A way of stating questions that permits either answers to the question or responses to random events, such as a coin toss. This approach is useful for forecasting socially undesirable behavior. For example, to forecast the profitability of a proposed chain of convenience stores, an analyst might need to forecast theft by store employees. Wimbush and Dalton (1997) examined ways to predict which job candidates might become thieves. Theft can be expected to vary by situation, by method of estimation, and by the definition of a theft (amount stolen and time period). Previous research led to widely varying estimates of theft, ranging from 28% to 62%. (Even at the lowest figure, this represents a major cost.) Wimbush and Dalton thought direct questions were not reasonable because people would lie. When they asked 210 employees on an anonymous questionnaire, “Are/were you involved in theft from your employer of from [dollar amount specified] in cash, supplies, or merchandise a month?” 28.2% admitted to theft. Wimbush and Dalton then used more appropriate methods. One was the Randomized Response Technique: they asked interviewees to flip a coin in a self-administered survey (only the respondent knows how the coin lands) and then answer the following question:
“If your coin flip is a head OR if you are/were involved in the theft from your employer of [dollar amount specified] in cash, supplies, or merchandise a month, please put an “X” in the box to the right.”
When asked this way, the estimated percentage of thieves was 57.9%. Another approach was the Unmatched Count Technique: they gave 353 respondents two sets of questions; half had five items and the other half had six items to choose from, such as “I have been to Spain,” or “I currently have one or more cats.” The sixth item was the question used above in the direct questioning approach. They asked respondents to indicate whether any of the items was true. The percentage of thieves, as estimated by the Unmatched Count Technique, was 59.2%.