Rein Taagepera’s latest book should make a splash. With Making Social Sciences More Scientific, he could open a new period in the history of the mathematics used in political science, which has been successively importing techniques from sociology, economics, and now from physics.
First, on the negative side, Taagepera expands on his criticisms of usual techniques in “the predominant current” in political science and other disciplines. He claims that “a cancer is eating at the scientific study of society and politics: excessive dependence on linear regression” and other statistical techniques of data analysis. Many scholars and students get prisoners of and addicted to “canned computer programs”, which actually are unable to supply answers to most interesting questions.
Tagepera’s positive program is this: “Social sciences must advance in two directions. First, they must go beyond statistical approaches, into model building. Second, they must clean up their use of statistics.”
Taagepera's approach contrasts with the typical advise by some teachers of a generation or two ago: “let data speak”; we shouldn't do it simply because data don’t speak if we don’t present appropriate questions. Statistics does not provide models, but only tools to estimate a previously given model. “Construction of explanatory models can precede systematic data collection”. Emphasis should be put on thinking, “of the type it cannot be abdicated to computers”.
At building explanatory models based on relationships between variables, we should not be restricted to additions and subtractions, but adopt a wider menu with multiplications and divisions, powers and roots. Taagepera’s book provides a very helpful analysis of the mathematical formats of relationships (including linear, exponential, logistic, etc.) that should be expected from constraints of the variable values. For instance, when the variables can take only positive values, linear regression should be carried out on their logarithms. The book gives also very specific and useful advice on how to use statistical techniques, ranging from how to address the problems of causality to how to publish regression results
A powerful motivation for Taagepera “more scientific” social sciences is his conviction that a better science would make for better politics. He notes that much of the current social science, limited to a restricted set of methods and non-cumulative results, is largely unimpressive for outsiders, socio-political decision-makers included, and has little impact on the real world. “How much attention do politicians pay to political science or other social sciences?” –he asks. “We all know”… The reason is that “to the society at large, quantitative social scientists presently seem no better at prediction than qualitative historians, philosophers, and journalists –they just look more boring”… In contrast, “science becomes useful to practitioners only when it has reached a somewhat advanced stage of development”, as happened, most remarkably, with physics regarding engineering, but also to different extents with medicine to biology, business to economics, and should happen more with politics regarding political science.
My own illustration:
Three scientific formulations on the relationship between electoral systems and party systems.
“I expressed [the electoral system] effects in the formulation of three sociological laws: (1) a majority vote on one ballot is conducive to a two-party system; (2) proportional representation is conducive to a multiparty system…
It is also clear that the relationship between electoral and party systems is not a one-way phenomenon; if a one-ballot vote tends toward a two-party system, a two party system also favors the adoption of a single ballot voting system.
“The total amount of explained variance is explained almost entirely by a single variable: the effective threshold [which is a function of the district magnitude M, later formulated as 75%/M+1]. Each percentage increase in the effective threshold reduces the effective number of elective parties by 0.06… All of the coefficients of the regressions of the dependent variables on the effective threshold… are statistically significant, usually at the 1 per cent level.”
Independent variable ----- Dependent variable
Effective threshold -------- Effective number of parliamentary parties
Adjusted R2 = 0.28
“When an assembly of S seats is elected in districts of M seats, the most likely number of seat-winning parties (No) is
Rein Taagepera, Making Social Sciences More Scientific. The Need for Predictive Models, Oxford University Press, 2008.
Gianfranco Pasquino said...
Just a (legitimate) question mark: ?