I came across a group of researchers doing precisely the kind of work that I have been advocating for on this blog. This is really exciting stuff that I am still working through, but I wanted to share it with you because they do a good job of introducing a conceptual framework for approaching visual patterns in complex data.
Responding to Stephen Wolfram‘s (very influential) 2002 book A New Kind of Science, Hudson et al. critique the contemporary statistical methodology in social sciences as inappropriate for most human behavior. Because of the complexity of social behavior, many simple questions in social science are irreducible—no matter how much computational power is applied, researchers are no closer to proving quantitatively the most intuitive of social acts. This creates, “a strange methodological anomie, where one uses these methods as if all is straightforward, while becoming ever more disengaged in one’s questions and answers from the reality wrestled with on a day-to-day basis by those who live within it.” (Hudson et al. 2005)
They instead suggest a pattern recognition approach to social science data:
In place of the continuous-variable mathematical structures that underlie classical mechanics and statistics, Wolfram’s approach focuses on the discrete transformation of patterns. Simple pattern-based models can, through iteration, produce surprisingly complex behavior in physical and biological systems. Biochemists, for example, search for patterns in amino acids as elements for understanding the functions of a strand of DNA, and then the patterns of those strands combine to produce the patterns formed by larger strands, then by chromosomes, then by the entire genome.
The problem is that the human brain is built around pattern recognition, while social science statistics methods tend to underutilitze that function:
[W]e as humans know from our own lives that…we [cannot] define every concept in terms of quantities. As Wolfram puts it, “it is in many cases clear that the whole notion of continuity is just an idealization—although one that happens to be almost required if one wants to make use of traditional mathematical methods.” (Wolfram, 2003, 729).
Luckily, our brains are excellent tools for analyzing patterns. A great deal of empirical research indicates that human psychology is actually structured to find meaning in patterns (Political Psychology 2003). Since our brains can function as a pattern analysis machine, then Hudson et al. argue that researchers should take advantage of that fact to present data intuitively:
Pattern recognition is the ability of an individual to consider a complex set of inputs, often containing hundreds of features, and make a decision based on the comparison of some subset of those features to a situation which the individual has previously encountered or learned. In problem solving situations, recall can substitute for reasoning. For example, chess involves a well-defined, entirely deterministic system and should be solvable using purely logical reasoning. Chess-playing computers use this approach, but Chase and Simon (1973) found that human expert-level chess playing is done primarily by pattern recognition.
This is why police departments still need detectives and the CIA analysts. Our brains recognize patterns using qualitative rules and heuristics in a way much more powerfully than any quantitative methods.
So how does Hudson et al. apply this approach? Since, “human understanding involves matching observed events to a pattern,” then, “the function of political discourse is to provide sufficient information—in the forms of declarative knowledge, event sequences and substitution principles—to cause the audience to understand (i.e. pattern match) the situation in the same manner that the individual transmitting the information understands it.”
What they are getting at is that there are ways to specify simple rules for behavior and then use the brain to draw out relevant patterns in that behavior. Using news reports of various interactions behavior between Israeli and Palestinian actors, the authors came up with a small set of simple rules that seemed prevalent in cooperation literature (tit-for-tat, olive branch—standard behavior under mutually destructive circumstances in a prisoner’s dilemna game, and so-called ‘meta patterns’ that involved complex patterns arising out of simpler ones)
This analysis produced a rich set of results. For example, we found three general results on TFT, the simplest of our rules. First, the TFT behaviors are generally, but not totally, symmetric in the dyad—generally when one side is engaging in TFT, whether cooperative or conflictual, the other side is doing so as well. There is no reason that this must be the case, but the fact that we observe it suggests that the two antagonists are implementing a classical TFT solution to the prisoners’ dilemma game.
My favorite part, however, is their data visualization. An integral element of pattern analysis, as I and the authors have both argued, is recognizing visual organization in complex data. As such, the authors created a web tool to visualize their dyadic relationships in a very original way (online at http://kennedyosx.byu.edu/).
It is particularly suited to recognizing patterns in behavior over time—yet unlike a time series visualization, the time frame is relational, in terms of consequences elicited by one party in response to actions (or signaling) by another. This makes their visualization particularly suited to pattern recognition, in that it does not require any log transformations to make relational organization recognizable.
I think that social scientists have quite a bit to gain from reading Wolfram’s book, taking instruction from Hudson et al. Although I think that their approach is probably too complicated for most users, it provides a structural framework for just the sort of approach I have been advocating: considering the brain as a pattern recognition machine in order to utilize its interpretive power in data analysis
Axelrod, Robert. 1984. The Evolution of Cooperation. New York: Basic Books
Hudson, Valerie, ed. 1991. Artificial Intelligence and International Politics. Boulder: Westview
Hudson, Valerie M, Philip A. Schrodt and Ray D. Whitmer “A New Kind of Social Science:
Moving Ahead with Reverse Wolfram Models Applied to Event Data.” Paper prepared for delivery at the Annual Meeting of the International Studies Association, Honolulu, March 2004.
Khong, Yuen Foong. 1992. Analogies at War. Princeton: Princeton University Press.
Margolis, Howard. 1987. Patterns, Thinking and Cognition: A Theory of Judgement. Chicago: University of Chicago Press.
Political Psychology. 2003. “Special Issue: Neuroscientific Contributions to Political Psychology.” Political Psychology 24,4.
- Social scientists explain many things – but can they explain themselves? | Kate Roach (guardian.co.uk)