R/data.R

In kleinschmidt/phonetic-sup-unsup: Supervised/unsupervised adaptation analysis

#' Subjects excluded from analysis (Experiment 1)
#'
#' Subjects excluded from analysis because they previously participated, or
#' because they didn't consistently classify the 0ms or 70ms VOT stimuli
#' sufficiently consistently (as extrapolated by a logistic GLM). Generated
#' automatically by \code{\link{exclusions}}.
#'
#' @format A data frame with 8 variables:
#' \describe{
#' \item{\code{subject}}{Anonymized MTurk worker identifier}
#' \item{\code{assignmentid}}{Unique ID for the excluded assignment}
#' \item{\code{lo0ms}}{GLM-extrapolated log-odds of /p/ response at 0ms VOT}
#' \item{\code{lo70ms}}{GLM-extrapolated log-odds of /p/ response at 70ms VOT}
#' \item{\code{loMinCorrect}}{The minimum log-odds of a correct response for
#'   either 0ms or 70ms}
#' \item{\code{exclude80PercentAcc}}{Logical: whether assignment is excluded
#'   based on endpoint accuracy}
#' \item{\code{submittime}}{Time assignment was submitted (from determining
#'   which assignment was submitted first for subjects with repeats)}
#' \item{\code{rank}}{Rank of this assignment based on \code{submittime}.
#'   Greater than 1 means a repeat to be excluded}
#' }
"excludes"

#' Supervised/unserupvised phonetic adaptation (Experiment 1)
#'
#' The whole dataset, parsed by \code{\link{load_and_parse}}.
#'
#' This data comes from a phonetic adaptation experiment, where each listener
#' hears a different distribution of VOTs and classifies each one as a /b/ or a
#' /p/.  The VOTs are presented in the form of /b/-/p/ minimal pair words
#' (beach/peach, bees/peas, and beak/peak).  On each trial, the subject hears a
#' word from one of these pairs with a VOT drawn from their specific
#' distribution, and click on a picture to indicate the word they heard.
#'
#' Subjects excluded from analysis are described in \code{\link{excludes}}.
#'
#' There are two, crossed conditions:
#'
#' \describe{
#' \item{Supervision}{For subjects in the 'unsupervised' condition, all trials
#'   were unlabeled, and either the /b/ or /p/ response was appropriate.  In
#'   the 'supervised' and 'mixed' conditions, some trials were labeled.  On
#'   these trials, only one of the response pictures matched the end of the
#'   word, effectively labeleing the VOT.  For instance, on an unlabeled
#'   beach/peach trial, the subject could click on a picture of a beach or a
#'   peach.  On a labeled beach/peach trial, there might be a picture of a
#'   beach and a picture of a peak, which labeled the VOT of the word as a /b/.
#'   The 'supervised' and 'mixed' conditions differed only in how the labeled
#'   trials were distributed over the different VOT values a subject heard.}
#' \item{VOT distribution}{Each subject heard one of five distributions of VOTs.
#'   All the distributions are bimodal, with the means separated by 40ms VOT,
#'   and differ only in the location: -10/30ms, 0/40ms, 10/50ms, 20/60ms, or
#'   30/70ms.}
#' }
#'
#' There are two different outcome variables:
#'
#' \describe{
#' \item{Classification}{Listeners' percepts measured by which of the two
#'   possible pictures they clicked on.  Variable \code{respCat} is a two-level
#'   factor (b or p), and \code{respP} codes this as a binary (0 or 1) value
#'   suitable for logistic regression.}
#' \item{Reaction time}{Listeners initiated each trial by clicking on a "light"
#'   in the center of the screen.  The sound file played immediately after this
#'   click.  Reaction time is coded in \code{rt} as the number of milliseconds
#'   between the start of the sound file and the response (click on picture).}
#' }
#'
#' @format A data frame with 89,244 observations of 28 variables:
#' \describe{
#' \item{\code{subject}}{Anonymized MTurk worker identifier}
#' \item{\code{assignmentid}}{Unique ID for the excluded assignment}
#' \item{\code{hitid}}{The (non-unique) ID of the HIT for this assignment}
#' \item{\code{experiment}}{Name of the experiment:
#'   'supervised-unsupervised-vot'}
#' \item{\code{accepttime}}{Time assignment was initially accepted by
#'   subject}
#' \item{\code{submittime}}{Time assignment was finished and submitted.}
#' \item{\code{block}}{The type of block for this dataset ('visworld')}
#' \item{\code{supCond}}{Factor: The supervision condition: one of 'supervised',
#'   'unsupervised', or 'mixed'}.
#' \item{\code{bvotCond}}{Factor: the location of the lower cluster of the
#'   bimodal input in ms VOT, either '0', '10', '20', '30', or '40'.}
#' \item{\code{trial}}{Trial number, starting at 0.}
#' \item{\code{stim}}{Numeric: Unique ID of the stimulus file played on that
#'   trial.}
#' \item{\code{stimfn}}{Factor: Filename of the stimulus.}
#' \item{\code{wordclass}}{Factor: Minimal pair for this stimulus, one of
#'   'BEACH', 'BEAK', or 'BEES'.}
#' \item{\code{respCategory}}{The \strong{intended} (correct) category of the
#'   response, \strong{not the listener's actual choice} (that's
#'   \code{respCat}).}
#' \item{\code{trialSupCond}}{Factor: Whether this \emph{trial} was 'supervised'
#'   (labeled) or 'unsupervised' (unlabeled).  Use \code{labeled} instead.}
#' \item{\code{targetId}}{The name of the image that the listener clicked on.}
#' \item{\code{targetPos}}{The location of the clicked image.}
#' \item{\code{clickx}}{Relative to left of experiment window.}
#' \item{\code{clicky}}{Relative to top of experiment window.}
#' \item{\code{tstart}}{System time (in ms) of trial-start click.}
#' \item{\code{tend}}{System time (in ms) of response click.}
#' \item{\code{rt}}{Response time, in ms (difference between \code{tstart} and
#'   \code{tend}.}
#' \item{\code{respCat}}{Factor: The category ('b' or 'p') of the listner's
#'   response (based on \code{targetId}).}
#' \item{\code{trueCat}}{Factor: The intended ('correct') category ('b' or 'p').
#'   Depends on the input distribution (\code{bvotCond}).}
#' \item{\code{vot}}{Numeric: The VOT of the stimulus for this trial (ms).}
#' \item{\code{labeled}}{Factor: Whether this trial was 'labeled' or
#'   'unlabeled'.}
#' }
#' 
"supunsup"

#' Non-excluded assignments (Expt. 1)
#'
#' @seealso Format described in \code{\link{supunsup}}
"supunsup_clean"

#' Data from excluded assignments (Expt. 1)
#'
#' @seealso Format described in \code{\link{supunsup}}
"supunsup_excluded"

#' Some reasonable priors on category means/variances
#'
#' These are based on two sources: Kronrod, Copress, and Feldman (2012, CogSci)
#' used a combination of categorization and discrimination data to estimate
#' underlying mean and variance for /b/ and /p/, assuming a mean of 60ms for /p/
#' (which is based on Lisker & Abramson, 1964). Goldrick, Vaughn, and Murphy
#' (2013, JASA) analyzed VOTs produced by 12 talkers producing 80 monosyllabic
#' words with word-initial voiced stops (only /b/ analyzed here).  Many talkers
#' prevoiced some or most of their /b/s, and these are included separately here.
#'
#' @format A data frame with 4 observations of 5 variables:
#' \describe{
#' \item{\code{prevoiced}}{Whether or not stats are for prevoiced stops (only for
#' goldrick et al. data}
#' \item{\code{mean}}{Mean VOT (in ms) for this category}
#' \item{\code{sd}}{VOT standard deviation (in ms; inferred or sample)}
#' \item{\code{category}}{'b' or 'p'}
#' \item{\code{source}}{Source of this estimate ('goldrick2013' or 'kronrod2012')}
#' }
"prior_stats"

#' Supervised/unsupervised phonetic adaptation (Experiment 2)
#'
#' Data from a follow-up experiment that manipulated /b/ and /p/ mean VOTs
#' separately, using only unsupervised distributional learning. There are five
#' conditions: (-80,50), (-50,50), (-20,50), (10,50), (10,80). The cleaned
#' dataset (with random responders removed) is \code{separatemeans_clean}, and
#' the excluded data is in \code{separatemeans_excluded}. Assignment metadata
#' (including audioequipment and useragent is in
#' \code{separatemeans_assignments}
#' 
#' Because of the greater separation between clusters this introduces, there's
#' an additional test phase after the main part of the experiment, where
#' subjects classify a flat continuum from -10ms to 50ms VOT, repeated 10 times
#' each.
#'
#' Subjects classifying at less than 80% accuracy as extrapolated by a GLM at
#' -10 and 80ms VOT are excluded in \code{separatemeans_clean}.
#'
#' @format The main dataset is a data.frame/tbl_df with 36,208
#'   observations of 57 variables.  It has the columns from
#'   \code{\link{supunsup}}, plus
#'   \describe{
#'   \item{\code{pvotCond}}{Mean VOT of /p/ distribution}
#'   \item{\code{is_test}}{TRUE for test block, FALSE for exposure block}
#'   }
#'
#'   The assignments metadata is a data.frame/tbl_df with 124 observations of 10
#'   variables:
#'   \describe{
#'   \item{subject}{Anonymized worker ID}
#'   \item{assignmentid}{MTurk assignment ID}
#'   \item{hitid}{MTurk HIT ID}
#'   \item{accepttime}{When assignment was accepted}
#'   \item{submittime}{When assignment was submitted}
#'   \item{audioequip}{Subjects' self-reported audio equipment}
#'   \item{comments}{Subjects' comments after completing the experiment. The
#'   prompt was "Enter any comments you have below.  Did anything seem weird?  Were
#'   you hearing anything besides the choices we gave you?"}
#'   \item{errors}{Not used in this experiment}
#'   \item{practiceResp}{Not used in this experiment}
#'   \item{userAgent}{The browser userAgent string.}
#'   }
"separatemeans"

#' Data from non-excluded assignments (Expt. 2)
#'
#' @seealso Format described in \code{\link{separatemeans}}
"separatemeans_clean"

#' Data from excluded assignments (Expt. 2)
#'
#' @seealso Format described in \code{\link{separatemeans}}
"separatemeans_excluded"

#' Summary of all Experiment 2 assignments
#'
#' @seealso Format described in \code{\link{separatemeans}}
"separatemeans_assignments"