In Selbosh/ref2014: Analysis of the UK Research Excellence Framework 2014

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  message = FALSE
)

library(ref2014)

There were 36 Units of Assessment (UoAs) in the REF. The names of each of these subject areas are stored in the vector UoA.

data(UoA)
cat(paste('1.', UoA), sep = '\n')

This makes it easy to look up the ID of a particular unit of assessment.

which(UoA == 'Mathematical Sciences')
grep('Math', UoA)

Or to select several related subject areas.

grep('Engineering', UoA, value = TRUE)

library(dplyr)
outputs <- tidy_outputs()
maths_results <- subset(results, `Unit of assessment number` == 10)
maths_outputs <- subset(outputs, uoa_number == 10) %>%
  mutate(row_id = row_number())

Key features in the submission data include the titles and DOIs of the research outputs. In the mathematical sciences data, there r round(100 * mean(!is.na(maths_outputs$doi)), 2)% of entries contain DOIs and r round(100 * mean(!is.na(maths_outputs$issn)), 2) have ISSNs.

## If any two entries share an ISSN, DOI or (standardised) volume title,
## then they MUST be collected into the same journal. 

# maths_long <- maths_outputs %>%
#   select(row_id, doi, volume_std, issn, isbn) %>%
#   tidyr::gather('identifier', 'value', -row_id) %>%
#   filter(!is.na(value)) # Important to avoid grouping by missingness

# library(igraph)
# connected_components <- maths_long %>%
#   select(-identifier) %>%
#   graph_from_data_frame %>%
#   clusters %>% membership %>% stack %>%
#   rename(journal_id = values, row_id = ind)

# maths_outputs2 <- maths_outputs %>%
#   merge(connected_components, by = 'row_id', all.x = TRUE)

maths_outputs2 <- cluster_outputs_by_journals(maths_outputs)

warning('Current implementation is fragile wrt column names')

After grouping submissions into journals, we have reduced the dimensionality of the data from r nrow(maths_outputs) papers to r n_distinct(maths_outputs2$journal_id) journals. There are r sum(is.na(maths_outputs2$journal_id)) submissions (r round(100 * mean(is.na(maths_outputs2$journal_id)), 2)%) that could not be assigned to a journal as they were missing volume title, DOI, ISSN and ISBN. Of these unassigned submissions, r with(maths_outputs2, sum(is.na(journal_id) & output_type == 'D')) were classified as journal articles.

Selbosh/ref2014 documentation built on Nov. 15, 2019, 4:27 a.m.