Commentary
Replication, robustness or methodological competition
Bibliographic Data
| ID | 21714375 |
|---|---|
| Authors | Harald Hammarström (0000-0003-0120-6396, Uppsala University, corresponding author) |
| Year | 2025 |
| Volume | 29 |
| Issue | 3 |
| Pages | 555-557 |
| Publication date | 2025-10-27 |
| Open Access | Yes |
| Type | ARTICLE |
| Venue | Linguistic Typology (JOURNAL) |
| Journal identifiers | ISSN: 1430-0532 • E-ISSN: 1613-415X |
| Publisher | Walter de Gruyter GmbH (PUBLISHER • DE) |
| DOI | 10.1515/lingty-2025-0041 |
| OpenAlex | W4413098795 |
| Language | EN |
| References cited | 4 |
The authors (Becker and Guzmn Naranjo 2025) frame their study as replication (of the methods part) of four previous studies in quantitative typology.They have adopted a rather broad definition of the term replication, and what they do more specifically is to test the robustness of these studies against a different method.They characterize robustness as the stability of the result of a study "across different methods used for analysis" and "under variations in experimental procedures".Unfortunately no qualification to "variations" or "different" is provided, but surely robustness to just any different method is not interesting.If a method is known to be better than another (for a given question), robustness to the poorer method is not relevant.Presumably the authors mean that if there are different reasonable methods where it is not settled which one is better (for a given question at the current state of knowledge), then robustness is of value.Ideally we should find the better method as far as possible and resort to robustness only in the special case that several methods are equally arguable.Most of the previous work highlighted as replications for robustness in typology would rather count as invalidated 1 using a more suitable method, namely one with more appropriate controls for independence against a previous one with weaker or no such controls.Rather than robustness, the primary question should be whether or not the method used by the authors is better (for the cases at hand).Indeed, the authors reveal this stance too when interpreting differences in the results. 2 But are the bias control methods employed by the authors arguably better than their predecessors?Phylogenetic regression is a relatively well-understood family of methods (see references cited by the authors) and conforms to the theory behind language family
Biology · Cognitive science · Epistemology · Computer Science · Language and cultural evolution · Philosophy · Psychology · Genetics · Virology
| Highly cited | No |
|---|