Computes the Field-Normalized Citation Impact (FNCI) for each work in the corpus. FNCI normalises a work's citation count by the average citation count of all works in the same field and publication year.
Usage
sm_metric_fnci(
corpus,
classification = c("openalex_concepts", "mesh", "manual"),
min_cell = NULL,
by_type = FALSE,
call = rlang::caller_env()
)Arguments
- corpus
An sm_corpus object.
- classification
Character; the classification system used to assign works to fields. One of
"openalex_concepts"(default),"mesh", or"manual".- min_cell
NULL(default) or a positive integer. When set, a baseline needs at leastmin_cellworks with a known citation count; works in smaller cells fall back to larger baselines (see Details), and the result gains the columnsn_cellandbaseline.- by_type
Logical. If
TRUE, baseline cells are also split by document type, normalised fromworks$type: articles and reviews get cells of their own and all other types getfnci = NA. Adds the columndoc_type. DefaultFALSE.- call
Caller environment for error reporting.
Value
A tibble with one row per work and the columns work_id, field,
year, cited_by_count, field_mean (the baseline: the mean citation
count of the work's cell) and fnci. The column set depends on the
arguments: by_type = TRUE adds doc_type ("article", "review" or
NA), and min_cell adds n_cell (integer; the number of works with a
known citation count in the baseline cell) and baseline (a factor with
the levels field_year, field_year_pm1, year and none; the level
the baseline comes from). An empty corpus gives a 0-row tibble with the
same columns.
Details
FNCI is defined as:
$$FNCI_i = \frac{C_i}{\bar{C}_{f,y}}$$
Where \(C_i\) is the citation count of work i and \(\bar{C}_{f,y}\) is the mean citation count of all works in field f published in year y.
An FNCI of 1.0 means the work is cited at the corpus average of its baseline cell (by default its field and year). Values > 1.0 indicate above-average impact.
For "openalex_concepts", the highest-scoring level-0 concept from the
concepts table is used. For "mesh", MeSH headings are used: concepts
of the "mesh" vocabulary of sm_concept_vocabularies(), also when
spelled "MEDLINE" (requires MeSH data in concepts; see below). For
"manual", a field column must already exist in corpus$works.
Small cells
A cell with few works gives an unstable baseline, and a work alone in its
cell gets FNCI = 1 by construction. When min_cell is NULL and more
than 25% of the works with a citation count sit in cells with fewer than
10 such works, a message (class scimapR_fnci_small_cells) gives that
share. sm_its(), sm_did() and sm_synth() (outcome = "cnci") and
sm_metric_summary() (metric = "cnci") show it too when they derive
impact from citation counts; their own min_cell argument is passed on
here.
Fallback baselines
With min_cell set, each work gets the most specific baseline that has at
least min_cell works with a known citation count, the work itself
included:
field_year: its field and publication year;field_year_pm1: its field in the pooled years y - 1 to y + 1 (truncated at the edges of the corpus);year: all works of its publication year.
If even the year has fewer than min_cell such works, fnci and
field_mean are NA, baseline is "none" and n_cell gives the size
of the year cell. Works without a field ("unclassified") start at the
year level, and works without a year get NA throughout.
In this mode field may list several categories separated by ;, as
the Web of Science WC field does. A work belongs to the cell of each of its
categories. For each category the most specific level that reaches
min_cell is found, and the work's baseline is the average of the
category baselines at the most specific level any of them reached:
baseline names that level and n_cell is the smallest of the averaged
cells. The baselines are averaged (\(C_i / mean(e_k)\)), not the ratios
(\(mean(C_i / e_k)\)). field keeps the original string. Without
min_cell the string is one field, as before.
Document types
With by_type = TRUE, works$type is normalised: lower-cased, its first
;-separated part, trimmed. "article" covers article, journal-article,
jour and research-article; "review" covers review and review article.
Works of any other or a missing type get fnci = NA (and baseline = NA),
with one message (class scimapR_fnci_other_types). The type stays part
of the cell at every fallback level. The PubMed and Semantic Scholar
fetchers record every work as "journal-article", so their reviews count
as articles.
MeSH fields
MeSH check tags (Humans, Animals, Male, Female, Pregnancy, the age groups, common laboratory and farm animals, and the historical periods) describe who or what was studied, and common descriptors of study design, statistics and laboratory material (such as Retrospective Studies, Risk Factors, Treatment Outcome or Cells, Cultured) describe how it was studied. Neither is ever used as a field.
sm_fetch_pubmed(), sm_read_pubmed_xml() and sm_enrich_concepts()
(source = "mesh") record each heading's major-topic flag as its score:
1 for a major topic of the article (starred in PubMed) and 0 otherwise.
A work gets its highest-scoring heading; among several, the one that
most works of the corpus carry (ties in alphabetical order). A work whose
headings carry no scores, as with MeSH terms from other sources, gets its
heading that is most frequent in the corpus, with one message (class
scimapR_fnci_mesh_unscored). A work with no other headings is
unclassified.
References
Waltman, L., van Eck, N. J., van Leeuwen, T. N., Visser, M. S., & van Raan, A. F. J. (2011). Towards a New Crown Indicator: Some Theoretical Considerations. Journal of Informetrics, 5(1), 37–47. doi:10.1016/j.joi.2010.08.001
Examples
corpus <- sm_example_corpus()
fnci <- sm_metric_fnci(corpus, classification = "openalex_concepts")
#> ℹ 100% of the works sit in field-year cells with fewer than 10 works, so their
#> FNCI baselines rest on very few citation counts.
#> ℹ Set `min_cell` of `sm_metric_fnci()` to fall back to larger baselines for
#> small cells.
head(fnci)
#> # A tibble: 6 × 6
#> work_id field year cited_by_count field_mean fnci
#> <chr> <chr> <int> <int> <dbl> <dbl>
#> 1 W000000001 clinical outcomes 2023 3 10.2 0.293
#> 2 W000000002 colorectal cancer 2020 9 12.2 0.738
#> 3 W000000003 gene expression 2024 28 11.3 2.47
#> 4 W000000004 spatial transcriptomics 2020 29 27.5 1.05
#> 5 W000000005 immune checkpoint 2020 16 16 1
#> 6 W000000006 immune checkpoint 2018 16 16 1
# Fall back to larger baselines for cells with fewer than 5 works, with
# articles and reviews normalised separately
fnci <- sm_metric_fnci(corpus, min_cell = 5, by_type = TRUE)
#> ℹ 14 works of a type other than article or review get no FNCI with `by_type =
#> TRUE`.
#> ℹ Type: "letter".
table(fnci$baseline, useNA = "ifany")
#>
#> field_year field_year_pm1 year none <NA>
#> 5 34 132 15 14