Field-Weighted Citation Impact (FWCI)

• Field-Weighted Citation Impact (FWCI) is a field-normalized indicator comparing a paper’s actual citations to the expected citations for similar publications based on subject, year, and document type.
• It standardizes citation performance across diverse disciplines, facilitating equitable research evaluation at article, author, and institutional levels.
• Computed using Scopus/SciVal data, FWCI plays a central role in bibliometrics despite debates over aggregation methods and potential metric-driven incentives.

Field-Weighted Citation Impact (FWCI) is a field-normalized citation indicator that compares the citations received by a publication with the expected citations for similar publications, where similarity is defined by subject field, publication year, and document type. In Scopus/SciVal usage, FWCI answers the question “Compared to similar papers worldwide, how heavily is this output cited?”; a value of 1.0 corresponds to the world average, values above 1.0 indicate above-average citation impact, and values below 1.0 indicate below-average impact (Ismailov, 26 Sep 2025). Although FWCI is fundamentally an article-level metric, it is routinely aggregated to authors, departments, institutions, countries, journals, and funding portfolios, making it central to contemporary bibliometrics and research evaluation (Hajikhani et al., 16 Jun 2026).

1. Definition and normalization logic

At its core, FWCI is the ratio of actual citations to expected citations for a comparable reference set. In its standard form,

$$\mathrm{FWCI}_i = \frac{C^{\mathrm{obs}}_i}{C^{\mathrm{exp}}_{f(i),\,y(i),\,d(i)}},$$

where $C^{\mathrm{obs}}_i$ is the citations actually received by publication $i$, and $C^{\mathrm{exp}}_{f(i),\,y(i),\,d(i)}$ is the expected citations for publications in the same subject field, publication year, and document type (Hajikhani et al., 16 Jun 2026). Equivalent formulations appear across the literature as $C_{\text{observed}}/C_{\text{expected}}$ or $c/c_0$, and they are treated as the same normalization idea in Scopus/SciVal, Clarivate’s CNCI, and related indicators (Leydesdorff et al., 2017).

The “field-weighted” part is essential because citation practices differ systematically among fields. Biomedical sciences tend to have much higher citation densities than pure mathematics, and long reference lists in one domain do not mean that raw citation counts are substantively comparable to those in another (Ismailov, 26 Sep 2025). Empirical work using inverse-probability-of-treatment weighting shows that field differences in citation rates remain even after balancing for co-authors, pages, references, journal impact factor, and other citation-influencing factors, which is why field-normalization is treated as a prerequisite for citation analysis rather than an optional refinement (Bornmann et al., 2019).

FWCI belongs to a broader family of field-normalized citation indicators. In the terminology used in the literature, it is mathematically aligned with the average normalized citation score or mean normalized citation score framework, in which each publication receives a normalized score $c_i/e_i$ and a set-level score is the average of these per-paper ratios (Waltman, 2015). This places FWCI within the same conceptual lineage as the CWTS “crown indicator,” although the latter is classically formulated as a ratio of summed actual citations to summed expected citations rather than as an average of per-paper ratios (Moed, 2010).

2. Computation, databases, and aggregation

FWCI is computed on top of Scopus data, typically via SciVal. For a given paper, the platform identifies a comparison set defined by the same subject area, publication year, and document type; computes the expected citations as the average citation count in that comparison set; and divides the paper’s actual citations by that expected value (Ismailov, 26 Sep 2025). In the FT50 journal study, the relevant field classification is Scopus’ All Science Journal Classification (ASJC), and “similar” explicitly means same ASJC field, same year, and same document type (Hajikhani et al., 16 Jun 2026).

Interpretation is standardized across studies. An FWCI of 1.0 means exactly world average citation impact for the relevant field-year-type cell; FWCI $>1.0$ means above-average impact, and FWCI $<1.0$ means below-average impact (Nazarovets, 2024). The same conceptual structure underlies CNCI, but with Web of Science categories and data rather than Scopus, which is why FWCI and CNCI are often described as parallel implementations in different commercial database ecosystems (Ismailov, 26 Sep 2025).

Although defined at article level, FWCI is frequently aggregated as the arithmetic mean of article-level scores:

$$\mathrm{FWCI}_{\text{set}} = \frac{1}{N}\sum_{i=1}^{N} \mathrm{FWCI}_i.$$

This mean-based aggregation is used for authors, institutions, countries, journals, and funding portfolios (Ismailov, 26 Sep 2025). The FT50 study, for example, computes article-level FWCI for each paper and then aggregates to the journal level by taking the mean FWCI over the 2005–2019 publication window (Hajikhani et al., 16 Jun 2026). The GSR ranking does the same at the venue level for 2022–2024 Article/Review papers, using the arithmetic mean of paper-level FWCI values (Yu, 17 May 2026).

The literature summarized here does not report a single invariant operational citation window. One study states that SciVal’s FWCI uses a four-year accrual window, while the Subaru Telescope study reports FWCI “calculated based on citations received within a fixed three-year window following publication” (Hajikhani et al., 16 Jun 2026). This suggests that reported implementations or analytical extractions may differ across studies or platform configurations. What remains constant across these accounts is the normalization by field, year, and type.

3. Institutional uses and evaluative roles

FWCI is widely used as an assessment device because it promises cross-field comparability. Institutions and agencies use it to identify high-impact papers, to classify outputs as above or below world average, and to compare authors, departments, or countries using a common normalized scale (Ismailov, 26 Sep 2025). In this sense, FWCI functions as a compact representation of relative scholarly visibility within a defined citation ecology.

Different studies show FWCI embedded in distinct assessment regimes:

Context	Role of FWCI	Reported finding
SASHE in Azerbaijan	Official scoring of overall FWCI/CNCI, top-10% papers, Q1/Q2 journals	FWCI/CNCI carry the greatest weight (Ismailov, 26 Sep 2025)
Ukrainian university ranking diagnostics	Normalized check on hump-forming papers near the h-core	Kremenchuk: 3.35; Dnipro: 4.32 (Nazarovets, 2024)
FT50 journals	Core indicator of scholarly influence	Span from about 8.4 to about 1.16 (Hajikhani et al., 16 Jun 2026)
Subaru Telescope study	Facility-level impact evaluation	Subaru papers exceeded FWCI 2.0 (Fujiwara, 26 Aug 2025)
Science Foundation Ireland awards	Funding-portfolio decomposition	Fitted mean FWCI 1.433 (Colgáin, 5 Feb 2026)
GSR venue ranking	Composite ranking term with weight 0.35	FWCI and IF2 together account for 70% of score (Yu, 17 May 2026)

The SASHE case is particularly explicit. Researchers are scored using overall FWCI/CNCI, the number of papers in the top 10% by FWCI/CNCI, and the number of publications in Q1/Q2 journals; FWCI and CNCI carry the greatest weight, making them central to hiring, promotion, and funding decisions (Ismailov, 26 Sep 2025). The FT50 study likewise treats FWCI as one of the three indicators constituting the scholarly impact dimension, alongside Top-10% rate and Field-Weighted Views Impact, and uses z-scored FWCI in a citation-impact composite (Hajikhani et al., 16 Jun 2026).

FWCI is also used diagnostically rather than purely evaluatively. Nazarovets used SciVal-derived FWCI, rank-citation curves, and self-citation shares to investigate anomalous h-index patterns in the Ukrainian Scopus ranking. The metric alone was not treated as decisive, but clusters of papers with very high FWCI values, humped curves, and unusually high self-citation rates were taken to “raise significant concerns” about the causes of the anomalies (Nazarovets, 2024).

Infrastructure and funding studies use FWCI to assess portfolio-level academic return. In the Subaru Telescope bibliometric analysis, Subaru-based publications from Japan, while accounting for less than 10% of Japan’s astronomy output, consistently achieved FWCI values above 2.0 and a much higher share of top-10% papers, which was interpreted as enhanced research visibility and impact (Fujiwara, 26 Aug 2025). In the SFI decomposition, 3,243 publications were analyzed through their FWCI values to assess whether a funding regime oriented toward socioeconomic impact still maintained strong academic interest (Colgáin, 5 Feb 2026).

4. Statistical properties and methodological debates

A central methodological issue is that citation distributions are highly skewed. Waltman’s review emphasizes that normalized indicators such as MNCS, and by direct analogy FWCI, are built on means even though citation distributions are extremely skewed and only a few papers receive very many citations (Waltman, 2015). Leydesdorff and Bornmann make the same point at the journal level: the classical impact factor is problematic partly because one uses the mean in a domain where distributions are not symmetric (Leydesdorff et al., 2010).

The SFI decomposition pushes this critique directly onto FWCI. There, FWCI values for SFI-funded publications are modeled as approximately lognormal after excluding publications with $C^{\mathrm{obs}}_i$0, with fitted parameters $C^{\mathrm{obs}}_i$1 and $C^{\mathrm{obs}}_i$2, implying a mean $C^{\mathrm{obs}}_i$3 (Colgáin, 5 Feb 2026). Because a lognormal distribution has mode $C^{\mathrm{obs}}_i$4 median $C^{\mathrm{obs}}_i$5 mean, the paper argues that FWCI normalized to a mean of 1 does not make FWCI $C^{\mathrm{obs}}_i$6 a “typical” paper. On the contrary, for realistic parameter values the mean corresponds to about the 72nd percentile, so an FWCI of 1 is not the median paper but a relatively highly cited one (Colgáin, 5 Feb 2026). One common misconception is therefore that “FWCI = 1” means “typical paper”; in the distributional sense discussed in that study, it means world-average mean performance, not median performance.

A second debate concerns aggregation. The common set-level FWCI is the average of per-paper ratios, whereas the CWTS “crown indicator” classically used a globalized ratio of sums,

$C^{\mathrm{obs}}_i$7

rather than

$C^{\mathrm{obs}}_i$8

Moed defends the globalized ratio as a measure of “oeuvre impact” for a coherent research group, while Waltman notes that contemporary practice generally prefers the average-of-ratios framework (Moed, 2010). FWCI, as typically reported, therefore captures average normalized impact per publication rather than normalized impact of the oeuvre as a whole.

A third debate concerns how field-normalization should be implemented. Source-normalization by fractional counting weights citations according to the reference-list length of the citing document, thereby reducing between-field variance in impact factors by approximately 81% in one empirical analysis, yet that same analysis concludes that fractional citation distributions cannot reliably classify specialties (Leydesdorff et al., 2010). This is relevant because FWCI relies on an external field taxonomy. Waltman’s review repeatedly notes that the quality of any field-normalized indicator depends on the quality of the field classification used to define the expected citation baseline (Waltman, 2015).

More recent work has proposed alternatives that model full citation distributions rather than dividing by a field mean. A Bayesian approach based on a Poisson–gamma mixture yields a Bayesian Impact Score,

$C^{\mathrm{obs}}_i$9

which is increasing and concave in citations and decreasing with age, thereby making the time dimension explicit in a way FWCI does not (Gómez-Déniz et al., 2024). This suggests that ratio-based normalization by a field mean is not the only mathematically coherent way to produce cross-field comparability.

5. Limitations, incentive effects, and controversies

The strongest criticisms of FWCI concern how it behaves once turned into a high-stakes performance target. Because overall FWCI is an average, every additional paper with FWCI below the current average reduces the aggregate score. The canonical illustration is the contrast between two researchers: Researcher A has ten papers, three with FWCI $i$0 and seven with FWCI $i$1, yielding

$i$2

whereas Researcher B has only the three coauthored papers with FWCI $i$3, giving

$i$4

Under FWCI-based evaluation, B appears much stronger than A despite A’s greater activity and broader contribution profile (Ismailov, 26 Sep 2025). The arithmetic mean therefore penalizes productivity whenever additional outputs are slow to attract citations.

This feature interacts badly with slow-burning work. The critique is explicit for deep theoretical research, especially in pure mathematics, where newly published papers in venues such as Annals of Mathematics or Acta Mathematica may initially have FWCI $i$5 simply because citations accumulate slowly. In the SASHE scoring scheme, such a paper contributes little via journal rank, contributes nothing to the top-10% segment until its FWCI becomes high, and lowers the average FWCI/CNCI that dominates the total score (Ismailov, 26 Sep 2025). The paper’s Azerbaijani mathematician scenario makes the distortion concrete: publication of a deep Q1 Annals paper can cause a researcher’s total evaluation score to fall.

Once careers depend on FWCI, the rational responses are metric-optimizing behaviors: avoiding topics that take time to gain recognition, focusing on subfields where citations accumulate rapidly, delaying or abandoning citation-risky projects, coauthoring strategically within citation networks, preferring journals with rapid review and high citation rates, self-citing systematically, and even publishing corrigenda or similar small items to generate additional citations (Ismailov, 26 Sep 2025). The critique is not that these behaviors are aberrant; it is that they are logical responses to a metric-heavy incentive system.

Evidence from anomaly detection studies reinforces the concern. In Nazarovets’ analysis of Ukrainian university rankings, hump-forming paper clusters at two universities had FWCI values of 3.35 and 4.32, but also self-citation shares of 53.4% and 41.5%, respectively. The author does not claim that FWCI proves manipulation; rather, high FWCI, humped rank-citation curves, and high self-citation percentages together “raise significant concerns” and require expert evaluation (Nazarovets, 2024). A related caveat appears in the FT50 study, which notes that self-citations had not been stripped from the indicators, including FWCI (Hajikhani et al., 16 Jun 2026).

Another controversy is whether FWCI captures genuine influence or merely citation-rich visibility. In the FT50 analysis, scholarly impact as measured by FWCI is largely independent of policy and patent influence: the journal-level Spearman correlation between FWCI and policy citations per publication is $i$6, between FWCI and patent citations per publication is $i$7, and nearly half of the journals change quartile when policy and patent indicators are added to the ranking (Hajikhani et al., 16 Jun 2026). High FWCI, then, does not imply broad societal reach.

A further distortion has been argued in the context of “grey” open-access publishing in computer science. That study does not compute FWCI directly, but argues that if large volumes of lower-influence publications from MDPI, Frontiers, Hindawi, and IEEE Access enter the same field baselines as traditional venues, expected citation values can be depressed, thereby distorting FWCI for both grey and traditional publishers (Cunningham et al., 2024). This suggests that field-normalization can itself be sensitive to structural changes in the indexed publication ecosystem.

6. Related indicators, alternatives, and responsible use

FWCI has several close relatives. CNCI implements the same observed/expected logic using Web of Science and Clarivate data rather than Scopus (Ismailov, 26 Sep 2025). MNCS expresses the same average-of-ratios construction in bibliometric theory (Waltman, 2015). RCR and FCR also normalize by a field-specific expected citation rate, though they define the reference set differently, with RCR using co-citation networks and FCR using the Dimensions field and age model (Gómez-Déniz et al., 2024). These indicators differ in field delineation and statistical construction, but they share the basic aim of removing structural field and time effects from citation counts.

Other alternatives move away from mean-based ratios. Percentile-based indicators, including citation percentiles, PPtop 10%, PPtop 1%, and I3/I3-N, normalize a paper’s position within a field-specific citation distribution rather than dividing by the field mean (Bornmann et al., 2019). In a convergent-validity study using F1000Prime as the peer-review baseline, PPtop 1% discriminated best among quality levels, while I3/N performed similarly to, and slightly better than, most other field-normalized indicators (Bornmann et al., 2019). Source-normalized indicators such as SNCS and fractional counting instead weight citations by the citing environment rather than by a cited-side expected mean (Leydesdorff et al., 2010).

The most consistent recommendation across the critical literature is not to abolish FWCI but to constrain its role. “Metrics Over Merit” does not propose a replacement formula; it argues that FWCI should not be treated as a proxy for merit, should be used as one indicator among many rather than the primary one, and should be combined with qualitative expert peer review, narrative CVs or portfolios, and context-specific indicators such as prizes, invited talks, software, or data contributions (Ismailov, 26 Sep 2025). The same paper recommends reducing the weight of FWCI/CNCI in institutional evaluation systems, acknowledging long citation lags and field characteristics, and aligning evaluation practices with scientific values such as depth, originality, and risk-taking (Ismailov, 26 Sep 2025).

A responsible interpretation of FWCI therefore requires several cautions. FWCI measures citation behavior, not depth, originality, or long-term importance (Ismailov, 26 Sep 2025). FWCI = 1 indicates world-average mean citation impact in a reference set, not median or typical impact (Colgáin, 5 Feb 2026). High FWCI can coexist with weak policy or technological uptake (Hajikhani et al., 16 Jun 2026). Unusual FWCI values, whether high or low, require contextual and expert interpretation rather than mechanical judgment (Nazarovets, 2024). Used cautiously, FWCI remains a technically coherent normalization device; used bluntly as a master metric, it can transform a comparative aid into a gatekeeping ratio.