J.J. Higa, L.A. Simm / Behavioural Processes 67 (2004) 501–509 503 Fig. 1. Illustration of test apparatus. Two glass tubes with cross-bars formed a square aperture or “hoop” through which the fish swam. The tubes contained an infrared light source and detector that detected movement through the hoop. A reflective mirror was mounted on a stand and located outside the tank. in individual 2.5 cm diameter glass tubes, separated by three glass cross-bars: two at the base forming a 3.75 cm square aperture (called a “hoop”) and one at the top stabilizing the tubes. The infrared light and detector created a beam that extended across the middle of the hoop. A mirror-box controlled delivery of a reinforcer (a reflective mirror). The box contained two 25W bulbs, arranged side-by-side, that back illuminated a one-way mirror attached to the front. The mirror measured 20 cm × 7.5 cm and was located on the long side of the tank, centered horizontally and 3.0 cm from the top of the tank. Turning the light bulbs off and on made the mirror reflective and transparent, respectively. An IBM-compatible computer and a program written in MED-PC controlled the experimental events and recorded breaks in the infrared beam as the fish swam through the hoop. To prevent recording one response more than once as the fish swam through the hoop, the program waited one second before detecting subsequent breaks in the beam. 1.3. Procedure The fish received training in preliminary study and were already trained to swim through a hoop 1. We exposed the fish to four FI schedules in the following order: FI 30, 120, 60, and 240 s. A program made available a reinforcer (20 s of a reflective mirror) for the first response after the FI requirement elapsed since completion of the previous reinforcer. Reinforcers were given immediately after the fish broke the infrared beam. Sessions consisted of 30 intervals and began with a 20 s presentation of the reflective mirror, marking the start of the first interval. Each condition was in effect for five sessions. 1 Ina preliminary study, the fish were given training with a reflective mirror until it consistently evoked the aggressive response. Next, the hoop was placed into the tank, just below the water level. The fish did not require specific training as they all readily swam through the hoop for mirror presentation reinforcers. 504 J.J. Higa, L.A. Simm / Behavioural Processes 67 (2004) 501–509 2. Results and discussion We analyzed the data and looked at several different measures of timing behavior. In all cases we used data from all intervals and all sessions for a total of 150 intervals for each condition, for each subject. 2.1. Cumulative records Fig. 2 presents cumulative records for individual subjects. Each record represents the mean time of a response in an interval. In some places one response (n) appears to occur at an earlier time on the x-axis than the preceding response (n − 1). Arrows in Fig. 2 point to examples of these instances. Backward excursions in the cumulative record occur because the data point represents an average of several responses. Hence, in some cases the average time of say the fifth response in an interval is shorter overall than the fourth response (e.g., see record for LB06). The first noticeable result is that some fish (LB01 and LB08) produced more responses than others (LB06 and LB10). Still, in most cases the fish responded at a higher rate towards the end of the interval than at the beginning, as indicated by changes in the slope of the records. Below, we report estimates of the point at which response rates change in an interval. Next, responding either increased gradually throughout an interval resembling an FI scallop pattern (e.g., LB08, FI 240 s) or remained high throughout an interval (e.g., LB10, FI 30 s). When comparing cumulative records across conditions, increases in the FI requirement produced lower levels of responding during the start of each interval. Similar patterns of responding during FI schedules have been reported in other species (e.g., Barnes and Keenan, 1993; Dews, 1970; Ferster and Skinner, 1957; Zeiler and Powell, 1994) as well as fish (e.g., Gee et al., 1994; Lejeune andWearden, 1991; Rozin, 1965; Talton et al., 1999). 2.2. Wait time and break point Fig. 3 (top) presents the meanwait time for each fish and the group across the different FI schedules. Overall, wait time increased with increases in the FI requirement. For example, mean wait time for LB01 increased from roughly 16 to 38 s across the FI schedules. A twoway repeated measures analysis of variance (ANOVA) comparing the main effect of the FI requirement and sessions indicated that the main effect of sessions was not significant, however, there were significant differences across the FI requirement, F(3, 36) = 23.726, p < 0.001 (alpha level set at 0.05). Pair-wise multiple comparisons (Tukey) indicated significant differences between the FI 240 and 30 s (q = 11.046, p < 0.001), FI 60 s (q = 9.269, p < 0.001), and FI 120 s (q = 8.006, p < 0.001) schedules. Other FI combinations were not statistically significant. Historically, there has been discussion over whether the relation between independent temporal variables and dependent measures such as wait time duration is best described by a proportional (e.g., Shull, 1971,1979; Starr and Staddon, 1974; Staddon and Cerutti, 2003) or power rule (e.g., Catania, 1970; Lowe et al., 1979; Zeiler and Powell, 1994). Given no significant differences in wait times among the smaller FI values, it is not possible to determine whether a linear or power function best describes the pattern of wait times because there are essentially two data points: one at FI 240 s and the other clustered at the smaller FI schedules. Another method is to look at wait time as a proportion of the FI value, given in Fig. 3 (bottom). The figure shows a pattern in whichwait time is a decreasing proportion of the FI value, as would be the case with a power function. However, although a one-way repeated measures ANOVA comparing relative wait time across changes in the FI requirement was significant, F(3, 9) = 26.169, p < 0.001, pair-wise comparisons show no significant differences between FI 60 and 120 s, and FI 120 and 240 s. It may be the case that wait times from fish are a less sensitive measure of timing than other measures such as break point. Talton et al. (1999) reported relatively small changes in wait time during different FI values, and found more systematic changes in break point times. Following the method reported by Talton et al. (1999),we estimated break points by first normalizing the pattern of responding in an interval, both in terms of time in an interval (x-axis) and the frequency of responses (y-axis). To do this we calculated the mean number of responses in bins of 1/10 the FI requirement, and normalized the overall level of responding based on the maximum generated in the interval. This was done for individual subjects. An advantage of this procedure is that it results in an equivalent number of points (10) over which to calculate a break point. Next, to calculate the break point, we used a linear regression analysis by J.J. Higa, L.A. Simm / Behavioural Processes 67 (2004) 501–509 505 Fig. 2. Cumulative records for each subject and condition. The data are based on all intervals and sessions. Each data point represents the average time of a response in an interval. which each normalized response-curve was fitted with two lines that intersected at bins 3–8 as possible break points. For example, we calculated the R2 of a line through bins 1–3 and a line through bins 3–10. In the next iteration we calculated the R2 of a line through bins 1–4 and 4–10, and so forth. The combination of lines that maximized the average R2 was used as an estimate of the break point; the value (i.e., bin number) was then multiplied by the bin duration that was used to generate the normalized curve for each condition so that we could represent the break point in terms of the actual time in an interval. The results of this analysis 506 J.J. Higa, L.A. Simm / Behavioural Processes 67 (2004) 501–509 Fig. 3. Mean wait time for individual subjects and the group as a function of the FI requirement (top). The data are also plotted as a proportion of the FI requirement (below). are presented in Fig. 4. As in the Talton et al. (1999) study, break point increased with increases in the FI requirement, and ranged from about 10 to 138 s, F(3, 9) = 20.403, p < 0.001, with significant differences between FI 30 s versus FI 240 s (q = 10.029, p < 0.001), FI 60 s versus FI 240 s (q = 9.032, p < 0.001), and FI 120 s versus FI 240 s (q = 6.804, p < 0.004). We next determined which of two functions best described changes in break point by calculating bestfit linear and power functions for individual animals as well as the group mean. The equations were: f (x) = mx + b (1) Fig. 4. Break point in an interval as a function of the FI requirement. Shown are the results for individual subjects as well as the group mean.