The role of misclassification in estimating proportions and an estimator of misclassification probability

TL;DRAbstract

Abstract. Dot grids are often used to estimate the proportion of land cover belonging to some class in an aerial photograph. Interpreter misclassification is an often-ignored source of error in dot-grid sampling that has the potential to significantly bias proportion estimates. For the case when the true class of items is unknown, we present a maximum-likelihood estimator of misclassification probability based on agreement between two interpreters. Two of the assumptions underlying the estimator are: (i) the probability that an interpreter makes a misclassification is constant, (ii) both interpreters have the same probability of misclassification. Simulation results suggest the estimator has acceptable performance when (ii) does not hold. This estimator can be used to investigate whether bias due to misclassification has exceeded a threshold, or to correct bias due misclassification.

Chat with Paper

AI Agents for this Paper