Post by johnb417 on Jun 7, 2012 8:53:41 GMT -5
Flawed population viability analysis can result
in misleading population assessment: A case study
for wolves in Algonquin park, Canada
Brent R. Pattersona,*, Dennis L. Murrayb
Ontario Ministry of Natural Resources,
Wildlife Research and Development Section, Trent University, DNA Building,
Accepted 6 December 2007
Available online 13 February 2008
For many populations and species, population viability analysis (PVA) plays a critical role in developing defensible conservation strategies and recovery plans. Although technical aspects of PVA have been well scrutinized, misapplication of PVA and misinterpretation of its results have received less attention. To illustrate potential hazards of improper use of PVA we reanalyzed data from a recent study on viability of wolves in Algonquin Provincial Park (APP), Ontario, Canada. The original PVA predicted extirpation of wolves from APP and prompted both a ban on wolf harvesting in a 10–16 km buffer zone surrounding the 7571 km2 park and an intensive research program to evaluate efficacy of the ban. Our reanalysis showed that limited and imprecise wolf population density and demographic rate estimates, as well as flawed population assessment and reconstruction methods, led to overly pessimistic evaluation of wolf population status in APP. In fact, our analyses suggest that wolves in APP are unlikely to decline significantly over the next 20 years. Further, contrary to earlier conclusions we suggest that rapid wolf population recovery following protection from human exploitation would be likely and readily detectable.
These findings highlight the need for adequate data, appropriate methodology, and proper analytical context when conducting PVA. Because the original PVA prompted substantial redirection of staff and financial resources from other significant conservation initiatives, we conclude that improper PVA may undermine execution of effective wildlife management and ultimately provide a disservice to conservation biology.
This paper serves as an example of how PVA may be misapplied and sometimes provides misleading results, and thereby sets a foundation for a broader discussion on the appropriate context for future risk assessment.
Theberge et al. (2006) inferred wolf population status by showing that median recruitment (0.14) was significantly lower
than mortality (0.30). However, note that recruitment values they used (Theberge et al., 2006: Table 2) were not those calculated as described in the methods and provided in their Table 1, and no explanation was offered as to how they were obtained. We repeated this comparison of median recruitment and mortality rate estimates using the ‘‘correct’’ recruitment values calculated as described by Theberge et al. (2006): (Table 1). We then further evaluated differences between recruitment and total mortality (separate analyses using data from both Tables 1 and 2 in Theberge et al., 2006) by examining the statistical distributions for each parameter.
Study periods reflect different time segments from the complete (1989–1999) dataset. We represent the number of inter-year growth rates as n.
Probability that, if resampled, the l would fall above specific thresholds is estimated from the normal distribution provided by l and r2.in APP (Theberge et al., 2006) could have been due to low statistical power and therefore conducted sensitivity and elasticity analyses on the stage-based (Lefkovitch) matrix for female-only portion of the wolf population. The matrix consisted of three age classes (pups, yearlings and adults); these
parameters are commonly used to characterize wolf population demography (see Fuller et al., 2003). The matrix was populated with annual adult survival from Theberge et al. (2006) but the remaining parameters could not be estimated directly from the Theberge et al. study. Wolf litter sizes were obtained from a study in APP by Mills (2006), whereas remaining Morris and Doak (2002) presented eight guidelines for objective and informative conduct of PVA: (1) avoid conducting a formal PVA if the amount
of data are inadequate; (2) do not present estimates of population viability unaccompanied by confidence intervals; (3) view viability metrics as relative rather than absolute gauges of population status; (4) do not project population viability far into the future; (5) always consider how potential determinants of population viability that have been omitted from the model might change risk assessment; (6) never base management
on estimates of the probability of absolute extinction; (7) whenever possible, consider multiple models to address ‘‘model uncertainty’’; and (8) consider a PVA to be a
work in progress, not the final word. Despite the above concerns, PVAs are rarely criticized for the context in which they are conducted and presented. For example, researchers seldom show evidence that the ‘‘population’’ under consideration truly is a discrete functional unit, implying that extrapolation of results of small-scale analyses to broader-scale populations may be problematic (Keppie, 2006). Similarly, limited attention has been directed at ensuring that PVAs meet not only minimum technical standards but also remain objective and impartial in their assessment. For instance, selective use of analyses to support a particular point of view (e.g. data dredging), or slanting interpretation of results to emphasize a perceived conservation need, are not typically scrutinized. Certainly, the issue of impartiality and objectivity in conservation biology is not new, and several studies provide guidelines to encourage maintenance of scientific integrity despite the pursuit of specific conservationobjectives (e.g. Longino, al. (2006) also evaluated the likelihood that a research program designed to measure efficacy of the harvest ban could actually detect a positive population response.
Our reanalysis reveals flaws in the original PVA, leading to contrasting conclusions regarding the current status and long-term prognosis for wolves in APP. This paper serves as an example of how PVA may be misapplied and sometimes provides misleading results, and thereby sets a foundation for a broader discussion on the appropriate context for future risk assessment. most important demographic parameter for monitoring wolf population response to the harvest ban. 3.4. Testing and falsifiability of the null hypothesis Not surprisingly, the power to detect a positive change in wolf
population growth following implementation of the harvest ban was highly dependent on both the assumed increase in growth rate (effect size), and assumed variability in this rate
(Table 5). When population parameters more consistent with the literature were used, the probability of detecting a signifi- cant change in population growth rate was reasonably high (Table 5). For example, using the observed post-ban rate of growth (l = 0.20), and assuming r2 = 0.20 (considerably higher than actually observed following the harvest ban), and setting the critical levels of the test (i.e. 1_a) to 0.95, 0.85, or 0.75, then
the probability of falsely concluding that the wolf ban had no positive effect on population growth rate, when in fact it had (Type II error), was 0.402, 0.189, or 0.107, respectively.
in misleading population assessment: A case study
for wolves in Algonquin park, Canada
Brent R. Pattersona,*, Dennis L. Murrayb
Ontario Ministry of Natural Resources,
Wildlife Research and Development Section, Trent University, DNA Building,
Accepted 6 December 2007
Available online 13 February 2008
For many populations and species, population viability analysis (PVA) plays a critical role in developing defensible conservation strategies and recovery plans. Although technical aspects of PVA have been well scrutinized, misapplication of PVA and misinterpretation of its results have received less attention. To illustrate potential hazards of improper use of PVA we reanalyzed data from a recent study on viability of wolves in Algonquin Provincial Park (APP), Ontario, Canada. The original PVA predicted extirpation of wolves from APP and prompted both a ban on wolf harvesting in a 10–16 km buffer zone surrounding the 7571 km2 park and an intensive research program to evaluate efficacy of the ban. Our reanalysis showed that limited and imprecise wolf population density and demographic rate estimates, as well as flawed population assessment and reconstruction methods, led to overly pessimistic evaluation of wolf population status in APP. In fact, our analyses suggest that wolves in APP are unlikely to decline significantly over the next 20 years. Further, contrary to earlier conclusions we suggest that rapid wolf population recovery following protection from human exploitation would be likely and readily detectable.
These findings highlight the need for adequate data, appropriate methodology, and proper analytical context when conducting PVA. Because the original PVA prompted substantial redirection of staff and financial resources from other significant conservation initiatives, we conclude that improper PVA may undermine execution of effective wildlife management and ultimately provide a disservice to conservation biology.
This paper serves as an example of how PVA may be misapplied and sometimes provides misleading results, and thereby sets a foundation for a broader discussion on the appropriate context for future risk assessment.
Theberge et al. (2006) inferred wolf population status by showing that median recruitment (0.14) was significantly lower
than mortality (0.30). However, note that recruitment values they used (Theberge et al., 2006: Table 2) were not those calculated as described in the methods and provided in their Table 1, and no explanation was offered as to how they were obtained. We repeated this comparison of median recruitment and mortality rate estimates using the ‘‘correct’’ recruitment values calculated as described by Theberge et al. (2006): (Table 1). We then further evaluated differences between recruitment and total mortality (separate analyses using data from both Tables 1 and 2 in Theberge et al., 2006) by examining the statistical distributions for each parameter.
Study periods reflect different time segments from the complete (1989–1999) dataset. We represent the number of inter-year growth rates as n.
Probability that, if resampled, the l would fall above specific thresholds is estimated from the normal distribution provided by l and r2.in APP (Theberge et al., 2006) could have been due to low statistical power and therefore conducted sensitivity and elasticity analyses on the stage-based (Lefkovitch) matrix for female-only portion of the wolf population. The matrix consisted of three age classes (pups, yearlings and adults); these
parameters are commonly used to characterize wolf population demography (see Fuller et al., 2003). The matrix was populated with annual adult survival from Theberge et al. (2006) but the remaining parameters could not be estimated directly from the Theberge et al. study. Wolf litter sizes were obtained from a study in APP by Mills (2006), whereas remaining Morris and Doak (2002) presented eight guidelines for objective and informative conduct of PVA: (1) avoid conducting a formal PVA if the amount
of data are inadequate; (2) do not present estimates of population viability unaccompanied by confidence intervals; (3) view viability metrics as relative rather than absolute gauges of population status; (4) do not project population viability far into the future; (5) always consider how potential determinants of population viability that have been omitted from the model might change risk assessment; (6) never base management
on estimates of the probability of absolute extinction; (7) whenever possible, consider multiple models to address ‘‘model uncertainty’’; and (8) consider a PVA to be a
work in progress, not the final word. Despite the above concerns, PVAs are rarely criticized for the context in which they are conducted and presented. For example, researchers seldom show evidence that the ‘‘population’’ under consideration truly is a discrete functional unit, implying that extrapolation of results of small-scale analyses to broader-scale populations may be problematic (Keppie, 2006). Similarly, limited attention has been directed at ensuring that PVAs meet not only minimum technical standards but also remain objective and impartial in their assessment. For instance, selective use of analyses to support a particular point of view (e.g. data dredging), or slanting interpretation of results to emphasize a perceived conservation need, are not typically scrutinized. Certainly, the issue of impartiality and objectivity in conservation biology is not new, and several studies provide guidelines to encourage maintenance of scientific integrity despite the pursuit of specific conservationobjectives (e.g. Longino, al. (2006) also evaluated the likelihood that a research program designed to measure efficacy of the harvest ban could actually detect a positive population response.
Our reanalysis reveals flaws in the original PVA, leading to contrasting conclusions regarding the current status and long-term prognosis for wolves in APP. This paper serves as an example of how PVA may be misapplied and sometimes provides misleading results, and thereby sets a foundation for a broader discussion on the appropriate context for future risk assessment. most important demographic parameter for monitoring wolf population response to the harvest ban. 3.4. Testing and falsifiability of the null hypothesis Not surprisingly, the power to detect a positive change in wolf
population growth following implementation of the harvest ban was highly dependent on both the assumed increase in growth rate (effect size), and assumed variability in this rate
(Table 5). When population parameters more consistent with the literature were used, the probability of detecting a signifi- cant change in population growth rate was reasonably high (Table 5). For example, using the observed post-ban rate of growth (l = 0.20), and assuming r2 = 0.20 (considerably higher than actually observed following the harvest ban), and setting the critical levels of the test (i.e. 1_a) to 0.95, 0.85, or 0.75, then
the probability of falsely concluding that the wolf ban had no positive effect on population growth rate, when in fact it had (Type II error), was 0.402, 0.189, or 0.107, respectively.