Add research workflow guide

This commit is contained in:
yila-research-ec85 2026-10-20 13:53:57 +02:00
parent 558c2b6d3b
commit 4a807cc592

View file

@ -1,3 +1,75 @@
# citation-chasing-guide
# Citation Chasing: Backward and Forward Searching for Literature Reviews
Backward and forward citation chasing for systematic and scoping reviews
Database searching finds most of the evidence for a review, but rarely all of it. Studies get indexed under unexpected terms, older work predates current vocabulary, and some relevant reports sit in journals that the main databases do not cover. Citation chasing, also called snowballing or citation searching, fills these gaps by following the links between papers. This guide explains backward and forward chasing, how to run them systematically, how to report them, and where tools help. It also mentions how an AI research assistant such as [Yila](https://yila.ai) can support organizing the results, while decisions about inclusion remain with the reviewers.
## Two directions
**Backward chasing** means reading the reference lists of included studies and relevant reviews to find earlier work. It is the oldest form of citation searching and the easiest to explain: if a study is relevant, the work it cites is a good place to look for more.
**Forward chasing** means finding later work that cites an included study. This became practical with citation indexes and is now available in several free and commercial tools. Forward chasing is particularly useful for finding follow-up publications, secondary analyses and newer studies that use a different vocabulary from the original.
Both directions can be repeated. Chasing the references of newly found studies is sometimes called iterative or multi-generation snowballing. Most reviews stop after one round, but iterating can matter in fields with poor indexing.
## Why it finds what searches miss
Search strategies depend on words. If a concept was described differently in 1995 than in 2020, or differently in nursing than in economics, a single search string will miss part of the literature. Citations do not depend on vocabulary. A paper cites another because the authors considered it relevant, whatever words either one uses. That makes citation chasing a useful complement to term-based searching, especially for interdisciplinary topics, methodological questions and complex interventions.
Evaluations of review methods have repeatedly found that a share of included studies, sometimes small and sometimes substantial, is identified only through reference lists or citation tracking. The proportion varies by field, which is a good reason to plan citation chasing from the start rather than treating it as an optional extra.
## Choosing seed papers
The papers you chase from are called seeds. Common choices are:
- all studies included after full text screening;
- key systematic reviews on the same or neighbouring questions;
- highly cited methodological or conceptual papers that define the topic.
Using every included study as a seed is the most defensible approach for a systematic review. For a scoping review with a very large number of included studies, a documented subset may be reasonable, as long as the selection rule is stated in the methods.
## Running backward chasing
Collect the reference lists of all seeds. Many databases and publisher pages offer exports; otherwise reference managers can often parse references from a PDF. Deduplicate the combined list against everything already screened in the main search. What remains goes through the same title and abstract screening as database records, using the same criteria and, ideally, the same duplicate screening process.
Record how many references were collected, how many were duplicates of records already screened, how many were screened and how many were included. These numbers feed the "other methods" side of the PRISMA flow diagram.
## Running forward chasing
Forward chasing needs a citation index. Options include commercial indexes, free scholarly graphs and search engines that show "cited by" counts. Each source covers a different set of publications, so results differ. Record which source you used and the date, because citation counts grow continuously and a forward search run in March will not match one run in September.
Export the citing records, deduplicate them against the main search and the backward chasing results, and screen them in the usual way.
## Practical tips
**Do it after full text screening.** Chasing from studies that will later be excluded wastes effort. Wait until the included set is stable, then chase.
**Use consistent identifiers.** DOIs and other persistent identifiers make deduplication across sources far more reliable than matching on titles.
**Keep the screening blind where you can.** Reviewers screening chased records should apply the same criteria as for database records, without assuming that a cited paper is more likely to be relevant.
**Watch for citation bias.** Papers with positive or striking results tend to be cited more often. Citation chasing therefore leans towards well-cited work. That is another reason it complements, rather than replaces, structured database searches.
**Set a stopping rule.** If you iterate, state in advance when you will stop, for example when a round yields no new included studies.
## Reporting
PRISMA 2020 and the PRISMA-S extension for search reporting ask reviewers to describe citation searching clearly. A good methods paragraph states which seeds were used, which direction or directions were chased, which tools or indexes were used, the dates, how many rounds were run, and how results were deduplicated and screened. The flow diagram should show records identified through citation searching separately from database records.
The TARCiS statement offers further guidance specifically on terminology, conduct and reporting of citation searching, and is worth consulting when planning the methods.
## Where tools help
Several tools now automate parts of citation chasing by taking a list of DOIs and returning both references and citing papers, already deduplicated. These save hours on large reviews. Reference managers remain useful for storing and tagging the results. An AI research assistant can help in adjacent ways: summarizing what a cluster of citing papers is about, helping draft the methods paragraph from your notes, or keeping track of which seeds have been processed. A tool like Yila fits this organizational role. The screening judgments, and the record of how many records came from which source, should still come from the review team's own documented process.
## A short checklist
1. Decide in the protocol whether you will chase backward, forward or both.
2. Define seeds and any selection rule.
3. Choose and record citation sources and dates.
4. Deduplicate against all previously screened records.
5. Screen with the same criteria and process as the main search.
6. Record counts at every step for the flow diagram.
7. Report tools, dates, rounds and stopping rules.
## Summary
Citation chasing exploits the network of references that links research together, finding studies that vocabulary-based searches miss. Run it after the included set is stable, apply the same screening standards as for database results, record every count, and report the method transparently. Done this way, it strengthens the completeness of a review at modest cost.