Vibrate-Thrust. 
To determine when any two behaviors performed by the same 
spider were significantly linked beyond chance expectation, the 
method developed by Andrew (1956) and employed by McKinney 
(1961) and Wilson and Kleiman (1974) was used. Two behaviors 
were considered significantly linked if the difference between the 
observed and expected values was greater than three times the 
square root of the expected value. In other words, the square root 
of the observed value was estimated as the standard error of the 
expected value, and a difference of more than three times the stand- 
ard error between the obtained and expected totals was regarded 
as significant with a deviation of 2.58 times the standard error cor- 
responding to p < 0.01 level of significance. With large samples 
the distribution was the same as for chi -square; however, Andrew’s 
method was not restricted by the assumptions of chi-square, name- 
ly it: (1) did not assume independence of each variable; (2) allowed 
expected frequencies of zero; and (3) allowed expected frequencies 
for five or fewer in more than 20% of the cells. Therefore, Andrew’s 
method had the same power as chi-square but allowed for the analy- 
sis of infrequently occurring behaviors, and did not assume that all 
