The Role of Starting Points to Order Investigation: Why and How to Enrich the Logic of Research Questions

Evolutionary Factors Argument

1. All evolutionary research programs using the MA approach are open to considering only a subset of known evolutionary factors.

2. No evolutionary research program should be open to considering only a subset of known evolutionary factors.

3. Therefore, no evolutionary research program should use methodological adaptationism.

4. Instead, all research programs studying evolution should be open to considering all known factors and use the EFF approach.

I deny that premise 1 applies to all approaches which should be considered methodological adaptationist and I resist inferring conclusion 4 from premises 1 and 2. I agree with Lloyd’s critiques of the particular research programs guilty of pursuing natural selection as the only potential cause in evolutionary biology. But Lloyd does more than critique particular adaptationist programs, she argues that methodological adaptationism qua methodological approach should be abandoned. This general claim follows if all methodological adaptationists dogmatically assume all traits have functions despite all evidence to the contrary. But I am not convinced that methodological adaptationism is itself the root of the problems and so I am not convinced that the LRQ appropriately characterizes all methodological adaptationist programs as MA. I aim to make space within the logic of research questions for some methodological adaptationists not using MA and for a plurality of good approaches to capture the actual methodological diversity in biology.

To pursue my critique of LRQ, I raise the issue of how to classify programs who temporarily or at first pursue natural selection as the cause of investigated traits. I begin by posing a dilemma to the LRQ as it stands now. I then analyze Lloyd’s texts to show that one can find support in Lloyd’s texts for both horns of the dilemma, making the results of using the LRQ to classify the selection-first approaches indeterminate.

Dilemma 1

• (1) All methodological adaptationists use MA.

• (2) Some methodological adaptationists use EFF.

In Favor of (1) All Methodological Adaptationists use MA.

And philosophers and biologists have argued that there is a certain approach to adaptationism, under which one just assumes temporarily that the trait is an adaptation, and proceeds from there, known as methodological adaptationism, that is both productive and harmless. I think this is false. Adaptationism is not harmless, even in its weakened methodological form. (Lloyd 2016, 130)

Lloyd clearly states the thesis that she objects to: adaptationism which looks for selection first is itself objectionable. The contrast is with an openly dogmatic adaptationism which outright claims to never look beyond selection hypotheses. But for Lloyd (2015), the selection-first adaptationism is itself problematic and objectionable:

it is still an open question whether the method outlined for the methodological adaptationist in practice ever allows non-adaptive explanations ever to win the day. (346)

The reason why she finds the selection-first adaptationists objectionable is that she doubts that they ever actually consider other evolutionary factors. In both this and the previous quotation, she is writing generally about methodological adaptationism and its characterization by the LRQ, not making claims about particular research programs.

Since Mayr and other methodological adaptationists admit the possibility of nonadaptive alternatives, it would seem that they admit, at least in lip service, that there should be nonadaptive answers on the list of possible answers to their question. Should nonadaptive answers belong on the methodological adaptationist list? Should the methodological adaptationist list perhaps look like the evolutionary factors list, but with a few more entries like: ‘The function of this trait is P’? The answer to both of these questions is no. (346)

Lloyd here considers the possibility that some methodological adaptationists should be classified as using EFF in LRQ. And she rejects the possibility for reasons that she argues for later in the paper. I will address her strongest reasons below in section 7. In short, she thinks that methodological adaptationism lacks a stopping rule telling them to move on when enough is enough.

Secondly, if, in attempting to pursue the evolutionary factors methodology, the research community gets stuck on the first, adaptationist step of pursuing only functional answers to the question of whether the trait has a function, then the other characterizations of the trait will not be pursued, and a balanced view of the weight of evidence will not appear. Logically, they would be behaving as methodological adaptationists. (Lloyd 2016, 150)

Here Lloyd makes the point that what matters is not what questions the researchers say that they ask but only the question they actually ask and so the answers they actually allow. She imagines someone talking as using the EFF but in practice using MA. This shows the deep interpretative issues facing scholars of science concerning how to associate an actual research approach with a research question on the LRQ.

In Favor of (2) Some methodological adaptationists do not use MA.

When we investigate the evolutionary origins of a given trait, as adaptationists we might often prioritize the functional factors, natural and sexual selection, as the factors of most interest in evolutionary research, and we might start with the question,“Does this trait have a function?” If the trait, after investigation, doesn’t appear to have a function, correlation with fitness, or design features, we might pursue other evolutionary explanations (Lloyd 2016, 132)

This comes in the context of explaining the EFF approach. Here she says that the EFF can prioritize adaptation and so certain factors over others. It is not clear how prioritizing is different from starting with adaptationism and how these selection-first EFF users avoid the pitfalls of methodological adaptationism. In terms of identification as EFF, it is clear when a program starts with selection and actually ends up supporting another kind of hypothesis. But not all cases are so clear and there will be cases which are not obviously EFF or MA.

Someone following the evolutionary factors framework, asks, quite generally:”What evolutionary factors account for the form and distribution of this trait,” or, for example, they can start with “Does this trait have a function?” This question has a series of possible distinct answers (that might be considered in any order, except that the adaptive answers usually go first in practice) (Lloyd 2016, 132)

Here she claims that the EFF allows considering factors in different orders and systematically prioritizing adaptation, which is just how defenders of methodological adaptationism characterize their approach.

We can easily use a more encompassing evolutionary and Darwinian method, the evolutionary factors approach, under which we can ask a variety of questions, including the adaptationist question (Lloyd 2016, 137)

Here she claims that the EFF allows the possibility of asking the adaptationist question. This is the first time she explicitly talks in terms of multiple questions.

So, it is not just that the methodological adaptationists are asking the wrong questions, but in addition, that they are mishandling a possible answer to the question that they are asking, the one that says “this trait does not have a function.” At that point, the methodological adaptationist could move on to a different, more productive question, such as “What other evolutionary factors may play a role in evolving this trait?” But, and this is the point, they do not. And this leads to a variety of scientific errors (Lloyd 2016, 142)

Lloyd again talks in terms of multiple questions and question switching. This is hard to interpret. Of course, MA in the problematic sense means never considering other factors. But what if the program she is imaging did exist, how would we characterize it? There are two possibilities. First, the program which first asked the MA question and the EFF question switched approaches from MA to EFF. Second, the program consistently follows the MA (or EFF) approach, which allows multiple questions and starting with certain factors.

Taking Stock

Both (1) and (2) have textual support and opposition. Lloyd’s official position is (1), that the methodological adaptationists which she has analyzed and critiqued use MA and that this is why their approach should be abandoned in general. In order to challenge horn (1), in the next section I provide a novel case study of successful approaches analogous to adaptationism which in practice move beyond their starting point. My case study further challenges horn (2) by showing that opposing approaches would be both classified as EFF. The LRQ has the consequence that all non-objectionable approaches are EFF. I therefore use my case study to reject the dilemma and suggest a way of enriching the LRQ with more structure to enable finer grade distinctions in approaches.

Community Ecology: Patterns, Processes, and Heuristics

Community ecologists are interested in describing, predicting, explaining, and controlling patterns of abundance and diversity of organisms that live in the same place. How much biodiversity is there, and what ecological factors maintain and modify it? Ecological communities occupy the range in between single populations and whole ecosystems. Community ecologists often restrict their domain to single trophic level communities, in order to exclude interactions involving predation and to ensure all species compete for the same kinds of resources (Mittelbach and McGill 2019).

Many of the basic patterns community ecologists study come from counting and sampling (Magurran 2004). Species richness counts record the number of species in a community. Relative species abundance distributions record the number of species and the number of individuals in a community. Ecologists study these patterns both individually (for a single community at a single time) and comparatively (across time, space, and scales—from square meter plots to entire continents). Many features of these patterns call for explanation: the number of species increases with habitable area, most species are rare, relative species abundance distributions have a characteristic shape, species correlate with certain resources, not every species lives everywhere it can, etc. Other significant patterns include resource-species correlations, from why certain species colonize open spaces to why conifers live in the taiga.

Ecologists have discovered many causal factors relevant to these patterns at many scales. Mark Vellend (2010, 2016) proposed a theory of ecological communities which unifies the many factors and processes investigated by community ecologists into four high-level processes: drift, dispersal, speciation, and selection. These processes are different ways of changing the composition and abundances of species. Drift is stochastic death and birth; dispersal is how organisms move across space; speciation is how populations of new species arise; and selection is the differential reproduction of certain individuals due to their species. Ecological selection is more commonly known as interspecific competition because it arises from competition between differentially adapted individuals of different species. I will refer to it as just competition.

Different research programs focus on different subsets of Vellend’s four processes. Programs interested in descriptive and predictive accuracy can formalize their chosen processes into mathematical models. They can then compare the models with abundance data and evaluate the model fit with statistics.² Whether a model fit is good or poor depends on the predictions of the model, the data, and the standards set.

We need to look at three heuristics scientists use to respond to model fits. I describe the heuristics abstractly here and in the next section show how they are used. A few background details about the heuristics first. These are heuristics in the sense of Wimsatt (2007), problem solving strategies with limited scope but great power when used in conditions they are adapted to. These three heuristics address a relative significance problem (Beatty 1997; Kovaka 2017, ch. 5). Relative significance problems concern the relative significance of multiple causal factors or processes to a pattern. The causal relevance of the causal factors or processes to the kind of pattern is accepted. A relative significance problem asks what the contribution of each of the factors is in particular cases. While any relative ratio of significances may be found, the simple way of describing these heuristics use the idea that some subset of processes “dominates” or accounts for more of the data than the remaining processes. Each of the following three heuristics are ways that ecologists have for making inferences about the relative significance.

Heuristic 1: Basic Explanation

From good model fit, conclude that the processes included in the model dominate the data.

This is the basic modeling inference, adapted to the case of relative significance of processes and models containing subsets of processes.

Heuristic 2: Baseline Modeling³

From poor model fit, conclude that the processes excluded from the model dominate the data.

The model playing the role of the baseline model is used to detect deviations from the data. These deviations are then ascribed to known or unknown processes excluded from the model. The lack of deviations from the baseline model are ascribed to the processes included in the model.⁴

Heuristic 3: Adding Complexity

From poor model fit, decide to augment the model with (some of) the excluded processes.

The shortcomings of the model are here rectified, not by citing additional processes, but my augmenting the model to include the excluded processes. Adding Complexity describes how models are developed over time, not used for explanations. The augmented models can then be compared with data and followed by any of these heuristics again.

Used in concert, these three heuristics enable research programs to respond productively to any confrontation of models to data and to develop over time.

Competitionism

Competitionism is the approach based on starting to investigate the role of interspecific competition in patterns of abundance and diversity. Competitionists then use Basic Explanation, Baseline Modeling, and Adding Complexity heuristics to move beyond interspecific competition.

Competitionist research programs tell the following basic story: Different species have different abilities to utilize resources. The species best able to utilize a resource will exclude its competitors from using the resource. In this way, species differences and resource distributions affect species abundances. Competitionism’s basic explanatory principle is the competitive exclusion principle: sufficiently similar species cannot coexist because one will out-compete and exclude the other. Their basic investigative heuristic is to seek relevant species differences when coexistence is observed. In this way, the competitionist’s starting point is interspecific competition based on species differences.

One competitionist research program is led by David Tilman (1980, 1982) and built around his Resource-Ratio, known as R* (“R Star”) Theory. The only factor considered in R* Theory is competition between species for limited resources. In R* Theory, each species has a minimum level of a resource (the R* value) that they require to at least maintain their population size. The Theory predicts that, when two species compete for a single limiting resource, the species with the lower R* value will exclude the other species. Species coexistence is possible, for example, when two species compete for two limiting resources and neither species has lower R*s for both resources (see (Miller et al. 2005, Wilson, Spijkerman, and Huisman 2007) for a general review of R* Theory).

An example of a basic explanatory success for this competitionist program is an experiment Tilman (1977) performed for his PhD. Tilman gathered two species of algae from Lake Michigan and measured their respective R* values for both phosphate and silicate inside of a chemostat (a laboratory environment in which the resource level can be kept constant over time). With these four R* values, Tilman made predictions from R* Theory about whether the two algae species would coexist or not in the chemostat given certain levels of phosphate and silicate. Tilman was able to produce both stable coexistence and competitive exclusion in the chemostat by varying the relative ratios of phosphate and silicate, confirming R* Theory. Using the Basic Explanation heuristic, Tilman argues that this the successful predictions of R* Theory show that interspecific competition is responsible for both producing coexistence and competitive exclusion (see also, e.g., Tilman 1986; Stanley Harpole and Tilman 2006).

Yet competitionists are not blind to everything but interspecific competition. Tilman has argued that some cases where his R* Theory fails to give accurate predictions show that additional processes such as immigration are important. For example, Tilman leads the University of Minnesota’s Cedar Creek Ecosystem Science Reserve, part of the Long-Term Ecological Research network, where he conducts grassland field experiments. Field experiments must control for invasion by wild species. Tilman (1994) argues that some cases where R* Theory poorly predicts abundance patterns actually show the effect of immigration on the experimental plots.

This is an example of the explanatory heuristic Baseline Modeling. The competitionists begin with a theory that includes only interspecific competition and test it. When the model fit is poor, rather than being disconfirmation of R* Theory, the competitionists conclude that other processes such as immigration not included in the model generating the predictions are playing an important role in structuring the observed patterns.

Tilman has developed R* Theory in a new direction due in part to the Invasion Paradox: Native species resist invasion from exotic species. But exotic species also invade and exclude native species. R* Theory explains the former well, but the latter shows the importance of immigration. Tilman’s (1994) Stochastic Niche Theory incorporates drift and immigration with resource competition. This is an example of the developmental heuristic Adding Complexity. Starting from only interspecific competition, the competitionists are led to include other processes such as drift in their theoretical toolbox.

Neutralism

Neutralism is the approach based on starting to investigate the role of drift, immigration, and speciation on patterns of abundance and diversity. Neutralists then use Basic Explanation, Baseline Modeling, and Adding Complexity heuristics to move beyond drift, immigration, and speciation.

Neutralist research programs tell the following basic story: Drift decreases the diversity of a community by slowly driving species extinct. Immigration increases diversity as individuals from the metacommunity, typically with greater diversity, slowly disperse into the local community. In the metacommunity, diversity is maintained against drift via speciation. Moreover, all individuals, regardless of species, are functionally equivalent or neutral. This has the effect of both excluding interspecific competition and making current abundances dependent on past abundances. Abundance patterns result from the relationship of the rates of the three processes of drift, immigration, and speciation in much the same way as the water level of a sink results from the relationship of the inflow and outflow rates. Neutralism’s basic explanatory principle is that even functionally equivalent species can coexist indefinitely. Their basic investigative heuristic is to seek the effects of community size, immigration, and speciation when coexistence is observed. In this way, the neutralist’s starting point is drift, immigration, and speciation based on species similarities.⁵

The best example of a neutralist research program is led by Stephen Hubbell (2001), who developed The Unified Neutral Theory of Biodiversity and Biogeography (“Neutral Theory” for short).⁶ Neutral Theory includes only drift, immigration, and speciation. Neutral models are individual-based models which assume neutrality and approximately fixed community sizes. Neutral Theory predicts that species abundance distributions will fit the statistical distributions produced by the neutral model. An example of a basic explanatory success for the neutralist program is its ability to describe the abundance patterns of a tropical forest. Hubbell co-founded the Forest Dynamics Plot Project at the Smithsonian Tropical Research Institute on Barro Colorado Island in 1982. They have complete census data for the 50-hectare plot for every 5 years. The neutral model describes the abundances and dynamics of the BCI forest plot well. Using the Basic Explanation heuristic, Hubbell and colleagues argue that the neutral model’s good model fits to that the patterns of abundance of trees on BCI show that these patterns are mainly the result of immigration balancing drift (Condit, Chisholm, and Hubbell 2012).

Neutralists are not blind to the effects of species differences and interspecific competition either. For example, Hubbell showed that while the species abundances in five forest plots from Central America can be explained using neutral models, there is a stronger correlation of the abundances of the commonest species across the five plots than is expected given a neutral model (Hubbell 2001, 337). Because the neutral model leaves out interspecific competition and poorly predicts the abundances of the commonest species, Hubbell argues that their abundances are dominated by interspecific competition. This is an example of the explanatory heuristic Baseline Modeling.⁷

Other neutralists have directly modified neutral models to relax neutrality and introduce interspecific competition. For example, Rosindell, Harmon, and Etienne (2015) use to the neutral model to predict the change in number of species over macro-evolutionary time. However, the neutral model predicts too many species at any given time than we observe in the fossil record. The ecologists therefore relaxed the neutrality assumption and introduced small selective differences into the neutral model. The small degree of interspecific competition included in the nearly neutral model reduces the expected number of species and gives more accurate predictions. This is an example of the developmental heuristic Adding Complexity.

Why Community Ecology is a Problem for the LRQ

The case of competitionism and neutralism is an example of relevantly different methodological approaches which are both open to considering, and have actually considered, all relevant factors. A philosophical account of methodological approaches should be able to account for this, but the LRQ cannot do so as it currently stands. To see this, let us ask how the LRQ should characterize competitionism and neutralism in comparison with the MA and (ecological) EFF. This leads to a second dilemma:

The Ordered Factors Proposal

Drawing on the importance of starting points, the LRQ needs to consider the way that approaches order their investigation of questions and factors. The Ordered Factors (+OF) proposal extends the LRQ to represent approaches using sequences of research questions or factors, the ordered sequences corresponding to the order of inquiry. The evolutionary and ecological factors frameworks characterizations remain unchanged. Now we can distinguish dogmatic methodological adaptationism from sophisticated methodological adaptationism. Dogmatic adaptationism uses MA. Sophisticated adaptationism can be formulated as:

Sophisticated Adaptationism

1. How can natural selection account for the form and distribution of this trait?

2. How can other evolutionary causes account for what natural selection cannot?

Accordingly, the approaches in community ecology can be formulated as:

Competitionism

1. How can interspecific competition account for the distribution and abundance of organisms?

2. How can drift, immigration, and speciation account for what interspecific competition cannot?

Neutralism

1. How can drift, immigration, and speciation account for the distribution and abundance of organisms?

2. How can interspecific competition account for what interspecific competition cannot?

Because with the LRQ, research questions imply their answers and their answers are just sets of causes or factors which are acceptable answers to the question, to shorten their description we can also represent approaches as sets of causes or factors, as in Table 1. This allows comparison of the various options of the views at a glance. Let round brackets mark ordered sequences and curly brackets mark unordered sets.

In characterizing a methodology with +OF, we need not find a linear ordering of all the factors considered. Neither adaptationism, nor competitionism, nor neutralism linearly orders all the factors it considers. Some methodologies may have no general ordering of their factors. They may have heuristics for when to apply one ordering and when to apply another. The +OF proposal does not impose order on the programs, it captures the ordering present in the methodology.⁸

Why Starting Points

Methodologies order factors because orderings are useful for biologists addressing relative significance problems. Competitionists and neutralists do argue sometimes as if they are engaged in a theory choice problem between whether interspecific competition or drift, immigration, and speciation is the one true set of processes responsible for abundance patterns. But biologists’ debates are not usually like the exclusive choices found in the Geocentrists vs. Heleocentrists. While biologists often argue in terms of why their pet processes are the most important processes, their actions show that they acknowledge that all the known factors are in play. Starting points help biologists to answer these relative significance problems.

By imposing an ordering on their inquiry, when biologists decide that they can no longer make do, they move beyond their starting point. As I showed with the competitionist and neutralist programs, biologists have at least two heuristics for this. In Baseline Modeling the processes excluded from the models and incorporated directly into the explanation of the pattern. The reason neutral models offer bad predictions of, say, which species of paramecium will exclude the other when they are isolated and given a constant level of nutrient is because interspecific competition is responsible for this exclusion. By finding that you cannot do without citing interspecific competition, you have better evidence that interspecific competition is playing an important role than if you merely made good predictions from a competition model. And similarly for Adding Complexity.

Bill Wimsatt (1987) argues that the initial assumptions research programs make do not directly aim at the truth but are strategic decisions about how to get closer to the truth later. This fits how many biologists describe their methodologies. As Tilman (1986) writes:

Thus, many of the broad, long-term patterns observed in plant communities may be the result of interspecific competition for limiting resources. Such correspondence does not mean that other processes are unimportant in structuring these plant communities, but the simple theory that predicts them may provide a framework for more detailed studies. (75)

Hubbell (2008) similarly writes that

the most rapid path to understanding ecological systems … is through an interaction of empirical science … with theories that start very simply, with few free parameters and assumptions, and add complexity reluctantly, kicking and screaming, only when absolutely necessary to obtain some desired level of fit to the data. Neutral theory is one such starting point. (143)

Lloyd cites Ernst Mayr (1983) writing

[The evolutionary biologist] must first attempt to explain biological phenomena and processes as the product of natural selection. Only after all attempts to do so have failed, is he justified in designating the unexplained residue tentatively as a product of chance.

She is right to press Mayr for waiting too long to look for alternative explanations, but in doing so she throws out the characteristic feature of adaptationism: taking natural selection as its starting point.

Given that starting points are useful, it is a further interesting question why there is an extant plurality of them across biological disciplines. Fully answering this question goes beyond the scope of this paper. But here is a clue. In the examples from ecology, the most important factor shaping each methodology is the initial target system of the program.⁹ Competition theory was built around creating artificial situations in a laboratory where a small number of species interact for controlled resources.¹⁰ Neutral theory was built around counting trees in tropical forests where ecologists have no control. Arguably, each program grew from systems in which their starting point processes dominate the other known processes. Over their development, these programs were driven beyond their starting points by the failure of nature to abide with what worked earlier.

Lack of Stopping Rules

In characterizing the LRQ, Lloyd anticipates the proposal I defend. Her strongest point against the +OF proposal is her denial of the existence of stopping rules. She argues adaptationists like Mayr, who say they only start with natural selection before moving to additional factors, cannot actually move on because they lack rules for when to stop looking for a given factor. This is not the statistician’s concept of a rule for when to stop collecting data and begin analyzing it. The concern is that, without formal standards for when to abandon the search for the function of a trait, adaptationists will under no circumstances consider alternative explanations. This would entail that starting points are not useful because they would not enable heuristics such as Baseline Modeling and Adding Complexity.

My response to this is twofold. First, this is not a special problem for adaptationists. Any methodology which entertains working hypotheses they are willing to revise faces the problem of whether to tinker with or abandon a given hypothesis that has not met the standards of evidence. The evolutionary factors framework also lacks a stopping rule for make decisions such as: Did this developmental constraint hypothesis perform worse than a sexual selection hypothesis because development was truly the better explanation or because they had not stated the developmental constraint hypothesis good enough?

Second, I deny that formal, algorithmic stopping rules are required. Scientists have only heuristics for when to move on. Good judgment should decide when enough failure is enough, and the development of a research program involves risk and trial and error. I agree with Lakatos that judgements about the health of a research program are only justified in hindsight after a long run.

Adaptationism Is Not Useful

Lloyd (2015) almost explicitly addresses the use of starting points in one place:

I do not mean to deny the common point about the division of scientific labor, by saying that it is a good idea for some to start by asking about the function of a trait …. The question is: is it useful for anyone to be a methodological adaptationist rather than following an evolutionary factors approach? (348, fn. 32)

She answers ‘no’ because she does not allow that methodological adaptationists can systematically entertain the factors included in the evolutionary factors view while imposing an ordering on the factors. But without recourse to the lack of stopping rules, this cannot be maintained. I have shown how starting points are useful, in part because they enable powerful heuristics for moving beyond one’s starting points. Further, the continued existence of adaptationism, competitionism, neutralism, and the many other research programs organized around a subset of factors in the hostile environments that are their larger disciplines is a good sign that such methodologies are useful in general. Specific criticisms are needed to undermine specific programs.

Division of Labor

In the last quotation, Lloyd also suggests that starting points are not enough to distinguish approaches because they are variations within one methodology engaged in “the division of scientific labor”. This response portrays research programs like the competitionists and neutralists as engaged as parts of a larger, unified program using the Ecological Factors Framework. But these programs do not only mutually benefit each other while working toward common goals. They work on only partially overlapping questions and offer competing answers to them. Programs in ecology interact in ways analogous to how species interact in a community.