R: Generate an adaptive or non-adaptive test HTML interface
mirtCAT
R Documentation
Generate an adaptive or non-adaptive test HTML interface
Description
Provides tools to generate an HTML interface for creating adaptive and
non-adaptive educational and psychological tests using the shiny package.
Suitable for applying unidimensional and multidimensional computerized adaptive tests
using item response theory methodology. Test scoring is performed using the mirt package.
However, if no scoring is required (i.e., a standard survey) then defining a mirt
object may be omitted.
Usage
mirtCAT(df = NULL, mo = NULL, method = "MAP", criteria = "seq",
start_item = 1, local_pattern = NULL, design_elements = FALSE,
cl = NULL, design = list(), shinyGUI = list(), preCAT = list(), ...)
## S3 method for class 'mirtCAT'
print(x, ...)
## S3 method for class 'mirtCAT'
summary(object, sort = TRUE, ...)
## S3 method for class 'mirtCAT'
plot(x, pick_theta = NULL, SE = 1, main = NULL,
par.strip.text = list(cex = 0.7), par.settings = list(strip.background =
list(col = "#9ECAE1"), strip.border = list(col = "black")), ...)
Arguments
df
a data.frame or list object
containing the character vector inputs required to generate
GUI questions through shiny. If factors are supplied instead of character vectors
then the inputs will be coerced using the as.character() function (set
stringsAsFactors = FALSE when defining a data.frame to avoid this).
Each row in the object corresponds to a unique
item. The object supports the follow column name combinations as inputs to specify the
type of response format, questions, options, answers, and stems:
Type
Indicates the type of response input
to use from the shiny package. The supported types are: 'radio' for radio buttons
(radioButtons), 'select' for a pull-down box for selecting
inputs (selectInput), 'text' for requiring
typed user input (textInput), 'checkbox' for allowing multiple
responses to be checked off (checkboxGroupInput),
'slider' for generating slider inputs (sliderInput), or
'none' for presenting only an item stem with no selection options. Note that slider
inputs require additional arguments to be passed; see ... instructions below).
Question
If df is a data.frame, a
character vector containing all the questions or stems to be generated.
If df is a list, then the commands must be suitable for output with
shiny (e.g., Question <- list(h6('Item 1'), list(h6('Nested', h4(' item 2')))))
Option.#
Names pertaining to the possible response
options for each item, where the # corresponds to the specific category. For
instance, a test with 4 unique response options for each item would contain
the columns (Option.1, Option.2, Option.3, Option.4).
If, however, some items have fewer categories than others then NA's can be used for response
options that do not apply.
Answer or Answer.#
(Optional) A character vector (or multiple character
vectors) indicating the scoring key for items that have correct answer(s). If there
is no correct answer for a question then a value of NA must be declared.
Stem
(Optional) a character vector of absolute or relative paths
pointing external markdown (.md) or HTML (.html) files to be used as item stems.
NAs are used if the item has no corresponding file.
...
In cases where 'slider' inputs are used instead only
the Question input is required along with (at minimum) a
min, max, and step column. In rows where the Type == 'slider' the
column names will correspond to the input arguments to sliderInput.
Other input column options such as step, round, pre, post,
ticks, inline, placeholder, width, and size
are also supported for the respective inputs.
mo
single group object defined by the mirt::mirt() function. This is required
if the test is to be scored adaptively or non-adaptively, but not required for general
questionnaires. The object can be constructed by using the
generate.mirt_object function if population parameters are known or by
including a calibrated model estimated from the mirt function with real data.
method
argument passed to mirt::fscores() for computing new scores in the CAT
stage, with the addition of a 'fixed' input to keep the latent trait estimates
fixed at the previous values. Default is 'MAP'
criteria
adaptive criteria used, default is to administer each item sequentially
using criteria = 'seq'.
Possible inputs for unidimensional adaptive tests include: 'MI' for the maximum
information, 'MEPV' for minimum expected posterior variance,
'MLWI' for maximum likelihood weighted information,
'MPWI' for maximum posterior weighted information, 'MEI' for
maximum expected information, and 'IKLP' as well as 'IKL' for the
integration based Kullback-Leibler criteria with and without the prior density weight,
respectively, and their root-nitems administered weighted counter-parts, 'IKLn' and
'IKLPn'.
Possible inputs for multidimensional adaptive tests include: 'Drule'
for the maximum determinant of the information matrix, 'Trule' for the
maximum (potentially weighted) trace of the information matrix,
'Arule' for the minimum (potentially weighted) trace of the asymptotic covariance matrix,
'Erule' for the minimum value of the information matrix, and 'Wrule' for
the weighted information criteria. For each of these rules, the posterior weight for
the latent trait scores can also be included with the 'DPrule', 'TPrule',
'APrule', 'EPrule', and 'WPrule', respectively.
As a safety precaution, if the
selected criteria do not weight by the posterior (and therefore do not exist for
extreme response styles) and less than 5 items have been administered then
the method is temporarily switched to the posterior weighting
until a variable response pattern is observed.
Applicable to both unidimensional and multidimensional tests are the
'KL' and 'KLn' for point-wise Kullback-Leibler divergence and
point-wise Kullback-Leibler with a decreasing delta value (delta*sqrt(n),
where n is the number of items previous answered), respectively.
The delta criteria is defined in the design object
Non-adaptive methods applicable even when no mo is passed
are: 'random' to randomly select items, and 'seq' for selecting
items sequentially.
start_item
two possible inputs to determine the starting item are available.
Passing a single number will indicate the specific item to be used as the start item;
default is 1, which selects the first item in the defined test/survey.
If a character string is passed then the item will be selected from one of
the item selections criteria available (see the criteria argument)
local_pattern
a character/numeric matrix of response patterns
used to run the CAT application without generating the GUI interface.
This option requires complete response pattern(s) to be supplied. local_pattern
is required to be numeric if no questions are supplied, and the responses must be
within a valid range of the defined mo object.
Otherwise, it must contain character values of plausible responses which corresponds to the
answer key and/or options supplied in df
design_elements
logical; return an object containing the test, person, and design
elements? Primarily this is to be used with the findNextItem function
cl
an object definition to be passed to the parallel package
(see ?parallel::parLapply for details). If defined, and if
nrow(local_pattern) > 1, then each row will be run in parallel to help
decrease estimation times in simulation work
design
a list of design based control parameters for adaptive and non-adaptive tests.
These can be
min_SEM
Default is rep(0.3, nfact); minimum standard error or measurement
to be reached for the latent traits (thetas) before the test is stopped. If the test is
multidimensional, either a single value or a vector of values may be supplied to provide
SEM criteria values for each dimension
delta_thetas
Default is rep(0, nfact); stopping criteria based on the change in latent
trait values (e.g., a change from theta = 1.5 to theta = 1.54 would
stop the CAT if delta_thetas = 0.05). The default disables this stopping criteria
thetas.start
a numeric vector of starting values for the theta parameters.
Default is rep(0, nfact)
min_items
minimum number of items that must be answered
before the test is stopped. Default is 1
max_items
maximum number of items that
can be answered. Default is the length of the item bank
quadpts
Number of quadrature points used per dimension
for integration (if required). Default is identical to scheme in fscores
theta_range
upper and lower range for the theta
integration grid. Used in conjunction with quadpts to generate an equally spaced
quadrature grid. Default is c(-6,6)
weights
weights used when criteria == 'Wrule', but also
will be applied for weighted trace functions in the T- and A-rules. The default
weights the latent dimensions equally. Default is rep(1/nfact), nfact),
where nfact is the number of test dimensions
KL_delta
interval range used when criteria = 'KL'
or criteria = 'KLn'. Default is 0.1
content
an optional character vector indicating the type of content measured
by an item. Must be supplied in conjunction with content_prop
content_prop
an optional named numeric vector indicating the
distribution of item content proportions. A content vector must also be supplied
to indicate the item content membership. For instance, if content contains three
possible item content domains 'Addition', 'Subtraction', and 'Multiplication', and the
test should contain approximately half multiplication and a quarter of both
addition and subtraction, then a suitable input would be
Note that content_prop must sum to 1 in order to represent valid population
proportions.
classify
a numeric vector indicating cut-off values for classification
above or below some prior threshold. Default does not use the classification scheme
classify_CI
a numeric vector indicating the confident intervals used to
classify individuals being above or below values in classify. Values must
be between 0 and 1 (e.g., 0.95 gives 95% confidence interval)
exposure
a numeric vector specifying the amount of exposure control to apply for
each successive item (length must equal the number of items).
The default uses no exposure control. If the item exposure
is greater than 1 then the n most optimal
criteria will be randomly sampled from. For instance, if
exposure[5] == 3, and critiera = 'MI', then when the fifth item is to be
selected from the remaining pool of items the top 3 candidate items demonstrating
the largest information criteria will be sampled from. Naturally, the first and last
elements of exposure are ignored since exposure control will be meaningless.
If all elements in exposure are between 0 and 1 then the Sympson-Hetter exposure
control method will be implemented. In this method, an item is administered only if it
passes a probability simulation experiment, otherwise it is removed from the item pool.
Values closer to 1 are more likely to appear in the test, while value closer to 0 are more
likely to be randomly discarded.
constraints
A named list declaring various item selection contraints for which
particular item, where each list element is a vector of item numbers. Unless otherwise stated,
multiple elements can be decalared (e.g., list(ordered = c(1:5), ordered = c(7:9)) is
perfectly acceptable). These include:
not_scored
declaring items that can be selected but will not be used in the
scoring of the CAT. This is primarily useful when including experimental items for
future CATs. Only one vector of not_scored elements can be supplied
excluded
items which should not actually appear in the session
(useful when re-testing participants who have already seen some of the items).
Only one vector of excluded elements can be supplied
independent
declaring which items should never appear in the same CAT session.
Use this if, for example, item 1 and item 10 have very similar questions
types and therefore should not appear within the same session
ordered
if one item is selected during the CAT, administer this
particular group of items in order according to the specified sequence
unordered
same as ordered, except the items in the group will be selected at
random until the group is complete
customNextItem
a more advanced function of the form
customNextItem <- function(design, person, test) to use a customized item selection
method. This requires more complex programming and understanding of mirtCATs internal elements,
and it's recommended to initially use a browser to understand the state
of the input arguments.
Use this if you wish to program your item selection techniques explicitly, though this
can be combined the internal findNextItem function with analogous inputs.
Function must return a single integer value
indicating the next item to administer or an NA value to indicate that the test
should be terminated.
shinyGUI
a list of GUI based parameters to be over-written. These can be
title
A character string for the test title. Default is
'mirtCAT'
authors
A character string for the author names. Default is
'Author of survey'. If the input is an empty string ('') then the author
information will be omitted in the GUI
instructions
A three part character vector indicating how to use the GUI.
Default is:
c("Instructions:",
"To progress through the interface, click on the action button below.",
"Next")
firstpage
The first page of the shiny GUI. Default prints the title
and information message.
list(h1('Welcome to the mirtCAT interface'),
The following interface was created using the mirtCAT package.
To cite the package use citation('mirtCATd') in R.')
If an empty list is passed, this page will be skipped.
begin_message
Text to display on the page prior to beginning the CAT. Default
is "Click the action button to begin."
demographics
A person information page used in the GUI for collecting
demographic information, generated using tools from the shiny package. For example,
the following code asks the participants about their Gender:
By default, the demographics page is not included.
demographics_inputIDs
a character vector required if a custom demographics
input is used. Default is demographics_inputIDs = 'gender', corresponding to
the demographics default
max_time
maximum time allowed for the generated GUI, measured
in seconds. For instance, if the test should stop after 10 minutes then the number
600 should be passed (10 * 60). Default is Inf, therefore no time limit
temp_file
a character vector indicating where a temporary .rds file
containing the response information should be saved while the GUI is running.
The object will be saved after each item is successfully completed. This is used to
save response information to the hard drive in case there are power outages or
unexpected computer restarts.
If NULL, no temp file will be created. Upon completion of the test, the
temp file will be deleted. If a file already exists, however, then this will be used to
resume the GUI at the last location where the session was interrupted
lastpage
A function printing the last message, indicating that the test has been completed
(i.e., criteria has been met). The function requires exactly one argument (called person), where
the input argument is the person object that has been updated throughout the test. The default function is
function(person){
return(list(h5("You have successfully completed the interface.
Click the action button to terminate the application.")))
}
css
a character string defining CSS elements to modify the GUI presentation
elements. The input string is passed to the argument tags$style(HTML(shinyGUI$css))
prior to constructing the user interface
forced_choice
logical; require a response to each item? Default is TRUE.
This should only be set to FALSE for surveys (not CATs)
stopApp
logical; use a stopApp() call after the interface has been completed?
Default is TRUE. However, when hosting an application on a remote server this should be set
to FALSE to allow a more graceful completion (in which case the last page will be displayed
until the browser tab is closed)
ui
a shiny UI function used to define the interface. If NULL, the
default one will be used. See mirtCAT:::default_UI for the internal code definition
preCAT
a list object which can be used to specify a pre-CAT block in which
different test properties may be applied prior to beginning the CAT session. If the
list is empty, no preCAT block will be used. All of the following elements are required
to use the preCAT input:
min_items
minimum number of items to administer before the CAT session begins.
Default is 0
max_items
max number of items to administer before the CAT session begins.
An input greater than 0 is required to run the preCAT stage
criteria
selection criteria (see above). Default is 'random'
method
estimation criteria (see above). It is generally recommended to
select a method which can deal with all-or-none response patterns, such as 'EAP'
or 'MAP', or in the multidimensional case 'DPrule' or 'TPrule'. Default is 'MAP'
response_variance
logical; terminate the preCAT stage when there is variability in the
response pattern (i.e., when maximum-likelihood estimation contains a potential optimum)?
Default is FALSE
...
additional arguments to be passed to mirt, fscores,
runApp, or lattice
x
object of class 'mirtCAT'
object
object of class 'mirtCAT'
sort
logical; sort the response patterns based on the order they
were administered? If FALSE, the raw response patterns containing NAs will be returned
for items that were not administered
pick_theta
a number indicating which theta to plot (only applicable for multidimensional
tests). The default is to facet each theta on one plot, but to plot only the first factor pass
pick_theta = 1
SE
size of the standard errors to plot. The default is 1, and therefore plots the
standard error. To obtain the 95% interval use SE = 1.96 (from the z-distribution)
main
title of the plot. Will default to 'CAT Standard Errors' or
'CAT ##% Confidence Intervals' depending on the SE input
par.strip.text
plotting argument passed to lattice
par.settings
plotting argument passed to lattice
Details
All tests will stop once the 'min_SEM' criteria has been reached or classification
above or below the specified cutoffs can be made. If all questions should
be answered, users should specify an extremely small 'min_SEM' or, equivalently,
a large 'min_items' criteria to the design list input.
Value
Returns a list object of class 'Person' containing the following elements:
raw_responses
A character vector indicating the raws responses to the respective
items, where NA indicates the item was not answered
scored_responses
An integer vector of scored responses if the item_answers input
was used for each respective item
items_answered
An integer vector indicating the order in which the items were
answered
thetas
A numeric vector indicating the final theta estimates
SE_thetas
A numeric vector indicating the standard errors of the
final theta estimates
thetas_history
A matrix indicating the progression of updating the theta values
during the test
thetas_SE_history
A matrix indicating the standard errors for theta after each
successive item was answered
item_time
A numeric vector indicating how long the respondent took to answer
each question (in seconds)
demographics
A data.frame object containing the information collected on the
first page of the shiny GUI. This is used to store the demographic information for each
participant
classification
A character vector indicating whether the traits could be
classified as 'above' or 'below' the desired cutoffs