Helena Montoya Calero

This research note introduces a 24-item survey instrument for measuring scientific decision-making (SDM) in organizations, reports diagnostic and within-study consistency checks, and applies it to a field sample of 3,178 respondents in 54 firms (analytic N = 2,745). Scientific decision-making is increasingly treated as a managerial resource, yet scalable tools to observe reported decision practices inside firms remain scarce. Variation is concentrated within firms: the overall SDM intraclass correlation is 0.018, implying that only 1.8% of variance is attributable to firm membership. Adaptive Learning, the dimension capturing whether managers revise beliefs in response to new evidence, shows the clearest top-tier contrast: Tier 1 managers score higher than frontline managers (unconditional Δ = .17; with education, tenure, and decision-area controls, b = .14–.16, p < .001, without and with firm fixed effects), while Causal Reasoning and Evidence Evaluation do not show comparable top-tier patterns. SDM is also positively associated with performance on a short within-study decision vignette (r = .19), with a 9.3-percentage-point top-versus-bottom-quintile gap on vignette accuracy. The contribution of this note is a survey module and a compact within-study benchmark.

Does theory-based entrepreneurship training work for everyone, or only for certain founders? This paper studies heterogeneous treatment effects of the IdeaBoosterLab programme, which trains early-stage entrepreneurs to apply the scientific method as a decision-making tool. We pool six randomised controlled trials conducted between 2017 and 2024 in the Netherlands, Italy, Colombia, the United Kingdom, Spain, and India, yielding 1,187 founding teams assigned to three arms: Theory-based training (TE), Evidence-based training (E), and a pure control group, observed over up to six follow-up periods. Using the GenericML framework (Chernozhukov et al. 2025), we estimate the Best Linear Predictor (BLP) of individual treatment effects, Group Average Treatment Effects (GATES) across benefit quantiles, and the Classification and Analysis (CLAN) of founder characteristics over-represented in high- and low-benefit groups. We examine heterogeneity along three dimensions: person-level traits (prior experience, age, education, serial entrepreneurship), venture characteristics (development stage, full-time commitment), and scientific-method absorption (theory and empirics stock scores). Preliminary draft, results pending.

Compares classical econometric approaches (LASSO) with LLM-based methods for predicting ordinal organizational variables. Explores linear probing, prompt-ensemble, and linear distillation techniques.