Changes in version 0.5-9 New Features: Bug Fixes: - compRelSEM() failed whenever the first composite had ordered= indicators (fixed on 14 February 2026, unfortunately right after it was available on CRAN). - compRelSEM() failed for lavaan.mi objects without higher-order loadings. Changes in version 0.5-8 (2026-02-14) New Features: - compRelSEM() and its documentation received a massive overhaul: - New argument W= available for users to supply custom weights via lavaan syntax - omit.indicators= argument deprecated in favor of using W= - All common-factor variance contributed to items in a composite is now (by default) considered true-score variance. - higher= argument deprecated. New general formula (see documentation Details) assumes any higher-order factor(s) to be the source of true-score variance. - New argument true= available for users to indicate which common factor(s) should be considered true-score variance - omit.factors= argument deprecated in favor of using true= - return.df= argument replaced by simplify= - Coefficients now returned (by default) in a list, with headers indicating how to appropriately interpret each coefficient. The verbose headers can be suppressed (mimicking old behavior) by setting simplify = -1L - miPowerFit() has been renamed to epcEquivFit() and its functionality has been substantially extended (see PR #151). - The output of epcEquivFit() now has its own class, epcequivfit.data.frame. A dedicated summary() method is provided, which reports both local classifications of fixed parameters and the resulting global equivalence decision. - By default, epcEquivFit() focuses on equivalence testing based on expected parameter changes (EPCs). The Saris, Satorra, and van der Veld (2009) decision rule based on modification indices and power is retained but is no longer treated as the primary result, as simulation results favor the EPC equivalence-testing approach. - The equivalence-testing procedure has been extended to include an underpowered classification when the EPC confidence interval (CI) is wider than the smallest effect size of interest (SESOI) region. - The arguments and the resulting column names with target.epc has been renamed to reflect the smallest effect size of interest (SESOI), consistent with terminology in the equivalence-testing literature. - Standardized EPCs for indicator residual covariances are now computed by scaling residual covariances by residual variances, rather than by total indicator variances as in lavaan and other SEM packages, providing a more appropriate effect-size representation for residual relationships. - A new function, epcEquivCheck(), is introduced to assess whether trivial misspecification is difficult to establish for a given model under the EPC equivalence-testing framework. - Clipboard utilities have been updated to accommodate these changes, and all relevant documentation has been revised to reflect the transition from miPowerFit() to epcEquivFit(). Removed Features: - runMI() was deprecated in 2024, now removed. Use the new R package lavaan.mi - efaUnrotate(), orthRotate(), oblqRotate(), and funRotate() have been deprecated since 2022 and are now removed. - measurementInvariance(), measurementInvarianceCat(), and longInvariance() have been deprecated since 2022 and are now removed. Bug Fixes: - compareFit() would throw an error if lavaan.mi package was not installed (fixed). - monteCarloCI() would lose parameter names in coefs= argument if there were no dimnames in ACM= argument. Fixed by PR #149. Changes in version 0.5-7 (2025-03-13) New Features: - auxiliary() has 2 new arguments - envir= is passed to do.call(), which prevents problems not finding the lavaan package when semTools is not loaded via library(). Thanks to Brian Keller. - return.syntax=TRUE will return model syntax for the saturated-correlates parameters, which can be added to the target-model syntax. This enables adding saturated correlates easily to a blavaan model. - New fmi(method="cor") option, to return the fraction missing information for correlation estimates. - Not necessarily a bug fix, but MASS::mvrnorm() has been replaced with mnormt::rmnorm() throughout the package. The latter generates data such that a sample of (e.g.) $N=11$ will have the same first 10 observations as when sampling only $N=10$. - New goricaSEM() function provided as a wrapper for the restriktor package. Deprecated Features: - efaUnrotate() deprecated, along with rotation functions orthRotate(), oblqRotate(), and funRotate(). Users can now directly use lavaan::efa() and its rotation= argument. Exploratory SEM (ESEM) is also supported by lavaan using special model.syntax() operators, as shown in its Issue #112. - runMI() and all related functionality are deprecated, now returning OLDlavaan.mi-class objects. All lavaan.mi functions, class, and methods are now available in the lavaan.mi package. Any semTools functions that can operate on either a lavaan-class or lavaan.mi-class objects (e.g., the compRelSEM() function) will still work with the new package's output. Bug Fixes: - Formulas implementing Bollen et al.'s (2012, 2014) BIC extensions have been corrected in the documentation and source code. - emmeans functionality made more robust: - to different ordering of names in a specified interaction term - to accommodate models fitted to incomplete data using missing="FIML" estimation - compRelSEM() no longer gives an error for a higher= order factor when indicators are categorical. A bug for single-group lavaan.mi objects was also fixed. - Fixed an old bug which prevented clipboard() from working, particularly on a Mac OS: https://github.com/simsem/semTools/issues/56 - Fixed a bug in the update() method for a measEq.syntax-class object, preventing an error that occurred when changing values/labels for a parameter across multiple groups. - Simple intercepts are now always returned by probe2WayMC(), probe2WayRC(), probe3WayMC(), and probe3WayRC(), even when the y-intercept is fixed in the fitted model. This is not actually a bug, but the old default behavior led to plotProbe() incorrectly plotting all regression lines through the origin (see https://github.com/simsem/semTools/issues/125 and https://github.com/simsem/semTools/issues/127). - Fixed a bug in monteCarloCI(). See issue #142 - Fixed a bug with nullRMSEA() requiring data= to be in the global environment; see issue #133. Thanks to Brian Keller. - lrv2ord() now works even with a single-variable model. - parcelAllocation() now allows for higher-order factors (see here). Known bugs: - compareFit() not currently working for lavaan.mi-class objects. Changes in version 0.5-6 (2022-05-10) New Features: - reliability() and reliabilityL2() deprecated. - Average variance extracted now has a dedicated AVE() function. - A new compRelSEM() function provides more comprehensive options for estimating composite reliability coefficients (various alphas and omegas) than the reliability() function. compRelSEM() also estimates reliability for a composite representing a higher-order factor (using the higher= argument to name such factors) and implements Lai's (2021) recently proposed composite-reliability coefficients for different types of multilevel constructs (named with the config= and shared= arguments). See https://github.com/simsem/semTools/issues/106 for a conversation with Lai about the implementation, as well as links to documentation of tests. - monteCarloCI() now works for lavaan.mi objects, which is much easier than combining multiple imputation with bootstrapping. - monteCarloCI() now optionally returns CIs for the standardized solution, only for lavaan objects (not lavaan.mi). - htmt() gains an argument htmt2=TRUE to use the geometric mean (default) rather than the arithmetic mean (which assumes tau-equivalence). - moreFitIndices() includes 3 new extended BICs discussed by Bollen et al. (2014) - miPowerFit() can now pass further arguments to lavaan::modificationIndices() via ... - Added a tag= argument to customize character used to flag preferred models in summary() method for FitDiff class (i.e., compareFit() output). Set tag=NA or NULL to omit tags. - plausibleValues() applied to blavaan objects now passes (via ...) arguments to blavaan::blavPredict(), just as it already passed arguments to lavPredict() for objects of lavaan or lavaan.mi classes. Thus, setting type="ypred" will return plausible values of latent item-responses underlying ordinal observations (analogous to Mplus option SAVE: LRESPONSES). Bug Fixes: - probe2/3WayM/RC(): When the moderator was listed first in nameX= and some (not all) slopes were labeled in the user's model syntax, an error was caused by creating incorrect labels. Fixes https://github.com/simsem/semTools/issues/103 - Also, probe2/3WayM/RC() help pages have a new Reference entry to a tutorial paper written about how to use them. Changes in version 0.5-5 (2021-07-07) New Features: - A new function lrv2ord() provides univariate and bivariate population parameters for multivariate-normal data that have been discretized but treated as continuous. - reliability() has new what=, omit.factors=, and omit.indicators= arguments to customize composite-reliability estimates. - SSpower() now allows multigroup population models to be specified with lavaan syntax via the popModel= argument. - SSpower() and findRMSEApower() functionality is now available with a graphical user interface: - Shiny app available at https://sjak.shinyapps.io/power4SEM/ - tutorial published open-access at https://doi.org/10.3758/s13428-020-01479-0 - monteCarloMed() has been replaced by monteCarloCI() because the method generalizes beyond evaluating mediation (indirect effects), and can be applied to any function of model parameters. monteCarloCI() now accepts a set of functions to sample simultaneously (e.g., both the indirect and total effects), rather than drawing parameters again for each function of parameters. More conveniently, if user-defined parameters are already specified when fitting the lavaan model, then monteCarloCI() obtains all necessary information from the lavaan object. Bug Fixes: - permuteMeasEq() returned an error when parallelType = "multicore". - reliability() could use the wrong thresholds when a sequence of ordinal indicator names overlapped (e.g., x1 and x10); now fixed. - probe*Way*C() functions for probing latent interactions formerly returned an error when users specified custom labels for the relevant regression parameters in lavaan syntax. And in multigroup models, only group-1 SEs were used regardless of which group= was specified. Both issues resolved. - The show() method for compareFit() now behaves as expected: nothing is printed when assigning compareFit() output to an object, but something is printed when there is no assignment. The summary() method prints either way and invisibly returns the object itself. Changes in version 0.5-4 (2021-01-11) New Features: - summary() method for lavaan.mi objects now passes arguments to lavTestLRT.mi() via .... Bug Fixes: - reliability() incorrectly used standardized thresholds rather than the model's parameter estimates (i.e., on the latent-response scales) to calculate omega for categorical indicators. - net() returned an error for models with categorical indicators, now fixed. - The wrong df were used to find confidence limits for RMSEA calculated from the scaled $\chi^2$ statistic in lavaan.mi objects. - emmeans methods appropriately deal with incomplete data. - probe2WayMC() and probe2WayRC() treated the wrong variable as moderator for point or SE estimates, depending on the order in nameX=. Changes in version 0.5-3 (2020-05-27) New Functions: - New discriminantValidity() function added. - Support functions for the emmeans package (see ?lavaan2emmeans for examples) New Features: - class?lavaan.mi methods and functions can optionally specify particular imputation numbers to omit, in addition to the general omission criteria in omit.imps=. Simply include specific imputation numbers in the omit.imps= vector. - Latent-interaction functions (e.g., probe2WayMC()) now work for lavaan.mi objects. The ?indProd documentation now includes an example adding product indicators to multiple imputed data sets. - The update() method for class?measEq.syntax has a new argument change.syntax. Users can pass lavaan syntax specifying an existing model parameter in order to change the labels or the fixed/free values. This provies some flexibility not found in the measEq.syntax() function itself (e.g., releasing an equality constraint in only one of >2 groups, whereas group.partial= can only release constraints across all groups). - https://github.com/simsem/semTools/issues/60 - The as.character() method for class?measEq.syntax now accepts the argument package = "mplus", which prints the syntax as an Mplus MODEL command. The as.character() method also has 2 new arguments: - groups.as.blocks=TRUE optionally prints multigroup syntax in "block" format, which enables users to hack measEq.syntax() to specify multilevel CFA models with invariance constraints. - params= allows users to select specific types of parameters to print, making it easier to check particular aspects of the model specification (e.g., params = c("loadings","lv.variances") in metric invariance models). - net() now accepts models fitted to categorical outcomes. - reliability() includes 2 new arguments: - default dropSingle = TRUE is consistent with old behavior. - "total" column no longer returned by default, or ever for 1-factor models. Users can request return.total = TRUE when multiple factors are multiple dimensions of a single scale composite. - New small-N corrections added to chisqSmallN() function. Also accepts lavaan.mi objects now. - efaUnrotate() now accepts summary statistics when data=NULL. Bug Fixes: - reliability() and maximalRelia() returned an error with categorical single-group models - https://groups.google.com/d/msg/lavaan/rPVEHUQjqVQ/SQaMrgn-AQAJ - reliability() only ignored higher-order factors without any observed indicators, and returned an error when first-order factors had categorical indicators. Both issues have been resolved: - https://github.com/simsem/semTools/issues/65 - fitMeasures() for lavaan.mi sometimes returned an error Changes in version 0.5-2 (2019-08-30) - Requires lavaan version 0.6-5 - Minor bug fixes - Addition of the plausibleValues() function to extract plausible values (i.e., multiple imputations) of factor scores from objects of class lavaan, lavaan.mi, or blavaan - Full support for lavaan.mi models fitted to multiply imputed multilevel data; resolved issue (#39). - Full support for fixed.x=TRUE and conditional.x=TRUE in lavaan.mi models, including std.nox solutions in summary(), modindices.mi(), and lavTestScore.mi(). - Added the omit.imps= argument to all lavaan.mi-related functions, optionally excluding solutions that did not converge, failed to estimate SEs, or contained NPD matrices (Heywood cases are a special case). Only the first 2 are excluded by default. - reliability(), reliabilityL2(), and maximalRelia() now accept lavaan.mi objects - Added (S)EPCs to lavTestScore.mi() output when epc=TRUE - Added (A)RIV/FMI to all pooled tests when available for lavaan.mi objects, to quantify additional uncertaintly in the test statistic due to missing data. - Allow multigroup models in plotProbe() and related latent-interaction functions. - standardizeMx() was deprecated in previous versions, now removed. Changes in version 0.5-1 (2018-09-24) - Requires lavaan version 0.6-3 - Minor bug fixes - The formerly deprecated lisrel2lavaan() function has been removed from semTools - compareFit() now accepts lavaan.mi objects returned by runMI() - For lavaan.mi objects returned by runMI(), the anova() method has been updated to behave more like lavaan's anova() method: - more than 2 nested models can be compared - fit indices are no longer an option, and must be requested using the fitMeasures() method - Given the previous addition of score-test functions modIndices.mi() and lavTestScore.mi() for lavaan.mi objects (parallel to modIndices() and lavTestScore() for lavaan objects), the remaining "trilogy" of tests in lavaan (lavTestLRT() and lavTestWald()) now have parallel functions for lavaan.mi objects: lavTestLRT.mi() and lavTestWald.mi() - lavTestWald.mi() implements what was formerly available in using anova(..., test = "D1") - lavTestLRT.mi() implements what was formerly available in using anova(..., test = "D3") - lavTestLRT.mi() cannot compare more than 2 nested models. The anova() method internally calls the compareFit() function to compare multiple lavaan.mi objects. - For all 3 tests (score, Wald, and LRT), the "D2" pooling method is an option, and there is a newly public function calculate.D2() that can be used to pool any set of Wald chi-squared or z statistics. - The runMI() function can now be applied to multilevel SEMs that can be fitted with lavaan() - Known issue (#39): fitted() and resid() methods will not yet work in models with both multiple levels and multiple groups. This will be resolved in a future version. - The 3 functions measurementInvariance(), measurementInvarianceCat(), and longInvariance() have been deprecated, redirecting users to the new measEq.syntax() function. It is much more general, capable of combining features of all 3 deprecated functions without their restrictions. - The function's primary purpose is writing syntax that users can read and edit. - The fitting of a model is optional, and fitting multiple models is not (yet) automated. See the ?measEq.syntax help-page examples for how to fit and compare several levels of invariance. - Find many more details posted on the lavaan forum (the Google group). Changes in version 0.5 Find our remaining version history on GitHub: https://github.com/simsem/semTools/wiki/Version-History Changes in version 0.5-0 (2018-06-27) - Requires lavaan version 0.6-1 - Minor bugs fixed - Minor convenience features added - Redesigned runMI() function, which no longer produces an object of class lavaanStar (that object class is no longer supported), which inherited from class lavaan. It now produces an object of class lavaan.mi, which inherits from lavaan's new lavaanList class (see the ?lavaanList help page for details). The reasons to redesign runMI() include: - The user was required to choose among available methods for pooling the chi-squared test statistic when fitting the model, and the baseline model was also fit so that incremental fit indices (e.g., CFI and TLI) could be calculated. This lead to more frequent convergence problems. - The lavaanStar class could inadvertently mislead users into thinking that certain results were available from multiple imputations that were not. For example, the modindices() function would return modification indices for the first imputation, but those were not appropriately pooled statistics. - The new runMI() no longer includes the chi= argument, because those options have been moved to an anova() method written for lavaan.mi objects. Additional methods have been written: see the class?lavaan.mi help page for a list of methods and details about their use. Additionally, appropriately pooled modification indices and (S)EPCs are now available for multiple imputations (via modindices.mi()), as well as a general score test via lavTestScore.mi(). - The parcelAllocation() has also been redesigned with new arguments to improve its flexibility and ease of use. Users are now required to provide lavaan syntax not only for the parcel-level model, but also for the item-level model. This allows parcel allocation to be automatically detected, no longer requiring users to provide a list of item-to-parcel assignments (see new examples on the help page). - The OpenMx enhancements in previous versions of semTools are obsolete now that OpenMx provides fit indices and standardized paths, so they have been removed. However, standardizeMx() is still available (temporarily deprecated) to provide standardized mean-structure parameters, until the OpenMx maintainers add that feature to OpenMx::mxStandardizeRAMpaths(). - measurementInvariance() and measurementInvarianceCat() now require users to name all arguments passed to cfa(), including the first argument: model=. This is to prevent errors that occurred when some previous users had passed arguments in a different order than expected, which should not be a limitation.