Monograph: Biases in the recovery and interpretation of micropalaeontological data

Bias caused by collecting and processing bulk samples is largely independent of what fossil clade or mineral is searched for. Instead, different methods bias the data to a different, frequently very large, degree. Furthermore, biases accumulate with each recovery step, and the sum may be extreme unless appropriate methods to minimize it are employed. The effects depend on what the data are used for, e.g. establishing range ends (zonal boundaries), taxonomy (frequencies as an aid to, for example, conodont apparatus reconstruction) and ecology (relative frequency in a fauna, frequency ⁄ kg, faunal diversity). However, the best published methods remove calcium carbonate and dolomite without bias. All rocks with such cement can be broken down without bias, and so can some claystones with little lime. The bias caused by concentration can be measured, kept low, and documented. Removal of clay is an exception: screening or decanting removes all small elements. Extraction methods should be stated in all publications so that the data can be assessed more fully and quoted properly. Information about the acid methods, screen hole diameter and collection size are especially important because these usually cause the greatest bias. Reliability of observed range ends increases with increasing number of specimens and with decreasing sample distance (recollecting near the boundary). Samples that are too small, yielding subadequate collections, can strongly bias placement of zonal boundaries and implied diversity. Not taking the uncertainty intervals of zonal boundaries into account may result in artificially extended observed ranges of other species. Methodological progress over the last 25 years has increased the potential average yield per hour of manpower over 100 times for samples yielding fewer than 100 elements ⁄ kg. This has made it possible to overcome most of the biases outlined herein. Similarly, taking the biases that are known or are likely to have affected data into account allows levels of precision in data to be evaluated and published.