In a commentary published as a ‘Review Article’, Randolph and Dobson (Reference Randolph and Dobson2012, hereafter ‘RD’) describe the dilution effect – a negative relationship between biodiversity and disease risk – as Panglossian. They agree that the dilution effect occurs in nature, but they object to specific pieces of evidence and to what they perceive as over-generalizations about its broad applicability. To justify their allegation of Panglossian thinking, RD present a caricature of dilution effect studies, suggesting that these studies are unaware of contrasting effects of biodiversity on disease and inappropriately applied to policy and management. The generality of the dilution effect, factors that strengthen, weaken, or reverse it, and the underlying causal mechanisms, are rich topics currently being addressed. Unfortunately, however, the RD commentary is replete with distortions and errors of commission and omission that undermine its merit. Below I list the most serious of these shortcomings in order to advance the dialogue on the importance and generality of the dilution effect.
DISTORTIONS
1. RD repeatedly call the dilution effect a ‘mantra’ (pp. 847, 848, 853 and 857), although their descriptions of the mantra differ at each mention. Each is placed in quotations but only one is attributed to a source, and this quotation is not present in the source cited (Keesing et al. Reference Keesing, Belden, Daszak, Dobson, Harvell, Holt, Hudson, Jolles, Jones, Mitchell, Myers, Bogich and Ostfeld2010). Their religious analogies extend to the unreferenced accusation that proponents of the dilution effect ‘preach that high biodiversity always protects against disease’ (p. 860). Neither such mantras nor preachings exist in the published literature. 2. RD repeatedly state that proponents have argued that the dilution effect is ‘universal’ (pp. 850, 853, 854), but this too is incorrect. To the contrary, it is obvious in RD's own descriptions of the dilution effect that universality is never claimed; indeed, RD borrow from some of the earliest descriptions of the dilution effect that delineate circumstances under which it is expected not to occur (e.g. Table 1, a restating of Ostfeld and Keesing, Reference Ostfeld and Keesing2000a). The dilution effect literature has been careful to describe conditions under which biodiversity is expected to amplify disease risk (Ostfeld and Keesing, Reference Ostfeld and Keesing2000a,Reference Ostfeld and Keesingb; Schmidt and Ostfeld, Reference Schmidt and Ostfeld2001; Ostfeld and LoGiudice, Reference Ostfeld and LoGiudice2003; Keesing et al. Reference Keesing, Holt and Ostfeld2006, Reference Keesing, Belden, Daszak, Dobson, Harvell, Holt, Hudson, Jolles, Jones, Mitchell, Myers, Bogich and Ostfeld2010; Pongsiri et al. 2009; Johnson and Thieltges, 2010). Examples of such an amplification effect, and neutral effects, exist (Keesing et al. Reference Keesing, Holt and Ostfeld2006; Ostfeld and Keesing, Reference Ostfeld and Keesing2012), but reduced disease risk appears to be a far more common consequence of high host diversity (Cardinale et al. Reference Cardinale, Duffy, Gonzalez, Hooper, Perrings, Venail, Narwani, Mace, Tilman, Wardle, Kinzig, Daily, Loreau, Grace, Larigauderie, Srivastava and Naeem2012).
ERRORS OF OMISSION
1. One important error of omission is the removal of data points from scatterplots (Figs 1 and 2) in order to reanalyse relationships within subsets of data. Data removal was not justified by statistical norms, e.g. status as significant outliers. Indeed, the data point removed from Fig. 2 (redrawn from Ostfeld et al. Reference Ostfeld, Canham, Oggenfuss, Winchcombe and Keesing2006), which shows a positive relationship between acorn production and subsequent Lyme disease risk, represents a masting year, which, although infrequent, illustrates highly influential ecological events that should not be obscured. 2. RD thoroughly discuss the roles of vertebrate hosts in both abundance and infection prevalence of ticks, arguing that while increasing total host density (across all species) might reduce infection prevalence, it must increase tick density. Unfortunately, there is no evidence to support this assertion, and this discussion omits the evidence (Keesing et al. Reference Keesing, Brunner, Duerr, Killilea, LoGiudice, Schmidt, Vuong and Ostfeld2009) that some hosts kill far more ticks (via grooming) than they feed, potentially reducing tick density. 3. RD repeatedly argue that host community composition rather than species richness is more informative in understanding the dilution effect, failing to recognize how thoroughly this idea has been explored. From the very inception of the dilution effect (Van Buskirk and Ostfeld, Reference Van Buskirk and Ostfeld1995; Norman et al. Reference Norman, Bowers, Begon and Hudson1999; Ostfeld and Keesing, Reference Ostfeld and Keesing2000a,Reference Ostfeld and Keesingb; Schmidt and Ostfeld, Reference Schmidt and Ostfeld2001; LoGiudice et al. Reference LoGiudice, Ostfeld, Schmidt and Keesing2003; Ostfeld and LoGiudice, Reference Ostfeld and LoGiudice2003), the composition of the host community has been identified as the most relevant metric of host diversity, as it provides more information than species richness alone. In dismissing species richness, RD fail to indicate that, whenever the pattern of community assembly or disassembly is nonrandom, species richness and community composition will be correlated. When patterns are highly nested, this correlation will be quite strong (Johnson et al. Reference Johnson, Preston, Hoverman and Richgels2013).
ERRORS OF COMMISSION
1. In critiquing an influential model of Lyme disease in eastern North America (LoGiudice et al. Reference LoGiudice, Ostfeld, Schmidt and Keesing2003), RD state that the structure of the model guarantees that high diversity of hosts will reduce tick infection prevalence, but this is incorrect. Ostfeld and LoGiudice (Reference Ostfeld and LoGiudice2003) showed that the same model is capable of producing a positive correlation between diversity and infection prevalence, but this occurs only when an unrealistic assumption (each host species has an equal probability of entering as communities are assembled) is made. RD also claim that this study assumed constant tick burdens on each given host, when in fact the study empirically determined tick burdens on each host. 2. RD claim that a subsequent model (Keesing et al. Reference Keesing, Brunner, Duerr, Killilea, LoGiudice, Schmidt, Vuong and Ostfeld2009) ‘assumed that constant total numbers of ticks are fed, with all ticks that are not picked up by one host type being re-distributed to others’ (p. 3). This too is incorrect. In the model of Keesing et al. (Reference Keesing, Brunner, Duerr, Killilea, LoGiudice, Schmidt, Vuong and Ostfeld2009), the percentage of ticks redistributed onto other hosts was systematically varied from 0 to 100%. 3. RD criticize the study by Ostfeld et al. (Reference Ostfeld, Canham, Oggenfuss, Winchcombe and Keesing2006), which compared the impact of varying deer density with that of rodent density in affecting tick density. They erroneously state that ‘direct estimates of deer abundance were attributes of the entire study site but were treated as characteristics of individual plots’ (p. 5). In fact, two estimates of deer density were used, one of which was specific to each individual plot, while the other was treated as a characteristic of all plots. Neither was positively correlated with subsequent tick abundance. 4. Moreover, RD's misunderstanding of this study's design leads them to falsely accuse it of inflating degrees of freedom and pseudo-replicating. In fact, the maximum-likelihood statistics used to compare models with and without deer accommodated the hierarchical design, and because specific hypotheses were not tested using a frequentist approach, degrees of freedom were not relevant. 5. Referring to several recent reviews of diversity–disease relationships, RD state: ‘Results consistently show that the effect of increasing biodiversity in all its various forms … is not uniform; amplification or neutral effects have been observed as much as reduced infection prevalence’ (p. 859). In fact, equal frequencies of amplification, neutrality and dilution are not shown in any of the reviews cited. Indeed, a recent review (Cardinale et al. Reference Cardinale, Duffy, Gonzalez, Hooper, Perrings, Venail, Narwani, Mace, Tilman, Wardle, Kinzig, Daily, Loreau, Grace, Larigauderie, Srivastava and Naeem2012) found dilution (a statistically significant, negative relationship between diversity and disease) in 91 of 107 (85%) statistical tests of plant diseases and in 30 of 45 (67%) statistical tests of animal diseases, for a total of 80% of all 152 tests reviewed.
CONCLUSIONS
As research on the dilution effect in a variety of disease systems has grown and its applications to policy and management are discussed, it is important to critique both the theory itself and the degree to which it is supported by evidence. Development of dilution effect theory has proceeded cautiously, with careful attention paid to alternative and complex effects of biodiversity. I am confident that continued research will reveal the degree to which it is general and useful. RD invoke Voltaire's infamous character Pangloss to imply that dilution effect theory is naïvely optimistic. They fail to recognize that, in fact, dilution effect theory has much more in common with the views of Voltaire's central character, Candide, who rejects Panglossian hyper-optimism to develop a pragmatic philosophy based on empirical observations. I hope that discussions of the generality and applicability of the dilution effect will continue more dispassionately.
ACKNOWLEDGEMENTS
I gratefully acknowledge Felicia Keesing and three anonymous referees for many helpful comments and Felicia Keesing for suggesting the title.
