Journal fees vs impact


When a research manuscript is finally ready for submission, its time for the delicate discussion of journal publication fees. Publishing ‘open-access’ (i.e., free for the public to read) attracts a higher charge than pay-per-view. In either case, the public, who in many cases paid for the research, also ultimately pay to read the results.

The strong influence that corporate publishing houses have on science, especially in biology, is due to their prestigious reputations. Highly read and highly cited journals are considered ‘high-impact’, and assigned an ‘impact factor’ (something approximating average number of citations per publication in the last year). This number is often used to judge the quality of a manuscript appearing in a journal, and affects the careers of its authors.

The impact factor system is flawed for many reasons that others have eloquently articulated, but given that the culture of impact factor-driven science still abounds, I wanted to compare the range of impact bang available for the public buck.

Here I investigate the relationship between impact factor and publication fees for a sample of Elsevier biological journals, using the web scraping and parallel processing functionality of R. ‘Scraping’ is the harvesting of data from websites by searching through the underlying html code to extract text.

Read remote-hosted data

First we use the rio() library to import an excel file of journal publication fees hosted at Elsevier.

library(tidyverse)

#thanks to https://stackoverflow.com/a/63910298

url <- "https://zenodo.org/record/1143853/files/elsevier_j.custom97.xlsx"

oa_cost <- rio::import(file= url) %>% select(1:5) %>% glimpse()
Rows: 2,761
Columns: 5
$ ISSN            <chr> "1876-2859", "1076-6332", "0001-4575", "0155…
$ `Journal Title` <chr> "Academic Pediatrics", "Academic Radiology",…
$ `OA Model`      <chr> "Hybrid", "Hybrid", "Hybrid", "Hybrid", "Hyb…
$ Currency        <chr> "USD", "USD", "USD", "USD", "USD", "USD", "U…
$ Price           <chr> "3000", "2750", "3550", "1100", "1800", "275…
names(oa_cost) <- c('isbn','title','model','curr','fee')

There are some journals with author fee exemptions which are irrelevant for the question at hand and are removed

oa_cost %>% filter(str_detect(title,'\r\n† fee not payable by author')) %>% select(title) %>% head()
                                                                                  title
1                            Acta Pharmaceutica Sinica B\r\n† fee not payable by author
2                                       Acta Sociológica\r\n† fee not payable by author
3                  African Journal of Emergency Medicine\r\n† fee not payable by author
4                             African Journal of Urology\r\n† fee not payable by author
5                          Ain Shams Engineering Journal\r\n† fee not payable by author
6 AKCE International Journal of Graphs and Combinatorics\r\n† fee not payable by author
oa_cost <- oa_cost %>% mutate(title = str_remove(title,'\r\n† fee not payable by author'))

Now to convert the journal titles to lower case, and select some biology journals using a soft match on ‘bio’,‘gene’,‘geno’ and ‘life’:

oa_cost_biol <- oa_cost %>% mutate(title_lower = tolower(title)) %>% 
  filter(str_detect(title_lower,'bio|geno|gene|life'))  


Scrape sciencedirect

To scrape the impact factor for each journal from the ScienceDirect database requires dynamically constructing the url, based on a template such as https://www.sciencedirect.com/journal/neurobiology-of-stress. The journal title will be lower case and hyphen-separated, appended after the final “/”. Based on trial and error, the string manipulation below to clean up commas, replace spaces with hyphens, and handle some edge cases, results in valid urls.

oa_for_scidir <- oa_for_url %>% 
  mutate(
  title_lower = str_replace_all(title_lower,' ','-'),
  title_lower = str_replace_all(title_lower,',',''),
  title_lower = str_replace_all(title_lower,'bba---','biochimica-et-biophysica-acta-bba-'),
  title_lower = str_replace_all(title_lower,'&','and'),
  title_lower = str_remove_all(title_lower,':')) %>%     
  filter(!str_detect(title_lower,'-part-|biologies'))


## Workshop rvest rvest is the main web scraping package in the tidyverse suite. It has handy functions to extract features from html, such as html_text() and html_table(). Searching the desired information in the website source code will help to identify the relevant html nodes. This can be tricky business but there is help available!

The solution below isn’t elegant but gets the job done for a test example.

library(rvest)

x <- "neurobiology-of-stress"

read_html(paste0("https://www.sciencedirect.com/journal/", x)) %>% 
  html_nodes('span') %>% html_text() %>% 
  as_tibble() %>%
  filter(str_detect(value,'Impact')) %>% 
  #retain only digits
  filter(str_detect(value,'\\d')) %>% 
  #add a column containing the journal name, stored in x 
  mutate(journal=x)
# A tibble: 1 x 2
  value              journal               
  <chr>              <chr>                 
1 7.197Impact Factor neurobiology-of-stress

This code is now assigned as a function called sd_scrape, and wrapped with tryCatch(), to make sure any errors in hitting the website are logged, rather than causing the job to fail completely:

sd_scrape <- function(x){
  
  tryCatch(
  #web scraping code
  read_html(paste0("https://www.sciencedirect.com/journal/", x)) %>% 
      html_nodes('span') %>% html_text() %>% as_tibble() %>%
      filter(str_detect(value,'Impact')) %>% 
      filter(str_detect(value,'\\d')) %>% 
      mutate(journal=x), 
  #if an individual site scrape fails, create a 1-row results table with 'error' and the journal name 
  error = function(c) tribble(~value, ~journal, "error", x)
    )
}


Parallelize with furrr

The next code block does a lot of heavy lifting, and will ideally only run once. At the outset is an if statement to make sure the rest of the code block runs only if the rvest output is missing. The scraping function is run using furrr::future_map(), which makes it run in parallel — ideally 4x faster than scraping each journal website one by one. I borrow heavily from this furrr tutorial.

library(furrr)

if(exists("oa_impact")){print('oa_impact exists!')}else{

# set up parallel processors
future::plan(multisession(workers = 4)) # parallelly::availableCores()

# scrape sciencedirect based on an input vector of journal titles, and store results in oa_impact
oa_impact <- future_map( oa_for_scidir$title_lower , sd_scrape) %>% 
  bind_rows() %>% 
  #remove text from value column
  mutate(value = as.numeric(str_remove(value,'Impact Factor')))
}


Join tables

Join the scraped impact factor and publication fee tables, and tag selected journals for increased plotting point size.

oa_cost_impact <- oa_impact %>% 
  left_join(oa_for_scidir, by = c('journal' = 'title_lower')) %>% 
  mutate(fee = as.numeric(fee)) %>% 
  #tag selected journals for increased plotting point size
  mutate(upsize = ifelse(journal %in% c('redox-biology','bioactive-materials',
                                        'international-journal-of-food-microbiology',
                                        'trends-in-cell-biology','current-biology'), 'y','n') )


Plot the results

This code fits a quadratic curve to the data, as well as jittering the points by $100 to handle over-plotting around x = $3000. The text aesthetic is not relevant for static ggplots, but required for interactive plotting below.

plot_scrape <- oa_cost_impact %>% 
    ggplot(aes(x = fee, y = value)) + 
    geom_point(data = . %>% filter(upsize=='n'), 
               aes(col = fee, text = title, shape = model), 
             position = position_jitterdodge(jitter.width = 100, jitter.height  =  0),
             alpha = 0.6) +
  scale_color_gradient(low = 'blue', high = 'red') +
  #ggrepel::geom_text_repel(data = . %>% filter(upsize=='y'), 
  #                         mapping = aes(label=title, col=fee)) +
  geom_point(data = . %>% filter(upsize=='y'), 
              aes(col = fee, text = title, shape = model), size=3) +
  stat_smooth(method  =  "lm", formula  =  y ~ x + I(x^2), 
              size = 1, se =  F, col = "dark blue" ) + #fits parabola
  xlab("cost USD$") + ylab("Impact Factor") +
  ggtitle("Publication Fee vs Impact Factor") +
  guides(shape = guide_legend(title = "")) # handles plotly legend overlap 

plot_scrape 


Interactive plots

The plotly() package allows conversion of static ggplots into interactive plots. The key is to include the tooltip, which will link to the text aesthetic in the ggplot call, and render the relevant text when the user hovers the mouse over a data point. ggplotly() renders legends slightly differently and can lead to crowding. This is handled with the plotly::layout() function.

Conclusions

The quadratic curve gives a reasonable fit and incorporates the points at x = $0. There is a large spread of impact factors around the line, meaning that the publication fee is a poor proxy for impact factor, and vice versa. Despite this, journals with higher impact factors tend to charge higher publication fees, and there are a good number of exceptionally expensive lower impact journals at the bottom right of the plot. Some extreme data points have been enlarged as plotly struggles with labels.
In so far as value can be measured in this context, better to be a bioactive materials or redox researcher than a food microbiologist!