How many specimens do I need? Sampling error in geometric morphometrics: testing the sensitivity of means and variances in simple randomized selection experiments

摘要: One of the most basic but problematic issues in modern morphometrics is how many specimens one needs to achieve accuracy samples. Indeed, this regularly posed questions introductory courses. There no simple and certainly absolute answer question. However, there are a number techniques for exploring effect sampling, our aim provide an example might function simplified informative way. Thus, using resampling methods sensitivity analyses based on randomized subsamples, we assessed sampling error horse teeth from several fossil populations. Centroid size shape upper premolar (PM2) were captured Procrustes geometric morphometrics. Means variances (using three different statistics variance) estimated, as well their confidence intervals. Also, largest population sample was randomly split into progressively smaller subsamples assess reducing affects statistical parameters. Results indicate that mean centroid highly accurate; even when small, errors generally considerably than differences among In contrast, estimation requires large samples tens (ca. >20), although requirement may be less stringent variance very small (e.g. populations underwent strong genetic bottlenecks). Variance either or can inaccurate samples, point makes it variable spatially chronologically well-separated populations, including two which distinctive consequence artificial selection. Likely, require <15–20 reasonable degree accuracy. analysis largely congruent with pattern suggested by bootstrapped intervals, observations previous study African monkeys. The performed, especially assessment, do not much time computational effort; however, they necessitate at least (50 more specimens). If type became common morphometrics, could effective tool preliminarily exploration results therefore assist assessing robustness. Finally, use studies increases, present case form part set examples allow us better understand estimate what desirable be, depending scientific question, data taxonomic level under investigation.