Student or educator? Sign up with your school email and get 6 months of Pro free.
Cited by 363

The DOI matches its Crossref record; 19 of the 19 references it cites were found in a registry or an index.

Transferability, sample selection bias and background data in presence‐only modelling: a response to Peterson et al. (2007)

Phillips, Steven J.·2008·Ecography

Abstract

The August 2007 issue of Ecography featured the paper “Transferability and model evaluation in ecological niche modeling: a comparison of GARP and Maxent” (Peterson et al. 2007), which raised concerns about the ability of the Maxent species distribution modelling method (Phillips et al. 2006) to predict species’ distributions in broad unsampled regions and suggested that GARP (Stockwell and Peters 1999) may be the “winner” at the “transferability challenge”. As one of the developers of the Maxent software, I'm pleased to have the opportunity to respond, and I thank the authors for sharing the data from their study. Transferability is an extremely important topic in distribution modelling, and is required by some of the most important applications of modelling. I shall address three issues: first, it is essential to clarify the difference between transferability and the problem of sample selection bias, since the study of Peterson et al. (2007) is better characterised as a study of sample selection bias rather than transferability. Second, when comparing modelling approaches, it is important to include enough species or replicates for the evaluation to have sufficient statistical power to give significant results: despite suggestive patterns in the quantitative results, the three species used by Peterson et al. (2007) were insufficient for achieving statistical significance, so their conclusions were based on subjective visual assessments. Third, since Peterson et al. (2007) focus on the interpretability of Maxent's cumulative output format, I give some clarification of Maxent's output formats, and mention a new logistic output format that may be easier to interpret. “Transferability” (also called “generalizability” or “generality”) concerns the ability of a model calibrated in one context to be “transferred”, i.e. to make useful predictions in a different context. There have been a number studies investigating transferability of species distribution models (Araújo et al. 2005, Graf et al. 2006, Pearson et al. 2006, Randin et al. 2006). Here we focus on presence-only modelling methods, which are increasingly being used to predict effects of global climate change on species’ potential distribution (Thomas et al. 2004) and for evaluating invasive potential of alien species (Peterson et al. 2003). Peterson et al. (2007) correctly observe that transferability is critical for both applications, and that a recent comprehensive comparison of presence-only methods (Elith et al. 2006) does not shed light on transferability. Most presence-only models, including those compared by Peterson et al. (2007), are based on distinguishing known occurrence sites for a species from “background” data (also known as “pseudo-absence data”) drawn from the study region. For these methods, transferral involves first creating a model using values of environmental variables at occurrence and background sites in one study region, then transferring the model by applying it to a second set of environmental variables with the same names, but describing a different region or time period (as in Peterson et al. 2003, Thomas et al. 2004). In contrast, Peterson et al. (2007) created models using background data from both the training region and the evaluation region combined. For the climate change application, this would be like mixing environmental data describing both current and future climate conditions while attempting to develop a model of a species’ current distribution. This is invalid, as future climate conditions have no bearing on current distributions. Similarly, when predicting an invasive species’ potential to invade an as-yet uninvaded region, its niche should be modeled using background data taken only from the native range of the species (as in Peterson et al. 2003), since environmental conditions in regions that the species has not had an opportunity to invade give no clue as to its environmental preferences. In their experimental design, Peterson et al. (2007) split occurrence data for three North American bird species into two subsets: “on-diagonal” (NW and SE quadrants of North America) and “off-diagonal” (NE and SW quadrants). They trained models using one subset of the occurrence data, and evaluated the predictions on the other; however, in all cases background data covers all of North America. There is no transferral involved, as model calibration and model evaluation use exactly the same environmental data. Rather, this is an extreme case of geographic bias in occurrence data. Sample selection bias is an important topic: occurrence data is typically strongly biased towards more accessible areas (Reddy and Dávalos 2003), and such bias can dramatically lower predictive performance of presence-only models (Phillips et al. unpubl.). However, sample selection bias and transferability are different topics, and either one (or both) may be relevant to any particular application of species distribution modelling. It is important

Overview

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1111/j.0906-7590.2008.5378.x

Cites

19 of 19 references verified0 citation errors · 0 not checkable

Statistics

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 19 references

Cited by

See all 363

Metadata

Authors
Phillips, S. J.
Venue
Ecography
Publisher
Wiley
Published
2008-03-28
Volume
31
Issue
2
Pages
272-278
ISSN
  • 0906-7590 (ISSN-L)
  • 1600-0587
OpenAlex
W2153183512
Cited by
363
References
19
FWCI
14.70
Type
journal article
Language
en
Open access
closed

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Transferability, sample selection bias and background data in presence‐only modelling: a response to Peterson et al. (2007). https://app.cochat.ai/papers/10.1111/j.0906-7590.2008.5378.x
The paper
Phillips, S. J. (2008). Transferability, sample selection bias and background data in presence‐only modelling: a response to Peterson et al. (2007). Ecography. https://doi.org/10.1111/j.0906-7590.2008.5378.x

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.