Observed Errors in Distance Estimation 2010-01-0046
In order to evaluate the variation in distance estimation accuracy, a survey was conducted during which 123 subjects estimated distances to static objects in a roadway setting. The subjects (which included many police officers) tended to underestimate distances to objects that were from 21 to 383 feet away; the average estimation error was −8.6% while the median error was − 22%. The variation in performance among individuals was extremely large, with extreme errors ranging from − 96% to + 811%. The distribution of error did not conform to a Gaussian (normal) distribution because of the skew of the observed error distribution towards large positive errors. Box plots were used to identify nine “outlier” respondents who produced a total of 15 error estimates which were extraordinary in their difference from the rest of the data. Tests for independence indicated that the likelihood of an error estimate being an outlier was not influenced by gender, age, or whether the subject was a police officer. When males were compared to females, both were found to have median estimation errors that were negative, corresponding to under-estimation. When nonparametric statistical methods were used to compare the median distance estimation errors, females' under-estimation errors were found to be larger (more negative) in a statistically significant way. In this survey, in general, the most accurate estimates were given by older males at short distances.
When asked to estimate the length of an average car, there were seven outliers. There was no statistically significant relationship between the likelihood of being an outlier and the subjects' gender, age, or whether they were a police officer. The median under-estimation errors for car length exhibited by females and younger respondents were larger, and were statistically significant.