Direct observation occurs when an observer classifies units of behavior with a category system at the same time that the behavior occurs. In contrast to observations made with recording media, the observer must keep up with the tempo of the events being observed. The preceding five minutes cannot be replayed to check the classifications which were made. However, this disadvantage may be offset by not needing expensive recording equipment to obtain observations, and by the increased amount of information which is available when the observer makes his encoding decisions.In this article we will examine direct observation procedures which can be used to measure classroom behavior variables. All of the observation procedures discussed here may be used live with a single observer using no electronic recording equipment. In the first part of this paper, the uses of direct observation will be explicated in the context of research studies and in evaluations of planned change. The second part will describe several observation systems in order to provide examples and to highlight a number of the important attributes possessed by direct observation systems. The final section of the article will discuss problems frequently encountered when direct observation is used.
Research Uses o/ Direct ObservationGenerating Prediction Variables. One of the research uses of direct observation has been to generate prediction variables. Typically, behavioral categories are combined in some predetermined manner in order to form operational definitions of the variables to be studied. A frequently occurring example is teacher indirectness, found in Flanders 1 and numerous other studies using his observation system or similar ones. This variable is defined as the ratio of the amount of lesson time spent in the categories: teacher accepts feelings, praises and encourages, accepts and uses student ideas, or asks questions, to the amount of lesson time spent in: teacher lectures, gives directions, and criticizes. Teachers' scores may be arranged into naturally occurring groups (e.g. high, medium, low) on the variable. Comparisons among groups' scores on the dependent measures are then made. Alternatively, the teacher behavior variable simply 474 EDMUND EMMER may be correlated with the dependent variables, assuming that the relationship is primarily a linear one. Usually some attempt is made to control for initial differences on particular variables through the analysis of covariance, partial correlation, or analysis of residual (regression derived) scores. Flanders, 2 in a series of studies in 2nd, 4th, and 6th grade classes, used a number of variables operationalized in his system to predict pupil achievement and attitude scores. Tisher8 used a modification of Smith and Meux's categories to classify a sample of nine science teachers into high, medium, and low groups according to the frequency of highercognitive behavior they required of their students. Withall's system was used to classify teacher socio-emotional behavior, and to group teachers (...