The process of numeric comparison was investigated. Four groups of 10 5s were asked to judge which of two digits or which of two dot patterns was numerically larger. Stimuli were either digits or dot patterns in familiar, unfamiliar, or random configurations. Mean reaction time was systematically related to the difference between logarithms of the stimulus values. A single numeric comparison process gave good account of the data for all Stimulus types. This process was well described by a random walk model with variable step size and fixed boundaries. Reaction time matrices were further analyzed using Kruskal's 1964 multidimensional scaling program MD-SCAL, and the recovered stimulus configurations were successfully simulated from a simple version of the model.

No abstract

Sumn2asy.-Rats and fish were trained to give a free operant response under partial versus consistent reinforcement and high versus low effort. Resistance to extinction was then measured in terms of number of responses to criterion ( R ) , time to criterion ( T ) , and rate to criterion (R/T). Effort failed to affect any of these measures in either species. In rats partial reinforcement produced greater resistance to extinction thzn consistent reinforcement on R and T, but not on R/T, whereas partially reirforced fish failed to differ from those consistently reinforced on R and T bur exceeded the consistently reinforced fish on R/T. The R and T measures were correlated in both groups of fish but only in the consistently reinforced rats. Partially reinforced fish were less variable than consistently reinforced fish on the T m t s u r e . It is suggested that the principal effect of partial reinforcement in rats is on mean time to extinction, but the effect in fish is on variabiliry of time to extinction.In their work on the partial reinforcement effect (i.e., increased resistance to extinction of a response resulting from intermittent omissions of reinforcement during training, PRE), Bitterman and his co-workers have studied several vertebrate and invertebrate species using both positive operant conditioning methods and negative classical conditioning methods. We are concerned here primarily with those studies in which positive reinforcement has been used with white rats and with the African mouthbreeding fish (Tilapia macrocephala). After training fish to suike at a target, training was continued, using a discrete trial technique and schedules of 50% and 100% reinforcement (Wodinsky & Bicterrnan, 1959). Response latency during subsequent extinction trials served as a measure of response strength. These procedures resulted in greater resistance to extinction in the 100% group than in the 50% group. Greater response strength for consistently reinforced Ss than for partially reinforced Ss also was demonstrated with widely spaced trials (Longo & Bitterman, 1960) and with extensive training (Wodinsky & Bitterman, 1360). The learning curves of the partial and consistent groups in the latter smdy also revealed a decrease in latency following nonreward after about 20 days of training at 20 trials per day. A later study demonstrated that this effect occurred whether the 50% schedule was random or alter-

No abstract

The advantages of loosely coupled hierarchical computer networks for psychological research are discussed. Hardware methods for interconnecting computers are considered, and software communication protocols are detailed. A hierarchy of interconnected computers is formed: (1) large campus computer for statistical analyses, (2) medium scale computer for data concentration and program development, and (3) laboratory computers for data collection. This hierarchy provides efficient computing at all levels and insures maximum computing power at minimum cost.The laboratory computer has emerged as a powerful flexible tool for psychological research. A natural extension of this tool is the interconnection of laboratory computers to increase this power. The connection of computers into networks has become a popular topic in computer science as well. Invariably, computers are interconnected to share the work of compu ting efficiently, and the resulting networks can vary in size from two minicomputers to a system as large as the ARPANET. When work is shared between computers, this is called distributed processing. The purpose of this paper is to discuss ways in which laboratory computers may be interconnected for communication with each other. It will then be argued that a hierarchical network provides the optimal organization for laboratory computing in psychology. Finally, the operation of two such systems at the University of Wisconsin are described.There are two basic methods for interconnection of computers. By the first method, the computers share one or more banks of memory. This may be accomplished by direct memory access channels or by multiport memories, but either method is expensive and requires the computers to be in close proximity. Furthermore, the operating systems in each computer must be closely interlocked to insure that memory is shared correctly, i.e., that one computer is not modifying a memory location as it is being read by the other. Operations of one computer depend so closely upon operations of another, that such computers are called tightly coupled. The second method of communication between computers requires that each computer act as a peripheral device to the other. There is simply a program module in each computer that handles input-output to the other computer, either by direct memory access or through registers. Such computers are called loosely coupled, and this organization places many fewer constraints on the operating systems of each computer. Long distance communication is possible