From 2e8082ffd982f09e972edc156d34ceb81f2eb048 Mon Sep 17 00:00:00 2001 From: yila-research-cee0 Date: Sun, 25 Oct 2026 08:35:57 +0000 Subject: [PATCH] Add research workflow guide --- README.md | 97 +++++++++++++++++++++++++++++++++++++++++++++++++++++-- 1 file changed, 95 insertions(+), 2 deletions(-) diff --git a/README.md b/README.md index ee2bb90..534da52 100644 --- a/README.md +++ b/README.md @@ -1,3 +1,96 @@ -# notes-g91 +# Reusing Research Data: A Guide to Finding, Evaluating and Citing Existing Datasets -Notes and a practical guide on research literature workflows. \ No newline at end of file +Every year, researchers deposit vast quantities of data in repositories, supplementary files and national archives. Much of it is never used again. Secondary analysis of existing data can answer new questions at low cost, enable replication, combine evidence across studies and give early-career researchers access to samples they could never collect themselves. Reusing data well, however, requires skills that are rarely taught: finding suitable datasets, understanding how they were produced, judging their quality, handling them ethically and citing them properly. This guide covers each of these steps and explains how a source-linked research assistant such as [Yila](https://yila.ai) can help you trace the publications and documentation behind a dataset. + +## Why reuse data? + +The benefits of data reuse are practical and scientific. Practically, existing data save time and money, avoid burdening participants with repeated data collection, and allow research on populations or periods that are no longer accessible. Scientifically, reuse supports reproducibility, allows pooled analyses with greater statistical power, and makes it possible to test ideas that the original researchers did not consider. + +Funders and journals increasingly require data sharing, and many now encourage or expect applicants to check whether existing data could answer their question before funding new collection. Demonstrating that you have considered reuse can strengthen a proposal. + +## Start with a question, then look for data + +It is tempting to find an interesting dataset and then look for something to do with it. That approach can work for exploratory projects, but it risks fishing for results. A better approach is to start with a clear research question and the variables you would need to answer it, then search for data that contain them. + +Write down: + +- the population and time period of interest; +- the key exposures, outcomes and covariates; +- the level of detail needed, such as individual records or aggregated statistics; +- any minimum sample size for the analysis you plan. + +This specification will help you judge whether a candidate dataset is fit for purpose. + +## Where to find data + +Data are scattered across many types of sources: + +- **General repositories** such as Zenodo, Figshare, Dryad and the Open Science Framework, which host data from all disciplines. +- **Disciplinary repositories** for specific fields, for example genomic, social science, ecological or astronomical archives. +- **National data services** that curate large surveys, censuses and administrative data, often with controlled access. +- **Supplementary materials** attached to journal articles. +- **Data papers** in journals dedicated to describing datasets in detail. +- **Data search engines** that index metadata across many repositories. + +Literature searching is also a route to data. Papers that report analyses often describe where their data can be obtained. Searching for your topic together with terms such as "dataset", "cohort" or "survey" can reveal data sources. Research assistants that link statements to their sources can help you move from a claim in a paper to the dataset behind it, but always confirm access details in the original publication or repository record. + +## Understand how the data were produced + +Before using a dataset, learn as much as possible about its origins. Key documents include: + +- the codebook or data dictionary, defining each variable and its coding; +- the study protocol or methods paper; +- information about sampling, response rates and weighting; +- any known errors, revisions or version history; +- licences and terms of use. + +Read the original publications that used the data. They often explain decisions that the documentation does not, such as how missing values were handled or why certain variables changed between waves. If the documentation is poor, contact the data creators; many are happy to answer questions. + +## Evaluate fitness for purpose + +A dataset may be excellent for its original purpose and unsuitable for yours. Ask: + +- **Coverage:** does the dataset include the population, time period and geography you need? +- **Measurement:** do the variables measure what you need, in a valid and reliable way? A proxy measure may introduce bias. +- **Completeness:** how much data are missing, and is missingness likely to be related to your outcomes? +- **Sample size:** is there enough power for your analysis, particularly for subgroups? +- **Linkage:** if you need to combine datasets, are there common identifiers and permission to link? +- **Quality:** were data checked and cleaned, and are there known problems? + +Document your assessment. It will be useful when you write the methods and limitations sections, and it helps reviewers understand your choices. + +## Handle data ethically and legally + +Reuse does not remove ethical obligations. Check whether your use is consistent with the consent participants gave and with the licence or data use agreement. Some data can be downloaded freely; others require an application, ethical approval, training or secure access environments. + +Protect confidentiality. Even de-identified data can sometimes be re-identified when combined with other sources. Follow the data provider's rules about storage, sharing and publication of small cell counts. If your institution requires ethics review for secondary analysis, obtain it before starting. + +Respect the work of the original creators. Some data providers ask to be informed of publications or offer co-authorship when substantial collaboration is involved. Read the terms carefully. + +## Plan the analysis before you look at the outcomes + +Secondary analysis is vulnerable to researcher degrees of freedom: with so many variables available, it is easy to try many analyses and report the ones that work. Protect against this by writing an analysis plan before you examine the outcomes, and consider pre-registering it. Registries increasingly accept plans for secondary data analysis, with a statement of what you already know about the data. + +Be explicit about which analyses were planned and which were exploratory. Exploratory work is valuable, but readers should know which is which. + +## Cite data properly + +Datasets are research outputs and deserve citation just like articles. A good data citation includes the creators, year, title, version, publisher or repository, and a persistent identifier such as a DOI. Cite the specific version you used, because datasets are often updated. + +Also cite the publications that describe the data, as requested by the data providers. Data citation gives credit to creators, helps others find the data and supports reproducibility. Many repositories provide a recommended citation; use it. + +## Make your own work reusable + +If your analysis produces derived data or code, consider sharing them where the original licence allows. Share code for cleaning and analysis so that others can reproduce your results. Document derived variables carefully. In doing so you contribute to the same ecosystem that made your research possible. + +## Common pitfalls + +- **Ignoring the documentation:** misunderstanding coding schemes or survey weights can produce wrong results. +- **Assuming quality:** published data can contain errors; check distributions and outliers. +- **Overlooking licences:** using data in ways the licence does not permit. +- **Fishing:** running many analyses and reporting only significant findings. +- **Missing citations:** failing to credit the data creators. + +## Conclusion + +Reusing research data is an efficient and responsible way to generate new knowledge, but it demands care at every step. Start with a question, search widely, read the documentation, assess fitness for purpose, follow ethical and legal requirements, plan your analysis and cite the data properly. Tools such as Yila can help you trace claims and datasets back to their sources, but understanding the data and judging their suitability remain the researcher's responsibility. Done well, secondary analysis extends the value of every dataset and honours the participants and researchers who made it possible.