Tuesday, July 28, 2015

Generals: summary and advice

Sometime within the first two years of a North American biology PhD, grad students take an exam that determines whether their research ideas hold water and whether they should continue with their PhD or leave. No pressure! This rite of passage is called the generals,  qualifying, or preliminary exam ("generals," "quals," and "prelims"), and it's analogous to a Masters defense in the European system. The specifics of the exam vary greatly between universities but tend to involve a written literature review and thesis proposal, as well as a multiple-hour oral exam by the thesis committee. (Sometimes there's a written exam, too.) The committee, which consists of 3-5 professors who read your proposal, will ask you questions for around three hours and then decide whether you should stay.

Preparing for generals in the month leading up to it was easily the most stress I've experienced in grad school and some of the preceding years. Yet, it didn't have to be. This post is a summary of what my generals was like and some advice based on what I learned. 

What my experience was like
Prelude to the prelims
I opened a Word doc called "draft 1" in Summer 2014, about a year ahead of the exam. This doc just served as a place to paste quotes from articles and write a sentence or two of methods when I didn't have any other work to do. The amount of writing in the 10 to 6 months pre-generals that made it into the final document was tiny, maybe 1-5%, but my mentality was that any bit I could add now would be an hour saved in the frantic last days before submission.

A lot of my thoughts on what exactly I wanted to study changed over the months, especially when I got preliminary data in November. It's one thing to say you're interested in collective antipredator behavior and another to specify that in chapter one, you'll study how alarm propagation in golden shiner schools changes under predation risk, and why this is important. For me, the stress began mounting as I prepared my second-year talk a month before the exam. (2nd-year Princeton EEB PhD students give a 25-minute talk to the department on their research plans.) How silly would it be to stand in front of the colleagues I respect and see every day and to fumble around with half-finished ideas? No, I needed a presentation that reflected how seriously I took my work, how much I had thought through my ideas, and how I wasn't walking down a research dead-end.

(Note: despite losing quite a bit of sleep the night before as I gave the presentation a much-needed redo, the talk went well and I got good feedback. I've used the talk multiple times since then to quickly show research plans to collaborators and colleagues and get discussion started. No need for all of these paragraphs to be somber!)

The last month
The talk symbolized an important transition in the PhD: I was no longer just interested in ideas; I had to be able to convince others that a potential project was not only possible but also worth studying. Over the next month, I read and wrote constantly. I wanted to make sure the ideas followed airtight logic, which meant spending a lot of time crawling through articles to find that one paper that showed that fish at the front of the school tended to be the hungry ones, or was it just the edge and not necessarily the front, and is the species closely enough related to golden shiners to be relevant.... Again, it's one thing to have a general idea of the literature supporting your ideas and another to have the details, and there are a lot of details about your species' life history, the methods researchers in the past have used that you plan to use, and the theoretical and empirical justification for your research avenue. With a week to go before the exam, I submitted a 46-page PDF to my committee and had to let out a laugh at the relief that it was no longer in my hands. (Note: many of those 46 pages were the 180 references!)

Over the next week, I just wanted to take the exam. I felt as ready as I could be, and I filled my time by skimming a behavioral ecology textbook and a few articles to make sure I was familiar with common topics. My exam was scheduled for the last day in the two weeks of generals dates, so I watched five of my cohort members celebrate the passing of their exams and grilled them on the details. All of them said it wasn't as bad as they'd worried, but they were glad to be done.

The exam
As is the unofficial tradition here, I brought some breakfast food for my committee and a short presentation with the major justification and goals for each chapter. (The exam started at 10am and went through lunch.) Before we started, I left the room and my committee discussed my proposal and what they wanted to ask me about. I returned, began my presentation, and was asked immediately why I was interested in heterogeneity, which was in my title. I answered, and then... the details get pretty blurry!

What I do remember is that the committee was very fair throughout the exam. They raised concerns, listened to my justifications, and told me when they disagreed with my reasoning, but they also told me when they thought an idea in my proposal was exciting, and they offered suggestions on how to make the experiments stronger. The mood was formal but almost relaxed, and it felt like a discussion on finding the best way to have an interesting research trajectory that could teach us something. I passed, but my committee asked for me to resubmit the methods section of my proposal, which we agreed needed some work, in a few weeks. I appreciated it, actually, because while I would incorporate their feedback anyway, it was a good additional incentive to synthesize it in one place.

And then... it was over! Really over. After the celebration that Friday, I had an incredibly quiet, anticlimactic weekend where I had to relearn work-life balance. When I took away the work I'd been doing for hours every day the past month, what was left? My favorite hobbies, including drumming to songs I love, had been largely neglected because I was "too busy," but doing nothing but hobbies on my free time felt strange without structure to my days. Once the concrete motivation of a pass-fail exam was removed, it took about 2 weeks to fully reignite  the curiosity and excitement for my project and to find a sustainable work-fun pace, where I wasn't working until 9pm every night, but I was still making progress every week that I was satisfied with.

Advice for PhD students
1) Start writing early 
This one might seem obvious, but not in the way you'd expect. I found it really helpful to have a place to store thoughts as far as a year in advance of the exam. The actual format and details of my project ideas changed up to the days before the proposal was submitted, but it was good to have something down before the pressure started building. This was helpful when I started writing the "actual" proposal because it felt like I was adding to a first draft of ideas instead of starting from scratch.

2) You're smarter than you think. It'll be alright
Writing a proposal for your PhD plans can shake up your confidence. You're constantly looking at holes in your logic that you're trying to fill: how long to wait between trials, is your species really the best for this question or are you just using what the lab studies, are your questions interesting to anyone besides you? These questions didn't seem too important before, but now your proposal feels bland without them. You need to deeply commit to your ideas to understand them well enough to convince others (and yourself) that they're important. There's a lot of self-doubt involved, but you're doing better than you think. There's a lot to learn, but you've also learned a lot in the past year and a half. Keep at it. One of my favorite quotes really applies here: "The reason we struggle with insecurity is that we're constantly comparing our behind-the-scenes to others' highlight reels." Everyone had to figure out the details of their projects, and it's natural to struggle a bit as you learn everything you'll need.

3) View the exam as a discussion
Your committee wants you to do well. Your department is training you to become a knowledgeable scientist, and the reason generals is hard is that there is a lot of nuance to doing good science. Disagreeing with each other is inherent to finding the best possible path to new knowledge. Your committee and you have the same goal, for you to do a PhD that answers interesting questions, and there's no point in being scared that they'll ask you something you don't know the answer to. (Spoiler alert: this is guaranteed to happen.) Take notes, ask them questions. This is probably the only time in your career that you'll have five incredibly intelligent people closely examining your work and giving you feedback for hours. Take advantage of it.

Good luck!
-Matt

Thursday, July 16, 2015

Behind the scenes: Couzin et al. 2011


The story behind Couzin et al. 2011:
“Uninformed individuals promote democratic consensus in animal groups"

Couzin ID, Ioannou CC, Demirel G, Gross T, Torney CJ, Hartnett A, Conradt L, Levin SA, Leonard NE. 2011. Uninformed individuals promote democratic consensus in animal groups. Science. 334: 1578-1580.
 
Conflicts of interest within groups are ubiquitous in the animal kingdom. Thirst will pull some gazelles to the watering hole instead of the grass patch, recent experience with predators will give some fish less incentive to enter open water, and differing knowledge about the geography of a region means migrating birds may disagree on where to roost. The benefits of group living, such as safety in numbers and enhanced environmental sensing, are often too high to afford individuals to split into subgroups: they have to stay together. Groups therefore often have to make consensus decisions, where the outcome affects everyone. How does this happen?

For years, scientists have focused on how the interactions between the group members who want to go to destination A versus destination B determine what the group decides. The role of members who don’t have a preference, however, remained overlooked.

“People tend to think of uninformed individuals as contributing nothing but noise to decision-making, whether in neural systems or animal systems,” said Iain Couzin, Max Planck Institute for Ornithology Director and professor at Princeton University. In the years leading up to 2011, Couzin felt that such group members had a much larger role to play in decision-making than just noise, however.       
 
Couzin and colleagues had touched upon the influence of leaders and naïve individuals in group movement in 2005 (“Effective leadership and decision-making in animal groups on the move,” Nature) before moving on to explore other questions, but the role of information heterogeneity among group members continued to intrigue him. In 2010, Couzin had been working on modeling the evolution of collective decision-making while Dr. Christos Ioannou, then a post-doc with Couzin and now at the University of Bristol, was examining how golden shiner schools decide which destination to swim to when group members are conflicted. The shiners had been trained to swim towards either a blue target or a yellow target, but to Ioannou’s surprise, he discovered a massive learning bias in the fish: they learned much more quickly and reliably to swim towards the yellow targets. Couzin and Ioannou weren’t sure whether to throw the data out or try to address the bias in the statistics. Similarly, the outputs of Couzin’s models were proving difficult to understand.
 
“There were lots of ups and downs – lots of light bulbs that turned out not to be light bulbs,” Couzin lab graduate student Andrew Hartnett said, who worked with Couzin on the models. “This type of work, theoretical collective behavior, requires a healthy dose of skepticism about your own work. Are our results real or just a trivial mistake, the result of our modeling choices?”
 
This skepticism motivated the creation of several models, each approaching the question of uninformed individuals’ importance from different angles, to ensure their results stood. “We threw the kitchen sink at the problem,” said Hartnett. “Iain and I talked about it constantly. He is great at listening and synthesizing ideas. It was a really fun way to do research.”
 
“We really wanted to make sure the models were coming to the same conclusions,” Couzin said. “We wanted relatively simple models that were revealing common principles and elucidating the mechanism for what we were seeing.”
 
A turning point in the study came to Couzin while walking for the bus. “I realized the learning bias in Ioannou’s experiment was a blessing.” This bias, he said, could allow them to unite Couzin’s models with Ioannou’s experiment. Because the shiners were exhibiting a learning bias towards yellow over blue, Couzin saw the opportunity to have groups of individuals with both mixed preferences and varied strengths of preferences.
 
“There was quite a delay actually in running the experiments to realizing that the model really predicted the naïve effect only when there is a bias,” Ioannou said. “This realization happened over time - Iain initially suspected it and shared this with me, and slowly, with more simulations and analysis, it became more certain.”
 
Once all three models and the experimental work began agreeing with one another, the authors realized they had discovered something monumental. Hartnett said, “Each model serves two purposes. The first is that different levels of abstraction reveal different pieces of the answer to: why is this happening? The second is that a common result across models with very different assumptions provides increasing evidence of the fact that this is a real phenomenon and not an artifact.” The experimental work provided a test of the models, showing they really do reflect the real world.
 
Golden shiners exhibit a learning bias: they learn to associate
yellow with a food reward much more strongly than they do
with blue. This bias was used to create groups with individuals
that not only differ in their preferred destination but also how
strong their preference is. Yellow-trained individuals would
cause groups to go to yellow targets, even if the groups were
composed of a majority of blue-trained individuals. Including
untrained individuals corroborated the models: uninformed
individuals return control to the majority preference.
Figure from Couzin et al. 2011.




This crucial role for unopinionated individuals is important because it explains why we don’t have an arms race for more and more opinionated groups in animal groups, including human societies. There is only so much one can gain by shouting a message; the group’s decisions will always be buffered by the members who are not heavily invested in one outcome or the other. “The tiny fraction that doesn’t care about the outcome of the group’s decision ends up determining the outcome,” Couzin said.

It’s also important to note that these results refer to the role of unbiased individuals, not unintelligent. The article, once published, generated much media attention claiming Couzin and colleagues were implying that unintelligent people are crucial for democracy. Couzin stresses an alternate point. “Much of the media claimed we were saying that less information is a good thing. As we report in the paper, having more information is a good thing. But animals are limited in their ability to get info, so groups tend to be heterogeneous.” This heterogeneity, as shown in Couzin et al. 2011, has a paramount effect on how groups reach decisions.

The article can be read here.