Elephants and Ivory Counting elephants ought to be easy. It is not. Counting elephants lost to poaching is even harder. But, without knowledge of both, how can we know whether banning the sale of ivory is helping to save elephants? Bob Burn explains the problems. Most people would agree that we should all strive to save elephants from extinction. It is not a controversial issue. Why then should elephants take up more time, and generate more heat, than any other item on the agenda when 118 delegates from the world s wildlife departments meet to set limits on wildlife trade? This is what happened at the Hague in June this year at the 14th Conference of the Parties of Convention on International Trade in Endangered Species (CITES). This article describes the role of statistics in lowering the temperature of the debate. What was happening at the Hague was the latest battle in a war between two broadly defined camps, both committed to conservation, but bitterly divided over how to respond to proposed changes in CITES policy. A brief historical digression may help to clarify this division. Prompted by international concern over the intensive poaching of elephants that occurred in the 1970s and 1980s, the CITES imposed a total ban on trade in ivory and other elephant products in October 1989. Criteria for allowing for a future partial relaxation of the ban for stable or increasing populations were agreed. After the ban, in some parts of Africa, notably in Southern Africa, elephant populations appeared to be increasing, although this was certainly not so for all populations. Indeed, some were becoming sufficiently large to present problems, with reports of habitat degradation and increasing human elephant conflict1. In the meantime, several elephant range states had accumulated significant stockpiles of ivory. In 1997, the CITES down-listed the African elephant from appendix I to appendix II for three countries in Southern Africa: Botswana, Namibia and Zimbabwe. This down-listing is CITES jargon means that some limited trade may be possible subject to strict controls, including export and import permits. Approval for the one-off sale of existing stockpiles in these three countries was also granted, and this permission was later extended to include South Africa. All revenue from these sales was to be set aside for elephant conservation. In response to reservations about these measures from some CITES member states (or parties ), it was also agreed that two monitoring systems should be established: Monitoring the illegal killing of elephants
(MIKE), for tracking elephant populations and poaching levels, and Elephant trade information system (ETIS), for monitoring the illegal ivory trade worldwide. The wildlife trade monitoring organisation TRAFFIC was mandated by CITES to manage ETIS. One of the objectives for the two monitoring systems (http://www.cites.org/eng/ res/10/10-10.shtml) included the following: assessing whether and to what extent observed trends are related to changes in the listing of elephant populations in the CITES Appendices and/or the resumption of legal international trade in ivory. In other words, the statistical analysis was to investigate whether there is a causal link between changes in CITES policy and changes in trends in elephant poaching and illegal trade in ivory. Interpretation of any such associa- Does all trading in ivory give a green light to poachers? tion lies at the core of the divergence between the two views. One side takes the view that any relaxation of restrictions on trading ivory amounts to a green light to poachers, and that any observed increase in poaching must be attributable to it. The opposing view would argue that there are many factors that could potentially explain an increase, including, in particular, demand for ivory, and that CITES listings are of no great interest to the poaching fraternity. It is this aspect of the difference in views that is of most relevance to the statistical analysis. A broader account of the divergence would have to mention that the proponents of the signals argument are predominantly those who subscribe to a zero-tolerance view on killing elephants. The other side embraces the view, to varying degrees, that it is legitimate to regard elephants as a natural resource and that the problem is to define sustainable levels of exploitation. The Statistical Services Centre of the University of Reading was contracted to develop the design and analysis of the MIKE and ETIS monitoring systems. A technical advisory group (TAG) was established, comprising scientists and elephant conservation specialists. The TAG also has two statisticians as members (Professor Ken Burnham and Dr Anil Gore), plus me as co-opted member. MIKE: monitoring elephant poaching Work began in earnest on the MIKE in 2000. It is an ambitious project, requiring primary data collection from sites in 28 countries in Africa and 12 countries in Asia. The data include estimates of elephant population sizes and data from anti-poaching patrols, from which poaching rates are to be estimated. Capacity to undertake the survey work varies enormously across elephant range states and the MIKE system s objectives include a substantial training component. Many of the population surveys are themselves quite large-scale projects. In Africa there are savannah elephants (Loxodonta africana) and forest elephants (Loxodonta africana cyclotis), a sub-species. The habitat of most Asian elephants (Elephas maximus) is not unlike that of the African forest elephant. Given their different habitats, different population survey methods are appropriate for each, and there are Forest elephants are hard to see; numbers are estimated from the dung they leave non-trivial problems with each method. Most savannah elephant populations are estimated by aerial surveys. Care is needed to avoid double counting and to account for the bias of not counting animals directly beneath the aircraft. Forest elephants are difficult to see in dense rain forest, so population numbers are generally estimated by an indirect method. First, line-transect surveys using distance sampling are used to obtain estimates of dung density2, 3. To derive elephant density from dung density, we need to estimate the dung decay rate. This in turn requires a small field survey; a robust method for this has been adopted by the MIKE system. Standardised survey procedures have been developed for both savannah and forest elephants, which have been adopted for routine use in all MIKE sites. Along with periodic estimates of elephant population numbers, the other key output required of the MIKE system is an estimate of the rate of poaching. Perhaps unsurprisingly, this has proved to be rather elusive. Elephant carcasses are encountered, of course, while conducting the population surveys, but the MIKE system provides much more information on carcass counts from data collected by antipoaching patrols. At present, a sound statistical analysis of these data is problematic, mainly because the sampling, such as it is, is generally purposive and opportunistic, completely lacking in sampling design. A further problem is that the patrols chief task is not to record the field data, but to find and catch poachers a difficult and dangerous job and not to record data; which must cast doubt on the quality of the data they do supply. The MIKE system is still at an early stage of its development, and it 119
Actual trade in ivory Seizures ETIS records is hoped that further data will become available to enable some sort of bias correction for the patrol data. A feature of the present analysis is to attempt catch effort modelling, rather like the way in which fisheries data are handled, to obtain estimates of poaching that can be compared across sites and years. Naturally, a major preoccupation has been to settle on an appropriate measure of effort for this analysis. There is still work to do on this. With the MIKE system in its current state, data have sometimes been rather fragmentary and uneven, coming on stream as and when the various monitoring tools are established. Against this backdrop, added to the fact that the delivery of outputs is largely driven by the CITES reporting cycle, it has sometimes been difficult to provide a complete analysis. In spite of these difficulties, a baseline analysis was produced this year and adopted by the CITES secretariat. ETIS: monitoring the illegal ivory trade Of the two systems, the ETIS has more extensive data, mainly because it builds on an earlier monitoring system that TRAFFIC established shortly after the 1989 ban. It currently contains 12 400 ivory seizure records from 82 countries. The statistical challenges are to use these seizures records to assess trends in the illegal ivory trade, and to identify the principal trade routes. Of course, the data have not been collected according to any sampling plan and there are potentially huge biases to account for. The seizures recorded in the ETIS are a subset of all seizures made, and these in turn are Law enforcement effort Law enforcement efficiency Reporting rate Figure 1. ETIS conceptual model: the shaded boxes represent observed data Legislation Corruption index CITES report Data collection a subset of the actual illegal ivory shipments (Figure 1). The number of seizures made by a country will depend on the law enforcement capacity and the efficiency with which these resources are deployed. Data on these have proved fiendishly difficult to obtain, so they are represented as unobservable nodes in Figure 1. Another unobservable is the reporting rate, i.e. the proportion of actual seizures that find their way into the ETIS database. The strategy for dealing with these latent variables was to seek proxy variables, which hopefully can, at least partially, account for the same variation. the CITES maintain records of wildlife trade legislation in each member state and this is used to construct a as a proxy for enforcement effort for each party. Efficiency is a function of good governance, so the index of corruption compiled by Transparency International (http://www. transparency. org) was used as a proxy for this. The seriousness accorded to the CITES by member states was used, in the form of a measuring their success in submitting reports, as a variable likely to be associated with enforcement efficiency as well as reporting rate. A further source of bias derives from the grossly uneven effort that TRAFFIC have made over the years in obtaining seizure information. In the early days of collecting seizures data, a serious monitoring role was not foreseen, and seizures data were obtained opportunistically, in some cases by targeting particular countries. A fairly elaborate data collection has been devised by TRAFFIC that accounts for this variable data collection effort. Of course, we shall never be able to estimate the absolute volume of trade from the seizure records, but we do assume that changes in seizures reflect trends in the underlying trade process. Bearing this in mind, estimates of trends in the illegal ivory trade have been obtained by adjusting for the effects of the proxy variables using random effects models and smoothing the results with generalised additive models 4. Aside from trends, the analysis of ETIS data has succeeded in identifying countries most heavily implicated in the illegal trade, both as sources of and ultimate destinations for ivory. A finding of interest is that those countries with the greatest volume of illegal ivory trade tend to be those with unregulated (or poorly regulated) domestic ivory markets. Some successes can be claimed: at the CITES 2002 Conference of the Parties, the ETIS analysis pinpointed China as among the most important sources of demand for illegal ivory, and Central Africa as the source for much of that ivory 5. At the next conference in 2004, China was able to report that it had tightened up controls on the illegal importation of ivory 6. However, the analysis presented at this year s conference indicates a sharp upturn in the trade, again attributed to a large extent to China, but also to a number of other countries. The end of the ivory war? For many years arguments about elephants have eaten up huge amounts of time in CITES conferences. Something changed this year at the Hague. Delegates from the African elephant range states got together and brokered a compromise agreement that finally won through. The agreement allows the proposed one-off sales of stockpiles in Southern Africa to go ahead, but there are to be no further requests put to CITES for the next 9 years. This is good news for everybody. First of all it is an African solution to an African problem, which thereby has an infinitely greater chance of long-term success than solutions that have been perceived as being imposed from outside. For us statisticians it offers us a breathing space from having to respond to ad hoc requests for analyses required to meet this or that political requirement in CITES. It looks as if there may be time to develop the science properly, at last. Some years ago we (rather optimistically) drafted a data analysis strategy for the MIKE system 7. This document outlined a global analysis of MIKE and ETIS data together, and aims to address the key question posed by the original MIKE ETIS objectives what are the factors affecting elephant poaching and 120
Figure 2. Tentative concept map of the MIKE and ETIS 121
elephant populations? A rather simplistic concept map of the main relationships is shown in Figure 2. Hidden by the apparent simplicity of this model are a number of complicating issues. There are discrepancies of scale: some of the variables are measured at site level; others at country level or above. The process is dynamic, modelled over time, and the time scales of variables are not all the same. Some variables 122 are inherently latent such as K, the carrying capacity and r, the intrinsic rate of increase of the population. In spite of the simplicity, it is tempting to try to use this as a starting point for further modelling. We have 9 years to come up with something better. legal Killing of Elephants, Ivory Trade and Stockpiles. IUCN/SSC African Elephant Specialist Group. 2. Buckland, S. T., Anderson, D., Burnham, K., Laake, J., Borchers, D. L. and Thomas, L. (2001) Introduction to Distance Sampling. Oxford: Oxford University Press. 3. Laing, S. E., Buckland, S. T., Burn, R. W., Lambie, D. and Amphlett, A. (2003) Dung and nest surveys: estimating decay rates. Journal of Applied Ecology, 40, 1102 1111. 4. Wood, S. N. (2006) Generalized Additive Models: an Introduction with R. Chapman and Hall CRC. 5. Milliken, T., Burn, R. W. and Sangalakula, L. (2002) Analysis of data from the Elephant Trade Information System (ETIS) (i) spatial aspects; (ii) trends analysis. In Proc. Conf. Parties to CITES (UNEP), Santiago, Doc 34.1. (Available from http://www.cites.org/eng/cop/12/ doc/e12-34-1.pdf.) 6. Milliken, T., Burn, R. W., Underwood, F. M. and Sangalakula, L. (2004) The Elephant Trade Information System (ETIS) and the illicit trade in ivory. In Proc. Conf. Parties to CITES (UNEP), Bangkok, 13, Doc 29.2A. (Available from http://www.cites.org/common/cop/13/ inf/e13-29-2a.pdf.) 7. Burn, R. W., Underwood, F. M. and Hunter, N. D. (2003) MIKE Data Analysis Strategy. Nairobi: CITES. References 1. Dublin, H. T., Milliken, T. and Barnes, R. F. W. (1995) Four Years after the CITES Ban: Il- Bob Burn is Principal Statistician at the Statistical Services Centre, University of Reading.