Field-weighted Citation Impact (FWCI) is a snowball metric that takes into account differences in publication type, field (or more specifically, subfield), and year of publication to help understand the citation impact of a particular publication.
It is calculated for a work as the simple ratio of citations received / citations expected in the year of publications and three following years.
To calculate the citations received (i.e., numerator) for a work, we sum the number of papers that cite that paper in the same year it was published and three years following.
To calculate the citations expected (i.e., denominator) for a work, we average the citations received in publication year and following three years for every publication with the same publication year, publication type, and publication subfield. Note: for work type = "article", we separate out journals vs. conference proceedings.
Is FWCI the same in other bibliometric databases?
Like other products, we calculate FWCI using the snowball metric recipe book. Therefore, the calculations are exactly the same. However, some differences in the underlying data exist that are worth considering when comparing to other datasets.
- Our database is more comprehensive than others and includes many works without citations---this drives down the average expected citation values so that works that do get cited are likely to have higher FWCI values in OpenAlex.
- We classify each work into a single primary subfield based on the text of the work and not the main fields of the journal it is published in (see our documentation on topics)---when works in a particular subfield are published in a journal with different primary subfields, this likely leads to differences in FWCI for that work in databases that use work-level vs. journal-level subfield classifiers.
- Our year of publication that is used is typically the date a publication is first online, but this is different from using the periodical date which can be in the year after a publication is online
Why don't all works have a FWCI value?
Not all work types are expected to receive many citations (e.g., paratext), and so when works in those types do receive citations, their FWCI can be outliers with extremely high values. Because a common use case is to find the average FWCI for an institution, these work types can have disproportionate effects. For now, we have chosen to omit those work types so that they are not included in averages and are also not zero (which would drive down averages). We may re-evaluate this in the future, particularly as dataset sharing and citing gains more momentum globally.