rechat

This R package provides useful functions for parsing and working with data from the rechat online chat platform.

Installation

The package can be installed from this repo, using the following code:

# install.packages("devtools")
devtools::install_github("willschulz/rechat")
library(rechat)

Getting Started

To get started, create a ReChat account at https://reso.chat/ and follow the instructions for running a study.

Parsing Chat Data

When you download chat data from your ReChat study, it would be in the form of a csv file. The following code will read this file into R and convert it to an R list:

chat_data <- parseChat("path/to/downloaded_file.csv")

Featurizing and Summarizing Chat Content

Users can conduct content analysis of chats by using the featurizeChat and summarizeChat functions, which can be used to apply any function to featurize chats at the message level, and summarize those features at the participant level, respectively.

Character Count

For example, we might want to summarize participants' loquaciousness by counting the number of characters they wrote during the chat. We can augment the chat_data with a message-level character count with the following code:

chat_data <- featurizeChat(chat_data, featurization_function = nchar)

Then, we can summarize character count at the participant level, and append a char_count variable to a dataframe survey_data using the following code:

survey_data <- summarizeChat(survey_data, chat_data, chat_feature_name = "nchar", summary_function = sum, na.rm = T, summary_feature_name = "char_count")

Note that in order to merge chat data to survey data, the survey should require participants to enter their chat completion code into a free-text box, and this survey variable should be named receiptCode.

Sentiment Analysis

More complex features can be constructed by writing custom functions to use as the featurization_function. For example, the following code can be used to summarize the overall sentiment of each participant's messages:

# Custom function that generates a mean message sentiment using the sentimentr package
mean_sentiment <- function(message){
  this_sentiment <- sentimentr::sentiment(message) %>% group_by(element_id) %>% summarise(sentiment = mean(sentiment)) %>% pull(sentiment)
  return(this_sentiment)
}

# Featurize chat_data with message-level sentiment scores
chat_data <- featurizeChat(chat_data, feature_name = "sentiment", featurization_function = mean_sentiment)

# Summarize participant sentiment by taking the mean over all their messages
survey_data <- summarizeChat(survey_data, chat_data, chat_feature_name = "sentiment", summary_function = mean, na.rm = T, summary_feature_name = "mean_sentiment")

Matching Discussion Partner Data

In many analyses, it is appropriate to treat observations as clustered at the chatroom level, which requires constructing an identifier for each chatroom. The matchAlters function adds a room_id column to the survey dataframe, by matching each receiptCode in the survey data (see above) to a receiptCode in the chat data:

survey_data <- matchAlters(survey_data, chat_data)

For dyadic chats, the matchAlters function also creates a column, alter_code, that identifies each participant's partner. If the researcher can use the getAlterVars function to look up the value of each participant's partner's variables, which may be useful in analyses of, for example, persuasion based on variables measured pre-chat:

survey_data <- getAlterVars(survey_data, var_names = c("treatment", "ideo_7", "affpol_pre", "male", "PID_6"))

Note that this function is not designed for group sizes larger than 2. For larger groups, it is recommended to first develop a model of participants' effects on each other, and then reshape the data as appropriate for one's modelling approach.