EcoSim: Null Models Software For Ecology

EcoSim allows you to test for community patterns with non-experimental data. EcoSim performs Monte Carlo randomizations to create “pseudo-communities” (Pianka 1986), then statistically compares the patterns in these randomized communities with those in the real data matrix. These null model tests have wide applicability in both applied and basic ecology. Kinds of questions you can ask with EcoSim are:

1) Is the species richness and evenness of unpolluted streams significantly different from that of polluted streams?

2) Is there evidence for unusual segregation in diet or perching microhabitats of co-occurring lizard species?

3) Does the taxonomic diversity of an avian island community differ from that of the adjacent mainland source pool?

EcoSim's Help file introduces you to null model analysis and shows you how to use EcoSim for null model tests to analyze your own data. Tutorials, default settings, and sample data sets are included with each model. EcoSim's input and output screens are friendly and easy to use.

Download EcoSim

EcoSim Update: 1 January 2012

This version of EcoSim is the most current and represents the original programming effort (funded by NSF for 7 years). It still works fine, although it is beginning to show its age, as the modules have not been substantively updated in 10 years. I will keep it available here at this website as long as it remains compatable with current versions of Windows. As always, I will continue to answer questions and provide advice about analyses with EcoSim.

Unfortunately, a "new" commercial version of EcoSim has recently appeared elsewhere on the web. The commercialization of EcoSim was done without my consent or approval, and I strongly oppose it. I had no involvement with the commercial product, and I cannot confirm the validity of any of the algorithms or modules contained in it.

To combat the commercialization of EcoSim, Aaron Ellison and I are initiating a new project. We are beginning to recode the routines in EcoSim to the R language and will release EcoSimR ("EcoSimmer") at this site. EcoSimR will consist of fully annotated R-script files, along with help files, tutorials, and sample data sets for you to work with. For modules that are computationally intensive (such as co-occurrence), we will code some functions in C for faster execution.

EcoSimR will offer four distinct advantages over the original EcoSim:

• EcoSimR will run on any platform that supports R (including Unix, Linux, and Mac) and will not be restricted to Windows
• EcoSimR will incorporate important updates to the null model algorithms based on the most current advances in the literature
• EcoSimR will include publication-quality graphic outputs with options to save as .pdf, .jpg, .tif, and other standard graphics formats
• EcoSimR will contain detailed, annotated code and complete stand-alone script files that do not have to be downloaded from an R library

We hope that EcoSimR will serve as a gateway drug and inspire you to do your own R-programming and modify the code to suit your specific needs!

Null Model Notes: 1 January 2012

• Niche Overlap. The niche overlap modules in EcoSim work well with one-dimensional unordered niche categories (such as dietary items or habitat types). However, for temporal niche overlap of ordered categories (months, times of the day), we recommend the ROSARIO software by Castro-Arellano et al. (2010).

• Co-occurrence. At the time of EcoSim's release, there was controversy in the literature over whether a "swap" algorithm or a "Knight's Tour" were the best procedures for creating a random matrix that preserves row and column totals (fixed-fixed). Since then, the swap implementation of the fixed-fixed algorithm has proved to be the best choice. In a series of papers with Werner Ulrich and colleagues (Ulrich and Gotelli 2007a, 2007b, Ulrich et al. 2009), the fixed-fixed algorithm shows good performance when tested against a variety of artificial benchmark matrices. Some of these matrices are purely random, others have segregated or aggregated sub-units embedded within them. The fixed-fixed algorithm is good at distinguishing random versus non-random patterns of species co-occurrence. However, when used with Stone and Robert's (1990) C-score (a popular choice in many published papers), the analysis cannot actually distinguish between segregated and aggregated structures! This is because any matrix with fixed row and column totals that contains some species pairs that are segregated must also contain some species pairs that are aggregated (Gotelli and Ulrich, in press). The C-score is an average of all the pairwise values for different species, so it reflects both positively and negatively associated species pairs.

  So what is the solution? Instead of calculating a statistical test for the entire matrix, the pattern can be tested for each individual pair of species in the matrix. But this solution generates a new problem: because there are S(S-1)/2 species pairs, the analysis would generate thousands of individual p-values (one for each unique species pair), even for a modest-sized matrix. This problem has also arisen in genomics and proteomics, where it is now possible to rapidly screen expression levels of thousands of genes. The empirical Bayes approach allows for a realistic adjustment and screening of the individual pairs to limit the rate of false positives (Effron 2005). Ulrich and Gotelli (2010) adapt these methods for co-occurrence analysis and apply them to a large set of empirical matrices from the ecological literature. Ulrich's Pairs software implements these procedures and represents an important step forward for the analysis of species co-occurrence. Be sure to check out Werner's many useful FORTRAN programs for ecological data analysis.

• Rarefaction. EcoSim's individual-based and sample-based rarefaction routines operate by taking random subsamples of the data and then calculating the mean and variance of diversity at smaller sample sizes. The estimate of variance derived this way is conditional on the larger sample from which the data are drawn. As a consequence, the variance estimate converges to zero as the sample size approaches that of the larger sample. However, it is better to view the data as themselves a sample of a larger community or assemblage. This unconditional variance does not converge to zero. Rarefaction with these unconditional variance estimators can be found in Rob Colwell's EstimateS and Anne Chao's SPADE software programs. Finally, Colwell et al. (in press) have conceptually unified the rarefaction model with theory on asymptotic species richness estimators, so that a single, smooth sampling curve can be drawn for interpolation to smaller sample sizes (rarefaction) and extrapolation to larger sample sizes (asymptotic estimators).

Literature Cited

Castro-Arellano, I., T.E. Lacher, Jr., M.R. Willig, and T.F. Rangel. 2010. Assessment of assemblage-wide temporal niche segregation using null models. Methods in Ecology & Evolution 1: 311-318.

Colwell, R.K., A. Chao, N.J. Gotelli, S-Y. Lin, C.X. Mao, R.L. Chazdon, and J.T. Longino. Models and estimators linking individual-based and sample-based rarefaction, extrapolation and comparison of assemblages. Journal of Plant Ecology, in press

Gotelli, N.J. and W. Ulrich. Statistical challenges in null model analysis. Oikos, in press.

Gotelli, N.J. and W. Ulrich. 2010. The empirical Bayes approach as a tool to identify non-random species associations. Oecologia 162:463-477.

Stone, L., and A. Roberts. 1990. The checkerboard score and species distributions. Oecologia 85:74–79.

Ulrich, W., M. Almeida-Neto, and N.J. Gotelli. 2009. A consumer's guide to nestedness analysis. Oikos 118: 3-17.

Ulrich, W. and N.J. Gotelli. 2007a. Disentangling community patterns of nestedness and species co-occurrence. Oikos 116:2053-2061.

Ulrich, W. and N.J. Gotelli. 2007b. Null model analysis of species nestedness patterns. Ecology 88:1824-1831.