The Lively Science


The Lively ScienceChapter 1Chapter 2: Experiments and Real WorldThe logicOut of the lab (the method of agreement)Experimental Control vs. Ecological ValidityNon-linear and Dynamic vs. Linear and Non-DynamicReductionismPredicting a "space of possibilities"CausePath Dependence

Chapter 1

In this book, Agar wants to highlight the difference between "behavioral social science or BSS and human social research or HSR" and the value that comes from taking the latter perspective.

In chapter 1, he provides a "preview of the book" which summarizes some of the differences between these two perspectives of conducting social science.

  1. With BSS, a hypothesis is framed before the research starts and it can’t change. In HSR a hypothesis changes as more is learned during a project, and new ones will appear and be considered well after the project starts.
  2. A BSS hypothesis is made up of “independent” and “dependent” variables. In HSR there aren’t any independent variables. There are, however, patterns that link variables together in many different ways, including feedback loops that undermine many statistical techniques.
  3. Where does a BSS hypothesis come from? Usually from a single academic theory. HSR will probably use several theories to formulate research questions and then make up a few more as the case requires and as more is learned during a project. BSS is theory testing; HSR is also theory-generating.
  4. BSS is empirical, maybe observational, like the earlier definition of “science” requires, but only within the limits of an interview or experiment or aggregate dataset designed and controlled by a researcher. HSR is open to information from any source, researcher designed or not.
  5. The BSS model requires assumptions like standardization—every subject sees the interview/experiment in exactly the same way—and ceteris parabus—nothing varies systematically except the variables in the hypothesis. HSR, in contrast, accepts that neither assumption is ever true in any human social world.
  6. To convert a human moment into a number, BSS assumes the number means something that it doesn’t necessarily mean to research subjects. HSR focuses more on discovering patterns that mean something to subjects rather than numerical measurements of variables.
  7. Both BSS and HSR worry about sampling, but BSS designs a sample before a project starts and HSR modifies it during a project as more is learned about variation among subjects that matter given a particular research question.
  8. Most statistics include prior assumptions about the variables, the samples and the population that BSS usually doesn’t investigate. In fact, popular statistical software, last time I looked, didn’t make it easy to examine a raw distribution of data. Worse, some statistics assume a normal distribution when in fact the variable of interest may well be distributed in many other ways. In HSR a frequent non-normal distribution is that a few patterns occur a lot and a few others are rare, something popularized recently as the “black swan” effect, also called by the unflattering name of “fat tails.”
  9. In BSS, assumptions about research subjects are often made to fit a theory’s premises. Consider the classic economics assumption, that an economic actor is a rational agent free of social influence making decisions on the basis of perfect information. Imagine an obsessively greedy Spock-like figure checking the Dow on the Enterprise computer. HSR modifies assumptions about subjects and their worlds based on what is learned from them. HSR subjects are more complicated and life-like in research reports than BSS subjects are.
  10. BSS aspires to a goal of objectivity, the notion that the researcher and his or her biographical and historical context have no influence on the research. HSR rejects this goal as delusional when humans research other humans. HSR science is, at its base, intersubjective, neither “objective” nor “subjective” in any simple way.

Chapter 2: Experiments and Real World

John Stuart Mill talked about induction — "the discovery of truths not self-evident" — as one of the most powerful tools human beings have to learn about the world. However, we are subject to many psychological biases so we need a more concrete logic and approach to go about this in the best way possible.

The logic

The two major types of induction that Mill talk about are what he calls:

  1. The Method of Difference — this is closest to experimental inference and is considered the "pinnacle "

    • When you compare several cases against one another and "the cases have to be identical to each other in every way but one" — how the varying item/variable interacts with the other variables can then be clearly understood and allows us to infer causal relationships
    • This creates the "BSS dream" which focuses on studying things within a highly controlled laboratory-like setting
  2. The Method of Agreement

    • While this method is not as powerful as the first, it still offers us a way forward when studying complex things like social systems which can't be controlled in the same way. In fact, most things in the world can't be studied well by the first method because it's impossible to control all the variables at play
    • Focuses more on observations

Out of the lab (the method of agreement)

The Method of Agreement relies mostly on observations.

While we can't make clear causal claims like we can with the Method of Difference, the Method of Agreement still allows us to learn from the observations that we make.

Mill infrequently referred to it as the "Indirect Method of Difference" and Agar refers to it as the "joint method." He describes it in the following fashion.

Here is a picture of how the join method works. Take a simple two by two table, like the example below, with X and Y being either present or absent. The natural human induction is to notice the "X and Y present" cell, the upper left in the diagram. It's not a bad ability to have if you're trying to make it through the day. If X and Y are really joined at the hip, though, then the two cells that show "one is present/the other is absent" — the lower left and upper right — should be empty. More than that, as the lower right hand cell of the table shows, if you don't get one then you can never get the other. If X and Y are always present and always absent together, then you have closed in on the traditional gold standard dream, even without an experiment.

(Recreation of the table reference above)

 Y PresentY Absent
X PresentManyNone
X AbsentNoneMany

Experimental Control vs. Ecological Validity

Agar points out here the common trade off between the laboratory's superior internal validity that comes with variable control and the ecological validity comes with studying social phenomena in the real world.

While the BSS camp tries to replicate the laboratory control, Agar suggests that this is not the correct way of going about things. We can study the real world because, "The real human social world isn't a lab. But it isn't random, either."

I liked the below anecdote...

The late Dell Hymes, grand old man of linguistic anthropology, used to joke with those who thought researching real life was too complicated a goal to aspire to. After a paper on the impossibility of figuring out the human social world was offered at a conference we both attended, Hymes said to the presenter, "You got here, didn't you?"

Non-linear and Dynamic vs. Linear and Non-Dynamic

Agar points out that studying social systems is much closer to weather forecasting than it is the the typical linear and controlled experiments. This is where chaos theory and complexity science comes in — small changes in the initial state of the system can have large changes in the


Mill stated that reductionism was the solution to all of our problems when it came to learning about the world. This, however, overlooks a great deal of problems. Simply reducing everything down to its parts often removes the fact that, "a chicken is more than a collection of feathers."

Agar points out that the HSR perspective is looking to understand the "patterns" of behavior while reductionism aims to understand the "parts." Of course, both of these perspectives need to be taken to fully understand a particular phenomena and these perspectives can help each other.

Predicting a "space of possibilities"

Agar discusses his past work in the illegal drug field where they develop the "trend theory" which provided a framework for different scenarios that might produce illegal drug epidemics in the future. He mentions this work to illustrate the point that systems which display non-linear dynamics can't be truly understood with the typical reductionist perspective. The best you can do is discover the larger human patterns, attempt to show how they interact and then make decent predictions in the near future based on the scenario at that time.


In this section, Agar discusses the idea of causation. Essentially, he points out that while there may be certain causes for things — e.g. a famine may cause people to move — they cannot be the end-all-be-all because (at least in the social world) causes may be sufficient but are not always necessary. For example, a famine may be sufficient to cause a man migration, however, war can also lead to the same result. So if you're trying to answer the question, "what causes mass migration?" looking for a specific cause may not be the best approach.

Path Dependence

In the following section Agar discusses the nature of studying dynamic, complex systems. Specifically, he talks about path dependence. "Path" in this instance has to do with the dynamic nature of a complex system — i.e., that we study the system over time. Non-dynamic perspectives will look at a system at time T. This isn't very useful for systems which are constantly evolving throughout time, so we studying them by looking at time T+1 or T+n. Thus, "the way they move and change over time is the path that they take." Path dependence, then, is the nature of a system to behave differently (i.e., take different paths) based on the initial conditions.

The phrase just means that with nonlinear dynamic systems, you always see a path, but it's not identical from one time to the next. Run the system again and you'll see it take a different path. But —and this is an important "but" — the paths will be limited by a bounded space.


The path dependence of a nonlinear dynamic system is important to carry forward into HSR epistemology. It radically Changs how human social science thinks about prediction. Path dependence means that, from where the system is at time T, you can't predict exactly where it will be in the future. It depends on what happens between T and that future. At the same time, by looking at many cases, HSR results will show a space that limits possible path trajectories. They will show how you can expect a certain range of things to happen. And you can guess that certain things probably never will. And you can guess that certain things probably never will. Isn't that sort of a prediction? It won't snow in the Sahara. And illegal drug that causes seizures won't take off.





Notes by Matthew R. DeVerna