Representation analysis of gender

Setups

library(tidyverse)
library(lubridate)
library(rnaturalearth)
library(wru)
source("utils/r-utils.R")
theme_set(theme_bw() + theme(legend.title = element_blank()))

Load data

Only keep articles from 2002 because few authors had gender predictions before 2002. See 093.summary-stats for more details.

load("Rdata/raws.Rdata")

alpha_threshold <- qnorm(0.975)
gender_df <- read_tsv("data/gender/genderize.tsv")

## 
## ── Column specification ──────────────────────────────────────────
## cols(
##   fore_name_simple = col_character(),
##   n_authors = col_double(),
##   genderize_sample_size = col_double(),
##   query_date = col_date(format = ""),
##   probability_male = col_double()
## )

pubmed_gender_df <- corr_authors %>%
  filter(year(year) >= 2002) %>%
  left_join(gender_df, by = "fore_name_simple")

iscb_gender_df <- keynotes %>%
  left_join(gender_df, by = "fore_name_simple")

start_year <- 1993
end_year <- 2019
n_years <- end_year - start_year
my_jours <- unique(pubmed_gender_df$journal)
my_confs <- unique(iscb_gender_df$conference)
n_jours <- length(my_jours)
n_confs <- length(my_confs)

Prepare data frames for later analyses

• rbind results of race predictions in iscb and Pubmed
• pivot long
• compute mean, sd, marginal error

iscb_pubmed <- iscb_gender_df %>%
  rename("journal" = conference) %>%
  select(year, journal, probability_male, publication_date) %>%
  mutate(
    type = "Keynote speakers/Fellows",
    adjusted_citations = 1
  ) %>%
  bind_rows(
    pubmed_gender_df %>%
      select(year, journal, probability_male, publication_date, adjusted_citations) %>%
      mutate(type = "Pubmed authors")
  ) %>%
  mutate(probability_female = 1 - probability_male) %>%
  pivot_longer(contains("probability"),
    names_to = "gender",
    values_to = "probabilities"
  ) %>%
  filter(!is.na(probabilities)) %>%
  group_by(type, year, gender) 

iscb_pubmed_sum <- iscb_pubmed %>%
  summarise(
    # n = n(),
    mean_prob = mean(probabilities, na.rm = T),
    se_prob = sd(probabilities, na.rm = T),
    # n = mean(n),
    me_prob = alpha_threshold * se_prob,
    .groups = "drop"
  )
# https://stats.stackexchange.com/questions/25895/computing-standard-error-in-weighted-mean-estimation

# save(iscb_pubmed, file = 'Rdata/iscb-pubmed_gender.Rdata')

Figures for paper

Figure 2: ISCB Fellows and keynote speakers appear more evenly split between men and women than PubMed authors, but the proportion has not reached parity.

fig_1 <- iscb_pubmed_sum %>%
  # group_by(year, type, gender) %>%
  gender_breakdown("main", fct_rev(type))
fig_1

ggsave("figs/gender_breakdown.png", fig_1, width = 5, height = 2.5, dpi = 600)
ggsave("figs/gender_breakdown.svg", fig_1, width = 5, height = 2.5)

## `summarise()` ungrouping output (override with `.groups` argument)

## # A tibble: 2 x 2
##   type                     prob_female_avg
##   <chr>                              <dbl>
## 1 Keynote speakers/Fellows           0.303
## 2 Pubmed authors                     0.277

Supplementary Figure S2

Additional fig. 1 with separated keynote speakers and fellows

fig_1d <- iscb_pubmed %>%
  ungroup() %>%
  mutate(
    type2 = case_when(
      type == "Pubmed authors" ~ "Pubmed authors",
      journal == "ISCB Fellow" ~ "ISCB Fellows",
      type == "Keynote speakers/Fellows" ~ "Keynote speakers"
    )
  ) %>%
  group_by(type2, year, gender) %>%
  summarise(
    mean_prob = mean(probabilities),
    se_prob = sd(probabilities)/sqrt(n()),
    me_prob = alpha_threshold * se_prob,
    .groups = "drop"
  ) %>%
  gender_breakdown("main", fct_rev(type2)) +
  scale_x_date(
    labels = scales::date_format("'%y"),
    expand = c(0, 0)
  )

## Scale for 'x' is already present. Adding another scale for 'x',
## which will replace the existing scale.

Mean and standard deviation of predicted probabilities

iscb_pubmed_sum %>%
  filter(gender == "probability_male") %>%
  gam_and_ci(
    df2 = iscb_pubmed %>% filter(gender == "probability_male"),
    start_y = start_year, end_y = end_year
  ) +
  theme(legend.position = c(0.88, 0.2))

## `geom_smooth()` using formula 'y ~ s(x, bs = "cs")'

## Warning: Removed 13171 rows containing non-finite values
## (stat_smooth).

Hypothesis testing

iscb_lm <- iscb_pubmed %>%
  filter(gender == "probability_female", !is.na(probabilities)) %>%
  mutate(type = as.factor(type)) %>% 
  mutate(type = type %>% relevel(ref = "Pubmed authors"))

scaled_iscb <- iscb_lm %>%
  filter(year(year) >= 2002)
# scaled_iscb$s_prob <- scale(scaled_iscb$probabilities, scale = F)
# scaled_iscb$s_year <- scale(scaled_iscb$year, scale = F)

main_lm <- glm(type ~ year + probabilities,
  data = scaled_iscb, # %>% mutate(year = as.factor(year))
  family = "binomial"
)

broom::tidy(main_lm)

## # A tibble: 3 x 5
##   term           estimate std.error statistic  p.value
##   <chr>             <dbl>     <dbl>     <dbl>    <dbl>
## 1 (Intercept)   -2.06     0.478         -4.30 1.67e- 5
## 2 year          -0.000271 0.0000320     -8.47 2.46e-17
## 3 probabilities  0.193    0.146          1.33 1.85e- 1

inte_lm <- glm(
  # type ~ scale(year, scale = F) * scale(probabilities, scale = F),
  # type ~ s_year * s_prob,
  type ~ year * probabilities,
  data = scaled_iscb, # %>% mutate(year = as.factor(year))
  family = "binomial"
)
broom::tidy(inte_lm)

## # A tibble: 4 x 5
##   term                 estimate std.error statistic  p.value
##   <chr>                   <dbl>     <dbl>     <dbl>    <dbl>
## 1 (Intercept)        -1.77      0.568        -3.11  1.85e- 3
## 2 year               -0.000291  0.0000383    -7.59  3.22e-14
## 3 probabilities      -0.992     1.29         -0.771 4.41e- 1
## 4 year:probabilities  0.0000787 0.0000846     0.930 3.52e- 1

anova(main_lm, inte_lm, test = "Chisq")

## Analysis of Deviance Table
## 
## Model 1: type ~ year + probabilities
## Model 2: type ~ year * probabilities
##   Resid. Df Resid. Dev Df Deviance Pr(>Chi)
## 1    153942     4582.8                     
## 2    153941     4582.0  1  0.86975    0.351

# mean(scaled_iscb$year)
# mean(scaled_iscb$probabilities)

The two groups of scientists did not have a significant association with the gender predicted from fore names (P = 0.18469). Interaction terms do not predict type over and above the main effect of gender probability and year.

sessionInfo()

## R version 4.0.3 (2020-10-10)
## Platform: x86_64-pc-linux-gnu (64-bit)
## Running under: Ubuntu 20.04 LTS
## 
## Matrix products: default
## BLAS/LAPACK: /usr/lib/x86_64-linux-gnu/openblas-pthread/libopenblasp-r0.3.8.so
## 
## locale:
##  [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C              
##  [3] LC_TIME=en_US.UTF-8        LC_COLLATE=en_US.UTF-8    
##  [5] LC_MONETARY=en_US.UTF-8    LC_MESSAGES=C             
##  [7] LC_PAPER=en_US.UTF-8       LC_NAME=C                 
##  [9] LC_ADDRESS=C               LC_TELEPHONE=C            
## [11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C       
## 
## attached base packages:
## [1] stats     graphics  grDevices utils     datasets  methods  
## [7] base     
## 
## other attached packages:
##  [1] gdtools_0.2.2       wru_0.1-10          rnaturalearth_0.1.0
##  [4] lubridate_1.7.9.2   caret_6.0-86        lattice_0.20-41    
##  [7] forcats_0.5.0       stringr_1.4.0       dplyr_1.0.2        
## [10] purrr_0.3.4         readr_1.4.0         tidyr_1.1.2        
## [13] tibble_3.0.4        ggplot2_3.3.2       tidyverse_1.3.0    
## 
## loaded via a namespace (and not attached):
##   [1] colorspace_2.0-0        ellipsis_0.3.1         
##   [3] class_7.3-17            rprojroot_1.3-2        
##   [5] fs_1.5.0                rstudioapi_0.12        
##   [7] farver_2.0.3            remotes_2.2.0          
##   [9] DT_0.16                 prodlim_2019.11.13     
##  [11] fansi_0.4.1             xml2_1.3.2             
##  [13] codetools_0.2-16        splines_4.0.3          
##  [15] knitr_1.30              pkgload_1.1.0          
##  [17] jsonlite_1.7.1          pROC_1.16.2            
##  [19] broom_0.7.2             dbplyr_2.0.0           
##  [21] rgeos_0.5-5             compiler_4.0.3         
##  [23] httr_1.4.2              backports_1.2.0        
##  [25] assertthat_0.2.1        Matrix_1.2-18          
##  [27] cli_2.1.0               htmltools_0.5.0        
##  [29] prettyunits_1.1.1       tools_4.0.3            
##  [31] gtable_0.3.0            glue_1.4.2             
##  [33] rnaturalearthdata_0.1.0 reshape2_1.4.4         
##  [35] Rcpp_1.0.5              cellranger_1.1.0       
##  [37] vctrs_0.3.4             svglite_1.2.3.2        
##  [39] nlme_3.1-149            iterators_1.0.13       
##  [41] crosstalk_1.1.0.1       timeDate_3043.102      
##  [43] gower_0.2.2             xfun_0.19              
##  [45] ps_1.4.0                testthat_3.0.0         
##  [47] rvest_0.3.6             lifecycle_0.2.0        
##  [49] devtools_2.3.2          MASS_7.3-53            
##  [51] scales_1.1.1            ipred_0.9-9            
##  [53] hms_0.5.3               RColorBrewer_1.1-2     
##  [55] yaml_2.2.1              curl_4.3               
##  [57] memoise_1.1.0           rpart_4.1-15           
##  [59] stringi_1.5.3           desc_1.2.0             
##  [61] foreach_1.5.1           e1071_1.7-4            
##  [63] pkgbuild_1.1.0          lava_1.6.8.1           
##  [65] systemfonts_0.3.2       rlang_0.4.8            
##  [67] pkgconfig_2.0.3         evaluate_0.14          
##  [69] sf_0.9-6                recipes_0.1.15         
##  [71] htmlwidgets_1.5.2       labeling_0.4.2         
##  [73] cowplot_1.1.0           tidyselect_1.1.0       
##  [75] processx_3.4.4          plyr_1.8.6             
##  [77] magrittr_1.5            R6_2.5.0               
##  [79] generics_0.1.0          DBI_1.1.0              
##  [81] mgcv_1.8-33             pillar_1.4.6           
##  [83] haven_2.3.1             withr_2.3.0            
##  [85] units_0.6-7             survival_3.2-7         
##  [87] sp_1.4-4                nnet_7.3-14            
##  [89] modelr_0.1.8            crayon_1.3.4           
##  [91] KernSmooth_2.23-17      utf8_1.1.4             
##  [93] rmarkdown_2.5           usethis_1.6.3          
##  [95] grid_4.0.3              readxl_1.3.1           
##  [97] data.table_1.13.2       callr_3.5.1            
##  [99] ModelMetrics_1.2.2.2    reprex_0.3.0           
## [101] digest_0.6.27           classInt_0.4-3         
## [103] stats4_4.0.3            munsell_0.5.0          
## [105] viridisLite_0.3.0       sessioninfo_1.1.1