| NEWS | R Documentation |
NPtest() can now be applied to matrices up to 256 columns (Fortran code changed).
More flexibility for the legend in plotINFO() added.
Full and 0 response patterns are not taken into account in SepRel().
Legend in plotjointICC now uses proper item labels (and better default colors).
Corrected item-test correlations added to itemfit.
Sort option in print method for itemfit added.
Additional tests in NPtest() added.
Toy datasets stored as data frames.
PersonMisfit() incl. S3 added.
sanity check for missings in split vector in LRtest.Rm() added.
smoother confidence bands added in plotGOF().
LRtest.Rm(): bugfix due to updates in base R.
plotGOF(): added arguments x_axis, y_axis, set_par, and reset_par to ease customization of the plot.
Imports functions from default packages packages, as required by the new CRAN-check.
implemented an anova() method for all models (except LLRAs, which have their own: anova.llra). see ?anova.eRm
added a function to compute separation reliability, as proposed in Wright & Stone (1999; see ?SepRel).
plotINFO() accepts more arguments via ... (see ?plotINFO).
fixed a bug in plotPWmap().
fixed a bug in the internal function get_item_cats() related to NAs (affecting LLRA and item-information functions).
switched encoding to UTF-8 to avoid problems with diacritics, etc. (e.g., Martin-Löf).
updated citations.
general improvements.
streamlining some functions, updating the vignette, etc.
warnings are now treated more consistently and can be suppressed with suppressWarnings()
the plotGOF() function was revamped and tidied up in a couple of places, e.g.:
the plotting order of elements in a non-interactive call has been changed to put confidence ellipses and lines in the background and text in the foreground.
the x- and y-limits are now dynamically computed by default, so that confidence regions and ellipses are inside the plotting region.
the leading “I” before item numbers has been removed for better legibility.
moved NEWS to the new fancy NEWS.Rd file/format
an error in test_info() was fixed.
eRm now depends on R >= 3.0.0
an error in Tpbis.stat was fixed.
a bug in person.parameter() has been fixed that caused the estimation to crash in some cases.
a bug in thresholds() has been fixed that caused the routine to crash.
the RaschSampler package has been merged into eRm for convenience (still available as a separate package).
the package is now byte-compiled by default.
some statistics added to NPtest(), T7 and T7a removed.
fixed a bug in plotPWmap().
fixed the mplot argument in plotting routines.
fixed the split-criterion "all.r" in LRtest().
deleted all usages of data() in examples, since eRm uses “lazy data.”
when calling NPtest(), the RaschSampler can now be controlled more specifically (burn_in, step, seed).
various improvements and bugfixes for LLRA-related functions.
person parameter values can be extracted now for all persons using coef().
Additionally, in-/exclusion of extrapolated values (for 0 and perfect scores) can be controlled via the argument extrapolated.
LRtest now computes standard errors (se = TRUE) by default.
plotDIF now plots “difficulties” for all models (formerly, “easiness” parameters were plotted for Rasch models).
minor bug fixed in plotGOF(), where on rare occasions confidence ellipses were plotted together with control lines (spotted by Peter Parker)
improved labelling in plotjointICC()
warning regarding group assignment when using median or mean split removed from MLoef()
modification in NPtest() to split long output lines
changed the delimiters of plotDIF() confidence intervals to pch = 20 (small bullet).
new experimental functions to calculate and plot item and test information (by Thomas Rusch)
bug fixed in the calculation of item and person Infit t and Outfit t (hint from Rainer Alexandrowicz).
eRm no longer depends on the RaschSampler package.
However, it must be installed to use NPtest().
changed the delimiters of plotDIF() confidence intervals to pch = 20.
new (wrapper) function LLRA() for fitting linear logistic models with relaxed assumptions including utilities for preparing data (llra.datprep()), setting up (build_W()) and modifying (collapse_W()) design matrices, comparing LLRA models (anova()) and plotting results (plotTR() and plotGR()) (by Thomas Rusch).
“exact” version of the Martin-Löf test for binary items and arbitrary splits added as method to NPtest().
in plotGOF() confidence ellipses can now be drawn for subsets of items, optionally using different colours
new function plotDIF() (by Kathrin Gruber): plots confidence intervals for item parameters estimated separately in subgroups, uses LR objects as input
adapted the MLoef() function to work with polytomous data and more than two item groups
error checks in NPtest:
0/full responses for items meaningless for NPtest,
group in method = "T4" must be of type logical,
specifying all items for T4 gives meaningless results.
warning regarding group assignment when using median split removed from LRtest() and Waldtest().
some modifications in plotPWmap(): horizontal plotting, different default plotting symbols, option to change size of plotting symbols
bug in MLoef() fixed (now using logs in calculating the person contributions)
eRm now depends on R >= 2.12.0
Latin1 encoding removed
bug in plotICC() (always same title) fixed
LLTM(), LRSM(), and LPCM() work now for repeated measurement designs with treatment groups and missing values.
Rename vignette to “eRm”.
new function plotPWmap() to plot Bond-and-Fox style pathway maps for the data by Julian Gilbey.
Since calculation of the t-statistics requires calculation of the kurtosis of the standardized residuals, according changes to itemfit.ppar(), personfit.ppar(), pifit.internal(), print.ifit(), and print.pfit().
plotPImap() patched by Julian Gilbey: length of item.subset did not match the documentation, warning stars did not all appear, pre-calculated person.parameter data can be passed to the function via pp, mis-ordered items can be coloured.
some minor bugs fixed.
the optimizer can be changed to optim() using fitctrl <- "optim" and reset to nlm() (the default) with fitctrl <- "nlm"
value of LRtest() now contains the list fitobj which contains the model objects according to the subgroups specified by splitcr
MLoef() no longer supports missing values
function invalid from package gtools integrated into eRm. eRm no longer depends on gtools.
for RM(), RSM(), and PCM(): eta parameters are now displayed as difficulty parameters; print() and summary() methods changed accordingly.
new labeling of eta parameters in RM(), RSM(), and PCM().
they now are labeled according to the estimated parameters for items (RM()), items + categories (RSM()), items x categories (PCM())
function MLoef() for Martin-Löf-Test added
df in personfit() and itemfit() corrected
the logLik() functions now extract the log-likelhood and df into objects of class "logLik.eRm" and "loglik.ppar" with elements loglik and df.
the corresponding print methods have been modified accordingly.
method coef.ppar() to extract person parameter estimates added
option for beta parameters added to coef.eRm
in confint.eRm: default parm = "beta"
minor modifications in the help file for IC()
plotPImap(): revised rug added, bug concerning item.subset fixed, minor modifications to enhance readability
minor modifications in plotjointICC(): allows for main title and colors, option legpos = FALSE suppresses legends, dev.new removed, legend = FALSE produced incorrect labeling
minor modifications in plotICC(): allows for main title and colors, default coloring with col = NULL instead of NA for compatibility, option legpos = FALSE suppresses legends, mplot is now FALSE if only one item is specified
plot.ppar: dev.new removed
option “visible” in print.ifit und print.pfit to allow for avoiding overly long output and for extraction of infit and outfit values (maybe changed to a coef method later)
strwrap() for NPtest print methods to break long lines
new methods IC.default() and pmat.default() for enhanced error messages
lazy loading package and datafiles