About network rewiring and dissimilarity (components)

Last year, I published a paper about interaction rewiring and how to partition network dissimilarity. From looking at the recent literature, it seems that there is still confusion about this topic (partly due to this paper by Tim Poisot).
The key cause of confusion may be that the purpose for calculating the rewiring component (called betaOS in Poisot’s paper and in my paper) is often not made clear enough. I see mostly two purposes: (case A) you want to quantify the degree of rewiring – then the original definition of Poisot et al. is what you need, or (case B) you want to quantify how much rewiring contributes to total network dissimilarity (implied by wording such as “rewiring drives dissimilarity”) – then the definition based on Novotny is for you. In my research, I am often interested in case A, but my above-mentioned paper focused on case B. Like in other areas of diversity partitioning, the core of the discussion is about denominators and the difference between dissimilarity (A) and additive components of dissimilarity (B).

Let me illustrate the denominator issue with a simple example: Imagine you go out for lunch and there are three options on the menu: Penne Arrabiata (vegetarian pasta), Spaghetti Bolognese (meat pasta) and Pizza Margherita (vegetarian pizza). To calculate the proportion of vegetarian options among pasta, you divide 1 by 2 (the denominator is 2): 50% of the pasta options are vegetarian. The percentage of vegetarian pasta among all options is 33% (the denominator is 3). The percentage of vegetarian meals among all options is 66% (the denominator is 3). All three numbers are informative, but they provide different information. However, you cannot calculate the percentage of vegetarian pizza as 66% – 50% = 16%; it is 66% – 33% = 33%. This shows that percentages are only additive if you use a common denominator, i.e. simple additive operations only work if you stick to percentages in the whole menu.

Back to ecological networks: “meat pasta” has to be replaced by “interactions that are shared between networks”, “vegetarian pasta” has to be replaced by “interactions that differ due to rewiring among species shared between networks”, and “vegetarian pizza” has to be replaced by “interactions that differ due to species turnover (in the wide sense)”. “Percentage” has to be replaced by “dissimilarity”, noting that the dissimilarity measures under consideration are just a special type of percentages.

You have to decide if you are interested in the proportion of pasta that is vegetarian (in the ecological network case: betaOS following Poisot; case A) or in the contribution of vegetarian pasta to all options on the menu (in the ecological network case: betaOS based on Novotny; case B). But I have yet to understand what the number 16% should tell me about the importance of vegetarian pizza on the menu (in the ecological network case: betaST following Poisot, which intends to estimate “the importance of species turnover in the overall dissimilarity”). If there is one vegetarian pasta and one vegetarian pizza, both contribute equally to the number of vegetarian options (in the ecological network case, this holds with betaST based on Novotny) – or would you say that the vegetarian options are mostly pasta and much less pizza (50% vs 16%)? Even beautiful maths proving that a denominator of 2 should be used for the pasta-specific question (case A) wouldn’t convince me that this denominator should also be used for the whole-menu question (case B).

Hope this post helps you think more clearly about what you want when estimating rewiring or partitioning network dissimilarity.


Our paper in Oikos is a “Top cited paper”!

Our paper on temporal scale-dependence of network structure led by you, published two years ago in Oikos, is having quite an impact. After being selected as Editors Choice, it became a top downloaded paper and then a top cited paper recently. As of now, it already has 31 citations in the Web of Science – so definitely a booster to Oikos’ Impact Factor 😉 If all papers in Oikos would be cited that much, it would have an Impact Factor of about 20! Congratulations, Benjamin!

What has happened the last years?

I heard there was some kind of pandemic… Have you also heard of it?

I’ve long been looking forward to writing my post-COVID post, but it never really felt like the time. Our university has just phased into the loosest COVID regulations since March 2020 – so maybe that’s the time. Well, over the past two years and a bit, my research certainly hasn’t progressed as planned (with two young kids, I never really had that feeling of an unexpected veeery long writing retreat…). But nevertheless, there have been some great achievements, and I don’t want to leave them unnoticed in this virtual space.

The greatest achievements were certainly those by my two PhD students:

Both have by now graduated and successfully defended their theses. Congratulations! They have published exciting papers, and more are on the way.

Some long-awaited papers have finally come out, including my first single author paper. Some others are still waiting to see the light of a journal… watch for updates here, which I plan to post more frequently again 😉

I’ve had the joy of digital teaching and zoom meetings (we’ve all finally learned how to use technology a bit better, and there are some advantages). But actually, I am longing a lot for real conferences and meeting real people (you?)!

Trees in IDENT are growing, and we (that’s Sylvie, I and a lot of students and hiwis) have never given up recording arthropods and damage on them:

New people have started (hola William!) and new concepts and ideas have developed. Let’s see what time will bring next 🙂

lab retreat 2019

The lab on our retreat in summer 2019 (Jochen, Sylvie, Tobias, Benjamin, Hai-Dong). It happened to be the day when all-time temperature records were broken in Germany, which explains the essential water bottles and the slight hint of dizziness. We still had a productive day with talks, discussion and R tipps, and even found 16 different types of species interactions on our little excursion. And last but not least, we talked about nothing (that is, what different ways of “nothing” you may find in data analysis, and when you have to switch between them).

A bipartite graph of the lab and keywords in 2018

Just because it is a nice illustration of topics we work on in the lab, I moved this here from the former “lab members” page:

A bipartite network of all current lab members and keywords for their research projects. People doing more similar work tend to be grouped closer together. Made with the bipartite R package, using this code: cols <- rgb(0:6/6, 0.1, 6:0/6, 0.7); plotweb(t(mydata), y.width.low=0.07, y.width.high=0.07, col.interaction=rep(cols,each=7), col.low=cols)

Lab get-together

The lab has grown! Just before the beginning of the field season, we got together last week to meet and talk about what everyone is going to study in their (upcoming) research projects. We found out that the different projects have quite a lot in common. We also talked about nothing for a while (that is, about the differences between 0, NA and other types of null); knowing something about nothing is quite important for good science…

Missing on these pictures is Rafael Bohn-Reckziegel, student assistant helping out with his amazing skills in insect sampling, data management and programming.

Welcome new lab members and good luck with your research!



PhD position available! time in ecological networks

(update Jan-2017: Position filled)

I have a PhD position (65% contract for 3 years) available in my lab (at the University of Freiburg)!

You will use models and fieldwork to investigate temporal dynamics in insect interaction networks (plant-pollinator, plant-herbivore, host-parasitoid). The main objective is to identify the functional relevance of temporal (daily, seasonal) structure of species interactions. The degree of modeling vs empirical  work is somewhat flexible.

Please apply before 5 December 2016.

Here you find the full ad as a pdf.

If you want to read more, this might be a good starting point: Fründ J, Dormann CF, Tscharntke T (2011). Linné’s floral clock is slow without pollinators – flower closure and plant-pollinator interaction webs. Ecology Letters, 14, pp. 896-904.

And somewhat more technical: Fründ J, McCann KS, Williams NM (2016). Sampling bias is a challenge for quantifying specialization and network structure: lessons from a quantitative niche model. Oikos, 125, pp. 502-513.

Another relevant paper with cool ideas and figures on the topic: Rasmussen et al. (2013). Strong impact of temporal resolution on the structure of an ecological network. Plos One

Simulate networks to understand sampling artefacts that obstruct the measurement of ecological specialization

With a little delay caused by my moving back to Germany, I’d like to to advertise my most recent paper (available in Oikos, includes a rich online appendix), which I find really important:

The network approach is currently popular in ecology, as it promises to manage the complexity of interactions among many different species. When network studies try to include all interacting species, some of them will inevitably be represented by few observations. Specialization, a major aspect of network structure, is overestimated with few observations: there is a sampling bias (in the sense of “bias due to sampling”). Trying to understand the functional relevance of biodiversity and the impact of environmental change on communities and ecosystems, more studies now compare different networks, which further limits how much effort can be spent to sample each network.

In this study, we developed a model that generates realistic quantitative interaction networks and used it to evaluate methods that try to overcome sampling bias in specialization estimates. We found that, unfortunately, all metrics and methods currently used for network analysis misrepresent true network structure when used on data with realistic numbers of observations. Although some metrics performed reasonably well, caution should be used when comparing empirical network patterns to theoretical predictions and when comparing different networks. Our model could be useful for carefully evaluating the potential for sampling bias for a given study and develop new methods to correct quantitative estimates of network structure. Our study also highlights the large potential for sampling bias in studies estimating specialization without a network focus.

Guest Post about modeling plant-pollinator webs on theoreticalecology

Together with Gita Benadi, I was invited to write a guest post on Florian Hartig’s blog theoreticalecology. We basically argue why a positive effect of nestedness on stability of mutualistic networks should not be considered an established fact, as long as it is not confirmed by dynamic models that take into account that mutualistic services are resources for which species may compete, in difference to current models, in which mutualistic services effectively reduce competition.

This post calls one of the most-cited ‘laws’ for plant-pollinator networks back into the arena, so check it out!