Title:	Measuring Discursive Sophistication in Open-Ended Survey Responses
Version:	0.1.1
Description:	A simple approach to measure political sophistication based on open-ended survey responses. Discursive sophistication captures the complexity of individual attitude expression by quantifying its relative size, range, and constraint. For more information on the measurement approach see: Kraft, Patrick W. 2023. "Women Also Know Stuff: Challenging the Gender Gap in Political Sophistication." American Political Science Review (forthcoming).
License:	GPL (≥ 3)
Encoding:	UTF-8
RoxygenNote:	7.2.3
Suggests:	testthat (≥ 3.0.0)
Config/testthat/edition:	3
Depends:	R (≥ 2.10)
LazyData:	true
Imports:	SnowballC, stm, stringr, tm, utils
NeedsCompilation:	no
Packaged:	2023-06-11 08:29:00 UTC; patrick
Author:	Patrick Kraft [aut, cre, cph]
Maintainer:	Patrick Kraft <kraft.pw@gmail.com>
Repository:	CRAN
Date/Publication:	2023-06-11 11:50:05 UTC

Cooperative Congressional Election Study 2018

Description

A subset of data from the UWM Team Content of the 2018 CCES wave. See Kraft (2023) for details.

Usage

cces

Format

cces

A data frame with 1,000 rows and 15 columns:

age

Age (in years)

female

Gender (1 = female)

educ_cont

Education level (1-6)

pid_cont

Party identification (1-7)

educ_pid

educ_cont * pid_cont

oe01-oe10

Open-ended responses

Source

https://cces.gov.harvard.edu/

Constraint Dictionary

Description

A sample of terms that signal a higher level of constraint between different considerations (combining conjunctions and exclusive words). See Kraft (2023) for details.

Usage

dict_sample

Format

cces

A data character vector with 4 elements:

conjunctions

also, and

exclusive

but, without

Compute discursive sophistication for a set of open-ended responses

Description

This function takes a data frame (data) containing a set of open-ended responses (openends) to compute the three components of discursive sophistication (size, range, and constraint) and combines them in a single scale. See Kraft (2023) for details.

Usage

discursive(
  data,
  openends,
  meta,
  args_textProcessor = NULL,
  args_prepDocuments = NULL,
  args_stm = NULL,
  keep_stm = TRUE,
  dictionary,
  remove_duplicates = FALSE,
  type = c("scale", "average", "average_scale", "product"),
  progress = TRUE
)

Arguments

Value

A list containing the measure of discursive sophistication and the underlying components in a data frame, as well as the output of stm::textProcessor(), stm::prepDocuments(), and stm::stm().

Examples

discursive(data = cces,
           openends = c(paste0("oe0", 1:9), "oe10"),
           meta = c("age", "educ_cont", "pid_cont", "educ_pid", "female"),
           args_prepDocuments = list(lower.thresh = 10),
           args_stm = list(K = 25, seed = 12345),
           dictionary = dict_sample)

Combine three components of discursive sophistication in a single scale

Description

This function combines the size, range, and constraint of open-ended responses in a single scale. See Kraft (2023) for details.

Usage

discursive_combine(
  size,
  range,
  constraint,
  type = c("scale", "average", "average_scale", "product")
)

Arguments

Value

A numeric vector with the same length as the number of rows in data.

Examples

discursive_combine(size = list(size = runif(100)), range = runif(100), constraint = runif(100))

Compute the constraint component of discursive sophistication

Description

This function takes a data frame (data) containing a set of open-ended responses (openends) and a dictionary to identify terms that signal a higher level of constraint between different considerations (usually conjunctions and exclusive words). It returns a numeric vector of dictionary counts re-scaled to range from 0 to 1. See Kraft (2023) for details.

Usage

discursive_constraint(data, openends, dictionary, remove_duplicates = FALSE)

Arguments

Value

A numeric vector with the same length as the number of rows in data.

Examples

discursive_constraint(data = cces,
                      openends = c(paste0("oe0", 1:9), "oe10"),
                      dictionary = dict_sample)

Compute the range component of discursive sophistication

Description

This function takes a data frame (data) containing a set of open-ended responses (openends) to compute the Shannon entropy in individual response lengths across items. The function returns a numeric vector of topic counts re-scaled to range from 0 to 1. See Kraft (2023) for details.

Usage

discursive_range(data, openends)

Arguments

Value

A numeric vector with the same length as the number of rows in data.

Examples

discursive_range(data = cces,
                 openends = c(paste0("oe0", 1:9), "oe10"))

Compute the size component of discursive sophistication

Description

This function takes a data frame (data) containing a set of open-ended responses (openends) and additional arguments passed to stm::textProcessor() and stm::prepDocuments() to estimate a structural topic model via stm::stm(). The results of the the structural topic model are used to compute the relative number of topics raised in each open-ended response. The function returns a numeric vector of topic counts re-scaled to range from 0 to 1. See Kraft (2023) for details.

Usage

discursive_size(
  data,
  openends,
  meta,
  args_textProcessor = NULL,
  args_prepDocuments = NULL,
  args_stm = NULL,
  keep_stm = TRUE,
  progress = TRUE
)

Arguments

Value

A list containing the size component of discursive sophistication as well as the output of stm::textProcessor(), stm::prepDocuments(), and stm::stm().

Examples

discursive_size(data = cces,
                openends = c(paste0("oe0", 1:9), "oe10"),
                meta = c("age", "educ_cont", "pid_cont", "educ_pid", "female"),
                args_prepDocuments = list(lower.thresh = 10),
                args_stm = list(K = 25, seed = 12345))

Compute number of topics based on stm results

Description

This function takes a structural topic model output estimated via stm::stm() as well as the underlying set of documents created via stm::prepDocuments() to compute the relative number of topics raised in each open-ended response. The function returns a numeric vector of topic counts re-scaled to range from 0 to 1. See Kraft (2023) for details.

Usage

ntopics(x, docs, progress = TRUE)

Arguments

Value

A numeric vector with the same length as the number of documents in x and docs.

Examples

meta <- c("age", "educ_cont", "pid_cont", "educ_pid", "female")
openends <- c(paste0("oe0", 1:9), "oe10")
cces$resp <- apply(cces[, openends], 1, paste, collapse = " ")
cces <- cces[!apply(cces[, meta], 1, anyNA), ]
processed <- stm::textProcessor(cces$resp, metadata = cces[, meta])
out <- stm::prepDocuments(processed$documents, processed$vocab, processed$meta, lower.thresh = 10)
stm_fit <- stm::stm(out$documents, out$vocab, prevalence = as.matrix(out$meta), K=25, seed=12345)
ntopics(stm_fit, out)

Compute Shannon entropy

Description

Internal function to compute Shannon entropy in relative word counts across a set of elements in a character vecotr. Entropy is re-scaled to range from 0 to 1. Function used in discursive_range().

Usage

oe_shannon(x)

Arguments

Value

Numeric vector with the same length as x.