Limit of detection values in data analysis: do they matter?

Data sets containing values below the limit of detection (LOD) are known as ‘censored data sets’. Such data sets are encountered regularly in most fields of environmental contaminant research. When analysing these data, it is necessary to decide how values below the LOD should be treated. A range of approaches have been proposed, and adopted to varying degrees, within the peer-reviewed literature. These include: (i) treating the LOD value as the absolute value (i.e. if the LOD value is <0.5 then the value of 0.5 will be used); (ii) treating the LOD values as zero; (iii) excluding the LOD values from the data set; and (iv) substituting the LOD value with a value between zero and the LOD value, most commonly a value equivalent to half of the LOD value. Whichever method is employed, the values used to produce a non-censored data set have limitations and deficiencies when it comes to subsequent data analysis and interpretation. At best, these approaches serve as a simplistic means for approximating reality but, at worst, they may result in unjustified conclusions being drawn.