Assessing change with short time-series data is difficult because visual inference is unreliable with such data, and current statistical procedures cannot control Type I error because they underestimate positive autocorrelation. This article describes these problems and shows how they can be solved with a new interrupted time-series analysis procedure (ITSACORR) that uses a more accurate estimate of autocorrelation. Monte Carlo analyses show that, with short series, ITSACORR provides better control of Type I error than all previous procedures and has acceptable power. Clinical examples also show that ITSACORR is easy to use and functions well with real data.

The principal aim of the present experiments was to assess whether punishment increased or decreased the rate of unpunished behavior (contrast and induction, respectively) for which reinforcement rate was held constant, with physical and nonphysical punishers (electric shock and response cost), pigeon and human subjects, signaled and unsignaled components (multiple and mixed schedules), and the presence or absence of a blackout period between components. Across the three experiments there were 20 punishment conditions. Induction was found in nine of those, less consistent response‐rate reduction was found in three, contrast was found in four, and in four there was no change in responding from conditions without punishment. Contrast occurred consistently only with multiple schedules during the first exposure to electric‐shock punishment. Induction and no change, however, were found with every combination of the independent variables studied. Four conclusions regarding the interactions between punished and unpunished responding emerged from the present results: (a) Both contrast and induction occurred with the reinforcement rate held constant and a blackout between components, (b) induction was more common than contrast, (c) contrast occurred only in the presence of a stimulus different from that correlated with the punisher, and (d) contrast diminished with prolonged exposure to punishment. None of the current theoretical accounts of punishment contrast can explain the present results.

No abstract

No abstract

Punishment improves discrimination learning, and programmed instruction is an elaborate form of discrimination training, so the present experiment assessed whether punishment also improves performance on programmed instruction. The cost of such improvement in terms of increased training time and dissatisfaction of subjects also was assessed. Three college students completed a computerized version of Holland and Skinner's (1961) programmed text. One subject received a two-component multiple schedule within a reversal design, and two subjects received the same twocomponent multiple-schedule without a return-io-baseline phase. During baseline, subjects received 5¢ for each frame completed, regardless of whether responses were correct or incorrect. Multipleschedule conditions were either baseline conditions or a loss of 5¢ for each incorrect response (punishment). Punishment improved performance by 10%, increased training time by 15%, and did not affect reported satisfaction. The most likely mechanisms for this improved performance are that punishment increased study time or the salience of stimuli. This experiment showed that punishment can improve performance by one letter grade without subject dissatisfaction or significantly increasing training time.