Human Production of “Random” Numbers

Abstract

Six statistically naive and six statistically sophisticated subjects each wrote a string of 2,520 random numbers. Although subjects did not use the ten digits equally often, there was not much agreement among them about which digits they preferred and which they avoided. There is much more consistency in the way subjects used pairs of digits and triplets. Subjects avoided repeating digits even when they were separated by a different third digit, and they used consecutive increasing, consecutive decreasing, and permutations of consecutive digits much more frequently than one would expect by chance. The best single predictor of a digit that a subject will use is the digit which has just been produced. Significant sequential dependencies among digits can be demonstrated for strings as long as 7 for statistically naive and as long as 5 for the statistically sophisticated subjects. To a lesser extent the statistically sophisticated subjects exhibited the same kinds of preferences and avoidances as the unsophisticated ones. There is a small but statistically significant negative correlation between numbers preferred in randomization and their ease of recall in short-term memory.