I am convinced that most people become scientists not for the big overarching aims of science, but for personal reasons. Because I love the outdoors, plants, working with data, and a very flexible independent job would be four of mine. Others love working with their hands, a certain form of status, just love their species, etc. But none of these are the overarching goals of science. And even if I don’t think overarching goals are why we get into science, I do think most scientists are bought into the overarching goals of science as well. Certainly I think most scientists see themselves as truth-seekers. Can we be more specific about the overarching goals of science? I am going to argue that there are three major overarching goals of science:
- Understanding – the answers to why questions. How does the world work. Why is the sky blue? Why are there so many species?
- Prediction – what will happen? Neptune must exist because of the weird deviations in the orbit of Saturn from expectations. How many species would I expect to find on an island of size X with climate Y?
- Description – the answers to what questions? What is the nature of the world around us? What is the gravitational constant? How many species are there in the world?
Of course most scientists buy into all three of these goals. But I believe that most scientists are attached to some of these goals more than others. So before reflecting on this further, if forced to pick only one goal, which one do you think most motivates your work as a scientist?
Now that I’ve forced you to commit, here a few thoughts:
- Obviously all three are important – I’m not down with X is the most important goal point of view. I believe science benefits from diverse methods and approaches. So science benefits from having people pursue all three of these goals. I truly believe which goal calls to you is a personal choice.
- Some fields of science (or at least some problems) don’t have to choose. Newton’s three laws blew everybody away because it simultaneously understood and predicted from hot-off-the-press descriptions from Brahe. But I would submit in a multicausal field like ecology, we rarely get to do all three at the same time. When 20 different forces are operating on the same system, understanding in detail even with quantitative models how one of those forces works, is not necessarily going to help you too much in being predictive about the full complex system in the field (where you don’t even know if that force is important or not). So I submit ecologists usually have to pick.
- It is quite common to hear just understanding vs prediction as goals. I feel like description has kind of faded into the background, although the strong movement for re-valuing natural history could be interpreted as an argument for description. I also think description sounds “qualitative” and therefore soft. But description can be highly quantitative which I think is one of the main hallmarks of science. And there are physicists whose life’s work is to put a few more decimal places on the gravitational constant or the charge of an electron. What would ecology be like if we pursued these questions more strongly? Would we still have order of magnitude error bars around the number of species on the planet?
- It is tempting to put these in a sequential order. First you describe. Then you understand. Then you predict. Except I think this is completely wrong. As noted, already, I don’t think understanding necessarily leads to prediction. So often in ecology correlative methods are more predictive than more mechanistic methods. And some forms of description only even make sense after we have clear understanding and models.
- I think it is fairly easy to define success of prediction (e.g. RMSE of predictions on data not used to calibrate the model) and description (how many significant digits can we put on a well defined quantity to measure). Defining success at understanding is much more slippery. I think some people would quickly say it is the same as mechanism (to which I might agree). But that is just passing the problem to another word. There is not even clear agreement among practicing ecologists (nor among philosophers) on what mechanism is (e.g. see Connolly et al 2017 vs McGill and Potochnik 2017 or McGill & Nekola 2010). I realize my views of mechanism are heterodox and not agreed with by most readers (or even co-bloggers). But the traditionalists have yet to give a really precise definition either. And to return to Newton, did he really understand gravity? or just give a really precise formula to describe gravity that is also predictive?
- With co-authors I have argued elsewhere that the best demonstration of understanding is prediction. If you cannot accurately predict, just how well do you understand a system? And that is a standard that for example physicists or meteorologists have successfully risen to meet. Conversely, it is popular among ecologists to claim that correlative methods cannot extrapolate – only mechanistic understanding can successfully extrapolate (Dunham & Beaupre 1998). But as I hinted above, I really think ecology would be best off if the advocates of these two goals recognized that in a multicausal world, these two goals often have little to do with each other and declared a truce.
- Nowhere did I mention models. Models are clearly an important part of science (see the McGill & Potochnik link above). But I would argue they are orthogonal to this discussion. Models can be present or absent in all three of the goals. Arguably all three are better when models are present most of the time. But that is a separate question from this one.
- And for the 1% of you who are wondering what I think. I started my career as a strong understanding first person. By the end of my graduate career I had evolved into what I would have said was a prediction first view. But I would now say that I am a description first, prediction second person and that that is what I’ve actually done most of my career. I have seen so many theories that seem detached from reality. If I want to know what really happens in the real world (and here I am giving primacy to a scientific goal that I recognize is personal and not global), my best bet is to go measure it.
What do you think? Am I missing a major goal? Do I have one too many goals? What big goal motivates you and why?
Dynamic Ecology 
Somewhere, Robert Peters is smiling:
Via Twitter:
Personally I care about understanding most. But I agree that there’s no objective way to rank these three goals in order of importance. I think science as a collective enterprise is best off if there’s a critical mass of scientists who highly value each goal.
One interesting thing to think about is cases in which, if your goal is X, you will fail in that goal if you don’t value Y or Z sufficiently. There are some obvious ones: if you misdescribe how the world is, you’re going to struggle to understand why it is the way it is or predict how it will be in future. And the post cites the claim that you might need a (mechanistic) understanding of how your data were generated in order to be able to make predictions that extrapolate to new contexts. I feel like there must be other examples, but I’m too tired right now to think of them.
Interesting. In my field (landscape perception) there is lots of description and then correlation or regression, but it rarely proceeds to prediction because investigators start from scratch again–there is rarely testing of the prediction. That is not the case in ecology?
I think the poll results match my understanding of ecology fairly well. Namely the strong majority of ecologists prefer understanding. There has been a movement of late for prediction, but it is still an open question whether it will really gain traction. But the starting over and failing to completely close out an attack on a question is I think also typical of ecology.
Via Twitter:
Lots of people say this sort of thing, but I confess it puzzles me. First of all because lots of technological advances provide us with more and different descriptive data than we used to have. Think of remote sensing for instance, or using cell phones and the internet to facilitate citizen science projects. The ability to collect more and different descriptive data is like the whole reason people are excited about most technological advances in ecology! Second, as Sam Scheiner’s data showed a few years ago, the majority of papers published in leading ecology journals report observational field data (as opposed to, e.g., field experiments, lab experiments, or theoretical models) that the authors collected themselves (as opposed to, e.g., meta-analyses of data collected by others). And as for novelty, the *natural history* sections of Am Nat and Ecology ask that the natural historical observations reported there be novel and interesting (so not, e.g., the first record of species X in locale Y). So I just don’t get it when people say that, at a systemic, field-wide level, descriptive observational work is undervalued in ecology or is being crowded out by technological advances.
I think what is missing in ecology is coordinated large-scale systematic observations. The breeding bird survey and BCI (and related forest plots) is about the only exception. If you look at how many papers came out of those two datasets its a pretty good argument that we need more data like that. And yet we have very little interest in funding such work. So I think it is fair to say that that kind of coordinated observations is undervalued.
NEON is gearing up to be just that: 47 geographically distributed platforms collecting 30 years of data; and within each year biological data sets generated across the season at multiple sub sites. That kind of coverage on things from phenology to biomass to diversity to activity, collected the same way, is as you said, pretty unprecedented. Definitely need more, and my hope is the NEON platform will be expanded to collect a wide range of data.
You don’t think that NutNet and its many imitators are changing that? Or do most of those projects not count because they involve too few species/sites or because they involve distributed experiments rather than distributed observational data?
NEON is certainly a big investment but I think the number of sites observed clearly reflects the priority of ecosystem science, not the much larger number of sites that could have been achieved at the same cost that would most benefit population/community/biodiversity ecology.
I think NUTNET and related are the best examples. But it is worth noting that NUTNET started based on the devotion of individual researchers, not a big investment by any funding agency.
NEON is a great example, but the efforts by people in GLEON are probably closer to what you’re describing. I think that the studies initiated by GLEON have largely been under appreciated, because many ecologists are unaware of a global network of (mostly) lake ecologists working together to solve large-scale issues in freshwater ecology. GLEON scientists use experiments, observations, big data, and modeling to answer questions.
I agree GLEON is more in the lines of what I’m thinking.
Via Twitter:
Counterpoint: Charles Darwin famously described his theory of evolution by natural selection as “a theory by which to work”. You’re misunderstanding Darwin’s approach to science if you think he inductively inferred evolution by natural selection from lots of descriptive raw data. Rather, his hypothesis about how the world worked guided him in which observations to collect (and what experiments to conduct), and how to interpret them.
The point is not proof by authority, or to suggest Darwin’s is the only way of working. But anecdotally, I think “describe, then predict” is a common view of how science ought to proceed (maybe I’m wrong that it’s common?). So I think it’s worth highlighting a counterexample.
I really find I cannot separate the three. As a field biologist, I always enjoy finding something new and unexpected (description), but I go into Nature with pre-formed templates/scenarios that I look to validate (understanding), with the gold standard of validation being the support of a non-trivial prediction.
Brian, your intro that caught my eye with your premise that personal reasons drive the choice to become a scientist. When I was a grad student at UArizona, EEB occupied a building with Cell/Neuro. On a lark, a colleague (RS) and me hung out in a corridor shared by both and asked grad students why they were scientists. All the Cell/Neuro students spoke of a fascination with pattern and how things were put together (a lot mentioned breaking their toys and trying to re-assemble them). But all of the EEB students evoked an early memory of an organism.
This may have been of its time (mid-80’s), but I think it described EEB students as having a rather unique personal motivation–encounters with nature–that don’t have a natural complement for quantum physicists or chemists (but very well might for astronomers and meteorologists).
I wonder if such early encounters with nature motivate EEB grad students as much nowadays, when children are less likely, it seems, to be kicked out of the house at an early age and left to explore the streets, parks, creeks, and suburban lawns, looking for entertainment?
I haven’t collected data as systematically as you have. But anecdotally I agree that the majority of ecologists are in the field because they love the system, while scientists in many other fields are there because of some form of driving curiosity. I wonder if you would have to go out of the physical sciences to the social sciences to find other fields that are as love-of-system oriented as ecology – I’m thinking maybe of psychology and economics. ut I’m probably speculating off the deep end here.
A few years ago I was in an Ignite session at the ESA meeting, I think on theory vs. data in ecology. There was a Q&A afterwards, all the speakers sat on stage and took questions from the audience. And in response to a question one of the other speakers (don’t recall who) ended her answer about how theoreticians and empiricists need each other with a passing remark about how natural history and love of wild nature are what get us all out of bed in the morning. Something like that. It was a universal claim about what motivates all ecologists. It will surprise no one who knows me, or who reads this blog, to learn that I couldn’t let it pass uncommented. So I said that, no, that’s one common personal motivation for doing ecology and that’s great. But there are other equally-great motivations, and that one source of friction between theoreticians and empiricists in ecology is that they too often don’t understand and *value* each others’ motivations. Not merely tolerate, *value*. My own motivation for doing ecology is well-articulated by Ben Bolker here: https://dynamicecology.wordpress.com/2013/12/03/hoisted-from-the-comments-ben-bolker-on-other-peoples-data/.
I confess I’m probably over-sensitive about this. Because I know my motivation places me in a minority among ecologists.
I think ecology as a whole would be better off if ecologists understood and valued one another’s motivations better. Or failing that, if they quit caring about each other’s motivations for doing ecology. Care about my ecology, not about why I do it. Because the quality of the science I do has squat to do with my personal motivations for doing it (ok, unless my personal motivations led me to falsify my data or something).
Understanding is what I seek when I wander around (why, why, why questions everywhere I look) and when I want to formally examine something via science. I get a kick out of a pattern showing up in the data, a graph, during a field study. Though I get a deep satisfaction out of making a model that seems decent and out of writing/synthesizing, moments of insight/understanding bring about a spontaneous joy that, for me, does not occur through prediction or description. It is also true, for me, that the degree of joy is not dependent on the perceived magnitude of the insight. Little discoveries are as much fun as larger ones.
The poll results suggest that this may be a common truth among researchers and perhaps among people in general.
I think that one thing missing from this list is application or seeing results go forward into “real world” change. As a conservation geneticist I hope that my findings can be used or applied to help in management decisions.
Yes! I second this perspective. My motivation for research is driven by the potential for application and collaborations with conservation managers.
Fair enough – I intentionally left out the basic/applied distinction because we’ve done that several times on DE. But of course I agree that motivates a lot of people.
I guess I would argue that “impact in real world” is strongly aligned with the prediction and description motivations but not so much on the understanding motivation (unless that improves prediction)
I think ‘prediction’ and ‘application’ are the same thing here.
1) I can predict something will happen (doing X will increase populations of an endangered species).
2) I create conditions for it to happen (implement conditions for X).
3) I check if it really happened (did populations go up?). This way I know if conservation/application worked, or not.
@SB – I agree with your analysis about prediction and application. I also think description and application go hand in hand. Basic questions like what is the abundance of deer or abundance and distribution of endangered species X and how are they trending or what is the population structure (via conservation genetics) of threatened species Y are all important to management and would have to be put in the description goal.
I was initially going to select “understand” but I went for “describe” as I was fairly sure it would be a minority viewpoint, and also because I wonder whether some people really appreciate the difference between describing and understanding, or if the difference is really as fundamental as we believe. It seems to me that “understanding how the natural world is structured by describing the patterns I see” is a perfectly valid way to do science. But this may not be “understand” in the sense that Brian means.
The other point I’d make is that there are certainly examples where we think we know the patterns and try to design tests to understand the processes behind those patterns, but actually the patterns are based on misconceptions or poor data or untested received wisdom. The idea that tropical species interactions are always more specialised is a case in point. If we take that as a given and design experiments to test that pattern, the results could be very misleading because by no means all tropical interactions are more specialised.
I think your second paragraph is a very good argument for the importance of the description goal. It kind of blows my mind how many basic facts we don’t know in ecology.
Good example of Jeff’s point from Ambika Kamath: https://ambikamath.wordpress.com/2017/03/14/how-do-we-know-what-we-know/
More broadly, I wonder how many zombie ideas in ecology get established because people overgeneralize from the first study or first few studies, especially if the first few empirical studies are accompanied by or inspire a bit of theory. Do zombie ideas form because “stylized facts” get enshrined too quickly as straight-up facts? (aside for readers who don’t know what I mean by “stylized facts”: https://dynamicecology.wordpress.com/2017/01/26/stylized-facts-in-ecology/)
Thanks for the shout-out, Jeremy!
Dear Brian,
I hooked up on this part: “It is quite common to hear just understanding vs prediction as goals. I feel like description has kind of faded into the background, although the strong movement for re-valuing natural history could be interpreted as an argument for description. I also think description sounds “qualitative” and therefore soft. But description can be highly quantitative which I think is one of the main hallmarks of science.”
I completely agree. One quick comment:
As the pool shows, a vast majority of ecologists prefer understanding over description. For this reason, it may be particularly challenging (although not impossible) to fund basic “descriptive” research, even if the proposed scientific program is creative and, yes, quantitative. For example, there are legions of papers on the “why” and “what will happen” of biodiversity, but comparatively little field effort on i) comparing methods of measuring biodiversity, ii) identifying sources of errors and biases across taxonomic and organisational levels, or iii) propagating uncertainities associated with biodiversity estimates. I know many ecologists who would refrain from funding such a research program because it involves no clear teastable hypotheses.
I would argue that climate change research has made huge progress in large part because they have refined their methods of measuring carbon fluxes and worked hard to increase the precision on their estimates. I do confess on this ground a biogeochemistry-envy 😉
I agree with everything you said.
As a biogeochemist/ecosystem ecologist, I think your envy is slightly misplaced. We have some fancy instrumentation now (e.g. increasingly numerous eddy covariance towers) but there is still a ton of uncertainty about, a) underlying mechanisms, b) effective representation of mechanisms in process-based models, and c) spatial and temporal heterogeneity in stocks and fluxes. At large scales, we have better constraints, but for many large ecosystems, and indeed biomes, we are relying on precious few plots and towers.
However, I do agree that focusing on measurement is a good thing!
Via Twitter. I was waiting for someone to make this joke. 🙂
Neptune was predicted based on anomalies of Uranus (not Saturn).
Thanks.
Brian, this won’t come as a surprise to you, I don’t think – but somebody still has to convince me that understanding can stand alone, that it doesn’t need to be supported on all sides by prediction. I buy that prediction doesn’t need understanding but not the other way around. Jeff
Brian, I loved this post, thanks! The only thing I would add is that our field would benefit if more ecologists were honest, and explicit, with themselves and their readers about their motivation. Is the purpose of your study to describe, understand or predict? It’s hard to do more than one, or at least do more than one effectively.
THIS. Retrofitting “hypotheses” to what’s really a descriptive study, thereby attempting to turn it into a prediction-testing study or understanding-promoting study, is a particularly common offense.
I fully agree. Its hard to know who to blame for not being fully honest and selfaware. In our current culture funding and publishing both judge description and even prediction pretty harshly relative to understanding. And when you see the poll results with a full 65% preferring understanding its not obvious how this changes.
I definitely don’t think most ecologists are being Machiavellian about it. I don’t think most of think to ourselves “I want to do some descriptive work, so I’ll spin it as hypothesis testing to get it funded and get it published in EcoLetts.” I think it’s more down to training. Beginning grad students have the importance of hypotheses and statistical hypothesis testing drummed into them. And so they feel like they have to have hypotheses even though they don’t *really* know what makes for a good scientific hypothesis, or how to use statistical null hypothesis testing effectively as part of a scientific research program, or about other ways to do good science.
In terms of how this changes, I don’t know either, but I hope this blog does it’s bit. This is a topic we’ve hit on before, and I’m sure we’ll return to it.
I do think the culture is changing a bit when it comes to description. I think there’s increasing value placed on descriptive work (especially “large scale” descriptive work) that quantifies some phenomenon across the entire planet. Brian, your working group that compiled and quantitatively summarized long-term monitoring data on local species richness is a great example.
Perhaps one thing that would help a little would be to talk about how to “pitch” papers like that. What is it about a descriptive study, presented as a descriptive study (rather than disguised as a hypothesis-testing study), that makes it attractive to other ecologists, attractive to selective journals, etc.? Just off the top of my head:
-having data for lots of species/sites/etc. impresses people.
-discovering a previously-unknown pattern or “stylized fact” can excite people. Gives theory a target to shoot at, something to explain.
-overturning people’s expectations is exciting. Brian, your working group’s Science paper on long-term changes in local species richness falls into this category. Lots of people just kind of assumed that local species richness is declining everywhere, so were surprised and interested (or even upset!) to learn that that’s not the case.
The idea isn’t to encourage people to bullshit or oversell their papers, obviously. A world in which everybody pretends their descriptive papers discovered amazing new patterns or refuted old ones isn’t better than a world in which everyone pretends their descriptive papers are really hypothesis-testing papers. The idea is just to wean people off the idea that you “have” to have hypotheses to get published in leading journals or just generally get other people excited about your work. One way to wean people off that idea is to point to actually-existing examples of high-impact descriptive work and talk about what made that work high-impact.
Perhaps I’m missing your point Jeremy, but it seems as though you are saying that descriptive (i.e. observational) research cannot be used to test hypotheses. If that were the case then whole fields of science could never test hypotheses, e.g. most of astronomy. Surely hypotheses can be tested by weight of evidence supporting a hypothesis as well as by experiment?
Oh, you certainly *can* use observational data to test scientific hypotheses. But that’s not what many descriptive papers in ecology are really doing, at least not very well. For instance, if your “hypothesis” is something like “variable X will be correlated with variable Y”, that’s not hypothesis testing. Not useful hypothesis testing, at any rate.
Jeremy – but as long as NSF and Ecology Letters give primacy to understanding over description and prediction, people are going to keep playing that game.
For that reason I wonder if the problem is really training the problem or if it is that the more senior established scientists driving those decisions need to become more open-minded?
For sure as you point I have gotten (with co-authors) a purely descriptive paper in Science and shortly thereafter a paper enumerating the kinds of descriptions we need in TREE. So it is not impossible. But I don’t think people who feel like publishing fundamentally important descriptive work in top journals or getting it funded is an uphill battle are wrong. And therefore established decision making scientists are responsible for incentivizing playing the deception game you & Peter describe (and which I agree is bad for science).
Good post! I agree with the opening sentence (“I am convinced that most people become scientists not for the big overarching aims of science, but for personal reasons”). However, regarding the 3 categories, I think I’m “none of the above.” I am most attracted to the PROCESS of science, i.e., the scientific method(s), i.e., a logical (though imperfect) way of getting closer to truths (large or small) in which I can participate. At some level, I find the process comforting: “I may not ultimately discover important or true answers, but I can at least PURSUE answers in a rigorous and thoughtful way.” I wonder how typical or atypical it is for someone to elevate the process above the end goals.
Great post and discussion! I do science basically to understand. Since I was a child, I’ve been always curious about “why” and “how”. Turning curiosity into a profession is quite a privilege, so I feel blessed.
I’d like to explain my vote for “other” in the list above. I think another common function of ecology (or evolutionary biology) is to understand ourselves by examining the lifestyles of other species. Sometimes this works extremely well, Andrea Wulf’s biography spends a great deal of time on Alexander von humboldt connections with leading intellectuals in other disciplines, and evolutionary/ecological problems have been integral to the development of game theory. Sometimes this leads to muddled generalizations or overconfidence, i think Jeremy has mentioned the book “the bet” describing Paul Ehrlich’s overly dire predictions of human population growth.
The trichotomy above is helpful, but I think that we are more likely to generalize in error when we fail to explicilty recognize that some of ecology is actually about indirectly understanding our fellow humans, not a direct study of the natural world. So I’ve helpfully added the term “sophia”.
Pingback: Happy 6th birthday to us! | Dynamic Ecology
Pingback: Friday links: how anger can lead to understanding, major revamp and expansion of NSF proposed, and more | Dynamic Ecology
I apologize for being late to the comments, I was just lead to this page by an organization I am interning for, who requested that I read this article. As I was reading, I was puzzled by a few things. First, it seemed your logic is a two way street when describing which order scientists should take when making their goals of prediction, description, and understanding. For example, you said if you cannot make good predictions, you do not understand something. However, that could be rephrased as, ‘if you don’t understand something, you won’t be able to make good predictions.’ Additionally, I wasn’t clear on your use of the word ‘understanding’. To me, that is an umbrella term that could mean a variety of things. For example, you can have a broad understanding versus a more detailed understanding. Personally, I don’t think you can make any prediction, nor understand any description, without a basic understanding of the topic at hand. Then through describing and predicting things, you begin to develop a more deeper understanding, which then allows you to make better predictions and descriptions, and it keeps cycling. To conclude, I believe you are right that a good scientist needs all three goals, however, I don’t think there is a fundamental order that needs to be followed because they all play off of each other. I’d be interested to hear your thoughts about this. Thank you for the thoughtful read!
Thanks for your comment. I use “understanding” in the pretty specific sense defined at the beginning of the post – i.e. answers to why questions.
I do think different scientists see the order of understand, predict, describe differently. Which is one of the things that makes science fun!
Pingback: Ask us anything: the future of machine learning in ecology | Dynamic Ecology
Pingback: Prediction Ecology Conference | Dynamic Ecology