Making Social Sciences More Scientific: The Need for Predictive Models

Taagepera argues that quantitative social science has become excessively dependent on regression and statistical approaches, which are essentially descriptive, and has neglected logical model building that is predictive in an explanatory way [citation:1][citation:5]. He rejects the view that social science's difficulties stem from measurement problems or too many variables. Instead, he contends that social scientists have been looking in the wrong direction: they ask 'What is?' rather than 'How should it be on logical grounds?' [citation:2][citation:3].

Taagepera generated twenty-five data points using Newton's law of gravitation and sent them to social scientists, from PhD students to full-time researchers, asking them to find the underlying pattern [citation:2][citation:3]. No one recovered the formula. Instead, they reported high R² values and statistically significant coefficients, which met standard social science criteria for a successful analysis despite being fundamentally wrong [citation:3]. The test demonstrates that statistical methods alone cannot discover the kind of relationships that predominate in physics and, Taagepera argues, can also be found in sociopolitical phenomena.

Taagepera advocates for multiplicative models over additive ones, arguing that outcomes often depend on the factor in shortest supply, which makes multiplication superior to addition in sociopolitical analysis [citation:9]. He calls for 'forbidden areas and anchor points' to ensure models do not predict impossible values like negative percentages [citation:9]. He also argues for interlocking models that trace sequential pathways between variables, and for symmetric regression instead of the skewed regressions that dominate current practice [citation:9]. These reforms aim to produce models that are logically consistent and quantitatively predictive.

Reviewers have praised the book's ambition and its constructive suggestions, calling it 'a pleasure to read' and 'fascinating' [citation:9]. The case for multiplicative models, interlocking models, and symmetric regression are identified as having major potential for improving the field . However, critics from the philosophy of science, particularly constructive realism, argue that society cannot be handled with strict theories similar to those of physics, and that other means often suit better for raising the applicational strength of social sciences than rendering them similar to physics through mathematical formalism [citation:6].

The subtitle, 'The Need for Predictive Models,' signals the book's central claim: social science should aim for prediction, not merely postdiction . Taagepera contrasts the vague statements typical of social science—'there is evidence in accordance with the hypothesis that X causes Y because X has a statistically significant effect on Y'—with the precise, replicable predictions he advocates, such as 'if V falls between 0 and 50%, X will increase by about 10%' . This distinction grounds his critique of regression and his call for logical model building.

Taagepera criticizes regression models with many contemporaneous variables on the right-hand side as 'gazpacho,' comparing them to a cook who piles all ingredients into a blender and calls the result a meal [citation:9]. The target is the additive model that treats variables as independently affecting an outcome, which Taagepera argues prevents researchers from describing sequential pathways between variables. These pathways are of great scholarly and practical interest because they allow prediction of how social processes unfold from an initial starting point to an outcome [citation:9].