Over‐reliance on Completeness exposes ultraconserved elements datasets to non‐randomly distributed missing data

As larger amounts of DNA sequence data become available, so does the need for heuristic metrics summarizing the quality of such datasets. This creates two seemingly conflicting outcomes for the phylogenomics community: so called ‘point‐and‐click’ algorithms make analysis of genomic data easier, but automation may perpetuate biases caused by uncritical reliance on algorithms. Here, based on nine ultraconserved element (UCE) datasets from arthropods, we demonstrate that relying solely on the Completeness metric—a common tool to report amounts of missing data (MD) and inform choices about filtering loci—underestimates the actual amount of non‐randomly distributed MD and fails to prevent possible biases introduced by it. By only counting terminals sampled for each locus, Completeness overlooks the amount of MD in loci containing sequences that are truncated or gappy. Thus, datasets assessed for data quality only by this metric, and reported to be ‘complete’, can retain substantial MD concentrated in some terminals with extreme amounts of it, which may affect estimation of parameters often interpreted as biologically meaningful measures by phylogenetic algorithms (e.g., branch lengths). We demonstrate this by showing that MD could drive the generation of long branches and affect the support values of nodes containing taxa with high amounts of MD. Finally, we offer six recommendations authors can employ alongside or alternatively to Completeness , to keep the community standards of data quality strong.