Data conventions explained
2025-10-23
VG.ASTR.Conventions.explained.RmdThis vignette explains the data conventions that ASTR follows.
Units
ASTR does not accept isotope, element, or oxide compositions without explicitly defined units. This design choice is intentional: units carry essential contextual information that determines how data can be transformed, compared, and interpreted. Missing or ambiguous units can lead to subtle but significant errors in downstream calculations, such as inappropriate normalization, invalid aggregations, or meaningless statistical analyses.
Measurement uncertainty
ASTR explicitly handles measurement uncertainty for all analytical data. Measurement uncertainty is a parameter encompassing the dispersion of analytical data both random and systematically. It provides a quantitative estimate of the quality of the measurement.
Random uncertainty: The standard deviation (SD) denotes the dispersion or variability of individual data points around the mean value of a statistically normal data set and expresses the precision of the individual data within the statistical sample. A small SD means that the individual data points are close to the mean value (low variability), whereas a high SD means that the individual data points are more widely dispersed (high variability). In a normal distribution, 1 SD covers the range of approximately 68% of all data; 2 SD covers the range in which approximately 95% of the data in the data set is defined. The ratio is expressed in the so-called 68-95-99.7 rule.
Systematic uncertainty: The standard error (SE) expresses the variability of the means of measurements (statistical samples) in relation to multiple repetitions (the statistical population). It hence expresses the accuracy of the measurement. The SE becomes smaller as the sample size increases, since higher repetition of measurements provides more reliable information about the true population mean. Confidence intervals are used as for the standard deviation. A 2-fold standard error (2SD) defines the range around the calculated mean that contains the true population mean with ca. 95 % probability.
Limit of detection and limit of quantification
ASTR allows for the intentional and meaningful handling of limits of detection/quantification. In each quantitative analysis, a threshold can be defined, below which a concentration of an analysed element can no longer be quantified. It is dependent on various influences, such as the instrument and method used (analytical sensitivity), the instrumental or baseline noise, the instrument stability, the element measured, matrix effects, measurement conditions including the laboratory environment (temperature, humidity), etc.
The limit of detection, usually expressed as LOD (or
other indicative notation), is defined as the smallest value that can be
reliably detected.
The limit of quantification (LOQ) is defined as the smallest amount that can be quantified with acceptable precision.
In archchem(), the limit of detection where indicated by
a below detection notation is automatically set to NA.
Users requesting a more advanced approach by valuing the
LOD in the ASTR package, e.g. for plotting functions, are
requested to implement their own lambda function redefining the
bdl_strategy.
Substitution methods could be e.g. dropping the left-censored value
by replacing it by NA or 0, calculating LOD/2 or LOD/√2,
skipping < of the left-censored value, or using regression models,
enhanced censoring calculations, or maximum likelihood estimates (Croghan & Egeghy,
2003; Giskeødegård & Lydersen,
2022; Helsel, 2006)