Abstract
Secondary calibrations (calibrations based on the results of previous molecular dating studies) are commonly applied in divergence time analyses in groups that lack fossil data; however, the consequences of applying secondary calibrations in a relaxed-clock approach are not fully understood. I tested whether applying the posterior estimate from a primary study as a prior distribution in a secondary study results in consistent age and uncertainty estimates. I compared age estimates from simulations with 100 randomly replicated secondary trees. On average, the 95% credible intervals of node ages for secondary estimates were significantly younger and narrower than primary estimates. The primary and secondary age estimates were significantly different in 97% of the replicates after Bonferroni corrections. Greater error in magnitude was associated with deeper than shallower nodes, but the opposite was found when standardized by median node age, and a significant positive relationship was determined between the number of tips/age of secondary trees and the total amount of error. When two secondary calibrated nodes were analyzed, estimates remained significantly different, and although the minimum and median estimates were associated with less error, maximum age estimates and credible interval widths had greater error. The shape of the prior also influenced error, in which applying a normal, rather than uniform, prior distribution resulted in greater error. Secondary calibrations, in summary, lead to a false impression of precision and the distribution of age estimates shift away from those that would be inferred by the primary analysis. These results suggest that secondary calibrations should not be applied as the only source of calibration in divergence time analyses that test time-dependent hypotheses until the additional error associated with secondary calibrations is more properly modeled to take into account increased uncertainty in age estimates.
Highlights
The differences between median age estimates were significantly younger on average in all replications based on the T-tests (P < 0.05; Fig 4), and all but three replicates remained significant at the alpha = 0.05 level when the conservative Bonferroni corrections were applied to correct for multiple comparisons
Greater magnitude of variation in age estimates were associated with estimates closer to the root of the tree (Fig 5), the opposite pattern was observed when values were standardized by the median node ages (Fig 6)
Increased variance of age estimates was determined for nodes closer to the root, in which the age estimates were mostly younger for the secondary estimates measured by the minimum, maximum, and median values from the 95% credible interval (CI) (Fig 5A–5C)
Summary
What is the cost of implementing secondary calibrations in empirical studies? If our aim is to test hypotheses with robust methods that produce reliable age estimates, the cost of applying secondary calibrations might be too high with our current methods
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
