mml.sdf now returns the scored data.
readNAEP now will attempt to read IRT parameters from the AM folder's
.dct file if not found in NAEPirtparams package.
gap now returns results (and does not give an error) when the standard deviation of a continuous variable (that does not have plausible values) is requested. Thanks to Christian Christrup Kjeldsen for identifying and documenting this bug.
edsurveyTable now returns when a line has n=1 respondents on it
cor.sdf now works with logical variables.
lm.sdf now has a
verbose function, some warnings now just show as messages (when
when an item is used in a call, it does not have omitted levels dropped by default.
should pass tests on all CRAN configurations
mvrlm.sdf had a bug when there was an item as a predictor that is now solved
it was not possible to use TIMSS overall scores with
mml.sdf without further editing the
testData, this has been fixed
some unquoted references in calls were not resolving correctly to columns on the data when they should have. This has been fixed.
mml.sdf to allow direct estimation using the
drawPVs to enable users to draw plausible values using
Dire after running
suggestWeights function assists researchers in deciding which weight to use for analyses with ECLS-K:2011 data. See the
suggestWeights documentation and vignette for more information.
summary2 output now shows zero weights without a colon.
it is now possible to subset an
edsurvey.data.frame to just one row with square braces.
readELS would sometimes re-cache files when they did not need to be. This has now been corrected.
readPIAAC is now set to validate variables with missing value labels undefined in the codebook. For missing label values it will now return the numeric value as the label instead of an NA value. Thanks to GitHub user webervienna for identifying and reporting this bug.
Added support for the National Household Education Survey (NHES) set of data. See
downloadNHES for instructions to locate and obtain the NHES data files, and see
readNHES for reading the NHES data for analysis.
Added support for the School Survey on Crime and Safety (SSOCS) set of data. See
downloadSSOCS for instructions to locate and obtain the SSOCS data files, and see
readSSOCS for reading the SSOCS data for analysis.
append.edsurvey.data.frame.list now uses
labB in preference to existing labels. It also now accepts label names other than
label. When the resulting labels are not unique, it generates new ones. Also, if a single
edsurvey.data.frame is added to an
edsurvey.data.frame.list, and the labels for the
edsurvey.data.frame.list appear to be automatically generated, then the labels are automatically generated for the output of the call to
append.edsurvey.data.frame.list as well. These changes should make the output of
append.edsurvey.data.frame.list seem more logical.
showCutPoints now returns one line per set of cut points and does not show the cut point labels.
updated warnings and errors when an
edsurvey.data.frame is passed to a vareity of functions that requires use of PSU or stratum variables but they are not on the data.
cor.sdf now allows dynamic variables for the variables being correlated (
searchSDF now generates a tabled output when passed an
readTIMSS verbose output (
verbose=TRUE) prints file merge details during data processing to the console. Additionally,
readTIMSS will print
dataLevel details for both processed and cached data for the
SD used to use the population variance estimator and now uses the sample variance estimator. This also affects standard deviations calculated with the
SD used to set
jkSumMultiplier to 1 by default and now uses the value from the
edusrvey.data.frame. While this was the documented behavior, it was counterintuitive since all other
EdSurvey functions use the value directly from the
achievementLevels with both
returnDiscrete = FALSE and
returnCumulative = TRUE it would return an error. It now properly returns.
levelsSDF updated for handling value labels containing a
= character in the label. The full label is now properly returned. This corrects erroneous warning messages related to recodes where
= character is present. Thank you to Nicole Guarino for identifying this bug.
Added support for the PISA Young Adult Follow-up Study (YAFS). See
readPISA_YAFS for more information.
readICCS now woks for 2016 ICCS data.
supports linking error in NAEP assessments that have combined PBA and DBA formats, including NAEP 2018 Social Studies (grades 8); 2019 Science (grades 4, 8, and 12); 2019 Mathematics and Reading (grades 4, 8, and 12). See the vignette for details on the implementation at https://www.air.org/sites/default/files/2021-07/EdSurvey-linkingerror.pdf.
mml.sdf a new function that estimates student scale scores using the marginal maximum likelihood regression estimation method (as the program AM does). This is an alternative method to the plausible values approach.
mixed.sdf now gives clearer information about groups and weights, similar to the updated
achievementLevels now accepts multiple sets of plausible values at once, allowing joint achievement levels to be calculated. Because of this the
cumulative results column names now include the name of the variable. So for a composite score it will now be labeled
composite_Level where it used to say
Level. Also the
cutpoints object now specifies the scores the cutpoints regard.
percentile now has a new method “symmetric” that is similar to “simple” but returns the negative result if all of the values are multiplied by negative one. This method is faster than “unbiased” which has a small bias correction.
percentile used to report the number of people who were more extreme than any percentile in the
nsmall column. This was not used and so was removed.
percentile now accepts a
dofMethod argument to allow either of the methods described in
WS for the Welch-Satterthwaite formula or
JR for the Johnson-Rust formula. This defaults to
JR but was previously
edsurveyTable now warns and removes rows with non-positive full sample weights or no outcome. The also warnings now include the number of rows removed by the filter. Non-positive and NA weights are removed only if the
TRUE. Missing outcome scores are removed only if
we simplified all the code in the statistical methods, including
after subsetting an edsurvey.data.frame, the cache (user defined variables) used to throw errors. This now works.
mixed.sdf used to not correctly identify covariances when outputting the variance tables. Thanks to Endraswara Wiryawan for identifying this bug.
readTIMSS previously attempted to merge on 2019 teachers using the student's grade level. This lead to many teacher variables not being merged on in AAD (Abu Dhabi, UAE), ARE (United Arab Emirates), and NZL (New Zealand). Thanks to Damian Campbell for identifying this bug.
readTIMSSAdv bug for 1995 data relating to storing column dimension information in cache.
EdSurvey is attached, it nolonger sets the options value of
timeout. Instead it only sets it to the larger of the current value and an hour when downloading files.
readTIMSS were updated to include 2019.
downloadPIAAC was updated so data/codebook downloads perform more robustly on different OS.
readPIAAC was updated so the number of replicate weights in the resulting
edsurvey.data.frame is accurate to the OECD source data. Before this number was hard coded, resulting in duplicate replicate weights.
readPISA added 2018 financial literacy.
downloadICILS now references instructions for downloading the 2018 ICILS data set, as the
readICILS function supported ICILS 2018 as of EdSurvey v2.4.0
$ on an edsurvey.data.frame with a weight variable, only the vector of full sample weights are returned now. This used to also return a matrix (when relevant) that also included the replicate weights.
readPISA now support the PISA 2018 Financial Literacy (FIN) database.
edsurveyTable and now supports surveys where the PSU and stratum variables varried by weight variable (usually longitudinal surveys) for Taylor series variance estimation.
mixed.sdf now has a
rq.sdf now returns the rho statistic
waldTest now accommodates JK1 denominator degrees of freedom explicitly and allows for cases where the F-test is not valid to return, but only returns the chi-squared test.
waldTest now covers
the vignette was expanded to include more links.
== (equal) and
!= (not equal) comparison operators now work on
edsurvey.data.frame.list objects to compare objects by their attributes (e.g., survey, year, country).
readNAEP now allows fr2 paths that are seperate from the usual location and is more robust to variation in capitalization, but requires that a file name be unique, regardless of capitalization.
getStratumVar are now more robust and return more informative warnings.
match.arg parameter matching for the following functions/arguments for usability:
readCivEDICCS (dataSet, gradeLvl),
readPISA (database, cognitive),
readTALIS (isced, dataLevel),
readPIAAC was fixed so it can now find cached files for
match.arg checks were added to better check user inputted data to the
readNAEP function has updates for the
frPath argument to correctly handle relative paths, path expansion (~), and additional file validation checks.
gap function now correctly parses expressions that use a variable in the global environment.
summary2 function now gives a clearer error when something that is not the name of a weight column is passed to the
ECLS_B is now a supported dataset with the
readECLS_K2011 now download and read in, respectively, the latest ECLS-K:2011 K-5 data.
readTALIS was updated to work with TALIS 2018.
downloadPISA added support for 2018 and now downloads 2018 data. OECD compressed the 2018 PISA with an algorithm not implemented in R (DEFLATE64), and so users are instructed to uncompress it outside of R and EdSurvey before using the data.
readPIAAC was updated to work with files OECD released in early 2020. This includes adding a
usaOption argument that allows the user to select between the 2012-2014 file and the 2017 file for U.S. data.
readICILS was updated to work with files IEA released early in 2020.
EdSurvey now supports NAEP assessments with no plausible value variables, such as some National Indian Education Study (NIES) data.
EdSurvey now supports TIMSS 2019 pre-release files, including bridge files.
readELS we bug an issue parsing
.sav variable format definitions relating to integer definitions with a display of more than 10 digits. We believe this does not change how EdSurvey reads in any public use files. The issue can arise if the user converts restricted use files with Stat/Transfer.
Some IEA datasets have convenience variables that EdSurvey incorrectly identified as plausible value variables that are discretized plausible values. EdSurvey already internally generates similar convenience variables based on the existing plausible values and does not use the IEA variables. As of 2.6.0, these columns are not linked as plausible variables and are ignored. Affected datasets include TIMSS, TIMSS Advanced, PIRLS, and ePIRLS.
cor.sdf now reports standard errors linearized back to the correlation space in the result value
se. It also reports the confidence interval in the Fisher-Z space as
ZconfidenceInterval and a confidence interval mapped back to the correlation space (without linearization) as
cor.sdf now allows users to select the Fisher-Z transformation or no transformation with the
transformation argument. The default of using the Fisher Z for Pearson and no transformation for other correlation types remains the same.
cor.sdf now allows the user to set a value of
jrrIMax to allow faster computation.
oddsRatio to use t-statistic instead of Z, and to identify the confidence interval in the linear space and then convert.
+ function now works on
edsurvey.data.frame.list objects to build an
?edsurvey.data.frame.list for examples.
%in% function now works on
edsurvey.data.frame.list objects, identifying if a survey is in an
?edsurvey.data.frame.list for examples.
edsurveyTable now warns when an
edsurvey.data.frame.list is passed to it with an
edsurvey.data.frame on it that cannot return valid results for the requested statistics. The output is
NA for these surveys.
gap now supports linking error in PISA for math, reading, and science when a user sets
contourPlot deprecated the argument
colors which only set density colors. Instead, the function now uses
pointColors, which set the density and point colors separately. This will cause calls to
contourPlot that use
colors to not function with 2.6.0.
summary2 call is made with a
light.edsurvey.data.frame, unfilled levels will not be returned in the table anymore.
Users can now assign a variable with $, including subsets. Variables on the original data cannot be removed but can be overwritten. All changes are for the session only and do not change the data on the disk.
getData now closes LaF connections regardless of how it exits (e.g., with an error). This should improve the stability of the package.
edsurvey.data.frame now can specify a
fr2path. This is used only for NAEP surveys and no functionality depends on it, but it may be used in the future. The value of
fr2path may always be set to
achievementLevels now warns when an
edsurvey.data.frame.list is passed to it with an
edsurvey.data.frame on it that cannot return valid results for the requested statistics. The output is
NA for these surveys.
percentile incorrectly calculated standard errors when the pre-multiplier for the replicate weights was not one. Users can see that for a survey using
getAttributes(data, "jkSumMultiplier") for their data.
cor.sdf previously failed to calculate some correlations when some records had non-zero full sample weights and zero replicate weights. Now results are still calculated.
With TIMSS data, when a user updated a teacher-level variable, it used to prevent the user from using the
edsurvey.data.frame for non-teacher-level analysis. We fixed this issue. Thanks to Anders Astrup Christensen for identifying this issue and sharing reproducible code that is now incorporated into the package as a test.
Added support for the 2007-2008 Beginning Teacher Longitudinal Study (see
readBTLS) for Waves 1 to 5.
readPISA uses less memory when reading in PISA 2015 data. This is because it now uses the
n_max arguments to read in portions of files. This should save time on read in.
Other PISA files were changed to use base R read functions, this results in a different ordering of columns in 2012 and 2015.
Some PISA 2003 variables were read in as an integer and are now in as a real values, mostly hours variables.
After the OECD files were updated,
readPIAAC now properly use the 2012 USA file for Cycle 1.
To allow both 4.0 and 3.6 compatibility, an
edsurvey.data.frame.list now stores
covs as characters in preference to factors.
EdSurvey is no longer imports
xml2. In addition,
RColorBrewer was imported but is now suggested.
ContourPlot now uses
heat.colors by default.
Several data files were stored in text format in functions and are now stored in
The news for 2.4.0 was edited to note that the
gap linking error in NAEP is only for mathematics and reading.
lm.sdf now rounds the printed coefficients matrix in the standard way.
Fixed a bug in
edsurveyTable that caused it to return incorrectly labeled output if multiple covariates were tabbed with weights.
A broken link in the vignette was fixed.
Added support for the 2009 High School Longitudinal Study (see
readHSLS). These do not support Restricted Use Data (RUD).
Added support for the 2002 Education Longitudinal Study (see
Added the ECLS Kindergarten Class of 1998-1999 Study. Thes datasets can be downloaded with
downloadECLS_K and read in with
readECLS_K1998. This was added in 2.3.0 but first added to the NEWS for 2.4.0.
Added support for ePIRLS (see
download_ePIRLS). This was added in 2.3.0 but first added to the NEWS for 2.4.0.
Added support for 2018 for the existing
readICILS function. Thanks to Jeppe Bundsgaard of Aarhus University, Danish School of Education, for contributing the code for this.
OECD moved the PIAAC data to a new location, and the
downloadPIAAC function now uses the new URL.
The PISA 2015 data cache could have been formed incorrectly; that is fixed. When PISA 2015 datasets are first used with 2.4 it will take time to re-cache the data. The process now also uses far less memory.
The PISA data had incorrect PSU and stratum variables for most years. They are all fixed except for 2000, which we do not believe has a PSU variable on the file. Several strata have only one PSU, so the design still needs to be edited by the user to get correct Taylor series sampling variance estimates.
mixed.sdf function now correctly aggregates results and has its methodology documented. It no longer supports binomial models and has several arguments deprecated as a result.
gap function argument
varMethod is deprecated. The function uses only jackknife variance estimation.
gap function now accounts for linking error between NAEP paper and digitally based assessments for mathematics and reading.
subset function used to fail when a global variable shared a name with a column on the data; it now works.
percentile function has been updated to a formula that generates survey percentiles that are robust to transformation. For example, if the values are multiplied by a constant, then the percentiles move by that same constant.
lm.sdf function header now prints the number of plausible values used (when they are used) as well as the number of plausible values used in the sampling variance (
levelsSDF function now makes more informative warnings when passed a
cor.sdf now prints a final new line.
searchSDF was called with more than one search string and the
levels argument was set to
TRUE, a useless warning was issued. The warning was removed.
EdSurvey functions wrap to the width of the console.
rq.sdf function adds quantile regression to the package. See the
rq.sdf documentation for more information.
getPSUVar were added to give users stratum and PSU variables for surveys and Taylor series analyses.
summary2 function now accepts a vector of variables.
searchSDF function now accepts a vector
string input so that results can be filtered.
The formula for degrees of freedom when Taylor series variance estimation is used has been updated. The new formula is derived in the statistics vignette. See https://www.air.org/sites/default/files/EdSurvey-Statistics.pdf.
PISA 2015 is now supported.
waldTest function allows the user to test composite hypotheses–hypotheses with multiple coefficients involved–even when the data include plausible values. Because there is no likelihood test for plausible values nor residuals, the Wald test fills the role of the likelihood ratio test, ANOVA, and F-test.
mvrlm function adds multivariate regression (a regression with multiple outcomes) to the package. See the
mvrlm documentation for more information.
Survey weighted mixed models can now be fit with the
mixed.sdf function. Both linear and logistic models can be fit. These models are limited to 2 levels (one level with random effects).
Regressions can now output standardized regression coefficients using
summary(myLm, src=TRUE). When the
lm.sdf call includes
standardizeWithSamplingVar=TRUE the standardized error of the standardized regression coefficient account for the sampling error and measurement error (when applicable). Otherwise, the standard deviations are assumed to be measured without error.
summary2 function to produce unweighted and weighted descriptive statistics of a variable in
Added $ variable access to edsurvey.data.frame. e.g. sdf$dsex
Updated covariance matrix estimation in the
vcov function for
glm.sdf to work when
varEstInputs was not returned
Added covariance matrix estimation for
lm.sdf when Taylor series variance estimation was used.
rebindAttributes function to make dplyr interaction more smooth. See the
rebindAttributes documentation for an example.
When printing an
edsurvey.data.frame it now says the survey, year, subject, and country at the top. The dimensions are moved down.
gap, achievement levels can be specified with partial matches. This helps out when achievement levels have long names.
Added the ECLS Kindergarten Class of 2010-2011 Study. Longitudinal datasets can be downloaded with
downloadECLS_K and read in with
PIRLS 2016 is now supported.
gap to report the number of primary sampling units (PSUs) used to calculate a statistic.
oddsRatio helper function for logit results to show odds ratios.
When running a Pearson correlation on a discrete variable,
cor.sdf function by default condenses the occupied response codes to be consecutive integers. This can now be turned off by setting the
condenseLevels argument to
FALSE so that the code book levels are used instead.
glm.sdf function now uses the
glm2 package to fit models. This package converges on a broader class of models.
EdSurvey package nolonger sets the number of threads used by the required
data.table package to one when EdSurvey is being attached. The issue is now fixed in
Across the download functions, when a file is missing the warning text was homogenized. Additionally, all download functions now support a
verbose argument that can be used to make downloads silent.
glm.sdf functions now accept formulas that use the
I() function or other unevaluated expressions to the left of the tilda. Previously only a single variable could be named.
recode.sdf now checks that each recode has only a
from in it.
edsurveyTable function now works without RHS variable, allowing the formula
y ~ 1 to return overall means.
percentile function used to produce output even if asked to produce a percentile outside of the valid range (0 to 100). Now it prints out message if users input at least percentile outside of the valid range, and stops if all percentiles are invalid.
getData function now removes rows with omitted levels after being recoded.
All SPSS (.sav) file reads using the
haven package set the
user_na = TRUE flag to ensure no defined missing/omitted values are automatically converted to
NA values prematurely.
For consistency with other download functions,
downloadPISA now uses a
years argument instead of a
edsurveyTable on an
edsurvey.data.frame.list used to have the potential to create invalid (unprintable) output if the factor levels did not agree on every element of the
edsurvey.data.frame.list. They now return printable output.
edsurveyTable could produce a standard error when there was data from only one stratum. It now produces an
NA standard error.
Works with the Trends in International Mathematics and Science Study (TIMSS), TIMSS Advanced, the Progress in International Reading Literacy Study (PIRLS), and the International Computer and Information Literacy Study (ICILS), International Civic and Citizenship Education Study (ICCS), the Civic Education Study (CivEd), the Program for International Student Assessment (PISA), the Program for the International Assessment of Adult Competencies (PIAAC), and the Teaching and Learning International Survey (TALIS).
International datasets can be downloaded with
International datasets can be loaded with
probit.sdf functions with support
for survey item responses.
gap codegap that compares the average, percentile, achievement level, or percentage of survey responses between two groups that potentially share members.
percentile that calculates the percentiles of a numeric variable.
showCodebook that retrieves variable names, variable labels, and value labels for an
edsurveyTable function for
significant faster computation with much smaller memory footprint. We also
made error messages and outputs more informative.
More informative error message and output for
glm.sdf are now S3 method extended from
stats::lm so users can call the function using
contourPlot for regressions diagnostic.
recode.sdf function for recoding levels within variables.
rename.sdf function for modifying variable names.
append.edsurvey.data.frame.list function to return
a list of sdfs from either an edsurvey.data.frame.list or a single edsurvey.data.frame.
Manual documentation was refreshed.
Moved vignettes to AIR webiste at https://www.air.org/project/nces-data-r-project-edsurvey or see links in the vignette inlcuded in this package.
Added a new vignette on “Exploratory Data Analysis on NCES Data” provides examples of conducting exploratory data analysis on NAEP data.
Added new vignette on “Calculating Adjusted p-Values From EdSurvey Results” to the AIR website describing the basics of adjusting p-Values to account for multiple comparisons.
Added new vignette on “Using EdSurvey to Analyze TIMSS Data” to the AIR website describing; an introduction to the methods used in analysis of large-scale educational assessment programs such as Trends in International Mathematics and Science Study (TIMSS) using the EdSurvey package. The vignette covers topics such as preparing the R environment for processing, creating summary tables, running linear regression models, and correlating variables.
Added new vignette on “Using EdSurvey for Trend Analysis” to the AIR website describing the methods used in the EdSurvey package to conduct analyses of statistics that change over time in large-scale educational studies.
Added new vignette on “Producing LaTeX Tables From edsurveyTable Results With edsurveyTable2pdf” to the AIR website detailing the creation of pdf summary tables from summary results using the edsurveyTable2pdf function.
Added new methodology documentation on “Methods Used for Gap Analysis in EdSurvey” to the AIR website convering the methods comparing the gap analysis results of the EdSurvey package to the NAEP Data Explorer.
Added new methodology documentation on “Methods Used for Estimating Percentiles in EdSurvey” to the AIR website describing the methods used to estimate percentiles.
Added new methodology documentation on “Weighted and Unweighted Correlation Methods for Large-Scale Educational Assessment: wCorr Formulas” to the AIR website detailing the methodology used by the wCorr R package for computing the Pearson, Spearman, polyserial, and polychoric correlations, with and without weights applied. See https://www.air.org/resource/report/weighted-and-unweighted-correlation-methods-large-scale-educational-assessment.
Fixed connection issue associated with closing connections to an
readNAEP function now works on a case sensitive file system.
Vignettes now should appear in numerical order on CRAN.
Examples now name most arguments.
Vignettes now name most argument.
some print functions had number of plausible values added.