Observation Error and Bias in Accounting Research
探讨会计研究中观测误差（测量不准确）如何导致研究偏差，通过实验测量误差率并展示其对结果解读的影响，最后提出控制方法。
Observation error refers to measurement inaccuracies made by an observer of some attribute of an object or event. Observation errors can introduce bias into research findings to the extent that the mismeasurements are nonrandom. While observation error has been studied systematically in broad social science contexts (e.g., Rosenthal [1966] and Barber [1976]), the topic has received scant attention in the accounting literature. Nevertheless, observation error in research data has been found to occur more often than is commonly believed. To illustrate, error rates of 3.3 percent, 2 percent, 4 percent to 24 percent, and .5 to 12 percent were reported by Snodgrass [1972], Conrad and Hull [1967], Klemmer [1959], and Silverman [1973], respectively, in tasks that involved comparing two sets of alpha-numeric symbols, a task somewhat similar to a verification task. An analysis by Rosenthal [1978] of 21 research studies indicated that about 1 percent of all observations were in error and that two-thirds of these errors favored the hypothesis being tested. Observation error may occur just as frequently in accounting research contexts. In the following section, we describe attributes of observations that have been associated with error in social science contexts and relate these to accounting research situations. Next, we describe an experiment designed to measure observation error in an accounting context, and then demonstrate how the resulting bias contaminates a conventional interpretation of the research results. The concluding section of the paper describes control procedures designed to reduce the incidence of observation error.