Research Article: Scientific discovery in a model-centric framework: Reproducibility, innovation, and epistemic diversity

Date Published: May 15, 2019

Publisher: Public Library of Science

Author(s): Berna Devezer, Luis G. Nardin, Bert Baumgaertner, Erkan Ozge Buzbas, Daniele Fanelli.

http://doi.org/10.1371/journal.pone.0216125

Abstract

Consistent confirmations obtained independently of each other lend credibility to a scientific result. We refer to results satisfying this consistency as reproducible and assume that reproducibility is a desirable property of scientific discovery. Yet seemingly science also progresses despite irreproducible results, indicating that the relationship between reproducibility and other desirable properties of scientific discovery is not well understood. These properties include early discovery of truth, persistence on truth once it is discovered, and time spent on truth in a long-term scientific inquiry. We build a mathematical model of scientific discovery that presents a viable framework to study its desirable properties including reproducibility. In this framework, we assume that scientists adopt a model-centric approach to discover the true model generating data in a stochastic process of scientific discovery. We analyze the properties of this process using Markov chain theory, Monte Carlo methods, and agent-based modeling. We show that the scientific process may not converge to truth even if scientific results are reproducible and that irreproducible results do not necessarily imply untrue results. The proportion of different research strategies represented in the scientific population, scientists’ choice of methodology, the complexity of truth, and the strength of signal contribute to this counter-intuitive finding. Important insights include that innovative research speeds up the discovery of scientific truth by facilitating the exploration of model space and epistemic diversity optimizes across desirable properties of scientific discovery.

Partial Text

Consistent confirmations obtained independently of each other lend credibility to a scientific result [1–4]. We refer to this notion of multiple confirmations as reproducibility of scientific results. Ioannidis [5] argued that a research claim is more likely to be false than true, partly due to the prevalent use of statistical significance and null hypothesis significance testing as method of inference. Recent theoretical research explored aspects of scientific practice contributing to irreproducibility. McElreath and Smaldino [6] modeled a population of scientists testing a variety of hypotheses and tracking positive and negative published findings to investigate how the evidential value of replication studies changed with the base rate of true hypotheses, statistical power, and false positive rate. Other studies found that current incentive structures may lead to degradation of scientific practice [7, 8]. Publication bias was also proposed to contribute to the transitioning of incorrect findings from claim to fact [9]. These studies focus on how structural incentives and questionable research practices (QRPs) influence reproducibility of scientific results within a hypothesis-centric framework, and how to improve statistical practices and publication norms to increase reproducibility. Under limitations of hypothesis testing [10], however, understanding salient properties of the scientific process is challenging, especially for fields that progress by building, comparing, selecting, and re-building models.

We adopt a notion of confirmation of results in idealized experiments and build a mathematical framework of scientific discovery based on this notion.

Our model-centric framework facilitates investigating the consequences of the process of scientific discovery, including reproducibility, as a system-wide phenomenon. System-wide reproducibility and its relationship to scientific discovery are largely unexplored topics. Navigating through numerous potential variables and parameters to create a realistic system rich in behavior whose outcomes are easily interpretable is challenging. Our model aims to create such a system by making design choices and simplifying assumptions. Among many results that we obtain, we report some intuitive results as reasonableness checks. These results connect our idealized system to reality. However, we highlight the results that seem counter-intuitive to us because we find them to be interesting patterns warranting further investigation. The implications and limitations of each specific result are discussed in the Results section.

First, we present results in a system with no replicator where properties of our scientific process can be obtained for all true models in our model space using Markov chain theory and computationally efficient Monte Carlo methods (S5 File). We use this computational advantage to gain insight into process properties and to inform ABM experiments for the system with replication, in which exploring all model space is computationally unfeasible. Second, we present results from these ABM experiments (S2 Table).

We studied the process of scientific discovery and reproducibility in a meta-scientific framework using a model-centric approach. We have chosen a model-centric approach because 1) it translates to scientific models directly, 2) it is a generic mode of inference encompassing hypothesis testing, and 3) model selection methods bypass difficulties associated with classical hypothesis testing.

 

Source:

http://doi.org/10.1371/journal.pone.0216125

 

Leave a Reply

Your email address will not be published.