How animals discriminate between stimulus magnitudes: a meta-analysis
Data files
May 05, 2025 version files 20.63 MB
-
README.md
7.83 KB
-
weber-repo-pub.zip
20.62 MB
Abstract
To maximize their fitness, animals must often discriminate between stimuli differing in magnitude (such as size, intensity, or number). Weber’s Law of proportional processing states that stimuli are compared based on the proportional difference in magnitude, rather than the absolute difference. Weber’s Law implies that when stimulus magnitudes are higher, it becomes harder to discriminate small differences between stimuli, leading to more discrimination errors. More generally, we can refer to a correlation between stimulus magnitude and discrimination error frequency as a magnitude effect, with Weber’s law being a special case of the magnitude effect. If more discrimination errors are made when stimulus magnitudes are higher, this could affect how signals evolve. However, the strength and prevalence of the magnitude effect across species has never previously been tested. Here, we conducted a meta-analysis to quantify the strength of the magnitude effect across studies, finding that, on average, perception followed Weber’s Law. However, the strength of the magnitude effect varied widely, and this variation was not explained by any biological or methodological differences between studies that we examined. Our findings suggest that the magnitude effect is commonplace, and that this sensory bias is therefore likely to affect signal evolution across a diverse range of biological systems. Better discrimination at lower magnitudes might result in signalers evolving lower magnitude signals when being discriminated is beneficial, and higher magnitude signals when being discriminated is costly.
Files provided in this dataset
Supplementary data
- Supplementary data 1: Summarises the data extracted from each dataset which was selected for inclusion, with the following headings:
- Reference: Label for this study, corresponding to the Y axis of Figure 3(a)
- Citation
- Species, Common name
- Stimulus: description of the stimulus for which discrimination ability was tested
- Modality: sensory modality (electroreception, hearing, smell, taste, touch, vision)
- Choice: choice type (whether the preference for the chosen stimulus was innate or conditioned)
- Task: task type (whether the subject had to determine which stimulus had a higher magnitude or simply whether the stimuli differed)
- Weber's law: whether the study mentioned Weber's law
- k, Lower bound, Upper bound: fitted value of k and its 95% confidence intervals. This is NA where a model could not be fitted.
- Model fitted, Model test statistic, Model p value: type of linear model fitted to determine the value of k, and the fitted model's test statistic and p-value. This is NA where a model could not be fitted.
- Included: Whether the k estimate was included in the meta-analysis or not
- Exclusion reason: Why the k estimate was excluded from the meta-analysis (if applicable), from the following reasons:
- k failed: a value of k could not be fitted
- AIC plot: the plot of AIC against k suggested problems with the fit
- response plot: the plot of the response against perceived stimulus difference suggested overfitting or other problems
- interval: the interval between the upper and lower 95% confidence intervals for the estimate of k was greater than 3, suggesting a low quality estimate
- Source: whether the dataset was found in the database search or the Akre and Johnsen 2014 review
- Data: the source of the data (either a repository reference, a figure number, or "Email" if the data were obtained by email)
- Search date: either the original 25/11/2022 search ("2022") or the 02/12/2024 second search ("2024")
- file prefix: prefix used to label all output files and diagnostic plots associated with this data
- Supplementary data 2: Full list of all studies screened in the literature search. The first tab is the results of the database search, and the second tab is the studies found in the Akre and Johnsen (2014) review.
- Authors, Title, Year, Link, Addresses, WoS.Categories, WoS.Research.Areas, Author.Keywords, WoS.Keywords, Scopus.Keywords: various paper metadata fields, saved from the database search
- title_pass, second_pass, final_decision: three consecutive passes made to determine whether the paper was relevant. Papers for which data extraction was attempted are marked with a "y" in all fields
- rejection_reason: the reason why the paper was not included in the meta-analysis (if applicable)
- data: how the raw data was extracted
- "available": raw data was available directly in a table, supplementary material, repository, etc.
- "figure": raw data was extracted from a figure using GraphGrabber (Benbow 2020)
- "provided": raw data was requested by email and provided by the author
- "unavailable": raw data was requested by email but was not provided by the author
- Source: whether the paper was found in the database search or the Akre and Johnsen (2014) review. Somewhat redundant because the tabs are already sorted
- n_duplicates.x (database tab only): how many duplicates were found of this study in the database search
- New (database tab only): papers from the 2024 search which were not found in the 2022 search are marked with "y"
- Reference_no (review tab only): the number of the study in the reference section of Akre and Johnsen (2014)
Analysis
- /code: contains code used to perform the analysis
- /1_extract_data: contains individual .R code files used to extract data from each dataset, the outputs of which are in results/data_extraction
- 2_combine_results.R: code to combine the outputs from /1_extract_data into a single file, results/k_results.csv
- 3_meta_analysis.R: code to perform the meta-analysis (using the output of combine_results)
- 4_litsearch_summary.R: code to summarise the literature search and count up the number of studies screened
- analysis_functions.R: functions used during the data extraction
- chisqgrid.R: function used to produce a grid of correlations between fixed effects
- /data: contains data used to perform the analysis
- study_list.csv: contains metadata for each study (used to produce Supplementary Data 1)
- raw data copied from original studies has been removed, but standardised versions used for analysis can be found in /results/standard_data
- /litsearch: contains the results of the literature searches conducted
- Supplementary Data 2 25-11-22.xlsx: initial literature search performed on 25/11/22. This is in the same format as the current version of Supplementary Data 2 in the main folder, but it is out of date - a second search was conducted in 2024 according to reviewers advice. This file therefore lacks some studies which are present in the current version of Supplementary Data 2.
- Supplementary Data 2 02-12-24.xlsx: second literature search performed on 02/12/24 after review. The search process used to generate this file is described in the paper. This is the current version of Supplementary Data 2, and identical to the file in the main folder.
- /results: contains the results of the analysis
- /data_extraction: data extracted from each study to be used in the meta-analysis, sorted into several subfolders:
- /models: contains models fitted to each dataset to estimate the value of k
- /plots: contains diagnostic plots for the models fitted to each dataset to estimate the value of k
- /k_fit: AIC value of the model plotted against k
- /k_response: subject response against perceived difference with k = fitted value of k
- /k0_response: subject response against perceived difference with k = 0
- /k1_response: subject response against perceived difference with k = 1
- /figures: contains figures produced during the meta-analysis
- /meta_analysis: contains tables produced during the meta-analysis
- colinearity.csv: colinearity between fixed effects
- k_species.csv: average value of k calculated for each species
- k_table.csv: same as Supplementary Data 1
- mods_table.csv: how k varies with fixed effects
- mods_vif.csv: Variance Inflation Factors for the fixed effects
- n_estimates.csv: counts up how many datasets follow Weber's law, near miss, opposite miss etc.
- variance_explained.csv: proportion of variance explained by each random effect
- /standard_data: data from each study in a standardised format, with the following column headers:
- a = magnitude of stimulus a;
- b = magnitude of stimulus b;
- abs_diff = |a - b|;
- mean_ab = mean of a and b;
- response = some measure of discrimination (varies between studies);
- N = no. of trials contributing to the response (if applicable)
- k_results.csv: all metadata from /data/study_list.csv combined with all models from /results/data_extraction/models
- /data_extraction: data extracted from each study to be used in the meta-analysis, sorted into several subfolders:
Installing kber
We created the package kber to aid analysis, which can be found here: https://github.com/mzw22/kber/
This package can be downloaded and loaded with the following code:
library(devtools) #allows downloads from github
install_github("mzw22/kber") #download package
library(kber)
