calc_stats provides an interface for calculating statistics/metrics on
model predictions and/or observed data. Supported statistics include
basic statistics on mean and standard deviation, Conditional Accuracy
Functions (CAFs), Quantiles, Delta Functions, and fit statistics. Results can
be aggregated across individuals.
calc_stats(object, type, ...)# S3 method for data.frame
calc_stats(
object,
type,
...,
conds = NULL,
resample = FALSE,
progress = 1,
level = "individual",
b_coding = NULL
)
# S3 method for drift_dm
calc_stats(object, type, ..., conds = NULL, resample = FALSE)
# S3 method for fits_ids_dm
calc_stats(
object,
type,
...,
conds = NULL,
resample = FALSE,
progress = 1,
level = "individual"
)
# S3 method for fits_agg_dm
calc_stats(
object,
type,
...,
conds = NULL,
resample = FALSE,
progress = 1,
level = "group",
messaging = TRUE
)
# S3 method for stats_dm
print(
x,
...,
round_digits = NULL,
print_rows = NULL,
some = NULL,
show_header = NULL,
show_note = NULL
)
# S3 method for stats_dm_list
print(x, ...)
If type is a single character string, then a subclass of data.frame is
returned, containing the respective statistic. Objects of type sum_dist
will have an additional attribute storing the boundary encoding (see also
b_coding). The reason for returning subclasses of data.frame is
to provide custom plot() methods (e.g., plot.cafs). To get rid
of the subclass label and additional attributes (i.e., to get just the plain
underlying data.frame, users can use unpack_obj()).
If type contains multiple character strings (i.e., is a character vector) a
subclass of list with the calculated statistics is returned. The list will
be of type stats_dm_list (to easily create multiple panels using the
respective plot.stats_dm_list() method).
The print methods print.stats_dm() and print.stats_dm_list() each
invisibly return the supplied object x.
an object for which statistics are calculated. This can be a
data.frame of observed data, a drift_dm object, a
fits_ids_dm object, or a fits_agg_dm object (see
estimate_dm()).
a character vector, specifying the statistics to calculate.
Supported values include "basic_stats", "cafs", "quantiles",
"delta_funs", "densities", and "fit_stats".
additional arguments passed to the respective method and the underlying calculation functions (see Details for mandatory arguments).
optional character vector specifying conditions to include.
Conditions must match those found in the object.
logical. If TRUE, then data is (re-)sampled to create
an uncertainty estimate for the requested summary statistic. See Details for
more information. Default is FALSE. Note that resampling does not work with
type = "fit_stats".
integer, indicating if information about the progress should be displayed. 0 -> no information, 1 -> a progress bar. Default is 1.
a single character string, indicating at which "level" the
statistic should be calculated. Options are "group" or "individual". If
"individual", the returned stats_dm object contains an "ID" column.
a list for boundary coding (see b_coding). Only
relevant when object is a data.frame. For other object types, the
b_coding of the object is used.
logical, if FALSE, no message is provided.
an object of type stats_dm or stats_dm_list, as returned by
the function calc_stats().
integer, controls the number of digits shown. Default is 3.
integer, controls the number of rows shown.
logical. If TRUE, a subset of randomly sampled rows is shown.
logical. If TRUE, a header specifying the type of
statistic will be displayed.
logical. If TRUE, a footnote is displayed indicating
that the underlying data.frame can be accessed as usual.
calc_stats is a generic function to handle the calculation of different
statistics/metrics for the supported object types. Per default, it returns
the requested statistics/metrics.
Basic Statistics
With "basic statistics", we refer to a summary of the mean and standard deviation of response times, including a proportion of response choices.
Conditional Accuracy Function (CAFs)
CAFs are a way to quantify response accuracy against speed. To calculate CAFs, RTs (whether correct or incorrect) are first binned and then the percent correct responses per bin is calculated.
When calculating model-based CAFs, a joint CDF combining both the pdf of correct and incorrect responses is calculated. Afterwards, this CDF is separated into even-spaced segments and the contribution of the pdf associated with a correct response relative to the joint CDF is calculated.
The number of bins can be controlled by passing the argument n_bins.
The default is 5.
Quantiles
For observed response times, the function stats::quantile is used with default settings.
Which quantiles are calcuated can be controlled by providing the
probabilites, probs, with values in \([0, 1]\). Default is
seq(0.1, 0.9, 0.1).
Delta Functions
Delta functions calculate the difference between quantiles of two conditions against their mean:
\(Delta_i = Q_{i,j} - Q_{i,k}\)
\(Avg_i = 0.5 \cdot Q_{i,j} + 0.5 \cdot Q_{i,k}\)
With i indicating a quantile, and j and k two conditions.
To calculate delta functions, users have to specify:
minuends: character vector, specifying condition(s) j. Must be in
conds(drift_dm_obj).
subtrahends: character vector, specifying condition(s) k. Must be in
conds(drift_dm_obj)
dvs: character, indicating which quantile columns to use.
Default is "Quant_<u_label>". If multiple dvs are provided,
then minuends and subtrahends must have the same length,
and matching occurs pairwise. In this case, if only one
minuend/subtrahend is specified, minuend and subtrahend are recycled to
the necessary length.
specifying probs is possible (see Quantiles)
Densities
With "densities", we refer to a summary of the distribution of observed or predicted data. For observed data, histogram values and kernel density estimates are provided. For predicted data, the model's predicted PDFs are provided.
Optional arguments are:
discr: numeric, the band-width when calculating the histogram or the
kernel density estimates. Defaults to 0.015 seconds
t_max: numeric, the maximum time window when calculating the distribution
summaries of observe data. Defaults to the longest RT (for observed data)
or the maximum of the time domain of a model (which is the preferred choice,
if possible). If necessary, t_max is slightly adjusted to match with
discr.
scale_mass: logical, only relevant if observed data is available. If
TRUE, density masses are scaled proportional to the number of trials per
condition.
Fit Statistics
Calculates the Log-Likelihood, Akaike and Bayesian Information Criteria, and root-mean squared-error statistic.
Optional arguments are:
k: numeric, for penalizing the AIC statistic (see also stats::AIC
and AIC.fits_ids_dm).
n_bins, probs: numeric vectors, see the section on CAFs and Quantiles
above
weight_err: numeric scalar, determines how CAFs and quantiles are
weighted. Default is 1.5.
When resampling = TRUE, an uncertainty interval is provided via simulation.
The default number of iterations is R = 100, which can be changed by
passing the optional argument R.
If resampling is requested, the returned stats_dm object contains the
column "Estimate", coding the interval. The interval width is controlled
via the optional argument interval_level, a single numeric value between
0 and 1 (default: 0.95). The interpretation of this interval depends on
the specific situation (see below).
Resampling at the Individual Level
If object is a drift_dm object (i.e., a single model created by
drift_dm()), synthetic data are simulated under the model, and
for each synthetic data set the requested statistic is calculated. The
interval then reflects the range of these simulated statistics. To determine
the number of trials for each synthetic data set, dRiftDM either uses the
observed data attached to the model (if available) or the optional argument
n_sim (passed to simulate_data()). Note that n_sim must be
provided if no observed data are available, and that n_sim always has
priority.
If object is a drift_dm object with attached observed data, resampling
is also performed for the observed data. In this case, trials are
bootstrapped, and for each bootstrap sample the requested statistic is
calculated.
If object is a data.frame, fits_agg_dm, or fits_ids_dm object,
resampling is performed for each individual if level = "individual". For
both models and observed data, synthetic or bootstrapped data sets are
generated as described above.
Resampling at the Group Level
Group-level resampling is possible only if object is a data.frame
(with an "ID" column), fits_agg_dm, or fits_ids_dm object. To request
this, set level = "group". Participants are then bootstrapped, and
for each bootstrapped sample the aggregated statistic is calculated.
Interpretation of Intervals
For level = "group", intervals represent bootstrapped confidence intervals
For level = "individual", intervals represent the variability in the
statistic when data for a single participant are resampled or simulated
under the model.
Note
For objects of type fits_agg_dm, which contain a mixture of group- and
individual-level information, the level argument only affects resampling
for the observed data. For the model itself, resampling is always performed
under the fitted model, in the same way as for a drift_dm object.
# Example 1: Calculate CAFs and Quantiles from a model ---------------------
# get a model for demonstration purpose
a_model <- ssp_dm()
# and then calculate cafs and quantiles
some_stats <- calc_stats(a_model, type = c("cafs", "quantiles"))
print(some_stats)
# Example 2: Calculate a Delta Function from a data.frame ------------------
# get a data set for demonstration purpose
some_data <- ulrich_simon_data
conds(some_data) # relevant for minuends and subtrahends
some_stats <- calc_stats(
a_model,
type = "delta_funs",
minuends = "incomp",
subtrahends = "comp"
)
print(some_stats, print_rows = 5)
# Example 3: Calculate Quantiles from a fits_ids_dm object -----------------
# get an auxiliary fits_ids_dm object
all_fits <- get_example_fits("fits_ids_dm")
some_stats <- calc_stats(all_fits, type = "quantiles")
print(some_stats, print_rows = 5) # note the ID column
# one can also request that the statistics are averaged across individuals
print(
calc_stats(all_fits, type = "quantiles", average = TRUE)
)
Run the code above in your browser using DataLab