Geographic Range and Sampling

In all sets of analyzed taxa, the dominant predictor of extinction risk was the correlated effects of geographic range and sampling that we refer to as overall commonness. Taxa seen in a greater number of sites and with wider geographic ranges had a lower extinction risk. To the extent that our sampling measures are a true proxy for sampling completeness, the result that overall commonness is the strongest predictor of extinction risk suggests that despite the intense sampling of graptolites as a group, the record for individual species is still fragmentary. In fact, the majority of species are seen in fewer than 10 sections (Fig. 10), and the distribution of observations is strongly skewed. When we restrict our analysis to a set of 272 taxa observed in at least 10 sections (“best sampled”), the resulting model has the third-weakest overall commonness factor. Instead, effects having to do with clade and cohort age were the dominant factors. Overall commonness in the best-sampled set explained only ~13% of the variance in taxon duration, approximately half as much as the average among the other sets of taxa. This suggests that half of the effect on extinction risk commonly attributed to geographic range may be a product of sampling effects, which is supported by the amount of variance explained by individual factors. The maximum pairwise distance, which has the lowest correlation with sampling measures (Fig. 4), explains approximately half as much variance as the first composite axis, overall commonness. Studies that have sought to distinguish between the effects of geographic range and sampling on extinction risk have most often suggested that the effects of geographic range are dominant over sampling (Payne and Finnegan Reference Payne and Finnegan2007; Powell Reference Powell2007; Harnik Reference Harnik2011; Finnegan et al. Reference Finnegan, Heim, Peters and Fischer2012). However, most focused explicitly on the best-sampled taxa, typically above species level, to maximize the number of observed occurrences and minimize the effects of sampling (Powell Reference Powell2007; Harnik Reference Harnik2011).

Figure 10 All 1114 taxa ordered by the natural log of the number of sites observed at (n.sect2).

Individual contributions from sampling and geographic range appear in nine of the taxon sets and explain a much smaller, although still significant, proportion of the total variation. Sampling, despite small geographic range, was seen in nine sets (Fig. 7) but without much apparent pattern. The impact of sampling in those sets is reflected in single-factor models in which number of localities, number of sections, and number of 5°×5° cells occupied are higher in rank for the basal paraclade, Neograptina, and post-Hirnantian sets than is the case for the other sets (Figs. 5, 6), suggesting these groups are poorly sampled. The reasons for this, particularly among the post-Hirnantian taxa for which graptolites are used extensively in biostratigraphy, may be because of the way graptolites are typically collected.

The majority of studies involving graptolites are interested in their use as biostratigraphic markers rather than paleoecological or evolutionary questions about the graptolites themselves. Consequently, investigators focus on collecting specimens that are well preserved and distinctive, rather than bulk samples that might reveal additional species. Focus is also given to locating the first and last occurrence of a species within a sequence to delineate biozone boundaries, with less emphasis given to documenting the persistence or abundance of a species between its first and last occurrences. These types of collection biases, typical of the graptolite occurrence data used to calculate durations, were not modeled or controlled for in this study and would likely explain additional variance between taxa. Particularly between those taxa considered to be stratigraphically important and all others. Differences between taxa that are numerically abundant versus rare might also be expected to correlate with extinction risk, as rare species are more likely to have their ranges within sections truncated and less likely to be collected in the field because the chances of finding well-preserved individuals are lower. In benthic taxa the effect of abundance on extinction risk has been found to be weak or not significant after controlling for geographic range (Harnik Reference Harnik2011; Harnik et al. Reference Harnik, Simpson and Payne2012; but see Simpson and Harnik [Reference Simpson and Harnik2009] for an exception). However, work on planktonic foraminifera found support for higher extinction risk in clades with low abundance (Stanley et al. Reference Stanley, Wetmore and Kennett1988).

Where it can be separated from sampling, the relatively small contribution of large geographic range to extinction risk in graptolites is unusual. On its own, geographic range accounted for less than 3% of the total variance in durations and was only a significant factor in five sets (Boyle, full set, Neograptina, post-Hirnantian, and NST). Large geographic range is typically highly predictive of reduced extinction risk across other clades (Jablonski Reference Jablonski2005 and references therein). The small effect in our analyses may reflect the impact of the strong correlation of variables in this study, where the explanatory power of large geographic range is tied up in the composite axis that we interpreted as overall commonness. However, there also is reason to believe that the contribution of geographic range to extinction risk in graptolites may genuinely have been lower than that in other groups (i.e., benthic invertebrates) because most planktic graptoloids were comparatively widespread. For example, Foote and Miller (Reference Foote and Miller2013), using great circle distances, reported that 98% of the benthic invertebrates they examined had a maximum range size of less than or equal to 15,442 km, whereas only 69% of the 1114 taxa examined here fall into this category using the maximum pairwise distance. In the best-sampled set only 35% of the taxa have geographic ranges smaller than 15,442 km. Graptolites are, therefore, generally much more widespread than benthic invertebrates. Graptolite geographic ranges are particularly large when one considers that the continents were confined almost exclusively to the Southern Hemisphere for much of the lower Paleozoic (Torsvik and Cocks Reference Torsvik and Cocks2013), which limits the maximum observed geographic range to essentially half the Earth’s surface. Given that most graptolite populations inhabited shelf-edge environs (Finney and Berry Reference Finney and Berry1997; Cooper et al. Reference Cooper, Rigby, Loydell and Bates2012), this observational limit may also closely approximate their original life occurrence limitations as well.

Thus, because the benefits of a large geographic range to decreasing extinction risk are logarithmic rather than linear (the largest effects arise from modest increases in small ranges; Foote and Miller Reference Foote and Miller2013: Fig. 3a), the influence of geographic range should be small compared with other factors given the large geographic range of most species. Perhaps not surprisingly, three of the analyses that exhibited strong effect of an individual geographic- range component (the post-Hirnantian, Neograptina, and NST sets) are also the sets that had the smallest average geographic ranges. The present data exhibit a significant trend toward smaller geographic-range size over time (Fig. 11), which may have been driven by the amalgamation of continents over the course of the early Paleozoic. The reduced average size of geographic ranges in later taxa, because of the logarithmic relationship between geographic range and extinction risk (Foote and Miller Reference Foote and Miller2013), increases the slope of the relationship and, thus, its observed strength. It is interesting to note that this interval of falling geographic-range size is also the same interval over which graptolite species diversity exhibits an unsteady but overall decline toward the ultimate extinction of the clade in the Early Devonian (see, for instance, Sadler et al. Reference Sadler, Cooper and Melchin2011).

Figure 11 Plot of the residuals from a correlation of the summation of a minimum spanning tree distance with number of observed locations versus time for the full set of 1114 taxa showing a significant negative trend (p<0.001).

It also possible that the relatively weak predictive ability of geographic range is attributable to the ecology of graptolites. Geographic range is expected to decrease extinction risk by decreasing the chance that single events will wipe out the entire population (Jablonski Reference Jablonski2005; Foote et al. Reference Foote, Crampton, Beau and Cooper2008). However, this expectation does not take into account the impact of geographic range on the likelihood of pseudoextinction, the apparent extinction of species when it evolves to a new form. The impact of geographic range on pseudoextinction depends on the distribution of a taxon within its geographic range. A relatively continuous distribution is likely to depress pseudoextinction by preventing isolation of populations (Lester and Ruttenberg Reference Lester and Ruttenburg2005), while a patchy or ephemeral distribution may actually encourage the formation of isolates and new species (Levin and Wilson Reference Levin and Wilson1976; Stanley Reference Stanley1979). A group with a high dispersal ability may achieve a large geographic range, especially when calculated using the union of all occurrences, and may have a high extinction risk due to the patchy nature of their distribution leading to pseudoextinction.

When geographic range-measures were examined as single-factor predictors of extinction risk, summation of a minimum spanning tree was often the highest-ranked model (Figs. 5, 6). This measure has not commonly been used as a measure of geographic range but has several advantages over other measures such as the maximum pairwise distance. Because this measure minimizes the distance between a series of points, it represents the minimum distance a taxon must have crossed, and this is particularly useful in this data set in which no corrections were made to avoid including uninhabitable areas (e.g., land). The data required to calculate summation of a minimum spanning tree is minimal, easily calculated in R, and can approximate complex geographic-range shapes. In contrast, the maximum pairwise distance is a poor predictor of extinction risk overall, probably because it is unable to account for complex geographic-range boundaries. Maximum pairwise distance ignores any range perpendicular to the bearing measured and is likely to greatly underestimate the true range of a species, particularly those that are wide-ranging or that have nearly circular or spatially complex distributions. It is also worth noting that some of the weakness of the maximum pairwise distance as a predictor of extinction risk might be due to the fact that we did not account for the possibility that a taxon might stretch around more than half the Earth’s circumference. This is likely to underestimate a taxon’s actual maximum pairwise distance and might weaken the relationship between this measure and extinction risk. However, when we tested this by analyzing the explanatory power of maximum pairwise distance among taxa with longitudinal ranges of less than or equal to 180°, we found a strong reduction in explanatory power of this factor compared with the full set and with random subsets that had a sample size equal to that of the longitudinally restricted set. This suggests that the taxa with large maximum pairwise distances, even if underestimated, are important contributors to the overall explanatory power of this variable and that the relatively poor performance of this measure is not a consequence of the way we measured it. Previous studies utilizing great circle distance, the equivalent of our maximum pairwise distance, have either used it as their only measure of geographic range (Harnik Reference Harnik2011), have not discussed differences between geographic-range measures (Kiessling and Aberhan Reference Kiessling and Aberhan2007; Foote et al. Reference Foote, Crampton, Beau and Cooper2008), or have found that the relationship between geographic range and extinction risk remained unchanged when other measures were used (Foote and Miller Reference Foote and Miller2013). However, none of these studies examined the actual magnitude of the effect of geographic range on extinction risk, and because of this, it is unclear whether maximum pairwise distance is a poor measure of geographic range in general or whether this weakness is particular to the geographically widespread graptolites.

In contrast to the several linear measures of geographic range, the convex hull was the only areal measure that we employed. It is also the only one that is commonly used for modern organisms and endorsed by the IUCN (2012) for evaluating extinction risk. In this study convex hulls hold an upper-middle rank among the geographic-range measures. The convex hull is known to have a bias toward overestimating geographic ranges, especially when distributions are discontinuous or have complex shapes (Burgman and Fox Reference Burgman and Fox2003), and this is certainly a concern in graptolites, which lived mainly along continental margins (Finney and Berry Reference Finney and Berry1997). The ranges calculated here are even more prone than usual to overestimation because we did not take into account taxa that extend across more than half the Earth’s surface. However, as with maximum pairwise distance, when only taxa with longitudinal ranges less than or equal to 180° are examined, the explanatory power of convex hull is substantially lower, beyond that expected by a reduced sample size alone. The convex-hull method is also very sensitive to sampling (Burgman and Fox Reference Burgman and Fox2003), as a single outlying occurrence can greatly expand the total area covered.

Unlike the results of previous studies (e.g., Foote and Miller Reference Foote and Miller2013), the association between geographic range and extinction risk was not stronger, in general, when distance measures were used rather than counts. The number of gyres occupied was a mid- to low-ranked factor in most sets, yet even at that rank, this factor performed surprisingly well given the uncertainties of reconstruction in paleoceanography of the Paleozoic (Christiansen and Stouge Reference Christiansen and Stouge1999). Another count measure, the number of regions occupied, was also a mid- to low-ranking factor, except for the Diplograptina, for which it ranked second. The general lack of signal for this factor could be attributable to the arbitrary definition of regions based on clusters of occurrences that reflect outcrop availability and study rather than the clustering of regions by taxon distributions. The final count measure, the number of 5°×5° cells occupied, effectively tracks both sampling and geographic range, consistent with previous uses of the measures (Kiessling and Aberhan Reference Kiessling and Aberhan2007). In two of the taxon sets (Boyle et al. Reference Boyle, Sheets, Wu, Goldman, Melchin, Cooper, Sadler and Mitchell2014 and pre-Katian), it is the highest-ranked single factor and is typically a high-ranking model.

Latitudinal range (and gyres associated with particular latitudes), in general may capture information about the ecological tolerance of graptolite species. In particular this measure may highlight a contrast between relatively eurytopic taxa versus those with limited climatic tolerance (either equatorial or polar), which are generally thought to have a greater extinction risk (Powell Reference Powell2007; Hopkins Reference Hopkins2011; Finnegan et al. Reference Finnegan, Heim, Peters and Fischer2012; Vilhena et al. Reference Vilhena, Harris, Bergstrom, Maliska, Ward, Sidor, Strömberg and Wilson2013). Contrary to expectations, however, latitudinal range was one of the least predictive factors in all of our analyses, with two notable exceptions. We first address the general question of why latitudinal range may not have been a strong predictor of extinction risk in graptolites and then return to these two exceptions to that generalization.

The general weakness of latitudinal and longitudinal ranges as predictors of extinction risk in our data set may reflect the phenomenon of apparent dispersal, that is, the ability, observed among many extant organisms, to disperse while seldom being able to persist in the colonized regions (Norris Reference Norris2000). Such an effect might be especially likely for blooming zooplankton, which likely included graptolites. In particular, shifts in the location of water masses can push planktic populations far beyond their normal latitudinal distribution (Bjørklund et al. Reference Bjørklund, Kruglikova and Anderson2012), and if these events are even moderately common on geological timescales, then the chance preservation and recovery of such ecologically rare but geologically common events might significantly expand the latitudinal range of species into regions never successfully colonized and thus distort the link between range size and extinction risk by giving an exaggerated impression of the realized niche area. Assessing this effect will require data on temporal patterns of site occupancy.

Despite our general concern about the fidelity of observed geographic range as a proxy for a species’ realized niche area, latitudinal range, along with the number of regions occupied, was the best single predictor of extinction risk in the Diplograptina and the Katian–Hirnantian taxon set (as noted above, all but a very few Katian–Hirnantian taxa are Diplograptina, and so these two sets overlap very strongly). That the Diplograptina exhibit significant effects on duration from both latitudinal range and number of regions might be indicative of a higher degree of provincialism and specialization within this group (e.g., Vandenbroucke et al. Reference Vandenbroucke, Armstrong, Williams, Zalasiewicz and Sabbe2009). Furthermore, the Late Ordovician was a time of global cooling (Trotter et al. Reference Trotter, Williams, Barnes, Lécuyer and Nicoll2008; Finnegan et al. Reference Finnegan, Heim, Peters and Fischer2012) and, potentially, a series of glaciation events led up to the Hirnantian glacial interval (e.g., Pope and Steffen Reference Pope and Steffen2003; Young et al. Reference Young, Saltzman, Foland, Linder and Kump2009; Melchin et al. Reference Melchin, Mitchell, Holmden and Štorch2013; Armstrong and Harper Reference Armstrong and Harper2014). A cooling environment would have established a steeper temperature gradient across the globe, leading to more variable environments during the Late Ordovician (Armstrong et al. Reference Armstrong, Baldini, Challands, Gröcke and Owen2009; Vandenbroucke et al. Reference Vandenbroucke, Armstrong, Williams, Paris, Zalasiewicz, Sabbe, Nõlvak, Challands, Verniers and Servais2010) and a restriction of habitat for the dominantly equatorial Diplograptina. It is also interesting to note that the Diplograptina have a unique PLSR composite axis that is a combination of moderate positive weighting on latitudinal range and a very strong negative weighting on shallow tolerance. We suggest that this may reflect the influence of paleoecology on extinction risk. In particular, it suggests that NST species had a lower rate of decrease in extinction risk with increasing range size compared with the steeper decline in extinction risk with increasing range size exhibited by the ST species. Such an effect is expected to be more pronounced during the Late Ordovician as climate shifted rapidly and ST taxa with restricted equatorial tolerances were driven to extinction. This would magnify the already existing association between latitudinal range and extinction risk in ST taxa while the NST taxa were likely to have inhabited deep-water environments and to have been driven to extinction by changes in ocean circulation (Finney et al. Reference Finney, Berry and Cooper2007), decoupling them from the normal association between latitudinal ranges and extinction risk.

Age Cohorts and Clades

The uniquely strong effect of latitudinal range on extinction risk during the Late Ordovician is an example of how different time periods may have very different selective pressures. This effect may be further exaggerated during mass extinctions, when major shifts in clade trajectories occur and have long-lasting consequences on macroevolutionary dynamics. The LOME, which by some data sets was the second-largest extinction event in Earth’s history (e.g., Bambach et al. Reference Bambach, Knoll and Wang2004; but see Alroy [Reference Alroy2008], for a different interpretation), permanently altered the path of graptolite evolution with a complete turnover in clades (Melchin and Mitchell Reference Melchin and Mitchell1991; Chen et al. Reference Chen, Melchin, Sheets, Mitchell and Jun-Xuan2005; Sadler et al. Reference Sadler, Cooper and Melchin2011). Similar climate shifts have been implicated as driving evolutionary dynamics in several marine clades (Jacobs and Lindberg Reference Jacobs and Lindberg1998; Rogers Reference Rogers2000; Steeman et al. Reference Steeman, Hebsgaard, Fordyce, Ho, Rabosky, Nielsen, Rahbek, Glenner, Sørensen and Willerslev2009; Polcyn et al. Reference Polcyn, Jacobs, Araújo and Schulp2013). Although climate change was most intense during the LOME, the environmental and biological effects of major cooling may have begun during the Katian (e.g., Pope and Steffen, Reference Pope and Steffen2003; Calner et al. Reference Calner, Lehnert and Nõlvak2010; Goldman and Wu Reference Goldman and Wu2010; Kajlo et al. Reference Kajlo, Hints, Hints, Männik, Martma and Nõlvak2011). To investigate this, we classified taxa according to whether they arose before or after the beginning of the Katian Stage and included this binning by taxon cohort age as a factor in the analyses. Taxon age cohort (pre-Katian, Katian–Hirnantian, or post-Hirnantian) was significant in six sets, most often in combination with clade. We discuss these combinations below.

As mentioned previously, because of the strong diversity turnover exhibited by the succession of graptolite clades during the Ordovician Period (Sadler et al. Reference Sadler, Cooper and Melchin2011; Cooper et al. Reference Cooper, Sadler, Munnecke and Crampton2014), the clade membership and taxon content of our temporal bins are tightly correlated. In many groups of organisms, including graptolites, individual clades have different extinction risk characteristics (Stanley et al. Reference Stanley, Wetmore and Kennett1988; Jablonski and Raup Reference Jablonski and Raup1995; Jeffery Reference Jeffery2001; Bapst et al. Reference Bapst, Bullock, Melchin, Sheets and Mitchell2012). We find clade membership to be a significant predictor of extinction risk, both alone and, more often, in combination with the age cohorts, in seven sets of taxa. Two of the sets in which clade was not significant were analyses of the restricted supraordinal-level sets, in which the effect of clade membership on extinction risk was already minimized. The tight correlation between age cohorts and clade is also seen in the full and best-sampled sets, in which the second composite axis of the PLSR analysis matches with the differences in geographic-range dynamics between the Neograptina, which dominated during the post-Hirnantian, and the Ordovician clades (the basal paraclade and the Diplograptina). This result is not entirely unexpected, since we picked the Katian–Hirnantian versus post-Hirnantian division expressly to coincide with the LOME and the associated change in clade diversity, but this new result does capture a distinct feature, which is that the taxa in the post-Hirnantian/Neograptina set have a significantly higher average extinction risk than did graptolites prior to the LOME event. This indicates that the LOME had a lasting effect on the macroevolutionary dynamics of the Graptoloida.

The basal paraclade consists of families that reached their peak diversity during the Early and Middle Ordovician and mostly preceded the origin of the axonophoran clades, Diplograptina and Neograptina. Of the three clades, this paraphyletic group had the greatest total variance in durations but the lowest percent variance explained with only two significant composite PLSR axes, which together explained just 28.11% of the total variance. The largest component was overall commonness, with the only other significant composite axis being a small addition from sampling measures. The basal paraclade taxa have been less intensively sampled than those of the other two clades, in part because rocks of this age are less accessible and interest in their use for biostratigraphy is somewhat lower, and yet they still tend to have longer durations than later taxa (which, of course, is partly why their biostratigraphic value is lower). The relatively low percent variance in durations explained in combination with relatively long durations may be an example of widespread dispersal of somewhat ecologically generalized, primitive planktic graptoloids into the new and relatively empty zooplankton communities of the Early Ordovician (Servais et al. Reference Servais, Owen, Harper, Kröger and Munnecke2010), coupled with the effect of the relatively widely dispersed continental masses (Torsvik and Cocks Reference Torsvik and Cocks2013). Most of the basal paraclade taxa are confined to the pre-Katian time interval, which, when analyzed independently, shows that family-level clade membership is strongly associated with extinction risk. This, in combination with the PLSR results of the basal paraclade, suggests that the constituent family-level clades have intrinsically different extinction risks as a consequence of some factor that we are unable to examine further with the data at hand.

The Diplograptina and Neograptina are broadly similar in their patterns of extinction risk. In both, 26–28% of the total variance in taxon duration is explained by the set of PLSR composite axes. However, the Neograptina have axes other than overall commonness (sampling and geographic range) that match with those in the post-Hirnantian taxon set. In both Diplograptina and Neograptina, age cohort is a significant factor associated with extinction risk. Because of the way the age cohorts were defined, most of the Neograptina fall into the post-Hirnantian set, whereas the Diplograptina are split more evenly between the pre-Katian and Katian–Hirnantian intervals. The significance of age cohorts in these two clades represents their interaction with the unique conditions surrounding the LOME. In the Diplograptina, the Late Ordovician taxa have shorter durations than their earlier representatives, and some of these short-lived Katian–Hirnantian taxa could be considered the graptolite equivalent of the benthic Hirnantia fauna; that is, they represent species that evolved in response to an environment changed by glaciation but that then went extinct when glaciation retreated in the latest Hirnantian (Rong et al. Reference Rong, Chen and Harper2002). On the other hand, the Late Ordovician Neograptina had longer durations than later representatives of the clade and were apparently able to tolerate and even thrive in the icehouse conditions, persist through the immediate postextinction interval, and subsequently diversify (Bapst et al. Reference Bapst, Bullock, Melchin, Sheets and Mitchell2012).

Although clade is a significant factor in seven of the taxon sets examined here, it is not clear what is driving the differences between family-level clades. Factors might include such things as colony size (similar to body size as examined in other clades; e.g., Jablonski and Raup Reference Jablonski and Raup1995; McRoberts and Newton Reference McRoberts and Newton1995; Jablonski Reference Jablonski2008; Liow et al. Reference Liow, Fortelius, Lintulaakso, Mannila and Stenseth2009; Harnik Reference Harnik2011), proximal development type, number of branches, or thecal type. Colony shape may be particularly important, as it has been linked to feeding efficiency in graptolites (Rigby Reference Rigby1991), although Bapst et al. (Reference Bapst, Bullock, Melchin, Sheets and Mitchell2012) found that variance in shape and theca size was not significantly correlated with extinction selectivity during the LOME. Because graptolites show a large degree of homoplasy in morphological characters, careful coding of biological characters that could then be included in an evaluation of extinction risk, including associated paleoenvironmental proxies (such as biomarkers, paleoredox, and paleoproductivity proxies) might be informative in determining the influence of clade membership itself versus morphological correlates and biofacies affiliation.

Biofacies Groups and Shallow Tolerance

The existence of distinct biofacies within graptolite faunas have long been recognized by graptolite workers (e.g., Berry Reference Berry1962), but their importance in graptolite evolutionary dynamics has not been clear. Early discussions of biofacies groups were concerned with their biological meaning, most importantly whether they represented horizontal (Cisne and Chandlee Reference Cisne and Chandlee1982; Finney Reference Finney1984, Reference Finney1986) or vertical (Cooper et al. Reference Cooper, Fortey and Lindholm1991; Cooper Reference Cooper1999; Chen et al. Reference Chen, Yuan-Dong and Mitchell2001) zonation of the water column. Recent reviews of the biofacies concept (Cooper and Sadler Reference Cooper and Sadler2010; Cooper et al. Reference Cooper, Rigby, Loydell and Bates2012; Goldman et al. Reference Goldman, Maletz, Melchin and Fan2013b) have favored the idea that group 1 and group 2 represent vertical differentiation (mesopelagic and epipelagic, respectively) with lateral zonation separating the wide-ranging group 2 epipelagic species from the shelf-endemic group 3 species (Fig. 1). We employ those interpretations in our discussion of biofacies effects below.

The same limited set of taxa analyzed in Boyle et al. (Reference Boyle, Sheets, Wu, Goldman, Melchin, Cooper, Sadler and Mitchell2014) was also examined here, with slight differences in results due to the addition of shallow tolerance as a factor. As in that preliminary study, our results show that biofacies group makes a substantial contribution (~9.5%) to extinction risk in this set of taxa. To examine the differences in biofacies groups more closely, each group was analyzed independently. Analyses of both the mesopelagic group 1 and the epipelagic group 2 taxa are similar to the results of most taxon sets in that the dominant axis is overall commonness, although explanatory power of this axis is particularly small in group 1 and large in group 2 compared with that factor in other sets of taxa (Fig. 7).

Although all single factors (i.e., lat.rg, n.bin, etc.) for group 2 taxa, with the exception of clade, were found to be significant predictors of extinction risk, similar to most taxon sets, this was not the case for group 1 taxa. Permutation tests indicate that none of the measured variables considered as predictors of extinction risk perform any better than random in group 1. One possible explanation for this result is that the sample size of biofacies group 1 (n=59) is too small to detect the effect of the causal factors that operated within this set. To test this possibility, we selected random samples of non–group 1 Ordovician taxa, matched to the size of group 1 (n=59) and tested each factor for significance. All single factors of sampling and geographic range were significant, but clade, age cohort, and shallow tolerance were not, probably because these last three were multistate factors that are more sensitive to small sample sizes than continuous variables. Thus, although small sample size may contribute to the failure to detect a significant effect on the measured properties on extinction risk in group 1 taxa, it is likely that the set of group 1 taxa are not, in fact, an ecologically coherent group at all, but instead contain a number of mesopelagic groups (clades?) that each have distinct extinction risk determinants and so resist any general description apart from their biofacies 1 affiliation and the significance of overall commonness, which is found in all taxon sets. Testing this hypothesis will require biofacies group identifications for a larger number of taxa and more detailed documentation of their ecological characteristics.

The biofacies group concept was established from examination of Early to Late Ordovician taxa (Cooper et al. Reference Cooper, Fortey and Lindholm1991; Cooper and Sadler Reference Cooper and Sadler2010) and currently it is not clear whether it can be extended to the Silurian Neograptina. However, in an attempt to increase the number of taxa assignable to biofacies groups, taxa were identified as ST (present in continental shelf deposits) or NST (absent from those shelf deposits). Ideally, these identifications should correspond to group 2 and group 1, respectively. However, when assignments are compared with taxa with a previous biofacies identification, only 60% match expectations. This lack of fit suggests that this crude assignment method does not replicate the biofacies group assignments of Cooper and Sadler (Reference Cooper and Sadler2010) exactly. Nonetheless, ST taxa had significantly longer durations than NST taxa, mirroring the distinction Cooper and Sadler (Reference Cooper and Sadler2010) found between biofacies group 2 and group 1 taxa, respectively (Table 3). As a single factor, shallow tolerance had very little, if any, explanatory power across the sets of taxa, but given the duration contrasts between ST and NST taxa, we nonetheless analyzed each state independently. Results show that ST taxa have exceptionally large contributions to extinction risk from family-level clade and taxon age cohorts (Fig. 7) compared with NST taxa. The reason for this large contribution of clade and age cohort in ST taxa is probably because the two largest clades, Monograptidae and Dichograptidae, are underrepresented in the ST group compared with random expectations. This creates a set that has a more even distribution of clade affiliation, allowing detection of subtle distinctions between clades that might otherwise be swamped by the largest clades.