Introduction

In a previous post, I looked at how House Representatives referred to the 45th US President on Twitter during the 116th Congress. Variation was super interesting, and demonstrated (in part) that party affiliation influenced how House members referred to 45. Namely, Republicans more frequently used referring expressions that highlighted the status of the presidency, eg, “POTUS” and “President Trump”, while Democrats more frequently used referring expressions that downplayed the status of the presidency, eg, “Trump”, the “Trump Administration”, and “Donald Trump”.

So, is this pattern a Trump Presidency anomaly, or is this par for the course? In other words, do lawmakers (or parties in the aggregate) change how they refer to sitting US Presidents depending on whether or not they are supportive of the President/President’s policies (ie, share party affiliation)? To investigate, we consider a collection of e-newsletters from US lawmakers circa 2009 to present day, spanning seven congresses and three presidencies. Results are intuitive, and fairly comparable to previous post findings.

Data set

The data set used here is a collection of e-newsletters sent out by US Lawmakers from 2009 onward, made available at DCinbox. Roughly 150K newsletters amounting to ~100 million words.

library(tidyverse)
setwd(ldir)
cpe <- read.csv('dcinbox_export.csv')
cpe <- janitor::clean_names(cpe)
cpe$doc_id  <- 1:nrow(cpe)
cpe$text <- cpe$body

Referring to American Presidents

So, lots of ways to refer to American presidents. Example (1) below presents a generic cline based on American English honorifics/political convention – expressions to the left of the cline are more deferential to the status of the office; those to the right less so. We can abide by these conventions, or flout them – both communicate. Example (2) illustrates this cline using as an example America’s current president.

1. President_FIRST_LAST > President_LAST > the president > FIRST_LAST > LAST_Administration > LAST

2. President Joe Biden > President Biden > the president > Joe Biden > the Biden Administration > Biden

The inclusion of “Biden Administration” (generically as “LAST_Administration”) is perhaps odd, as it more generally refers to the executive branch. But it is an interesting variant, and one that has become more frequent among US lawmakers. Note: “POTUS”, “Mr. President”, and (eg) “Mr. Biden” are not especially well-attested in the e-newsletter corpus – so, they are not included here.

Below, we construct a dictionary of referring expressions for the last three American presidents. This allows us to make referring expressions generic across administrations, as well as filter references to only sitting presidents. In other words, we want to ignore references to former President Obama, eg, during the Trump presidency, eg.

presidents <- c('Barack Obama', 'Donald Trump', 'Joe Biden')
lnames <- gsub('^.* ', '', presidents)
            
refs <- list(President_FIRST_LAST = paste0('President ', presidents),
             President_LAST = paste0('President ', gsub('^.* ', '', presidents)),
             FIRST_LAST = presidents,
             LAST = lnames,
             LAST_Administration = c(paste0(lnames, ' Administration'),
                                     paste0(lnames, "'s Administration")),
             the_president = 'the president')

congs <- list(OBAMA = c(111:114),
              TRUMP = c(115:116),
              BIDEN = c(117),
              GENERIC = c(111:117))

pres <- c(rep(toupper(refs$LAST), 6), 'GENERIC')

pres_forms <- cbind(pres, stack((refs))) %>%
  rename(token = values, gen = ind) %>%
  left_join(stack(congs), by = c('pres' = 'ind')) %>%
  rename(congress = values) %>%
  mutate(token = toupper(gsub(' ', '_', token)))

presx0 <- gsub('_', ' ', unique(pres_forms$token))

pres_forms %>% head() %>% knitr::kable()

Term extraction

The workflow for identifying/extracting relevant referring expressions is summarized as follows: tokenize each newsletter, combine tokens comprising multi-word referring expressions, filter tokens to only those included in our dictionary, and cast tokens to a data frame.

text_fn <- function(x){
    
    x2 <- text2df::tif2token(x1) 
    x3 <- text2df::token2mwe(tok = x2,
                             mwe = subset(presx0, grepl(' ', presx0)))
    
    x3 <- quanteda::as.tokens(x3)
    x4 <- quanteda::tokens_select(x3,
                                  pattern = gsub(' ', '_', presx0),
                                  selection = 'keep')
    text2df::token2df(x4)  
    }

We can then distribute this workflow across multiple cores:

batches <- split(cpe[, c('doc_id', 'text')], 
                 ceiling(seq_along(cpe$doc_id)/10000)) 

####
clust <- parallel::makeCluster(7)
parallel::clusterExport(cl = clust, 
                        varlist = c('batches', 'presx0'),
                        envir = environment())
anno2 <- pbapply::pblapply(X = batches,
                           FUN = text_fn,
                           cl = clust)
parallel::stopCluster(clust)
#####

pdata <- anno2 %>%
  bind_rows() %>%
  mutate(token = toupper(token)) %>%
  group_by(doc_id, token) %>%
  count() %>% ungroup()

Historical reference to the US President

pdata0 <- pdata %>%
  mutate(doc_id = as.integer(doc_id)) %>%
  left_join(subset(cpe, 
                   select = c(doc_id, 
                              chamber, 
                              congress, 
                              bio_guide_id, 
                              party))) %>%
  left_join(pres_forms) %>%
  filter(!gen %in% c('POTUS', 'Mr_LAST')) %>%
  filter(!is.na(pres)) 

Here, we compute the relative frequency with which each US lawmaker uses each referring expression per congress.

pdata1 <- pdata0 %>%
  group_by_at(vars(-doc_id, -n, -token)) %>% 
  summarize(n = sum(n)) %>%
  group_by(chamber, congress, bio_guide_id) %>% 
  mutate(pres_ref = sum(n)) %>%
  ungroup() %>%
  mutate(per = round(n/pres_ref, 3)) %>%
  left_join(pres_forms %>% filter(pres != 'GENERIC') %>%
              select(congress, pres) %>% distinct() %>% rename(label = pres)) %>%
  filter(party != 'Independent') 

As an example, relative frequencies are presented below for Marsha Blackburn – Republican Senator from Tennessee – during the 117th Congress. The junior Senator has referenced the 46th POTUS n = 157 times in her e-newsletters. She has referred to 46 as simply “Biden” a plurality (33.1%) of the time; she uses the expression “Biden Administration” in 24.2% of total references. In only 17.8% of occasions does the Senator refer to 46 as “President Biden”.

pdata1 %>%
  filter(bio_guide_id == 'B001243' & congress == 117) %>%
  select(gen:per) %>%
  knitr::kable()

In contrast, Senator Blackburn referred to the 45th POTUS as “President Trump” in 62.7% of references-to-president during the 116th Congress.

pdata1 %>%
  filter(bio_guide_id == 'B001243' & congress == 116) %>%
  select(gen:per) %>%
  knitr::kable()

pdata2 <- pdata1 %>%
  group_by(chamber, congress, party, gen, label) %>%
  summarize(per = median(per)) %>% ungroup() 

The plot below illustrates the median relative frequency for each referring expression by congress (111-117) and party – for members of the House.

pdata2 %>%
  filter(chamber == 'House') %>%
  ggplot(aes(y = per, 
             x = congress, 
             color = party
             )) + 
  
  ggthemes::scale_color_stata() +
  geom_point(aes(y = per, 
                 x = congress, 
                 color = party,
                 shape = label
             ), size = 2.5) + # 
  geom_line() +
  #theme(axis.text.x = element_text(angle = 90, hjust = 1))+
  scale_x_continuous(breaks=seq(111, 117, 1))+
  ggtitle('Referring to the sitting US President: House Representatives') +
  facet_wrap(~gen) #geom_smooth(se = T) 

An aggregate perspective

A slightly different perspective is presented in the plot below, in which overall mean prevalences by party and congress for all US lawmakers are summarized. Perhaps not surprisingly, a quick glance at these profiles very quickly sheds light on the political affiliation of the sitting President per congress.

pdata0 %>%
  group_by(congress, party, gen) %>% 
  summarize(n = sum(n)) %>%
  
  group_by(congress, party) %>% 
  mutate(pres_ref = sum(n)) %>%
  ungroup() %>%
  mutate(per = round(n/pres_ref, 3)) %>%
  mutate(gen = ordered(gen, levels = c('President_FIRST_LAST',
                                       'President_LAST',
                                       'the_president',
                                       'FIRST_LAST',
                                       'LAST_Administration',
                                       'LAST'))) %>%
  filter(party != 'Independent') %>%
  ggplot(aes(x = congress, 
             y = per, 
             group = gen,
             fill = gen)) +
  geom_col(show.legend = T, 
           alpha = 0.85,
           width = .7) +
  
  scale_x_continuous(breaks=seq(111, 117, 1)) +
  facet_wrap(~party, ncol = 1)  +
  ggtitle('Referencing the sitting US President: US Lawmakers')

Summary

So, some interesting patterns in how lawmakers have referred to US Presidents over the last seven congresses. Most lawmakers use all the variants investigated here, and often the exigencies of text constraints and information flow will dictate to some extent which referring expression is used – which we have not considered. That said, when viewed in the aggregate, the patterns described here demonstrate how lawmakers exploit lexical variation (in the honorific system) to purposefully communicate stance with readers of their newsletters. And, importantly, as lawmaker support for (or stance towards) a President changes (per an election), so to does their choice of referring expressions.