where x is the FI value and m, b, k, and n are parameters
whose values were based on the results of a program
in SigmaStat® that uses a Marquardt–Levenberg algorithm to determine the parameter values that minimize the sum of the squared differences between the data and predictedwait time durations. The results are presented in Table 1. Both functions fit the data from individual subjects well, with R2 values ranging from 0.946 to 0.992 for a linear function and from 0.958 to 0.994 for a power function. For a majority of fish (LB01, LB08, and LB10) a power equation resulted in a higher R2 value, however the difference was not statistically significant. Table 1 Parameter and R2 values from best-fit linear and power functions to individual and group break point for the different FI reinforcement schedules Subject Linear Power R2 m b R 2 n k LB01 0.974 0.637 − 14.609 0.994 1.428 0.0569 LB06 0.992 0.905 − 26.087 0.990 1.375 0.103 LB08 0.955 0.276 5.739 0.958 0.819 0.798 LB10 0.946 0.655 − 21.913 0.992 1.805 0.007 Group 0.988 0.618 − 14.217 0.999 1.369 0.076 J.J. Higa, L.A. Simm / Behavioural Processes 67 (2004) 501–509 507 Summarizing, both wait time and break point increased with increases in the FI value, with more consistent changes in break point. Furthermore, the break point pattern was well described by both a linear and power function. Still, for three of four subjects, the best-fit linear equation for the break points resulted in a sizeable negative y-intercept, which underestimates the results from the smaller FI schedules, hence it may be the case that a power function is a better description of the break point pattern. Interestingly, Zeiler and Powell (1994) found similar results in a study examining temporal performance of pigeons tested using a peak procedure with FI values ranging from 7.5 to 480 s. A possible explanation for our results is that the FI values chosen did not produce consistent and large enough changes in the pattern of swimming responses. Specifically, we may be seeing a floor effect in response latencies. Two fish, LB06 and LB10, often producedwait times longer than 30 s during the FI 30 s schedule, which indicates that the actual interval duration was longer than the programmed requirement. It may be that larger FI values (longer than 240 s) are necessary when studying interval timing effects in Bettas responding by swimming through a hoop for visual reinforcement. An even wider range of intervals (we currently tested up to an eight fold change in the interval duration) may be necessary for characterizing performance during FI schedules. 2.3. Scalar timing Another common timing result is a constant coefficient of variation (CV) across a range of interval values, a form ofWeber’s law referred to as scalar timing (e.g., Gibbon, 1977; see also Staddon and Higa, 1999 for discussion of differences between proportional and scalar timing), which is observed under most steady state conditions, although there are some data showing that it is sometimes violated at relatively shorter (e.g., milliseconds) and longer interval durations (e.g., Crystal, 2003; Gibbon et al., 1997; Zeiler and Powell, 1994). We checked for the scalar property in wait times using a method described by Lowe and Wearden (1981): we plotted the standard deviation of the wait times against the mean wait time and conducted a linear regression analysis (Fig. 5). Each point is based on the data from a session. Although overall levels of each measure varied across subjects (note x- and y-axis scales), all showed a positive relation between standard deviation and mean wait time. The regression analysis indicates that a line fit the data fairly well, with R2 values ranging from 0.76 to 0.83 and slope values varying from 0.61 to 0.93. Furthermore, while a majority of the intercepts were negative, none were significantly different from zero, as confirmed by a t-test (LB01, p = 0.59; LB06, p = 0.60; LB08, p = 0.67; LB10, p = 0.52). Intercept values that are near zero are excepted if wait time is a constant fraction of the mean, which is a form of scalar timing. Lowe and Wearden (1981) reported similar results from a regression analysis on data from rats and pigeons: Slope values ranged from 0.45 to 0.98 with negative intercepts for three of four data sets that did not differ significantly from zero. Thus, like some birds and mammals, the wait times of Bettas also appear to follow a scalar timing rule. Prior studies have examined learning processes in Bettas using a method that involves visual reinforcement and a swimming response (e.g., Lattal and Metzger, 1994;Wirth et al., 2003). These studies show, for example that the stimulus control of swimming is similar to that found in other fish (e.g., goldfish, Yarczower and Switalski, 1969) and other species (e.g., pigeons, Reynolds, 1961). The present study extends what is known about Betta behavior to interval timing. Overall, our results suggest that Bettas were sensitive to the different times at which a reinforcer was available. In particular, we observed cumulative patterns of responding that were, in many cases, positively accelerating as the interval elapsed. This result supports the results from other species, including other fish species, in which an FI scallop was reported (e.g., Dews, 1970; Talton et al., 1999). Furthermore, although wait times did not differ significantly during the smaller FI values, wait times were much longer during the largest FI schedule (FI 240 s). Break point in an interval, too, increased with increases in the FI schedule. Although studies with fish (e.g., Talton et al., 1999) have reported proportional timing effects with goldfish responding for food, we found only provisional evidence of a linear relation betweenwait time and break point and changes in the FI value. Our analysis suggest that linear and power functions both provide suitable descriptions of the results. Lastly, we found evidence of a scalar timing effect when comparing mean wait times against the 508 J.J. Higa, L.A. Simm / Behavioural Processes 67 (2004) 501–509 Fig. 5. The relation between the mean and standard deviation of wait times for individual subjects. The data are from each session and all conditions. The straight line represents the best-fit line from a linear regression analysis. Also shown are R2, slope, and intercept values for each line. standard deviation, which is comparable to results on wait time and FI schedules in other species (e.g., Lowe and Wearden, 1981). Summarizing, the present study demonstrates that it is possible to study timing behavior in Bettas, using mirror presentation reinforcers for swimming through a hoop. Bettas showed timing effects consistent with other timing studies that use different procedures and species. However, it is not yet clear whether a proportional or power function best characterizes changes in wait time or break point. It may be that our results represent species differences or is limited to the particular method used (i.e., the type or duration of the reinforcer or the response requirement). In addition, unlike most studies with rats and birds, our conditions lasted for only a few sessions at a time, and it may be that chronic exposure is necessary for both properties of interval timing. The present study provides evidence supporting a method that researchers may use to investigate these and other factors involved in the timing behavior of fish and evaluate theories about the underlying mechanism.