In the comments below we, the authors of Wittemyer et al. Science 2008 321:123-126 provide a) an expression of our concern about the process which has led to the appearance of the Joppa et al paper in PLoS One, b) a brief response to the challenges put forward by Joppa et al, pointing out flaws in their analyses that have led to a result that diverges from our own, and c) a response, follow up analyses and discussion concerning Nelson et al’s important comments on problems of scale.

a) An open dialogue?
In the interest of constructive scientific exchange, offering the target authors the opportunity to rebut a sharp critique, as exemplified by Joppa et al.’s response to our paper, is standard protocol. The ideal platform for such an exchange would be to have both papers (response and rebuttal) published in tandem, examples of which abound in the literature. However, we were not notified by PLoS One of the submission and acceptance of this critique until its publication and subsequent attention in the conservation news.

Upon inquiry to the editors of PLoS One regarding our desire to have a platform that is on equal footing of the published manuscript, we were informed that PLoS One does not have a mechanism to publish the type of response article that we seek to produce. We found this to be contradictory given that Joppa et al.’s manuscript was clearly a response to our original paper. In the interest of fairness and constructive exchange, it is odd that a scalding rebuttal of an influential paper would not be paired with any form of counterpoint or defense.

We were informed recently that our only recourse is to comment on the paper challenging our work online, hardly an equal platform. A comment was submitted stating the history of the debate among these authors. The comment was censored by the PLoS One editors, and removed. We follow that comment with this one in the hope that it will escape censorship.

The progression of this exchange is deeply concerning to us and others with whom we’ve spoken, and does not foster the open scientific dialogue so rightly trumpeted by the Public Library of Science.

b) Response to Joppa et al.
In their PLoS One response to our paper in Science, Joppa et al. have simply repeated analyses reported in the supplemental online materials of our original paper but with the caveat of minimizing the area excluded as urban in their derivation of rural growth rates (i.e. maximizing the rural growth rate by including substantial urban regions). As reported in our original paper, our analysis using compatible data sets (the UNEP vs. UNEP comparison after excluding urban extents) shows average buffer growth rates in countries with more than one PA are significantly higher than corresponding national rural averages (Z = 137.5, n = 37, p < 0.05). By failing to exclude the full extent of urban population growth, they have biased upward baseline rural growth rates and thereby changed the outcome of the rural vs. PA buffer comparison. We have addressed the critiques of Joppa et al. (and in part those of Nelson et al. addressed below) in more detail, including figures, which is freely available upon request.

c) Response and discussion related to Nelson et al.’s comments
The recent addition of comments by Nelson et al., if correct, render the Joppa et al. reanalysis and parts of our original paper inconsequential. Nelson et al. state rural growth rates can not be derived from the UNEP data in any way, and state that comparing results derived using the UNEP data to any other UN census database is not possible (apparently even if all data sets are from identical sources). We checked the compatibility of these two datasets in our original analysis and found that on a country by country basis, UNPD and UNEP total population growth rates (i.e. not attempting to remove urban areas from either dataset, but extracting the country growth rates by extracting the data on a country basis) are not significantly different (countries with greater than one park: Wilcoxon Z = 6.5, n = 37, p = 0.54; or p = 0.55 with all 45 countries), with UNPD growth rates on average being slightly higher (+ 0.01%) than UNEP. We assumed that this meant we were safe to compare the two datasets that are derived from identical national census data.

How do Nelson et al.’s comments relate to the rural growth rates we calculated? Our growth rates were calculated by summing the number of people attributed to different pixels within some spatially defined polygon. Often these polygons covered multiple administrative units (obviously in the case of our ecoregion analysis that is discussed below, but also in the case of the 10km buffer area). Growth rates were then calculated from these sums, in essence taking a weighted average of growth rates of the administrative units overlapping with the polygon where the weighting is by the relative number of people (density) from the different administrative units that were within that polygon. Clearly, it makes sense to exclude the areas with the highest density (those with urban extents) when comparing our 'rural' protected area buffers with areas farther from parks, thereby reducing the weight of growth rates from administrative units with large urban populations that are growing at the fastest rates. Nelson et al. make the case that coarser level analyses are required rather than the fine scaled ones we initially reported. Interestingly, we find the same result when we performed this analysis on a coarser 50 km buffer (though we note this complicates the analysis in that PA buffer growth rates now include urban areas—our primary concern about the moving buffer analysis of Joppa et al.), with significantly faster growth rates near PAs relative to more distant areas (Z = 131.5, n = 37, p < 0.05).

Even more interesting, results from our analysis where we control for the impact of ecoregions show significantly faster growth around protected areas relative to areas farther from parks that are of the same ecological type. The majority of ecoregions are much larger areas than either a 10 or 50 km buffer region, thereby incorporating input from numerous administrative units. Yet again, our finding that parks had higher human population growth rates was supported by this coarse-scale analysis.

Nelson et al. state that only growth rates at the level of administrative units can be used in analyses. In the supplemental information of our original paper, we explored human population growth and urbanization at the level of administrative units, stating: “As a result of the model structure used in the demographic databases, the degree of urbanization in a nation’s districts is likely to be positively correlated with growth rates, a potential source of bias in our results if PAs were aggregated in more urbanized districts. Paired analysis by country showed that the number of urban centers did not differ significantly between districts with or without PAs (Wilcoxon test: Z = 3.5, n = 45, p = 0.515). Interestingly, the average size of urban centers was significantly larger in districts without PAs (54,000 people) than in districts containing PAs (38,000 people) when collated by country (Z = -308.5, n = 45, p < 0.001). These results indicate district urbanization is not likely to bias our sample of PA buffer growth rates upward relative to rural growth rates.”

Furthermore, comparing growth rates (after removing urban extents) of administrative units with and without the focal PAs provided additional support for our conclusions, with a significant majority of countries (24 versus 13) demonstrating faster growth rates in administrative units containing PAs compared to those without PAs (Z = 113.5, n = 37, p < 0.05). Since removal or inclusion of urban extents should not influence the growth rates of administrative units, as stated by Nelson et al., this demonstrates administrative units with the rural PAs we studied are growing faster than other administrative units. Again, this is clearly not because these administrative units are more urbanized, as our results indicate the opposite that these administrative units contained similar numbers of urban areas, but significantly smaller urban areas than those farther from protected areas.

Possibly the most thorough test we were able to conduct compared growth rates calculated from village-level census data with estimates derived from the UNEP database in 10 km buffers around the 6 PAs studied in Ghana. This test, in effect, compares the modeled data with true fine scale census figures—the ideal standard for our analysis. As presented in the SOM, growth rates from these sources did not differ. Ideally, we could have conducted similar validations for the other 44 countries in our study, but, unfortunately, growth rates at such fine spatial scales were not available.

Using our best efforts to verify the scale of our analysis, we repeatedly found that growth rates near protected areas were greater than other similar regions in the country (including when the ecological context of areas was controlled for). Perhaps all these results are simply a random effect of the data modeling procedure for the UNEP data, though these modeled data are based on distance to market centers which is primarily a function of the degree of urbanization in a district and its infrastructural development. At the very least, our results then indicate that rural infrastructure development is greater in the vicinity of parks, which is a reflection of faster human population growth. We do not and have never discounted the problems of working with coarse, modeled datasets and have stated openly that our results show a broad pattern that must be tested with finer-scale studies. Most importantly, it is only through fine-scaled studies that the mechanisms behind demographic change on the boundaries of parks can be understood. But if the multiple analyses stated here are any indication, accelerated human population growth on the borders of protected areas is occurring in many areas. The conservation and development communities should take note.