The differential biology reader

One of the most disappointing critiques of animal personality research is that the ratings we use are "subjective" (I prefer the term impressionistic) and that we are just not very clear about what it is we are studying.

how is ‘behaviour’ defined, and how should it be defined? We outline what characteristics we believe a scientific definition should have, and why we think it is important that a definition have these traits. We then examine the range of available published definitions for behaviour. Finding no consensus, we present survey responses from 174 members of three behaviour-focused scientific societies as to their understanding of the term. Here again, we find surprisingly widespread disagreement as to what qualifies as behaviour. Respondents contradict themselves, each other and published definitions, indicating that they are using individually variable intuitive, rather than codified, meanings of ‘behaviour’.

We offer a new definition, based largely on survey responses: behaviour is the internally coordinated responses (actions or inactions) of whole living organisms (individuals or groups) to internal and/or external stimuli, excluding responses more easily understood as developmental changes.

While behaviours necessarily rely upon internal information processing by the individual (e.g. cognition and endocrine signalling), we do not consider the processing alone to be a response, and therefore do not include it as behaviour. Shettleworth (1998, page 5) defines cognition as ‘the mechanisms by which animals acquire, process, store and act on information from the environment’. In behaviour, we include the action, but deem the processing as necessary but not sufficient. While this runs counter to the views of 80% of our respondents, we think it is illogical to include cognitive processing while excluding other forms of internal information processing such as genetic expression cascades or endocrine feedback. Information processing may be a necessary substrate for behaviour, but we do not consider it a behaviour by itself.

The curious researcher may find they spend quite a bit of time playing with tools that, while painting potential, do not involve anything that you absolutely must understand by tomorrow. As a student and distracted generalist I am often in this position. In such circumstances the last thing I want are tools that hate me.

I have been thinking about lovable tools in two contexts recently. The first was a conversation with @NicMcPhee and @moorejh about reusable genetic programming tools, particularly for symbolic regression. All too often (and this was my own experience as a former computational synthesist) workers in evolutionary computation produce code that, while giving important and publishable results, is not reusable by anyone. Your technique does me little good if I cannot apply it to my own data.

The second context concerns Bayesian data analysis. My formal education spent about a week on Bayesian inference, but I was never forced to reckon with actually analyzing my own data. Frequentist methods have been serving me fine, so this is not a priority. Still, as something I want to understand, I am forced to play with it.

As for learning by doing, BUGS seems to be the standard form. But several problems are present. The most salient is that BUGS appears to be Windows-only (correct me if I am wrong). The software also seems to be generally cryptic when it comes to error detection. Yes, it is free software as Gelman notes, but if I am going to invest in a proprietary language for model-building, I want something robust (contrast this with Mplus; I tolerate learning its syntax only because the program is exceptionally well-designed).

What is working well for me now is pymc, a Python module for Bayesian model building and checking. I don't know anyone who is using pymc for getting work done, but as it is based on Python, the barrier for starting is very low (the documentation is also friendly). Even if you don't know Python, learning it is not going to hurt you, or be a waste of time (psychologists take note).

Building a tool on top of an already rich environment is crucial to flexible data analysis. I don't know how you get data into BUGS, but I'm guessing you can't make it to talk directly to PostgreSQL.

Genetic programmers, take note. If your technique is so wonderful, make a lovable tool that folks like me can use to analyze their own data.

photo cc-by lingualx

The Map Scroll has the scoop. Richard Florida's Who's Your City has posted new personality maps that now include Canada. 
These maps are an update to those originally drafted for Jason Rentfrow's paper on geographic variability.

An ethogram is a catalogue of discrete behaviors typically employed by a species. Traditionally animal behavior has been recorded by observing study individuals directly. However, this approach is difficult, often impossible, in the case of behaviors which occur in remote areas and/or at great depth or altitude. The recent development of increasingly sophisticated, animal-borne data loggers, has started to overcome this problem.

The typical behaviors extracted were characterized by the periodicities of body acceleration. Each categorized behavior was assumed to correspond to when the bird was on land, in flight, on the sea surface, diving and so on. The behaviors classified by the procedures accorded well with those independently defined from depth profiles. Because our approach is performed by unsupervised computation of the data, it has the potential to detect previously unknown types of behavior and unknown sequences of some behaviors.

   

Click here to download:

Personality_in_featureless_env.zip (1830 KB)

Is behavior just dictated by your situation rather than anything innate (personality or character)? A subset of social psychologists may work as if this is the case, but let us turn our attention to fruit flies and prisoners of the future.

This comes from the work of Björn Brembs, who asks a simple question about the generation of behavior. Is behavior the result of a simple map from sensory input, disrupted only by random noise at each juncture in the system, or there an internal initiator of spontaneous behavior?

Brembs walks you through the experiment showing how flies behave in a completely featureless environment. With constant visual sensory input, they still attempt to turn and roll about in the air.

we detected a non-linear signature in the fly behavior. Such a signature can only be found in systems whose indeterminate behavior is not due to noise but originates in their design.

If you found yourself in such an environment, what would you do (think of the prison in THX 1138). There is no sensory feedback, although you can still attempt to move around. Would you sit down, run, jump. Cry? Squeal (in delight)? Is there any chance of finding robust individual differences in what humans would do in this interesting although ecologically invalid environment?

Maye A, Hsieh C, Sugihara G, Brembs B, 2007 Order in Spontaneous Behavior. PLoS ONE 2(5): e443. doi:10.1371/journal.pone.00004433

Wrong Tomorrow attempts to bring accountability to pundits by tracking and verifying predictions. But this might not be possible in the general case if the experts are working against you:

Forecasting plays a vital role in human activity. Consumers, managers, and politicians make their decisions in part based on their anticipation of future events…The complexities in properly anticipating future events may encourage decision makers to seek experts' advice. The main difficulty is, however, that professional forecasts may not be reliable. If an expert is informed (i.e., he knows the relevant odds), then he can reveal the relevant probabilities to decision makers. However, if an expert is uninformed (i.e., he knows nothing about the relevant odds), then he may mislead the decision makers. A fundamental question is therefore how to determine whether experts are informed.

A test either rejects or does not reject each expert based on the observed data and the profile of the probabilities announced by the experts…However, consider the case in which all experts are uninformed (i.e., they do not know anything about true probabilities). We show that they can still independently produce false forecasts that are likely to both pass the test, no matter how the data evolve in the future. Hence, the data may not suffice to effectively discredit uninformed, but strategic, experts.

Olszewski and Sandroni. Manipulability of comparative tests. Published online before print. doi:10.1073/pnas.0812602106 PNAS 106 (13) 5029-5034

photo cc-by dsevilla

photo cc-by kacascade

Unintentional modernism via the image function.

I have been toying with Clojure, a modern Lisp dialect that runs on the Java Virtual Machine. It has been almost 10 years since I've worked properly with Lisp-style functional programming (in my case, Scheme). As all of my programming ventures these days center around R and Ruby (with the occasional frightening encounters with my first programming language, C), it is refreshing to get back into real functional programming.

I am examining Clojure as a language for individual-based modeling. The big daddy of these methods is Swarm, which is written in Objective-C. As a language, Clojure provides a number of features that are particularly suitable for individual- or agent-based simulation.

The first of these is the iterative nature of REPL (the Read-Evaluate-Print Loop) which makes it possible to interact with and reprogram your system on the fly. You can examine and modify your code while it is running. Compare this with the write-compile-run-debug-compile-run cycle of a language like Objective-C.

Second, functional languages in general are suited to the task of modeling. When building models, your focus should be on small units of input–output that build upon each other.

Yet Clojure offers many advantages over its dialectical ancestors. The first is immutable data structures. This forces you to think about everything as input and output (good for modeling) while making it easier to think and reason about (good for understanding your model) your program. The differentiation that Clojure makes between identity and state is also a good fit for modeling. Clojure is also built for a concurrent world. Clojure can handle a lot of different parts of a program simultaneously reading and writing changes to the world, allowing you to focus on each part of your system as individuals rather than worrying about the mechanics of making them all work together. Clojure also keeps vectors and hash tables as built-in, both convenient data structures for modeling tasks.

There are a number of other features in Clojure that I don't completely understand but look promising, like watchers for exploring and reporting program state.

The primary disadvantage is having to start from scratch. With mature systems like Swarm is that lot of the hard thinking has been done for you. Raw ingredients provide the tastiest environment for bugs. Another problem with Clojure is that its specification is still under development; it is a programming language in flux.

But…the demo program for concurrency in Clojure is an ants simulation. That has to be a good sign for future individual-based modelers.

photo cc-by gwen