Age, Income, and Delay Discounting Across Adulthood

This repository provides reproducible R and Python workflows for the data and analyses reported in Wan et al. (2024), Psychology and Aging.

Wan, H., Myerson, J., Green, L., Strube, M. J., & Hale, S. (2024). Age-related differences in delay discounting: Income matters. Psychology and Aging, 39(6), 632–643. https://doi.org/10.1037/pag0000818

Research question

The study examined whether income qualifies age-related differences in delay discounting. Adults ages 35–50 and 65–80 with annual household incomes below $50,000 or above $80,000 completed two established measures: the Monetary Choice Questionnaire (MCQ) and an adjusting-amount procedure.

The central finding was convergent across procedures: older adults discounted delayed rewards less steeply than younger adults in the lower-income groups, whereas the higher-income age groups did not differ. The results are consistent with the proposed buffering hypothesis, in which age-related emotional stability may reduce sensitivity to scarcity-related distress. Because the design is cross-sectional and emotional stability was not measured directly, the evidence does not establish developmental change or a causal mechanism.

Repository contents

Path	Purpose
Data on OSF	Deidentified study data; intentionally not distributed through GitHub
Analysis/CODEBOOK.md	Variable definitions, category coding, and outcome construction
Analysis/analysis_R.qmd	Primary R reproduction of the published model structure and sampling configuration
Analysis/analysis_python.py	Version-controllable Python source for the independent PyMC translation
Analysis/analysis_Py.ipynb	Jupyter notebook generated from the Python source
Figure/	Figures associated with the publication
Presentation/BBC.pptx	Conference presentation archived with the project
renv.lock	R package lockfile
requirements.txt	Pinned direct Python dependencies
CITATION.cff	Machine-readable citation metadata

The data are not included in this GitHub repository. Obtain the study data from the OSF project, which preserves the data, original code, and study materials. The analysis-ready workbook reflects exclusions applied upstream: 55 recruits did not meet an age-income cell, five failed an attention check, and four participants who reported nonbinary gender were excluded under the publication's binary-gender analysis, yielding the published analytic sample of 359.

Reproduce the analysis

Clone the repository and run commands from its root.

Download the analysis-ready workbook from OSF and save it locally as Analysis/Data_AgeIncome.xlsx. This path is ignored by Git so the data cannot be committed accidentally.

R (published implementation)

The R workflow most closely reproduces the published analysis.

install.packages("renv")
renv::restore()

The default render validates the sample, reconstructs the measures, and checks the reported non-MCMC statistics:

quarto render Analysis/analysis_R.qmd

To refit all Bayesian models with the published configuration (10 chains, 10,000 iterations per chain, 5,000 warmup):

quarto render Analysis/analysis_R.qmd -P run_models:true

The full run is computationally intensive and requires a working CmdStan installation. Model caches are written locally and are intentionally excluded from version control.

Python (independent translation)

Python 3.11 or 3.12 is recommended.

python -m venv .venv
source .venv/bin/activate
python -m pip install -r requirements.txt
jupyter lab Analysis/analysis_Py.ipynb

The notebook runs deterministic data checks and composite-score analyses by default. To fit the full PyMC model set:

AGEINCOME_RUN_MODELS=1 jupyter lab Analysis/analysis_Py.ipynb

The Python workflow is a transparent cross-language translation, not the software used to produce the article's tables. Small posterior differences are expected across probabilistic-programming systems and random seeds.

Reproduction benchmarks

The workflows assert the following published values before model interpretation:

• final sample sizes: 90 younger/lower-income, 97 younger/higher-income, 86 older/lower-income, and 86 older/higher-income participants;
• cross-procedure correlations: r = .84 ($30) and r = .86 ($80);
• composite focused correlations: age within lower income, r = .202; age within higher income, r = .041; income among younger adults, r = .163; income among older adults, r = .003;
• variance explained by the composite age, income, and interaction model: 2.84%; and
• continuous-age slope in the lower-income groups: 0.0115 standardized units per year.

The published omnibus age-by-income interaction was not statistically significant (p = .122). The paper's primary evidence comes from theory-driven focused contrasts; the repository does not relabel the omnibus interaction as confirmatory evidence. The hypotheses and analytic plan were not preregistered.

Transparency and responsible reuse

The study was approved by the Washington University in St. Louis Institutional Review Board (IRB No. 201806131). The design, hypotheses, and analytic plan were not preregistered. Data collection occurred through Prolific in 2022 and was restricted to U.S. IP addresses. Study data should be accessed through OSF rather than redistributed through this repository.

Code in this repository is released under the MIT License. The article, publication figures, presentation, data, and third-party materials may be subject to separate copyright, consent, or reuse conditions; the MIT License does not override those conditions. Consult the OSF record and the relevant publisher before redistributing non-code materials.

Contributing

Reproducibility corrections are welcome. See CONTRIBUTING.md for the expected workflow and reporting standard.