Flawed Analysis? A Response to Myhrvold

Gregory M. Erickson,Peter J. Makovicky,Brian D. Inouye,Chang‐Fu Zhou,Ke‐Qin Gao

DOI: https://doi.org/10.1002/ar.23187

2015-01-01

The Anatomical Record

Abstract:Nathan Myhrvold contends in the recently published letter to the editor (Myhrvold, Anat Rec 2015;298:489–493) that he could not replicate the results from our growth analysis for the ceratopsian dinosaur Psittacosaurus lujiatunensis that was published in The Anatomical Record (Erickson et al., Anat Rec 2009;292:1514–1521). Myhrvold raised the same concerns about this study in an earlier publication (Myhrvold, 2013) and in the popular press (Chang, 2013) to which we responded (e.g. Erickson et al. [accepted]). In our re-examination we found that we did not provide sufficient data for others to easily replicate each of the many steps that went into making this taxon's growth curve. This unfortunately led to Myhrvold attempting to recover our results using values largely garnered from the graphics. This should have produced similar results but it appears he utilized different subsets of our data in fitting growth curves than we described, and does not appear to have used the same Logistic function we applied. We also discovered a few transcription mistakes in data reporting that fortunately were not utilized in our growth curve analysis, but may have contributed to Myhrvold's misunderstanding about our study. Here we: (1) explain the steps and reasoning behind our methods; (2) provide the data utilized in making the growth curve so that others can use them in their own work, and (3) correct the presentation errors. In our study, we sought to determine the population demographics for Psittacosaurus lujiatunensis from the Lujiatun Bed of the Lower Cretaceous, Yixian Formation in China (Erickson et al., 2009). We histologically aged and estimated the body mass for 22 individuals spanning sizes from hatchlings to some of largest specimens known. The latter (8–11 year olds) show clear signs of nearing full adult size in the form of fused neurocentral sutures (Brochu, 1996; Irmis, 2007), and advanced histological maturity [attenuating growth lines, a shift to avascular bone in the outermost growth zones, thick endosteal deposits, and osteoclastic remodeling; (Francillon-Vieillot et al., 1990; Castanet et al., 1993; Reid, 1997)] (Fig. 1). The relationship between femoral length and age from these specimens was then used to estimate ages for an additional 58 specimens that we examined from the population. The demography, not the growth curve, was used to make an ecological life table and describe the survivorship patterns. We also used the 22 histologically aged specimens to make a growth curve. This helped to: (1) confirm that the dinosaurs came from a catastrophic assemblage (Voorhies, 1969). [The animals appear to have died simultaneously in an ash fallout; (He et al., 2006)]; (2) inform us as to the type of life table analysis to conduct; static versus dynamic (Deevey, 1947); and (3) determine the maximal growth rates so that this taxon's rates could be compared to those for other nonavian dinosaurs and living vertebrates (Erickson, 2014; Grady et al., 2014). This last point required a biologically reasonable estimate of adult mass for comparison on a log-log plot of vertebrate growth rates standardized to body size [figure 6 in Erickson et al. (2009)]. Psittcosaurus lujiatunensis osteology and histology showing the slowing of growth. Skeleton (left) of LMP-R00117 (Liaoning Paleontological Museum, Shenyang Normal University, Shenyang, China), one of the largest out of hundreds of specimens known for this taxon (or = orbit, rh = right humerus, ju = left jugal, ax = the axis or first cervical vertebra). It shows signs of advanced somatic maturity such as fused cervical neurocentra. The arrows point to the fully closed suture for the neurocentrum of the fourth cervical vertebra (c4). The fusion line that once spanned between the arrows is completely capped by periosteal bone and is no longer visible. The histological section of the humerus (right) from PKUVP 1054 (Peking University, School of Earth and Space Sciences, Beijing, China), an animal just 8-years-old. It already shows signs of advanced maturity including extremely attenuated growth line spacing (stacked dark lines spanning left to right at the top of the image), large osteoclastic erosion rooms forming haversian canals (ovoid structures at the bottom of the graphic), and diminished vascularization towards the periphery of the element (top) with the final zones (i.e., bone between the growth lines) being nearly avascular. All of the larger specimens we sampled show similar signs of being nearly full-grown. These data were used to deduce that, of the commonly utilized equations used in zoology to describe growth, only the Logistic provided a realistic estimate of full adult size needed for making interspecific growth rate comparisons in our study. We fitted sigmoidal functions to the growth data because growth throughout development is predominantly sigmoidal in crocodilians (living archosaurian reptiles) (Wilkinson and Rhodes, 1997; Richardson et al., 2002; Erickson, 2014) and birds (living archosaurian dinosaurs) (Ricklefs, 1967), which constitute the phylogenetic bracket for nonavian dinosaurs. This growth pattern has also been demonstrated in both ornithischian and saurischian dinosaurs using growth series where older adult specimens were available (e.g., Erickson et al., 2004; Bybee et al., 2006; Cooper et al., 2008; Lehman and Woodward, 2008). On the basis of both lines of evidence, it is most parsimonious to assume dinosaurs grew in the same manner (Erickson, 2014; Grady et al., 2014). Of the common functions traditionally utilized in zoological research: Logistic, Gompertz, and von Bertalanffy (Ricklefs, 1968; Peters, 1983), we found that the Logistic and Gompertz functions provided comparable best R2 fits, but the latter produced a three-fold greater adult mass. The von Bertalanffy growth curve provided poor statistical fit with regard to R2 and was biologically unrealistic in being almost linear across the range of sampled specimens (see Supporting Information). We inferred that the Logistic result was biologically more realistic in predicting maturity at a size consistent with the histological and osteological signs of advanced maturity in the older specimens we sampled. Furthermore, P. lujiatunensis is the most richly collected dinosaur in the world, with hundreds of representative specimens in museum holdings. In no case has a specimen of larger size than the asymptotic value we predicted been found. Myhrvold initially tried to fit growth curves to all 80 specimens that we examined. Although it is stated in our paper only specimens that were histologically aged and body mass determined were used in making the growth curve. Furthermore, our graphic [figure 6 in Erickson et al. (2009)] shows that only a small subset of the 80 dinosaurs was utilized. Upon finding his growth curve did not match ours, Myhrvold correctly reasoned that we must have used just the histologically aged specimens. However, in his second analysis he used data for just “20” histologically aged animals when in fact there were 22 in the study. Furthermore in attempting to replicate our study, Myhrvold does not appear to use the same growth function we utilized. As stated in our article we used a Logistic function. His fitting of a variety of alternative functions resulted in curves that not surprisingly vary greatly from ours. In retrospect we realize that we did not provide sufficient data for others to reconstruct the many steps that go into making a growth curve for a dinosaur. Here we provide: a corrigendum to Table 1 that: (1) clarifies which specimens were used in describing the growth curve; (2) includes the mass estimates for the largest specimens used in the developmental mass extrapolation exercise; (3) provides sources of measurement variance including the entire range of femoral length measurements/estimates we made and number of growth lines we observed for each specimen; (4) lists the specimen age estimates; (5) gives the values we specifically utilized in making the growth curve; and (6) corrects two minor transcriptional errors (IVPP 14155 was mismarked as having been histologically aged and PKUVP 1058 femoral length was 52 mm, not 30 mm). In the Supplement we provide the statistical packages and analyses we undertook in making the growth curve. Myhrvold states that a number of our data points are missing from the growth curve graphic we provided [figure 6 in Erickson et al. (2009)]. As stated in the text and in our figure legend, our graphic only depicts the small subset of animals whose ages were determined from histology. His replicated figure [figure 1 in Myhrvold, (2015)] includes the 58 animals that we did not directly age. In our reanalysis we did however discover that there is an omitted data point in the figure 6 graphic from Erickson et al. (2009) (the individual LPM R00142 aged at 2 years). Fortunately, this datum was properly utilized in our growth analysis. Other data points that Myhrvold suggests are missing represent individuals of similar size and age whose dots on the graphic partially or completely overlap. To avoid future confusion, here we provide a corrigendum figure 6 with all 22 individuals depicted. Points that partially or completely overlap are clearly identified. Myrhrvold correctly points out that the regression equation we reported produces a growth curve that plots left of that depicted on the graphic in figure 6 in Erickson et al. (2009). The equation was incorrectly transcribed. We reran our analysis using a maximum likelihood algorithm in R v3.0.1 [R Core Team, (2013)] and constructed confidence intervals using the R package bbmle [Bolker and R Development Core Team, (2014)]. This recovers a best fit of 37.48/(1 + exp(−0.55 × (Age – 9.03))). We welcome the opportunity to correct this discrepancy here (corrigendum figure 6). Finally, Myhrvold contends that our life table results should be viewed with suspicion because he concludes that our growth curve results were in error. The fact of the matter is that an ecological life table uses only the population age and size information, not growth rates derived from growth analyses (Deevey, 1947; Voorhies, 1969; Ebert, 1999). (Note: Myhrvold mistates what Steinsaltz and Orzak (2011) concluded about our life table analysis for the tyrannosaur Albertosaurus sacrophagus (Erickson et al., 2006), implying the similar analysis for P. lujiatunensis was also suspect. Steinsaltz and Orzack (2011) did not conclude the statistics were “invalid.” The major conclusion of the tyrannosaur study, that there was low mortality during midlife and higher rates past adulthood, actually received statistical support in their analysis. A limitation was a lack of statistical power [owing to small samples sizes that are inherent in any dinosaur histological analysis; (Sander et al., 2011; Erickson, 2014) to accurately describe the shape of the curve. We deduced on ecological reasoning that it was a Type B1 survivorship curve (Pearl and Miner, 1935; Deevey, 1947), consistent with known curves for wild animal populations that show the dual-stage mortality pattern that we, and Steinsaltz and Orzak, found). In summation, we welcome this opportunity to again address the problems noted by Myhrvold and clarify what we did and why. Myhrvold's failure to replicate our growth curve findings stems partly from our not disseminating sufficient data to replicate the many steps that go into a growth curve analysis for a fossil vertebrate, coupled with his using apparently different data and functions than we described. The transcription mistakes we made in our presentation, which we correct here, were not implemented in the growth curve analysis and did not compromise the conclusions we made about P. lujiatunensis biology. The authors thank the editors of The Anatomical Record and reviewers of this exchange for helping us to improve the manuscript. Additional Supporting Information may be found in the online version of this article. Supporting Information Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.