Distributed Ecology

Sheep cyclone problem...a start.

2012-02-24T10:42:00.000-08:00

The other day the Oikos blog posed this problem from the "Just simple enough" blog by Amy Hurford. I'll quote the problem here.

Sheepclone!

"What type of individual movement rules are needed to produce a Sheep cyclone?"

Well, I decided this would be a good time to refresh my Netlogo knowledge and give it crack. I'll be the first to admit this is a sub-optimal solution. But I thought with the combination of flocking behavior and avoidance rules I might be able to recreate the cyclones. Well, it turns out that was not the case. I built upon the preexisting flocking model in Netlogo. I added a car and some walls and the avoidance behavior. I controlled how far and wide the sheep could see in front of them and how many degrees off their current heading they would turn. I also differentially weighted their perception of "walls" and "cars". They would turn away from a car earlier than they would a wall. Unfortunately I didn't quite get the cyclone effect. Mostly it seemed to generate some sheep in front and some behind the car just sort of buzzing around with the occasional circle.

So what did I learn from this? The rules are a bit more complicated than I first thought. Of course I could just program the sheep to run around the car, but really what's the point of that? The purpose of agent based modeling is to reproduce a given pattern with a plausible set of underlying rules, not just recreate a specific behavior. I believe that two things need to happen to generate the behavior. One is change in flocking rules. The idea would be to create a panic state where the sheep no longer flock the way they nornally do but always follow their nearest neighbor. Next they need an escape behavior by turning away from a wall next to them. This behavior would generate the left turn around the car after the sheep has passed it but still has the alley to its right. I haven't had time to program these more complicated rules, but I'll get around to it. You can see a couple of youtube videos I made here and here, and then download and run my Netlogo code from Github.

Monty Hall meets a python

2012-02-20T23:29:00.003-08:00

Ahh, the Monty Hall problem. A classic problem that begins many statistics classes involving conditional probabilities. I first learned about it in my Bayesian statistics class in grad school, and I've seen it many times since. The problem in a nutshell is that imagine you are on the gameshow "Let's make a deal". You are presented with three doors, one has a car behind it, the other two goats. You choose your door and Monty Hall reveals one other door that does not have the prize behind it. Now you have a choice, you can either stick with your original door, or switch. The question is, what should you do to maximize your chances of winning the car? Intuition says that your odds are 50/50 after he eliminates the door, so it doesn't matter. But if that was the case it wouldn't be a famous problem. No instead if you stick withy our answer your odds of winning the car are 1/3, and 2/3's if you switch. One can arrive at this easily enough through a variety of methods (see the Wikipedia page on the problem), but an easy way to arrive at these conclusions is via simulation.

I recently stumbled upon a blog by Corey Chivers on called bayesianbiologist and I was reading through some of his posts and I came across his R code (very elegant code I might add) for simulating this problem. I had done this years ago but currently I'm trying to expand my horizons beyond R and a work on other languages. Currently I'm interested in Python so I decided now that I've got NumPy/SciPy/matplotlib up and running with Eclipse I thought I would write a version in Python. Its certainly not the most elegant code, but it was interesting writing something I could bang out in 10 minutes in R in a different language. You realize that how you would do something in R frames the way you think about problems. Seeing how constrained my thought process was makes a good case for learning other programming languages. I hope that it will give me a broader perspective about how to approach computational problems. You can see my full code here at my git, or below. As its written now, it runs 100 simulations 1000 times and produces a histogram of the probability of winning the car.



'''
Python version of the Monty Hall problem
by EM Hart 2/20/2012
Change scenario by changing the code 
in strat_dictionary (0,1,2)

'''

#from matplotlib import pyplot
import numpy.random as np
import numpy
from scipy import *
from matplotlib import pyplot

#create an array for Wins
pwins = zeros(1000)
#####Create an an array of all possible values
potential = array([1,2,3])
#####Change this to change your strategy
strat_dict = ["Stay","Switch","Random"]
master_strat = strat_dict[2]
for k in range(1000):
    wins = zeros(100)

    for i in range(100):

####Assign a prize value
        prize = np.random_integers(1,3,1)
###Now make a guess
        guess = np.random_integers(1,3,1)
   
    
####Now we need to figure out which of the doors are revealed
        if prize == guess:
            reveals = potential[where(prize!=potential)]
            reveals = reveals[np.random_integers(0,1,1)]
####Here is where I might have used which in R
        if prize != guess:
            reveals = potential[where(prize!=potential)]
            reveals = reveals[where(guess != reveals)]
####This formulation allows me to have Random strategy
        if master_strat == "Random":
            strat = strat_dict[np.random_integers(0,1,1)]
        if master_strat == "Stay":
            strat = "Stay"
        if master_strat == "Switch":
            strat = "Switch"
        
#Now its simple if we just stay       
        if strat == "Stay":
            guess = guess
####Switch is a bit more complicated, this is a very inelegant solution compared to R
        if strat == "Switch":
            switch = concatenate((guess,reveals))
            for j in range(3):
                exc = potential[j] in switch
                if exc==False:
                    guess = potential[j]

        
    
        if guess == prize:
            wins[i]=1

        pwins[k]= wins.sum()
        

pwins = pwins/1000
pyplot.hist(pwins,100)

pyplot.show()

Fast paced food-web plotting action

2011-12-20T14:52:00.000-08:00

A simple foodweb

One of my interests is in food web topology and food web dynamics. My research in experimental ponds has left me with 2430 food webs to make sense of, and one way to facilitate understanding that many networks, and really any food web, is through visualization. There are fancy ways to visualize large webs such as FoodWeb3D, but A). I usually don't have huge webs, and B). I've become a died in the wool ggplot2 convert, so I wanted to be able to create plots using ggplot2. What I needed was a function that would return a ggplot2 object that was a barebones plot of my web. Before I get into the details of the plotting function, it's worth a few moments to go over what a food web diagram looks like and what it visualizes. Each point is a species and each line is saying that one species eats the other. The diagram on the right is a visualization of an S x S square matrix containing 1's and 0's where S is the number of species and a 1 in a column means that the species in column number x eats the species in row number y. The central task is how do I take that square matrix and convert it to ordered points and then plot it? The first job is simply to create a set of nodes, and to do this I wrote a function where the input was the number of points you want, and it returns that number evenly spaced along the unit circle.


########## support function create.xy
########## returns regularly spaced circular coordinates for the size 
########## of your web

create.xy <- function(po){
  degs <- seq(0,2*pi,by=(2*pi/(po)))
 return(cbind(cos(degs),sin(degs)))
}

Network diagram with color!

The actual plotting function works by looping through each column of the matrix and determining which points have 1's (indicating consumption) and then repeating xy coordinates for the consuming node (species) in a matrix, and then creating a matrix of xy coordinates of the nodes consumed. A tricky part is that these two matrices need to be combined into a single data frame and properly ordered. I achieve this by indexing each point in the consumer matrix with an odd number, and each point in the consumed species matrix with an even number. Then after combining the two different matrices, I can order them by the index and convert it to a data frame object so ggplot2 can plot them. On top of the matrix of 1's and 0's the function requires a series of color labels for each consumer node. It can be all black, but it allows for plots with different colors such as the one at right. I've used the convention of red for predator links and green for herbivory links, but you could use whatever colors. One problem is that if there are lots of reciprocal links (Species A eats B, but B also eats A) and each has their own color, then the colors will mix in the figure (red and blue become purple in the above case), so be wary of using too many colors.

A time series of food webs

I also stripped each plot of any labels and axes, this is easy enough to modify in the code where I call ggplot.Other things like adding line weights for interactions strengths, arrows for directionality, etc can all be modified within the ggplot options. One of the most flexible parts for me was that the function returns a ggplot2 object which I can then use to create series of plots. For instance I'm looking at food web dynamics within my ponds so like to visualize a time series of webs. Using the arrange() function with a list of ggplot2 objects I can easily do this. The function will also return the raw data frame that can be used for any other plotting you might want to do. You can see a Gist of all the code here, or add to it on the github page. Here are some more examples of plots below. Note that as the number of species increases the webs become impossible to discern, such as the 249 species Caribbean food-web


20 species niche model	249 species web, Bascompte et al 2005	40 species random web

Plotting implicit functions in R

2011-11-11T12:06:00.000-08:00

So in prepping for my latest manuscript on population dynamics I have been creating all the necessary figures. One of them I considered was a 2-d surface plot of a modified Ricker equation showing the transitions from extinction stability, and stability to limit cycles. Inconveniently though the only way to do this is with an implicit function. Since becoming a fully ggplot2 convert, I wanted to be able to do it in ggplot2. Let's start with a simple definition so we're all clear. Here's an explicit function.\[y=2^2+x^2\]It's explicit because y is stated explicitly in terms of x, i.e.\[f(x)=2^2+x^2\]Yet sometimes its not possible to define y explicitly in terms of x. Implicit functions are formally defined as equations that satisfy the condition:\[R(x,y)=0\]A good example is the unit circle. The unit circle can be found in the equation:\[f(x,y)=x^2+y^2\]And unit circle is created when:\[f(x,y)=1\] Therefore we can draw the unit circle using the equation:\[x^2+y^2=1\]\[x^2+y^2-1=0\]This is the implicit function for the unit circle. You can solve it explicitly for y, but it requires two functions.\[y=\sqrt{1-x^2}, y=-\sqrt{1-x^2}\]So, if you have an implicit function you want to plot, how could you do it in R? The easy answer is to numerically solve your equation over a specific range, but then how can you plot it?

my.fun <- function(x,y){y^2+x^2 -1}
x<-seq(-1.5,1.5,length=1000)
y<-seq(-1.5,1.5,length=1000)
z<-outer(x,y,my.fun)
contour(x,y,z,level=0)

That will produce a graph that looks like the figure on the left, whereas the figure on the right was created with the function I wrote called imp.solve().

my.fun <- function(x,y){y^2+x^2 -1}
to.plot <- imp.solve(x,y,my.fun)
#Plot with ggplot
ggplot(to.plot,aes(x=x,y=y))+geom_point()+xlim(range(x))+ylim(range(y))

The first example produces a quick and dirty plot that is all fine and good, but its a surface plot. That means you're stuck with the "0" on the line denoting the surface is only drawing areas where 0 is. On top of that you lose lots of other control over what you can draw and what if you you want to use ggplot2, or some other package, or you actually want to know what the numerical answer is? My function parses

the output from the outer() function and returns a dataframe you can easily plot in ggplot2. It can handle most polynomial functions, although I'll leave you in charge of ordering your own output for line plots in scenarios with multiple x's at the same y point. Here's another example with a cubic function. \[x^3+y^3-1=0\]

An ecological example

So why does this matter in ecology? Often you might want to create a surface plot for transitions between stable states and unstable states in with a discrete time equation. Here's an example from my own work. I'm trying to recreate a figure such as this from a paper by Aviles 1999. Here is the figure at right. The function is a modified Ricker equation:\[N_{t+1}={N_t}^{(1+\gamma)}e^{(-c+N_t)}e^r\]Here gamma is the allee effect term, c is the strength of density dependence and r is the population growth rate. Both gamma and r are bifurcation points and we can use an implicit function to draw the boundaries between say stability and extinction at a constant c. So while implicit functions aren't the most common thing to come across in ecology, they do arise with some mathematical models. In this example the transition between stability and extinction is given by the equation:\[ce^{(\gamma - r)}-\gamma=0\]Holding c constant at .01, the following code will generate our figure.


##########First set up the parameters to vary over########
y<-seq(0.01,.9,length=1000)
x<-seq(-2,8,length=1000)
###################Define the function############
my.fun <- function(x,y){.01*exp((y-x)/y)-y }
to.plot <- imp.solve(x,y,my.fun)

ggplot(to.plot,aes(x=x,y=y))+geom_line()

This is the same as the the first line in the above figure from the paper. You can find the full code for this function over at my github page as Imp.solve.R. So while a previously quick and dirty solution existed in R, here is an improved version with greater plotting flexibility. Happy implicit function graphing!

#SciFund Rocket's ahead on Rocket Hub

2011-11-01T07:46:00.000-07:00

I don't have a project in here but I think everyone should check them out and support them. This may be how research can be funded in the future, so check out science crowdfunding in action.

Oh Hi I'm still here

2011-11-01T07:38:00.000-07:00

Hi blog world...I haven't left, but my wife has become seriously ill and she and I have been in and out of hospitals the past month, with more to go. It's sad to have worked hard on the blog, but on the list of boats to keep afloat the blog is not at the top of the list.

Check it out I'm on #SciFund

2011-10-18T09:34:00.000-07:00

Thanks to Jarrett and Jai for letting me post on the #SciFund Blog. http://scifund.wordpress.com/2011/10/18/how-a-scifund-proposal-is-different-from-an-nsf-proposal-or-what-can-we-learn-from-beer/

Climate change, range shift simulation modeling and the limits of its usefulness

2011-10-10T15:25:00.000-07:00

Two new papers just came out in Ecology Letters that caught my interest about range shifts. The first is by Early and Sax called: Analysis of climate paths reveals potential limitations on species range shifts. and was highlighted by Nature. The next article was by Sheldon et al, also in the early online section titled: Climate change and community disassembly: impacts of warming on tropical and temperate montane community structure. Both papers focus on range shifts and both would qualify as "3M" as they use collections of published data and simulation models. Full disclosure as well I did my dissertation on climate change and I currently am a post-doc who only does individual based and simulation modeling. I really enjoyed Early and Sax's paper because it used simulation to develop a novel point that would be difficult to demonstrate empirically. Its long been known that species ranges will shift as climate changes and we can use what are called bioclimate envelope models to make predictions about where species will go. These models essentially say that where a species is now is a good indicator of the climate 'envelope' that it can survive in and it will move geographically to follow that envelope. Early and Sax ask, if you're a salamander in California, and suddenly Oregon starts looking pretty good thermally speaking in 2075, can you get there? The short answer is most likely "No". Not because of physical barriers but because of climate barriers. This is because as Oregon gets gradually warmer, it will also have cold years too. So while a salamander can survive the range of temperatures in California, as climate change happens in Oregon only some of those years will be within the range of temperatures our intrepid salamanders want to live in. So if you have a favorable climate patch in Oregon, but only a narrow corridor to get to it, just because future climate models predict that a lizard could get there doesn't mean it will. Of course many people have pointed out the link between dispersal rates and distance and actualized range shifts. The novelty of Early and Sax's work is that they point to the importance of a species ability to persist in unfavorable edge habitats.

Potential habitat and habitat that can actually be reached

They make a compelling case that just because a habitat will become suitable based on a bioclimate envelope model doesn't mean that the the species can make it there. The first factor affecting persistence is going to be climate variability in movement corridors. Furthermore they point out that persistence be impacted by other things. For instance competitors and predators of a given focal species are also on the move and depending on how well they can track a shifting climate. Obviously if you're a manager planning for climate change, you'll need to consider how best to make sure that species of conservation importance can best make it through these corridors. One problem with this work that it relies on bioclimate envelope models which have been criticized for the past decade or so (i.e. Pearson and Dawson 2003). The authors actually go to great lengths to justify their use of envelope models, and I can't fault them too much. I wouldn't hang too much on these kinds of models either, but I think they work fine for the authors purposes. Actually the unmentioned corollary of this work is biotic interactions and priority effects (i.e. Chase 2007). As lots of different species are moving from place to place, the new communities that are assembled in these new ranges could be different than the originating communities. Community assembly under climate change could be a fruitful thing to look at in conjunction with bioclimate envelope models and range shifts. This brings me to the next paper using simulation to look at climate change because it focuses on community disassembly.

This is work by Sheldon et al. is an example of a simulation paper that I thought wasn't particularly interesting. Feel free to turn that criticism right around on me because she has a paper in Ecol. Letters and I don't have any. Pressing forward though. The premise of the paper is straightforward. There will be more warming in northern latitudes, so you expect a greater amount of community disassembly. Back up for a moment our author tells us. Tropical species have much narrower thermal ranges therefore even though there's greater change in the far north, there may be a bigger impact in the tropics. To get at this they use already published transects of a variety of species (n = 53) that spanned 90 degrees of latitude, 45 in each direction. They then created a simulation to examine how montane communities disassemble with changing temperature.

Kuala Lumpur montane forest

In this case a community is a published mountain transect. The simulation uses variety of dispersal mechanisms that allow species individually to move up (or down but mostly up) a mountain. Some species have higher dispersal capacity and can make it to these new elevations that are within their thermal tolerance. Others won't be able to disperse or their range will "move off" the mountain because the mountain is too small. In short Sheldon et al. found that montane communities in the tropics were actually more sensitive to temperature changes compared to temperate communities, e.g. there was more community disassembly. This I do think is an interesting result theoretical result, but I don't find it compelling on its own. I would have been much more convinced if the authors had included some actual data to compare their model to. My problem with this is that its really just recapitulating what we already know. Its a well known truism that species will have individualistic reactions to climate change, causing novel community assembly (and disassembly in this case). This idea has been around for at least 20 years. I think the important question is the follow up: "What does community assembly and disassembly mean in this brave new climate world?". Ok, so we already know that community disassembly is happening (before this paper), so what are the implications of this? If community disassembly is happening faster in the tropics what will this mean for tropical vs temperate communities? I wanted either further simulation or some real data. I was left feeling that lots of work went into this simulation but the project was only half finished. I don't like to criticize other people's work too much (a problem of ecologists being too nice?), but again it seems like the authors went to great length to demonstrate something with a simulation that I don't think was that novel relative to the Early and Sax paper.

Let me restate my disclosure of earlier, I am in a post-doc that is all modeling and simulation. One of things that haunts me is the thought that: "I'm just making shit up". Ok, its grounded in actual data and theory, but I'm nonetheless plagued by the dangers of circularity. I can program a simulation to mimic a pattern, but does that mean that my simulation is an actual mechanistic recreation of the processes that produced the actual data? I can come up with a hypothesis and write a simulation to support it, but does that mean my hypothesis is supported? I don't have an answer to this question, but I do worry that simulation only projects can drift dangerously close to being circular (Like I said it can keep me up at night). I think that simulation is a powerful tool for ecologists, but in the melee of trying to pump out as many papers as possible it can get overused. These two papers highlight that. Early and Sax's paper I thought was an excellent use of simulation and I really enjoyed the paper. Sheldon et al's on the other hand was a good simulation but needed more follow through, preferably in the form of actual data, to have punch. In the end without that empirical data to support it, it felt like it was just a bit too circular for me to trust the result based on simulation alone.

Early R, & Sax DF (2011). Analysis of climate paths reveals potential limitations on species range shifts. Ecology letters PMID: 21955643

Kimberly S. Sheldon, Sylvia Yang, & Joshua J. Tewksbury (2011). Climate change and community disassembly: impacts of warming on tropical and temperate montane community structure Ecology LettersClaim token: sciseekclaimtoken-4e9468ef61429Sample HTML to include:

sciseekclaimtoken-4e9468ef61429

Post-Doc opportunity

2011-09-29T10:53:00.000-07:00

Here's a chance to post-doc with Jeremy Fox (the ever thoughtful man behind the curtain at the Oikos Blog), which would be awesome. In the words of the Oatmeal it would be: "Like making out with a unicorn". Here is the ad.

"I am once again seeking applicants for the Killam postdoc. This is a competitive, 2-year postdoc awarded by the University of Calgary in any academic discipline. The award provides a salary of $45,000 CAD/year, health benefits, and a research/relocation allowance of $6000.

The awardee would be expected to develop their own independent research on some topic of mutual interest to themselves and their faculty mentor (that would be me). I welcome applicants interested in pursuing fundamental research in any area of population, community, or evolutionary ecology. I would be particularly interested in applicants interested in collaborating with David Vasseur (Yale University) and I on theoretical and/or empirical research on spatial synchrony, but my interests are by no means limited to this topic.

To be eligible, you must have completed your PhD sometime after Sept. 1, 2009, or else anticipate completion by Sept. 1, 2012.

Application deadline is Dec. 15, 2011 (that's the departmental deadline; one applicant will be selected by my dept. for consideration at the university level in Jan. 2012). The application needs to include a research proposal; prospective applicants should contact me well in advance to discuss this.

For more on the Killam postdoc, see http://www.grad.ucalgary.ca/awards/award_competitions/killam_postdoctoral_awards

For more on my lab, see http://homepages.ucalgary.ca/%7Ejefox/Home.htm

Please circulate this in your dept., and bring it to the attention of any potential applicants."

Urban evolution revisited.

2011-09-22T14:50:00.000-07:00

Last month I posted about an article in the New York Times on urban ecology and evolution. Since then two interesting papers have come out about natural selection and urban environments. The first is by Halfwerk et al. from PNAS titled: Low-frequency songs lose their potency in noisyurban conditions and a second one also about birds by Rodewald et al. in Ecology: Dynamic selective environments and evolutionary trapsin human-dominated landscapes. Both deal with selective forces that urban environments impose on bird species. Halfwerk et al. worked with Great Tit's

Tasty bug!

in the Netherlands. Birds that rely on communication via songs have always had to deal with how environments alter their songs, dealing with reverberations in different forest types as an example.Halfwerk et al. examined the affects of anthropogenic noise on female preference for males. Cities typically have low frequency background noise and birds, and in order to cope with this male Great Tit's have change the frequency of their songs to reduce spectral overlap with ambient noise. Halfwerk and colleagues found that males tend to use low frequency songs right before egg laying and that females show higher fidelity when attracted by low frequency songs. But in urban environments there is a rapid signal degradation of the low frequency songs, so if males want to be heard by more females they have to use high frequency songs. The authors argue that this creates a trade-off between natural selection and sexual selection. There is natural selection they argue for high frequency songs because it maintains a strong signal efficiency in urban environments but sexual selection for low frequency songs due to female preference and fidelity. In this study the urban birds are worse off because they are more likely to be cuckolded due to the use of high frequency songs, but in Rodewald et al's work on Cardinals in urban environments the opposite is true.

Rodewald et al. examined how urban vs rural environments altered the relationship between plumage coloration and annual reproductive output in the northern Cardinal. They found that Cardinals in urban environments were had no relationship between annual reproductive output and redness (as measured from photographs) but in rural environments there was a negative relationship between redness and reproductive success. Cardinals gain most of their coloration from carotenoids and in typical rural environments these are hard to come by so redness becomes a typical sexually selected trait. Only the strongest cardinals can gain access to the best resources (and carotenoids) and females prefer those cardinals that are the most red. The catch is that in Ohio, where this study took place, there is a common invasive called Amur Honeysuckle (Lonicera maackii). In urban areas this is very common so combined with access to bird food there is no relationship between fitness and redness. In urban environments then there is relaxed sexual selection for coloration. The situation is worse though. In rural environments there is still honeysuckle, but its much more rare and as a consequence the fittest males gain access to it.

Red makes me strong!

But its what Rodewald calls an "evolutionary trap". Males will choose to nest in it, but its lower to the ground and nests in the honeysuckle are more likely to be preyed upon, especially early in the breeding season. Therefore in rural environments, redness is actually correlated with reduced number of successful reproductions in males. So in this Cardinal system in Ohio, urban birds have an easier time because everybody can be red. One thing that's unclear to me from the study was that I thought if nesting in the invasive honeysuckle causes reduced number of offspring in rural environments, how come urban males don't have the same problem? Regardless of this question though, both studies show how anrthopogenic forces can alter established sexually selective relationships. In the case of the Great Tit's, it was forcing males to change their song to a less desirable one, and in the case of the Cardinal urban environments let every male appear to be a good mate choice.

Both studies offer evidence of how urbanization changes the fitness landscapes for organisms. There isn't necessarily anything very novel about this in and of itself, in fact just look at pigeons or house sparrows as examples of birds and urbanization. What's really interesting about both studies is that they show the wide variety of complex interactions that happen with specific facets of urbanization that we rarely consider changing that fitness landscape. What remains to be seen in both systems is what is the eventual upshot of these altered relationships. Will new sexually selective relationships evolve in each species such that there becomes a high degree of "local adaptation" for mating preferences in urban and rural birds? That could be something interesting to study in the future. What are new selective relationships? We can all see the old ones becoming less important because we already know what they are. For instance in the Great Tit its not that females actually prefer high frequency singing males, its just that those are the ones they can find. In fact Halfwerk points out that male-male communication usually uses songs in that frequency range, so will urban environments change bird communication completely then? It would be interesting to see someone make specific predictions about what they expect to see these new relationships would be and to go out and test them.

References

Halfwerk W, Bot S, Buikx J, van der Velde M, Komdeur J, Ten Cate C, & Slabbekoorn H (2011). Low-frequency songs lose their potency in noisy urban conditions. Proceedings of the National Academy of Sciences of the United States of America, 108 (35), 14549-54 PMID: 21876157

Rodewald, A., Shustack, D., & Jones, T. (2011). Dynamic selective environments and evolutionary traps in human-dominated landscapes Ecology, 92 (9), 1781-1788 DOI: 10.1890/11-0022.1

A recipe for success?

2011-09-20T07:45:00.000-07:00

Earlier I wrote a post titled Optimal ecologist strategy. My argument there was that to be a successful modern ecologist you needed strong computational skills because your access to modern analytical tools is limited if you don't have good math skills and good programming skills. Well on Friday I dropped by my old advisor Nick Gotelli's office to chat about what kinds of confidence intervals to use in a simulation. We ended up having an hour long chat and I proudly told him about my adventures in python and my belief in the importance of programming skills. He told me that really it didn't matter if I could program or not. In his opinion what makes a good faculty hire for an ecologist position is strong empiricism and the ability to carry out field experiments.

Don't get trapped, find that maximum!

I was a bit surprised at this because I've been laboring under the belief that computational skills are really important and as we move into an era of "Big Data", faster computers etc... these skills become more and more important in getting a job. His point though was that engineers and computer scientists will always be out pacing me because these methods are their bread and butter skill set. He is of course right, for an ecologist I'm a pretty good statistician and a pretty good programmer, but judged by a statistician and a computer scientist my skills at both are mediocre at best. On top of that he argued that more and more people are becoming 3M people, so the competition for those jobs is getting harder and harder. So what do people think is the best strategy for making yourself a good job candidate? We only have a fixed number of hours in a day, so what skills are the most important for ecologists to focus on?

Git some....

2011-09-12T11:41:00.000-07:00

I'm moving all my projects to Git Hub, and maybe some old ones too. My main problem is that I'm vexed by a good memory of when I was young and had a good memory. The result is that I rarely document anything and just think "I'll remember that" Now that I'm not in my 20's anymore I've noticed this ability has faded a bit but I'm left with pages of old undocumented code. I may be too embarrassed to put all my tragically undocumented code up on github, but I'm moving my current project on there. You can check out my github here.

Paperless academic work flow, alternatively "save a tree..."

2011-09-06T10:27:00.000-07:00

...eat a beaver? That's what Google suggests when I type in "Save a tree", but for the purposes of this post I'm going to finish the sentence with "buy an iPad." Today I'm going to talk about my paperless workflow. This is something I've wanted for a long time and I think this is a reasonably good approximation of my dream of a paperless life. Here is what you'll need:

1. Drop box account (Free)
2. Mendeley account (Free too!)
3. Tablet with PDF annotation app (I use an iPad2 with iAnnotate)

Revenge of the terminal

I'm guessing except for #3 most people already have the components installed on their machine. Lets begin with Dropbox. If you don't know about it already, Dropbox is essentially a networked drive that you can access via their website and mobile device apps (any iOS or Android device). What's really amusing about Dropbox, and probably even more so for people a bit older than me, is that it essentially is a return to mainframe style computing with a terminal and no local storage.What's great is that if you're like me and you work on a desktop in your office and a laptop on the road or at home, I only need to maintain one version of all my code projects and word documents. For instance I run all my projects in Eclipse off of Dropbox folder. Once you have Dropbox installed on your local computer, you'll end up with a folder in my documents that looks something like this:

Ok, so now you've got your dropbox account setup, you should go over to Mendeley and download their software and install it on each computer you're going to use. I use two different machines so I have one on my Windows workstation and my MacBook Pro. Once you've downloaded and installed Mendeley, you'll need to link Mendeley with your Dropbox account. This is pretty simple, you can see it on YouTube here. What you need to be careful is that you don't set-up a watch list if you're syncing across multiple computers. This is because you'll end up with duplicate PDF insanity. So you've got everything linked together, now what? Well what you need is quality annotation app for your tablet. I like iAnnotate as I said before, but whatever you use be sure it truly annotates the PDF. Some of the less expensive tablet annotation apps work by converting your PDF to an image and then effectively drawing on it. This is not what you want. At this point there's a reasonable question to ask: "Why am I doing all this if I only have one machine?" Dropbox is what allows for integration of your tablet annotation app. So the first thing you'll do is add a citation to Mendeley (I use Google scholar and the one click importer). Next associate the PDF with the citation in Mendeley. You can drag and drop a PDF into Mendeley, but I've met with limited success depending on the journal, of the software accurately importing the citation correctly so my preference is to actually get the citation right and then associate the file. So at this point Mendeley has a PDF associated with a citation that's actually stored on Dropbox. The next step is to make sure that your annotation app is set-up with Dropbox. This will vary by app so I'll leave that up to you, but its usually rather simple. Next download the paper in your app:

Now you can open up your paper on your tablet:

Next you can highlight, take your notes, etc,

After you've read your paper, all you have to do is upload it back to your Dropbox folder, keeping the name the same as it is associated with Mendeley. iAnnotate has a feature to update a file, so as long as you get it on to the iPad via the iAnnotate / Dropbox link, you can use one click to update the file. You can then open it in Mendeley (or any PDF reader) and all your annotations will be there. I made the red highlights in iAnnotate:

You can also highlight in Mendeley, but that doesn't actually modify the PDF. The yellow highlights can't be seen in any PDF viewer but the Mendeley one.

If you want to be able to see your Mendeley highlights you'll have to use a small work around. Essentially you have to export your highlights as a new pdf, and then delete the old one on Dropbox and then put the exported one with the saved highlights back into your Dropbox folder. After that you'll have to delete the file from your iPad and then redownload it. It will look something like this with iPad marks in red and Mendeley exported ones in yellow:

If you're in a lab that doesn't use Mendeley the lack of ability to have highlights from the actual client doesn't really matter much. If your group does, then the work around may be too much of a pain for this to work for you. So lets sum up what works and what doesn't.Pros

A paperless work flow that lets you highlight and take notes on pdf's from your citation manager
Minimal expense if you already own a tablet and works way better than having one pool of PDF's for your reader app and another for the citation manager
Can work across multiple computers

Cons

Mendeley highlights aren't easily importable to other PDF readers
Syncing problems can happen because you're using distributed services that weren't truly meant to work together (like you can't use the watch folder feature of Mendeley)
The complicated work arounds needed to make the whole system work

What would be great is if the Mendeley ipad client essentially functioned like iAnnotate, or there was some integration. I would certainly shell out $10 to have a good iPad client like iAnnotate that seamlessly worked like this current system. I think Mendeley is a great system for lab groups to have, but if my lab actually used it I wouldn't be able to have this kind of workflow. I hope in the future Mendeley actually gets a better iPad client or works with existing ones for good integration, I could cut Dropbox out of the equation. I'm going to go on one more aside before I wrap this post up. I think there are lots of (I shudder to use this term) Web 2.0 (yuck I hate that word) and socially distributed tools that can really help lab groups function. Things like a lab wiki, Mendeley group to share papers, or collaborating on manuscripts using Google docs to name a few. I've mostly tried to use Google docs with older faculty members and met with tremendous resistance. I know that this far from always the case, but in my experience at UVM, we might as well be living in 1985 the way we collaborate. Its a brave new world with lots of sharing tools that can really help scientists share ideas and collaborate and I don't feel like we use enough of them in my limited time in academia. I'm sure this will change with time but, to borrow a term from Nick Gotelli, mostly through "demographic attrition". Hopefully we won't need to wait for a slew of retirements though to see these tools make into established lab groups.

The disqus migration

2011-09-05T10:09:00.000-07:00

I don't really see much comment action on my blog, and when I went to comment on it myself I got a glimpse into a possible reason. Of course given my massive readership and witty prose, I immediately discounted the fact it just might possibly be that not very many people read my blog and find it interesting enough to comment on. Now that I successfully falsified the two simplest hypotheses, I began my search for the more nuanced answer. I came to the realization that blogger's comment system is as efficient at commenting as an elderly woman is at paying for that one item at the express lane in the grocery store with a check at the register thats being run by a teenager who only knows about checks from movies they used to watch out of their parent VHS collection. So I installed Disqus. Now all I need to sit back and watch those comments fly in. Also if anyone was looking at the blog this morning it may have seemed like some sort of theme short circuit happened as I tried mucking with the CSS and HTML to add some custom "reply to comments" code but it didn't work.

Exponential growth of Ecolog

2011-09-02T13:57:00.000-07:00

Just for kicks I took the recently posted subscriber data on Ecolog and made this figure and fitted an exponential growth curve to it. I assumed an initial subscriber base of 100 and estimated monthly growth to be r = 0.0281. Update To give a bit more meaning to what that r means, Ecolog has a doubling time (as I calculated it) of ~31 months. By April of 2014 there will by 26,000 subscribers. I'll have to check back in with David to see if that actually holds true.

Tenure track position in systematics at the University of Vermont

2011-08-22T12:19:00.000-07:00

We are hiring a professor in insect systematics here at UVM. On top of UVM being a great university to work at Burlington is an amazing and special place to live. In fact I'm about to pack it in and go jump on my sailboat on this beautiful windy August day.If it were winter I'd be getting up earlier on Saturday than I ever do during the week to skin up the mountain to get some fresh tracks.

So what are you waiting for insect systematists?

Update
Another reason to consider UVM and Burlington, this rainbow over the lake as I went for my lakeside run down the city bikepath.

Assistant Professor in Systematics

Department of Biology
University of Vermont
Burlington, Vermont

The Department of Biology of the University of Vermont seeks applications for a tenure- track Assistant Professor position in Systematics and Evolutionary Biology of arthropods, especially insects. The position will be open in the fall of 2012. The successful candidate will have expertise in classical and molecular systematics, including analysis of complex data sets. Candidates pursuing phylogenomics and innovative methods in bioinformatics in combination with taxonomy are especially encouraged to apply. Department information at: http://www.uvm.edu/~biology/.

All applicants are expected to: 1) hold a Ph.D. degree in relevant disciplines and have two or more years of postdoctoral experience; 2) develop a competitively funded research program; 3) teach undergraduate courses (chosen from among general biology, evolution, systematic entomology, and others in the candidate's expertise); 4) teach, mentor and advise undergraduate and graduate students; and 5) oversee a natural history collection of historic significance.

Candidates must apply online: http://www.uvmjobs.com/. On left see "Search Postings" then find "Biology" under "HCM Department" then posting 0040090 (first posting). Sorry, but we cannot supply the direct link because it will time out.

Attach a cover letter with a statement of research focus and teaching interests (one document), a curriculum vitae, representative publications, and the contact information of three references.

Review of applications will begin on September 15, 2011, and will continue until the position is filled. Questions and up to three additional publications may be directed to Dr. Jos. J. Schall: jschall@uvm.edu.

The University of Vermont recently identified several "Spires of Excellence" in which it will strategically focus institutional investments and growth over the next several years. One spire associated with the position is Complex Systems. Candidates whose research interests align with this spire are especially encouraged to apply http://www.uvm.edu/~tri/.
The University seeks faculty who can contribute to the diversity and excellence of the academic community through their research, teaching, and/or service. Applicants are requested to include in their cover letter information about how they will further this goal. The University of Vermont is an Affirmative Action/Equal Opportunity employer. The Department is committed to increasing faculty diversity and welcomes applications from women, underrepresented ethnic, racial and cultural groups, and from people with disabilities.

Human population dynamics

2011-08-22T10:43:00.000-07:00

One of my least favorite questions I get asked as an ecologist is: "Why does this matter?". I won't let myself get sucked into too much of a diatribe, but I always turn into a curmudgeonly old scientist when I hear this. The unspoken assumption is that only things that apply human welfare "matter". So it was with great interest that I read Alan Berryman and Maurcio Lima's newest paper in Oikos: "Positive and negative feedbacks in human population dynamics: future equilibrium or collapse?" They essentially make the point that we can apply the tools of traditional population dynamic analysis to human population data to look for different feedbacks both positive and negative.

In this paper they rely on the R function which is (N is the population size):

It defines the population rate of change over some census period d which is usually 5 years. If you've ever read any other papers by these guys, or any of Berryman's books, this will be very familiar. They then go through and identify different periods of positive growth (cooperation) and negative growth (density dependence) with population size and try and relate those to different periods of human history. They also look at the rates of change for the rates of change to see how rapid changes in growth rate happened (a 2nd derivative of sorts). They then take a special interest the last 45 years for a variety of different specific populations such as sub-Saharan Africa, Japan, China, India, the US and fit generalize additive models (GAM) to them. Finally they use partial least squares (PLS) regression to fit composite variables to growth rates to see what correlates with recent rates of change, things such as fresh water, arable land, GDP, food production etc.

What's particularly interesting is that they see such a strong signal above. These are simple plots of growth rate vs population size and for global population size you can see a strong density dependence and a strong cooperation pattern in up until the middle ages and then again from the Enlightenment through the industrial revolution. Apparently there wasn't enough data from the late middle ages to fit a model just to that period but it does show an expected negative trend they associate with the little ice age and the black plague. They argue that the 2nd positive feedback (1700-1960) is sort of an Allee effect where cooperation leads to a positive relationship between population size and growth rate. The GAM results show that sub-Saharan Africa has a hump shaped relationship with population size showing a shift from cooperative behavior to negative density dependence over the past 45 years but is negative for all other populations. Their PLS results show that economic factors tend to be negatively correlated with population growth rates, but positively with measures of resources. This all begs two main questions. Does technology facilitate population growth (the enlightenment through industrial revolution) and why do we see a decline now and is that decline a sign of a stable equilibrium point? If you've got any experience with population dynamic models, you'll know that a single time lag model of negative density dependence will lead to a single non-trivial equilibrium point. They don't really answer them but end ambiguously, saying that the recent logistic growth that they observe does not necessitate that we are reaching an equilibrium as a species. Instead they note that as we deplete resources we could observe a 2nd order or higher feedback that could cause the human population to crash.

I have to say I was surprised to see we are in a period of density dependent growth. I doubt this paper will get any popular press, but I found it interesting example of ecological methods applied to human populations. I thought the GAM portion didn't elucidate much but otherwise a very interesting read. Its not the only time this has been done, you can see some work done by Peter Turchin on the same subject. Perhaps one of the greatest problems with tackling this topic is the question of scale. If I'm interested in the dynamics of mice in Sonoran desert, I can make certain "global" assumptions about rainfall, resources, etc...that make up the entire habitat of my study population. But human populations are all over the globe and you could easily have a disaster in one part of the world that doesn't necessarily show a signal in this level analysis. A tsunami in Japan doesn't necessarily impact population growth in the US. Although thats not always the case. Another component for humans is the globalization of energy and food markets. As we become a more interconnected human population, a disaster in some place like the middle east could truly disrupt population growth in the US through increases in food prices. A final thought is that ecological models assume that all individuals want to maximize their reproduction and don't because of some resource limitation. Human beings on the other hand have complex psychological/sociological reasons for not having children, so a couple may have the resources to reproduce but opt not to. In part Berryman and Lima get at this with their PLS analysis, but it is an underlying assumption of their model that can be violated by people.

Urban evolution in the mainstream

2011-08-15T08:31:00.001-07:00

Urban ecology and evolution has always been one of those things that fascinates me but that I've never really had much experience with. As a native of rural Vermont, my exposure to nature has always been what you would call "traditional New England". Boyhood swims in the creek flipping over rocks to look for salamanders, hikes through the woods where my dad taught me all the different tree species (This was important since we heated our house with wood we logged ourselves so I would need to know this to understand what trees to cut down because they provided the most heat). A nice article has appeared in the New York Times science section on evolution in urban environments. The article cover lots of ground. It discusses work by Jason Munshi-South on population structuring on white-footed mice caused by population isolation due to urbanization. This is at neutral loci, so of course its an open question as to how adapted these populations have become to urban life.

In the next study the article takes on is one of my favorites. It basically finds the genetic basis and mechanism for PCB resistance in Hudson river Tomcod. This work by Isaac Wirgin at NYU and collaborators is really fascinating. The Hudson river had PCB's dumped into it from 1947 - 1976. Wirgin screened fish populations all up and down the east coast, up to the maritime provences of Canada. He used a gene that is well known to cause the toxicity of PCB's in animals, AHR. Essentially this gene regulates the methylation of xenobiotic enzymes, and these are toxic themselves. The researchers found two variants of this gene, one that is normal in non-polluted sites north of the Hudson, and one that is 100x less sensitive to PCB's around the Hudson. They also look at haplotype diversity and find no evidence of a genetic bottleneck and also see reproductive isolation between the southern (Hudson river and south) and the northern estruarine populations. This evolution must have been quite rapid, less than 60 years obviously. They end the article saying that selection most likely acted via embryonic mortality and deformations and that there is most likely a cost to this decreased sensitivity. The article continues to touch on other topics like the evolution of bacterial resistance in urban hospitals, ant communities on road medians in Brooklyn and invasive species such as Oriental bittersweet. I think there are two corollaries to this article.

The first is that its great to see the idea that humans are part of nature, not different from nature. I think too often the general public draws this distinction between nature (i.e. forest, jungle, ocean) and us (cities, farms, etc..). So an article pointing out that cities are nature, just a new kind of nature (and by nature here I'm referring mostly to the abiotic environment), is great I think. How are road medians any different than small islands? If you're an ant they aren't that different. So I've enjoyed seeing that. The second is a point that I'm sure was hit home hard at ESA (I wish I could have made it) this year, and has been on the rise in the ecological literature of late; that is the intersection of ecology and evolution. Imagine an ecologist who hypothesizes that tomcod abundance will be lower in impacted environments vs pristine environments, and uses the Hudson river estruary as an impacted site. He/She goes out and takes abundance measurements and concludes there is no difference (this is obviously a very simple design for the purposes of our thought experiment), toxins are A-OK. This inference we know is false because of the previous study, and our current experiment is predicated that ecological and evolutionary time scales are different. But our experiment is not robust to this assumption. So what are we to do as ecologists? That's something I don't have an easy answer to. Being cognizant of it is start though. Perhaps something to tackle in some future posts.

Azimuth blog

2011-08-03T08:27:00.000-07:00

Here's a blog that I came across while looking up some things on network theory, its called Azimuth written by John Baez a mathematician over at UC Riverside (but currently in Singapore). While its often very math heavy posts that are outside of my depth, I've really enjoyed reading some of his posts on climate change change like this one on the ice ages and other biological topics like phylogenetics . So its definitely a blog worth checking out.

Optimal ecologist strategy

2011-08-02T12:02:00.000-07:00

So I'd like to pick up with a thought from an earlier post. The past 2 days I've had the pleasure of working down at the University of Florida with my boss Leticia Aviles and our collaborator Jose Miguel Ponciano. Working with Jose Miguel made me think about a question posed by Caroline Tucker and Marc Cadotte over at the EEB and Flow blog: what is the optimal strategy to be an ecologist? The underlying dichotomy posed by Likens and Lindenmayer is between empirical ecologists and the 3M's of modeling, meta-analysis and data mining. My original thinking was an optimal strategy is what Tucker and Cadotte refer as the "glib answer"; to become less specialized and more generalized. They suggest that instead ecologists continue to specialize and focus on collaboration between specialized experts. While I am actually part of such a collaboration at the moment, I still find their answer unsatisfactory, and here's why.

To begin with I'll abandon the notion that the divide between the field ecologist and a 3M ecologist is one of empiricism vs theory. I agree with one astute commenter on the original EEB and Flow blog post who noted that a 3M person is more often someone undertaking a statistical exercise than a true theoretical one. So now lets assume the division is more between the field ecologist and what I think of as a statistical/mathematical gymnast. As modern ecologists though I believe we need to gain a larger and larger mathematical and statistical skill set and still be good field/ empirical people. This is because we don't live in what I think of as an "ANOVA world" anymore (thats not a dig about ANOVA in any way though, I just think of it as a very common and typical statistical test) . The analytical tools available to us now are far more advanced than say 20 years ago (I'm imagining this in part because I was only 12 20 years ago). Its no longer sufficient to take some data and plug it into JMP. Let me illustrate with an example that is both trivial and complicated. Imagine a hierarchical model normal model with j groups and i observations.

I can easily fit this simple hierarchical model in JAGS (or WinBUGS for windows users) with the following BUGS code:


model {
#fit the data model
 for (i in 1:N) {
  y[i] ~ dnorm(y.hat[i], tau)
  y.hat[i] <- mu[groups[i]] 
  }
#model the hierarchical mean and variance

 for(i in 1:j){
  mu[i] ~ dnorm(mu.a,tau.a)
  }
#Set priors for data model
 tau <- pow(sigma, -2)
 sigma ~ dunif(0, 100)
#Set priors for hierarchical model
    tau.a ~ dgamma(.1,.001)
 sigma.a <- 1/sqrt(tau.a)
 mu.a ~ dnorm(0, .0001)
}

Next we can call this model from R using the package rjags, and then analyze it with coda.



library('rjags')
library('coda')
#Number of groups J
J <- 6
#Number of observations per group K
K <- 5
#Total number of observatinos
N <- J*K
#Vector of hierarchical means
mus <- rnorm(J, 50, 25)
#generate a data vector
my.data <- vector()
my.groups <- vector()
for(i in 1:J){
  my.data <- c(my.data,rnorm(K,mus[i],5))
  my.groups <- c(my.groups,rep(i,K))
}
hierEx.mod <- jags.model('simpmean.bug',data=list("y"=my.data,"groups"=my.groups,"N"=N,"j"=J),n.chains=3)
update(hierEx.mod,10000)
hierExsamp <-coda.samples(hierEx.mod,c('mu.a','sigma.a','mu','sigma'),n.iter=1000,thin=10)
summary(hierExsamp)
plot(hierExsamp)

This output gives:

Parameter	Bayesian Estimate	Simulated Value	True value
theta[1]	63.5 ( 59.5 , 66.9 )	63.3	60.1
theta[2]	73.8 ( 70.2 , 78 )	73.8	71.4
theta[3]	104 ( 100.3 , 107.6 )	104.5	104.1
theta[4]	61.8 ( 57.7 , 65.7 )	62	58.4
theta[5]	64.4 ( 60.5 , 68.5 )	64.3	62.9
theta[6]	4.9 ( 0.9 , 8.9 )	4.7	8.3
sigma_y	4.3 ( 3.3 , 5.9 )	3.9	5
mu	59.3 ( 23.8 , 88 )	60.8	50
sigma_theta	33.6 ( 18.5 , 79.7 )	30.4	25

It should be obvious that the hierarchical estimates (with credible intervals) are very close to both the true values and the actual expected values calculated from the simulated data. This is of course a very trivial problem that's so simple its really only useful for illustrative purposes. But this whole techniques relies on a relatively new and non-trivial set of tools such as R, BUGS, as well as all the advancements in the math behind Bayesian statistics and MCMC algorithms. If you're an ecologist who wants to take advantage of these tools, you'd need a working knowledge of probability theory, as well as programming skills in two scripting languages, R and BUGS. Can you analyze data with traditional methods and canned software? Sure, and I would never say that an elegant experiment is ever diminished by simple statistics, and advanced statistics and modeling don't make a bad experiment good. But that said, good experiments analyzed with canned software are limited by the assumptions of the procedures that are available. I think in 2011 ecologists need to be more general because of the advances in computational methods. Skills like programming are needed for stochastic simulation, and a good understanding of calculus and probability theory are needed to make use of state-space and other hierarchical models. I would argue that these skills are necessary to enable collaborations with other specialists.

I had a great example of that these past few days working with Jose Miguel. He helped us create very complicated likelihood functions for our spider metapopulation data with math that was way beyond me, but its up to me to take these likelihoods and fit them using MCMC methods. This collaboration seems like the sort of thing that Tucker and Cadotte were suggesting is fruitful among specialists, but I would argue I'm far more of a generalist and that collaboration works better because of that generalism. I think its that generalism that will allow for great collaboration, so while I might not be able to write my own genetic algorithms, I do know enough about them and programming that I think I could collaborate with a computer scientist on an ecological problem using them. About two years ago I observed the inverse as a part of a working group using modern computation methods (ANN's, GA's, classifier algorithms etc..) on a long term plankton data set. The problem was that the group was headed by a senior ecologist who was very steeped in empirical methods and natural history, and otherwise consisted of engineers. The knowledge gap was so big that despite the best intentions not much really came from it.

So what is the best approach for a younger ecologist looking to have a successful career? I would argue its still being a generalist where you are always trying keep up with the statistical and computational advances in the field (although I'm not successful yet so maybe this isn't the best idea). Working with Jose Miguel brought the limitations of my own generalism into sharp relief, but also made me realize that as long as generalists don't ever think they can do it all, there is a strength in being to work with people from a variety other disciplines.

A shout back to Jabberwocky Ecology

2011-08-02T11:53:00.000-07:00

Before I start my main post I'd like to thank Ethan White over at Jabberwocky Ecology for the shout out and inspiration to blog about ecology in general. Ethan's blog was the first one I really read, and gave me a sense of the potential of blogs.

The buzz about the 3M's

2011-07-28T08:30:00.001-07:00

There has been quite an uproar in the last few days ove a piece by Gene Likens (of Hubbard Brook fame) and David Lindenmayeron the empirical divide in ecology. In short the piece says that we live (Ecologist's that is) in a world that incentivizes rapid publication since publication number is the metric that many funding agencies and hiring committees use. To rise to the top of an ever competitive field, ecologists bump their pub numbers using meta-analysis, modeling and data-mining (hence the 3M's). They believe this is leading to a decline in natural history and taxonomy and a preference for those kinds of papers. They cite evidence of this in lower citation rates of empirical research compared to 3M papers. The future they foresee is one where those 3M papers, which rely on data, eventually are not possible because we've swung so far to the modeling side of things we have no data left to use. A well thought out response to this is from Caroline Tucker and Marc Cadotte over on the EEB and Flow. They observe that many papers in leading journals are still based on primary empirical data collected by the researchers, and that hiring committees often expect candidates to have a field program and they can't get by just on modeling papers and know how (based on MWC's personal experience). They also point out that bridging the divide between field ecologists and data modelers (not truly theoretical ecologists as a commenter noted) can lead to great papers and of course we need both. The question they ask is what is the most optimal strategy for an ecologist to take? And so I think here we come to the "great divide".

I have to agree with the EEB and Flow post, I don't really see a divide philosophically, I see an unfortunate system no one is happy with but we are all part of. My own "optimal strategy" has to become a generalist despite: "Given this double-edged sword, what is the optimal strategy? The glib, easy answer is that ecologists need to become less specialized, to do both theory and empirical work, if they want a successful career" I work in the field, but I try to have experiments that free me to do modeling work. I always try and help friends with analysis and modeling problems and its sometimes resulted in joint publications with me as the 3M person. That said, I've also been on the other end where I'm the primary author working on a modeling paper and asked other friends to be authors because of their in-depth natural history knowledge. So if I think its good to be both, that being both results in more collaborations, whats the problem? Young ecologists (myself, and probably anyone without tenure) are fully caught up in the race to get funding, get jobs etc...Older ecologists on the other hand (Likens, etc, or "luminaries" and Ethan White calls them) really don't have such worries and it is easy to stand back and criticize (albeit subtly) 3M people. In fact Ethan, whom is a very bright and personable guy that I met at ESA last year, seems to have taken the greatest issue with this. Ethan is spearheading a project, the ecological data wiki, which I think an undertone of the Lindenmayer and Likens paper is to criticize this and say its favoring 3M people and letting the parasitize empricists. I think this project is great, and I really support open science and here's where I fundamentally disagree with the original article. As long as a system is set-up to allow coauthorship of papers and we have a culture of collaboration and openess I think an open data system is great. The system will rapidly break down though the moment someone graciously posts data and then suddenly sees someone else with a paper from it. I think the ecological data wiki is the future, and I hope it works.

On a final note, whats perhaps the most fascinating thing about all of this is how its played out on blogs. All these great blogs have had a flurry of action responding and unifying the online ecological community. I would't have even heard of this because I never read the ESA bulletin, but I do read blogs. I think ecology blogs are truly a new and important way to carry out discourse that opens the doors to many more voices than just select publication by famous scientists in bulletins. This great community and its response has been whats most impressive to me.

Rising like a...

2011-07-28T08:24:00.004-07:00

Well, I've realized that I need to up the quality of this blog for it to be something I'm really proud of, so I've ditched my current theme for a new one. I was using this as a forum solely to explore current ecological literature. That worked for a bit but fell to the wayside. So I've decided to make it more general. I chose distributed ecology as a title because I like the idea of ecological thought like distributed computing. Lots of us scientists like little nodes around the web thinking and processing ideas into something great. So to that end I've come up with the apt but uninspired title "Distributed ecology". I might change it if I come up with something more pithy, or just leave it if I become famous in the ecology blog world. If I were a betting man, I'd bet on the former though :).

A useful paradigm obfuscated.

2011-07-18T15:13:00.000-07:00

I spent most of today working on the first 3 chapters of Andrew Royle and Bob Dorazio's book Hierarchical Modeling and Inference in Ecology: The Analysis of Data from Populations, Metapopulations and Communities. I have a bit of a love/hate relationship with this book. I guess I should start with the love part. I like the overall modeling philosophy they present, and I have to say I think they do a good job of it. It comes across philosophically more lucid than other books I've read (Jim Clark's book for example). Conceptually I like middle ground of a process model and an observational model that is all fit simultaneously. The problem though is that the book can't decide who its audience is. I program quite a bit and it provides nice bits of code that are easy to follow. Part of what I was working on today was transitioning from WinBUGS and R to R and JAGS, an opensource Gibbs sampler with BUGS like syntax. What really blew me away was the mathematical part of the book. At times it was very straight forward at others I felt like I needed at least a masters in statistics to understand what they were doing (I don't have that). Often times as I was reading it, they would make statements like: "The trinomial distribution can be rearranged to yield the following joint likelihood". As a reader suddenly you're thinking "whoa whoa whoa what just happened there? I have no idea how they got to that point".

I understand if you're a statistician, perhaps you're used to manipulating likelihoods, but I am not. So I think what they intended chapter 2 to be, but its way too basic I think and then suddenly ramps up to more complicated Bayesian statistics. This is the heart of the problem with the book. At times the book is a simple practical guide to hierarchical modeling for the average ecologist, but out of nowhere jumps to being a technical book for statisticians. Gelman and Hill in their book Data Analysis Using Regression and Multilevel/Hierarchical Models present a mathematically sparse but incredibly useful compendium for how to create mulitlevel models in WinBUGS. Compared to this the Royle and Dorazio book is far more complicated. I wish they had included more step-by-step math. How do I build the likelihood functions, how do I manipulate them? I think another edition with a bit more of that step-by-step math help would be very useful and give the book a broader appeal.

Hurray for being a doctor

2011-06-13T13:21:00.001-07:00

While I feel very foolish that I might be called "Dr. Hart" by anyone, I think its convenient when you want to strengthen a point you're trying to make. To that end I just got a letter published in our local weekly paper in part I'm guessing by the fact that I signed it "Dr. Edmund Hart". Here is the story about a suspicious form of chiropractic treatment called "NRT". Below is my response to it. It had to be less than 250 words, so I kept it brief.

"I find it hard to believe that you would publish an unquestioningly positive piece about quackery. The only quote you have regarding the validity of NRT is: “but just because it’s not scientifically proven doesn’t mean that it doesn’t work.” Actually that’s precisely what it means. As a scientist it’s infuriating to see “reporting” on a contentious practice without thinking about the scientific method. Hypotheses begin with an assumption of a null model, i.e. that the differences between groups are no greater than can be attributed to randomness. So when a study uses statistical methods to detect an effect of a medical treatment and doesn’t falsify the null hypothesis, it means the method doesn’t work. Does that mean that people will never see results personally? No, only that the probability of seeing a result is no more likely compared to receiving a placebo. A search of the scientific literature finds many papers showing no support for NRT.

Furthermore you seem surprised that she successfully discovers you eat too much sugar and drink tap water. It’s like a medium in a crowded room saying: “I’m hearing something from the beyond from a John.” It works because there is a high probability that someone present has a relative named John, it is a common name. This kind of soft reporting is dangerous because it at best sends people to fraudulent treatments who will lose money, and at worst sends people with real problems to someone who isn’t a real doctor.

Lazarus

2011-06-13T12:44:00.000-07:00

Like so many blogs, this one started out with a bang and then ended with a whimper, and then sputtered along, with various attempts at resurrection. I have littered the information highway with many blog corpses, but this is one I would like to keep alive. You might say..."But Ted, you're presenting an obscure message (Ecology) in a medium past its prime (the blog, twitter's ugly and older sibling), who cares." One reason is that I believe that twitter is useful for somethings, it is not really useful to explicate science. At best it will point you towards longer form science articles, in science I think the blog is still king. Regarding the obscure topic, well that is my plight / joy in life so not much can be done to change that now. But now that I've said my spell to revive the blog, on with the show!

I wanted to stick with my theme of writing about papers, so I will keep going with that theme, but I'd also like to be a bit more expansive in my thinking about ecology. To that end I hope this post meets both goals. The paper I'm reviewing today isBUGS in the Analysis of Biodiversity Experiments: Species Richness and Composition Are of Similar Importance for Grassland Productivity. I came across this as I was writing another paper about the use of Bayesian ANOVA's. I'll give a brief overview. Essentially the authors discuss that biodiversity function experiments that have some metric of function such as NPP, biomass, etc, and try and see the link between that metric and something like species richness, diversity, composition etc, tend to use 1 of 2 approaches. Those are traditional mixed ANOVA's and GLM mixed models. Species composition is usually a random effect whereas richness is a fixed effect. Depending on how you treat these effects (mixed or random), you can get different answers. Secondly, there is disagreement about the importance of species composition depending on if it is an ANOVA or GLM model.

To the rescue, the multilevel ANOVA championed by Andrew Gelman. It seems taken for granted by the authors that this is the best approach (I agree though). It turns out that this approach sidesteps the whole mixed / random effects debate / difficulty in interpretation by treating all effects as random. It resolves the conflict of the previous two methods by agreeing with the GLM approach that species composition is important. Now I think Bayesian ANOVA's are great. I published a paper using them last year. I think this paper is great because its one more demonstration of the flexibility and usefulness of Bayesian ANOVA's, and worth a read.

The mundane and its importance

2011-05-20T14:13:00.000-07:00

So I just spent the past 6.5 hours programming a macro for a friend in Excel. She was tasked by her advisor to take copious amounts of abundance data and convert it to presence absence data at different distances from a focal point. The problem was two fold. 1). The distance from the focal point changed with every data sheet and 2). The sheets were entered by people using no standard template. Instead they were haphazardly entered following data sheets exactly, but those sheets changed year to year. One thing I learned some time ago was that good field data sheets make awful electronic datasheet formats. Really given how much data they had it was clear what they really needed was an Access database. I spent my time helping a friend because those few hours of my time probably saved her 30 - 50 hours of had formatting in excel. But I was amazed that a senior researcher had let their research program turn to such a mess. Its a good lesson that I think is worth it for all researchers.

I think as academics we are sort of one person shows. We have to program our analyses, manage our own data, run the whole shop. If you're not good at those things, you can waste vast amounts of time. This lab had spent countless hours and who knows how much money paying grad students to waste their time curating data by hand into a format that then had to be reformatted again by hand. I'm hardly a proficient programmer in VBA, but even a brute force attempt can save weeks worth of labor. By being organized with your data you can be far more productive over time.

The end of field work,

2010-08-30T10:53:00.000-07:00

I've just concluded 17 weeks of field work, and it has been quite a haul. Its a particularly monumental conclusion because its the end of 3 years of work, and my last trip out there will be in a few weeks to destroy the sites. But now that I'm no longer in the field, I'm excited to get back to writing, and move those several manuscripts in the pipeline out into the wider world. My other focus is landing a post-doc. I found that in the middle of my PhD I had a sort of tunnel vision because I was focused exclusively on my project and all things that pertained to it. Whats really exciting (read also as scary) about a post-doc is that I can change directions, and move to explore new aspects of ecology. My main interests as I look to the future are in no particular order: metacommunities / spatial coexistence mechanisms, modeling and computational methods, macroecology, food-webs / network theory applied to ecology.

While I know I lack expertise in some of these subjects, they still broadly interest me and I think what unifies them is that they all can provide insight into community ecology using modern computational methods.Thats what gets to the heart of what I'm interested in all of these different areas. I want to use advanced computational methods (agent based models, network theory etc...) to make generalized predictions about ecology and then test those predictions with field data or experiments. I think that modern computational methods provide a way to take large complex data sets and problems and synthesize them in ways that simpler analytical expressions can't do. Things like agent based models allow us to take qualitative descriptions and theories and then simulate quantifiable output and test parameter sensitivity and assumptions. I think that what I'm truly interested in using this paradigm (which is the rather simplified flow chart image) to answer a wide variety of ecological questions that pique my interest.

Phylogenetic diversity metrics for ecological communities: integrating species richness, abundance and evolutionary history

2010-04-19T08:14:00.000-07:00

Ecology Letters, (2010) 13: 96–105

So this is a bit outside of my usual interests, but I think its important for me to be familiar with lots of different aspects of ecology. I saw a few talk about Phylogenetic diversity (PD) last year at ESA, but I didn't know much about it so I only really registered that these indices were out there and not much else. In this manuscript, Cadotte et al develop a series of three indices that integrate abundance information and PD indices. They then test out their indices with an idealized community with three different abundance distributions across the same phylogenetic structure. Finally they compare their index and a variety of other PD indices with a series of plant communities.

Their three indices are phylogenetic abundance eveness (PAE), imbalance of abundances at the clade level (IAC), and abundance weighted evolutionary distance (AED). PAE and IAC were the easiest to understand I found. PAE is simply how are branch lengths and abundance related. If the index is less than one you have high abundances at short branch lengths, and if its greater than 1 you have high abundances at long branches. If its equal to one the community is perfectly even. IAC creates a null distribution of abundance splits at each bifurcation in the tree. So if you have an N of 100 individuals, at the first split you expect it to be 50/ 50 and then at each further on down clade split it would be 25/25. The idea here is that if a particular clade does well in an environment you would expect that abundances would be skewed towards that clade. The AED index is a bit more complicated, but it is an entropy style index similar to the Shannon index. It works similarly to the Shannon index, but instead of just species abundances, it includes phylogenetic information.

These indices all fit within an existing PD index framework which they compare their own indices to with a dendrogram. I'm not really familiar with them at all so I can't really comment. There are of course some obvious caveats to all of this. To begin with this all assumes you have well resolved phylogenies for all your taxa, although the authors note you could use simple family level phylogenies. This makes sense because once you get below that level polytomies can become a problem. Another problem is that where do you get these phylogenies. Most phylogenies I've seen are for taxa families, so if you have a community with bears, bees, and flowers, how do these indices work? I'm sure thats a question thats been looked at in the literature, but I've not read enough about the topic to know. I found reading this paper to be very informative in terms of helping me learn more about PD, even though it doesn't much relate to my own work.

Application of an evolutionary algorithm to the inverse parameter estimation of an individual-based model

2010-04-14T10:42:00.000-07:00

Ecological Modelling 221 (2010) 840–849

So agent based models has been something I've really wanted to move into, but because I've decided that to make that shift would basically be another dissertation its really moved to the back burner. I did make a foray into last year though and gave a talk on it at ESA, and one of the real challenges, something that took weeks and weeks was parameter calibration. In this paper, the authors address that problem. They take a well established IBM and apply an evolutionary algorithm (EA)to estimate parameters. I have to say that this is a really neat approach. I won't go into too many details because much of the paper is devoted to the guts of their EA that I don't need to go into. In brief what they try and do is use an EA to parameteritize a fish IBM, where the output is biomass of 12 species after 200 years. In the manuscript they compare biomass outputs for manually calibrated parameter values, EA estimated values and expected values.
As you can see, they find very accurate estimates with the EA algorithm, and in some cases they even get better results than with the hand calibrated method. They also note that the EA solutions took a week, and the manual calibration took 6 weeks. At the same time they don't mention how long the EA took to write or if its generalizable to any IBM they program. But if it were, the EA would be a great time saver in terms of IBM optimization. I think this is a really great direction to go in for IBM's especially serious IBM's where you need information about parameter uncertainty, and some of their output can even be used as sensitivity analysis. A pretty cool study overall, and I think maybe something to consider for my post-doc.

Patterns and causes of species richness: a general simulation model for macroecology

2010-04-08T11:06:00.000-07:00

Ecology Letters, (2009) 12: 873–886

So its been a bit since I've written anything, I was in Florida last week and I haven't found much recently thats caught my attention since I've returned, so I've decided to go a bit deeper beyond literature from 2010. As I begin to seriously apply myself to my post-doc search, I am beginning to try and read literature by potential post-doc advisors. In this case it's Carsten Rahbek from theCenter for Macroecology, Evolution and Climate at the University of Copenhagen. He collaborates pretty heavily with Nick, so I hope that will at least lend me a bit of credence. This is a good paper to start with because it gives a perspective on new directions for the field of macroecology.

It begins by addressing what constitutes macroecological data and some of the short comings of current macroecological curve fitting methods. This just seems like a fancy word for regression. Essentially the data is a set of grid cells with species richness counts and another set of grid cells with environmental data, log transformed and then *presto*, regression! Now some of the serious short comings of this approach is that there are all sorts of autocorrelation problems with environmental data and other geometric quantities such as latitude. So what's a macorecologist to do? The approach outlined by the NCEAS group, led by Nick, is to develop a general simulation model (GSM). This is essentially an agent-based or cellular automata model with a variety of "knobs". The authors outline 3 potential knobs: dispersal kernel, speciation model, and environmental gradient. They posit at its simplest state you can have simple off-on switches for null / mechanistic models for each of these processes. They then outline how each of the unique 8 states of each model correspond (or don't) to a model in the literature. They finally outline a criteria for how to assess model fit using a modified mean square error and a minimization of Kluback-Leibler distance.

I love this sort of proposal because I'm a big fan of agent based model in ecology and certainly want to move in that direction for my post-doc. I believe that these sorts of models built from the ground up are easy to control, have transparent assumptions and allow for "experiments" where empirical work cannot logistically be conducted (like in macroecology). The problem is that with model such as the idea that you could toggle things such as environmental gradients, "Environmental gradients: 0 = colonization and ⁄ or range expansion into all grid cells equiprobable, 1 = probabilistic colonization and ⁄ or
range expansion into grid cells as a function of measured environmental variables", seems a bit simplistic. While I understand the 0 state, the 1 state could take on a variety of forms, i.e. the kind of "function of measured environmental variables". This remains one of the challenges of simulation modeling, that the possibilities are almost endless. Other than that I enjoyed this paper, and papers like these are always good to read because they point out the various holes in a field that need work.

Habitat isolation moderates the strength of top-down control in experimental pond food webs

2010-03-26T14:21:00.000-07:00

Ecology, 91(3), 2010, pp. 637–643

Ahhh, what a pleasure it is to read a paper in your field. I can just cruise through it, I know most of the references already, a nice change from my past few manuscripts where I was going outside of my knowledge base. This manuscript by Chase et al. in the most recent issue of Ecology was of particular interest to me because my dissertation is based on work in mesocosms. The authors are examining the relationship between isolation and food web structure on species richness in aquatic mesocosms (1000L tanks). Basically they have a two factor design; focal ponds are either isolated or connected (three other close ponds) and either have ambient levels of predator taxa, or they are artificially reduced. They then use MANOVA and ANOVA analysis to see the effect of the treatments on a variety of response variables: herbivore species richness (SR), predator SR, predator biomass, and 3 others. The take home message is that there's almost always higher richness and more biomass in connected versus isolated habitats. You might say "So what?". Well the other thing they have is the reduced predators, so you see less herbivore biomass in the high predation treatments, but a higher species richness. This is used as evidence of a classic indirect effect. The other aspect of the paper is they break down herbivores and predators by functional group, and you see highly differentiated responses between connected and isolated habitats.

The authors then posit two mechanisms for the higher diversity of herbivores in the connected / high predation treatment. Either indirect effects of a keystone predator like dragon fly larva, or intraguild predation causing an overall reduction in the actual predator pressure on herbivores. They then surmise that its the former because you see a "trophic cascade" in the removal treatments. Overall, I have to say I love the elegance of the experiment. Its clear, straight forward, and easily digestible. But I feel like there could be so much more to how they carried out the analysis. It reads like a quaint reenactment of Ecology papers of the 90's. Elegant design, simple statistical analysis, and large effect sizes. At the same time, we know so much more about space and community structure that a major short coming is that it just washes over metacommunity theory with a simple reference to the Holyoak et al 2005 book on the subject. I'd say I hope they take this further, using experimental mesocosms to explicitly test metacommunity hypotheses...but that's what I want to do :).

The implicit assumption of symmetry and the species abundance distribution

2010-03-25T10:00:00.000-07:00

Ecology Letters, (2008) 11: 93–105

A consequence of trying to read everyday is that you will cover a wide variety of topics. The upshot is unfortunately that you can get lost in many of the details in a topic outside of your area of expertise. Since much of my recent work has been in population dynamics, reading a paper about that subject is pretty straight forward for me. This paper by Alonso et al. is focused on the sampling theory and once he moved beyond a simple example with a multinomial likelihood, I began to get a bit lost. The qualitative aspects of the paper made solid logical sense though. First though, lets just take a step back and cover "What is a species abundance distribution (SAD)?"

At its simplest, it is a way to quantify the biodiversity of a community. Essentially you collect a vector of species, and create a histogram of it. What this creates is a histogram with abundance categories on the x-axis and the number of species in each category on the y-axis. Now usually you don't get exact numbers of the same species, so often times people will use "bins" on the x-axis of ranges of abundance. A good introduction to all of this is McGill et al (Ecology Letters, (2007) 10: 995–1015). The problem has typically been that we can statistically describe these broad community level patterns of abundance, but rarely do we understand the mechanism that creates them. Regardless of this problem (which the authors mention on the outset) another problem is that the sampling theory that underlies these SAD patterns makes two very strong assumptions that are highly unlikely to be true. 1). The abundance distribution for every species within the sampled community is the same, i.e. same birthrates, deathrates, life history, etc... They term these assumptions "observer independent". 2). All the species within the sampled community are equally easy to sight/capture/etc... These are termed "observer dependent" assumptions. Together these assumptions make up what is known as species symmetry. The authors then go on to develop a mechanistic "asymmetric" sampling theory. Here's the part where I get lost, and if you've got more of a background in this sort of thing than I do, I'm sure it will make sense.

What was most interesting about the paper is that much of the conclusion seems to be aimed at rebutting attacks on Neutral theory based of SAD's. They reject some of the criticisms of McGill and say: "Evidently, it is a trivial
truism that the consistency of a single empirical pattern with
a quantitative prediction from a single model is insufficient
to accept that the basic processes generating this pattern in
nature are precisely those assumed by the model (McGill
2006)."

They then go on a long discourse about scientific theory and weights of evidence from multiple studies. While I think its all fine what they write about testing model predictions etc etc in a general framework, I think in the end the almost "political" nature of the paper detracts from it. It would have been great if the discussion focused more on their theory in the paper, and less on the politics of the Niche-Neutrality debate.

March insanity

2010-03-22T14:15:00.000-07:00

So I took a brief break from ecology yesterday to pursue another project, modeling the NCAA bracket. I decided to run a stochastic simulation of the tournament. I know this has nothing to do with ecology, but I believe its a good exercise in modeling. Modeling has very little to do with the specifics of a process or problem and everything to do with the abstract framework you put those details into. Data is data be it from an experiment or the NCAA, and this is an exercise in being given a problem, and using the information and tools you have to solve that problem. Ok, enough pontificating, on with the modeling.

So here's what I did. I used data from the Pomeroy rankings and the Log 5 rule to calculate a binomial probability of one team beating another. The problem with this is that you don't have any data to actually model, its only a predictive probability. What I wanted to do was use a simple Bayesian Beta-Binomial model and integrate prior information from my more NCAA savy friends. I used this model to simulate the NCAA bracket 10,000 times for each friend and with a flat prior. So here's what I did 10,000 times.

First I used the Pomeroy probability to simulate 34 games between any two teams meeting in the tournament. I used that simulated data in the beta binomial model as my number of successes and failures. I then asked my friends to provide me with two probabilities for each team, a probability of them making it to the final four and a probability of a team winning it all. When teams encountered each other in the simulation I calculated the odds of one team beating another and then converted those odds into a probability and then used an algorithm to solve for the parameters of a beta-distribution. I scaled the beta parameters to reflect the confidence my experts had in their picks. In the tables you can see that priors can have a pretty heavy influence on picks. Here are my tables, each number represents the probability that a team makes it to a given round of the tournament.

Flat priors.

Combined priors

Here is the code in R for the model

Neutrality and the Response of Rare Species to Environmental Variance

2010-03-16T08:29:00.000-07:00

Benedetti-Cecchi L, Bertocci I, Vaselli S, Maggi E, Bulleri F (2008) Neutrality and the Response of Rare Species to Environmental Variance. PLoS ONE 3(7)

This is technically yesterday's article, but I didn't have a chance to finish the post on it. I like Benedetti-Cecchi's work and I haven't seen much on neutral theory out there recently, so I decided to read this paper. The premise is simple. They use two different data sets, one an observational and one experimental to test the decay in similarity between rocking intertidal communities against simulated neutral communities. The first observational part uses a 16 year time series to test if the observed decay in similarity is different than expected under ecological drift. The second experimental part uses a classic Benedetti-Cecchi approach of crossing treatments in the timing and intensity of the variance of a disturbance. His supposition is that if neutrality holds, the temporal variance shouldn't matter, only the intensity of the disturbance since this only effects births/deaths and colonization, and not the timing. I find this second supposition somewhat tenuous and the authors do as well, but try and support it by stating: " However, the extent to which neutrality can be tested by examining assemblage responses to environmental stochasticity is unclear [27]. We substantiate our conjecture by showing with simulations that temporal variance of disturbance has no effect on species rank-abundance curves under neutrality, either under constant or variable rates of immigration [28]. Furthermore, we consider that temporal variance of disturbance elicits non-neutral responses only if it modifies species abundance distributions by influencing the probability of occurrence of individual species or groups of species, consistently with the definition of neutrality." This to me seems to be close to a tautology.

The authors then find that their data isn't different from expectations of the neutral model. They make this assessment based on mantel tests of different distance metrics at each time step. They also do some simulations with their experimental data to create a null distribution based on the neutral model to test their experimental data against. Finally they note that rare species are have a density dependent advantage in variable environments because of high growth at lower numbers.

I like much of what they have to say about neutrality, and I think neutrality is more common than many ecologists admit. The part that I'm not such a fan of though is that they test only the neutral model. Many of the early challenges to Neutral theory, such as the one made by Brian McGill in 2003 in Nature focus on comparing neutral with other models such as the Log-normal. I think that the paper would have been stronger with a test of multiple models. Multiple models can match data, so you need a stronger test of model fit than a simple null hypothesis test.

Unifying Life-History Analyses for Inference of Fitness and Population Growth

2010-03-12T11:58:00.000-08:00

Unifying Life-History Analyses for Inference of
Fitness and Population Growth

vol. 172, no. 1 the american naturalist july 2008

I have no paper for yesterday because I was at a conference all day, instead I'm reading a paper that my friend Matt Kaproth sent to me. I must admit I really didn't give this the read that it deserved. In it Ruth Shaw and her colleagues provide several examples of using a new set of models they've called "Aster" models to estimate lifetime fitness. They point out that current models of fitness either ignore the fact that they violate the assumptions of those models (i.e. normality, etc...), or analyze components of fitness individually ignoring the obvious causal dependency between say the juvenile survivorship of a plant and the number of offspring it produces (you can't reproduce if you don't survive). Aster models allow a researcher to integrate these different life history stages that have different sampling distributions (i.e. binomial and Poisson) into a single model. They provide an R package that can implement this new method. Now, I've given a rather crass summary of the paper, but I have to admit my first breeze through this left me wanting to work some examples to understand the method better. Also much of this literature is "theoretical" to me in the sense that I've read about it, but rarely actually practiced the examples. Therefore I think I'll have to have some addendums to this post once I begin to understand Aster better. Until then I think anyone who estimates selection gradients or really does any work with actually measuring selection should check these methods out.

Temporal shift in density dependence among North American breeding duck populations

2010-03-10T07:56:00.000-08:00

Temporal shift in density dependence among North American
breeding duck populations
Ecology, 91(2), 2010, pp. 571–581

Well, I'll begin my second post with an observation that will soon become obvious to anyone that reads this. I'm interested in Bayesian methods, complex systems tools (specifically agent based models), population time series analysis and metacommunities. But I have a real soft spot for time series papers, and hopefully (fingers crossed) I'll have a couple floating out there myself (currently 1 in review, 1 in prep). I'll try to not skew too heavily to just reading papers about time series methods, but this is so close to my own current manuscript, I have to read it.

This paper left me with mixed feelings. On the one hand I read it very carefully (hence why this took me so long) and followed up with skimming some of the references. On the other, I felt like they had lots of time series smoke and mirrors to say not very much in the end. I think if anything this has shown me the importance of clarity in writing and making your point, but I owe the paper a more thorough review than that.

The authors fit three different time series models to time series data from 10 different duck species from a large scale breeding duck survey that took place from 1955 to 2005. They assert that there has been a shift in density dependence that happened in 1980, with in increase in density independence after 1980. They also show that population parameters are not related to "evolutionary strategy" as measured by a statistic called the LHI (life history index). Furthermore they show that population's shifting parameters are unrelated to carrying capacity or mean abundance.

Using the 10 different duck species time series data they fit three different models, a density independent, density dependent and a theta-Ricker. Curiously they also fit a Gompertz model and Gompertz state-space, but don't report the results from those models, except in an appendix, and go to great lengths to make the case that their data are robust to measurement error and that all the error they observe is process error. They find that when they split the data they see the above mentioned shift from density dependent to density independent processes. I think they have a good point, but I'm perplexed by their choice of figures and tables. They leave the most informative table in the appendix, one that shows the shifting delta AIC's and AIC weights for each segment. I think also some sort of figure showing mean abundance and the shift would have been good. Instead I believe the focused too heavily on the diagnostics of the time series and not the data themselves. I don't doubt the authors conclusions, but I'm surprised that the paper was as unclear as it was.

On the application of multilevel modeling in environmental and ecological studies

2010-03-09T13:23:00.000-08:00

On the application of multilevel modeling in environmental
and ecological studies
Ecology, 91(2), 2010, pp. 355–361 (citation is a link to the paper)

I just finished reading this paper by Qian et al from the latest issue of Ecology about the use of multilevel modeling. The paper goes over methods about the uses of multilevel modeling (which can be done with likelihood or Bayesian methods) in ecology with two worked examples. They make a really great point that hierarchical modeling allows you to include information in your model at different scales into a single model. The paper did not offer anything new to me, since much of it is drawn from Gelman and Hill's 2007 book: Data Analysis Using Regression and Multilevel/Hierarchical Models.

I'd already read that book as well as Qian's paper on Bayesian ANOVA's and given a talk on that. If you're unfamiliar with these methods, I think this paper is a good jumping off point, and at least an introduction of the Gelman and Hill book to ecologists. And while they don't do anything earth shattering in the paper, they explicitly state the importance of publishing the method to "normalize" it, and have it be more accepted by ecologists, and I think that I've already had two papers rejected in part because reviewers couldn't understand or see a need for Bayesian methods. Finally, I think one of their great points is this statement, which I must say I agree with: "In the
course of developing the models presented in this paper,
we have not used the usual means for judging whether a
model is ‘‘significant’’ (the F test or similar tests); rather
we qualitatively interpreted all fitted models to see if
they make scientific sense. As our improved datagathering
technology produces larger amounts of data
with increasingly more complicated structure, statistical
inference based on hypothesis testing will be increasingly
inadvisable. We cannot summarize a complicated
relation into a simple dichotomy of null and alternative hypotheses"

This is a fast read and I think a worthwhile one for ecologists to familiarize themselves with methods that will no doubt become standard in the years to come.