CoChat

Cross-Validation Approaches for Replicability in Psychology

Open access
Koul, Atesh, Becchio, Cristina, Cavallo, Andrea · 2018 · Frontiers in Psychology
Cited by 143

The DOI matches its Crossref record; 7 of the 7 references it cites were found in a registry or an index.

Abstract

IntroductionThe ability to replicate a scientific discovery or finding is one of the features that distinguishes science from non-science and pseudo-science (Bazerman, 1989; Dunlap, 1925; Popper, 1959). In the words of Popper (Popper, 1959), “only by [such] repetitions can we convince ourselves that we are not dealing with a mere isolated ‘coincidence’, but with events which, on account of their regularity and reproducibility, are, in principle, intersubjectively testable”.Recent years have seen a rising concern over the reproducibility of psychological science (Doris, 2015; Open Science Collaboration, 2015; Pashler and Harris, 2012), with some authors even claiming that most of the published research findings are irreproducible (Ioannidis, 2005). A systematic study designed to assess the reproducibility of psychological science (Open Science Collaboration, 2015) found that the mean effect size of the replicated studies was half of that of the originally conducted studies, and, even more strikingly, only 36% of replication studies had statistically significant results. To counter the concerns generated by these results and improve the quality and credibility of the psychological literature, two sets of strategies have been identified: i) improvement in research methodologies; ii) promotion of replication attempts. Accumulating evidence is the scientific community’s method of self correction. When available, this is certainly the best available option for improving the scientific process. Unfortunately, replication attempts are not always feasible. For example, they may not be feasible for large clinical-epidemiological studies. Also, replicating studies that test extremely rare events or conditions (e.g., neuropsychological conditions) may be difficult or simply impossible. Therefore, the question arises as to the generalizability of replication attempts. Are there measures that one could adopt in such cases to ensure a high level of reproducibility despite the impossibility of reproducing the original study? In this opinion article, we propose the incorporation of cross-validation techniques in single research studies as a strategy to address this issue. The first section of the article (section 1) describes in brief what counts as a replication. In section 2, we introduce the concept of cross-validation and how this technique can be utilized for establishing replicability. Next, we describe different cross-validation schemes (section 3), and finally, we conclude by highlighting some of the platforms for implementing cross-validation (section 4) and certain limitations (section 5).1. ReplicationSome controversies surrounding replication attempts seem to arise from the very notion of ‘replication’ – the purposes of the attempt and when would it count as being successful (Simonsohn, 2013; Verhagen and Wagenmakers, 2014). Accordingly, multiple propositions of scope, functioning and typology have been identified (Hüffmeier et al., 2016; LeBel et al., 2017; Lykken, 1968; Schmidt, 2009; Schmidt and Oh, 2016). A basic distinction is between direct replication and conceptual replication attempts (Zwaan et al., 2017). Direct replications are replication attempts that aim at reproducing the exact effects as obtained by a previous study, incorporating the exact experimental conditions. In contrast, conceptual replications examine the general nature of the previously obtained effects, while aiming at extending the original effects to a new context. For instance, a conceptual replication of an effect originally found in a monolingual population would attempt to replicate the effect in a bilingual population (e.g. influence of multiple language use on a particular phenomenon). This article suggests a third typology of replication: simulated replication. When due to practical or methodological constraints direct replication and conceptual replication are not feasible or difficult to perform, simulated replication – we contend – provides an alternative approach to put the replicability of research findings to the test. 2. Simulating replicability via cross-validation techniquesSimulated replicability can be implemented via procedures that repeatedly partition collected data so as to simulate replication attempts. Formally, this is referred to as cross-validation. Cross-validation entails a set of techniques that partition the dataset and repeatedly generate models and test their future predictive power (Browne, 2000). The partitioning can be performed in multiple different ways. The general format is that of a ‘leave k-observations-out’ analysis. In such an analysis, the entire dataset is typically divided into k smaller observations (e.g., trials or participants). k-1 observations (training dataset) are used to generate and train a model. The validity and generalizability of the generated model is then tested on the kth observation (test dataset). Cross-validation has the computational advantage that it avoids fitting a

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.3389/fpsyg.2018.01117

Cites

7 of 7 references verified0 citation errors · 0 not checkable

Statistics

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Cited by

See all 143

Metadata

Authors
Koul, A., Becchio, C., & Cavallo, A.
Venue
Frontiers in Psychology
Published
2018-07-02
Volume
9
Pages
1117-1117
ISSN
1664-1078 (ISSN-L)
OpenAlex
W2808252767
Cited by
143
References
7
FWCI
11.59
Type
journal article
Language
en
Open access
gold

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.