A recurrent controversy in the experimental analysis of behavior has been the degree to which one can generalize from the results of animal experiments to human behavior (Pierce and Epling pp 162-166; also see Hyten and Madden, 1993 for discussions of the issue). It has been claimed that human behavior under fixed-interval schedules of reinforcement shows different response patterns than do the results of the classic animal research. In particular, it is claimed that the human data typically do not show the 'scallop' (gradual increase in response rate until reinforcement) regarded as typical in the animal/experimental literature (see below).I have some doubts concerning this. I do not question the basic results; I'm sure that the researchers doing the human experiments did, in fact, observe the response patterns that they describe: typically "break-run" patterns as opposed to the "scallop" found in typical animal research (such as the Behavioral Experiments FI experiment). I do question whether this represents an inherent difference in the way human and nonhuman subjects perform under a given set of reinforcement contingencies; I suspect that the difference may be due to differences in experimental procedures in the human and animal research (but see Cole, 2001).
Conventionally, responding maintained by Fixed Interval schedules of reinforcement is characterized by a scallop. That is, there is typically a long pause following reinforcement, followed by a gradual increase in the rate of response until the next reinforcement. This pattern is typically shown by laboratory animals. Although this may not be true after extended training. It is also not typical of many human studies.
A "break-run" response pattern on a cumulative record is one where there is a pause after reinforcement and then an abrupt shift to a high, steady rate of response. There is no gradual acceleration period between the pause and the high rate of response at the end of the interval as there is in the fixed interval "scallop" produced in classic animal laboratory experiments. This is often found with human subjects.
One way to look at response rates in more detail is to view the rate of response as the average time between responses. We can then go beyond simply looking at the average to looking at how the times between responses are distributed (e.g., how many responses occur less than one half second after the previous response, how many between 0.5 and 1.0 seconds, etc). This is referred to as an "interresponse time" (or IRT) distribution (see pp 167-170 in Pierce and Epling for a description of IRT analysis).
Using this type of analysis, a break-run pattern should show most responses occurring at one of two IRTs; a very short one (high rate), or a very long one (pauses). This can be seen for minimal instructions in the example illustrated below.A scallop, on the other hand (shown for detailed instructions), should show a broader distribution of IRTs, corresponding to the gradually decreasing time between responses which produces the gradual increase in rate referred to as "acceleration". Remember, the slope of the cumulative record is a measure of the rate of response at that point. A gradual change in slope means a gradual change in rate, thus a number of different response rates over the course of the interval, and therefore a broad range of interresponse times.
Remember that short IRT's mean responses occuring close together in time, and thus at a high rate. A long IRT means a long time between responses, and thus a low response rate.Note that this refers to individual intervals! If individual intervals show break-run patterns, a summed cumulative record may still show a gradual increase if the runs start at different times during the intervals.
A second question concerns the effects of instructions on behavior. There is some indication that verbal instructions concerning the reinforcement contingencies in effect produce different patterns of response than does direct contact with those contingencies without instructions.To address these questions, we will use as our experimental setting a simple video game programmed on the Macintosh microcomputer.(available at http://www.mankato.msus.edu/dept/psych/ programs.html as RatRaceFI). This game involves a 5 X 5 matrix on the screen. A rodentoid symbol is located in the upper left corner of this grid; the subject's task is to gain points by moving the rat to the lower right-hand corner of the grid by pressing the <Q> and <P> keys. A point is recorded each time the rat reaches the lower right-hand corner.
The reinforcement schedule manipulation involves the relationship between key responses and the movement of the rat. We will define our operant as a press of either key; which key is pressed will be irrelevant to the schedule. Key presses will be reinforced on a fixed interval schedule of approximately 20 seconds.We will add a one-second limited hold contingency (also instructed) in order to eliminate possible timing behavior such as clock watching or the emission of homogenous sequences of responses (foot tapping, for example).
The effect of instructions will be addressed by running the subjects first with only general instructions, and then adding more specific instructions concerning the fixed interval reinforcement contingencies.Your subjects will be two experimentally naive human beings. Their task will be to play the computer game for three fifteen minute sessions (on three separate days).
On each day you will run two consecutive subsessions: a two point warmup session and a three point data session. Since each move of the 'rat' takes about 20sec, and 8 moves are required for a point, a point takes about three minutes. Therefore, each session will take about 15 minutes.
The reason for dividing the session into halves is so that we can use the first half of each session as a warmup, and record the data for only the second half of each session (i.e., the three point subsession).
The data itself will take two main forms, as shown in the illustrations above. The Segs column divides the summed intervals into sixteenths: the first bin is the total number of responses in the first 1.25sec of all of the intervals, the second bin the total number of responses in the second 1.25sec, and so forth. These can be used to plot a summed cumulative record, as illustrated below.
The IRT's column presents an IRT distribution of all responses, as explained in Pierce and Epling. Now we are not concerned with where in the interval responses occur, but rather with how much time elapses between successive responses. This data will be presented in the form of a histogram; illustrated above. Note that I've only plotted 8 IRT columns. You'll plot 16 (I'm trying to make a small graph readable).
We will be using a type of single-subject experimental design sometimes referred to as "multiple baseline"; one baseline for each of two subjects. Since this design introduces an experimental variable (verbal instructions) at different points in the experiment for the two subjects, it minimizes the likelihood that a change in response that occurs when the experimental variable is introduced for each 'subject is due to extraneous factors.
For the first session, both subjects will be given only general instructions (e.g.; "Your task is to earn points by moving the rat from the upper left-hand corner of the screen to the lower right-hand corner by pushing these buttons.").
For the second session, one subject will be given a more detailed description of the reinforcement schedule, while the second subject continues on the baseline (general instruction) condition.
For the third session, both subjects will be given detailed instructions.
Of course, you will 'debrief' your subjects at the end of the third session, informing them as to the purposes of the experiment and answering any questions that they might have.
Present your data for each subject individually in the form of graphs and tables. Your graphs should depict the cumulative fixed intervals (the first segment, the first plus the second segment, the first three segments, and so forth...)and the interresponse time distributions for the second half of each session (the three point subsession), for each subject, as described above.
Since there will probably be a wide range of values on the vertical axis of the IRT distributions you might want to use a logarithmic scale as I've illustrated above; one where the distance between 1 and 10 is the same as that between 10 and 100, 100 and 1000, etc. The easiest way to do this is to either buy semilog graph paper, or to use a graphing program that gives you the option of a logarithmic vertical axis. I've provided a sheet of the appropriate graph paper on the back of this handout. Make copies of it (you'll need one for each subject).
Combine all three graphs for each subject in a single figure. Do this for both the IRT data and the summed cumulative records. This makes it easier to see the effects of the experimental variable (the verbal instructions) on the subjects' behavior.
You should have a total of four Figures in your report:Summed Cumulative Record for Subject 1.
Summed Cumulative Record for Subject 2.
IRT distributions for Subject 1.
IRT distributions for Subject 2.It is best to present both graphs (the Cumulative record and the IRT distributions) for a given subject on the same page. This makes it easy to see the effects of your experimental variables.
You should also include a Table containing the Indices of Curvature (proportion of responses in the last quarter of the intervals). Calculate this by summing the cumulative number of responses in the last four of the segments and divide this number by the total number of responses. Append the rest of the data displayed on the computer screen to your report.
Finally, add a Discussion section commenting on your Results. How does the fixed interval behavior produced by this experimental situation compare to the behavior patterns Pierce and Epling describe?
Hyten, Cloyd and Madden, Gregory J. (1993). The Scallop in Human Fixed-Interval Research: A review of Problems with Data Description. The Psychological Record, 43, 471-500.Cole, M. R. (2001). The long-term effect of high- and low-rate responding histories on fixed interval responding in rats. Journal of the Experimental Analysis of Behavior, 75, 43-54.
4 cycle semilog graph paper for use with IRT distributions
You should have a sheet like this for each of your subjects::
