Critical bug fix:
calculateVifModeB() did not calculate the VIFs for modeB constructs correctly because of a bug in the calculation of the R^2. PLEASE REVIEW YOUR CALCULATIONS in cSEM version < 0.3.1:9000! (thanks to @Benjamin Liengaard for pointing it out).
predict() no longer silently returns empty predictions when
.test_data does not contain rownames.
Bug fix: calculation of the MSE in
modelSelectionCriteria() resulted in a vector of incorrect length. In some cases this affected the computation of “GM” and “Mallows_cp”.
Add lifecylce badges to postestimation functions.(#376)
Some arguments accepted by
... argument had not been documented properly. This has been fixed. See
args_assess_dotdotdot for a complete list of available arguments.
testMGD() gains a new
.output_type argument. By default (
.output_type = "structured"), the standard output is returned. If
.output_type = "structured", however, a tibble (data frame) summarizing the test decisions in a user-friendly way is returned. (#398)
print.cSEMAssess() no longer prints zero for VIF values of constructs that are not part of a particular structural equation.
calculateVIFModeB() now returns a matrix with the dependent construct in the rows and the VIFs for the coresponding weights in the columns. Previously, the output was a list.
Combine functions for surface, floodlight and simple effects analysis in the
doNonlinearEffectsAnalysis() function; Breaking: functions
doSurfaceAnalysis() have been removed!
Progress bars are now supported for every function that does resampling. Progress bars are fully customizable via the
progressr framework created by
progressr::handlers("void")and then run your csem commands. (#359)
Fix bug in the computation of the Bc and Bca interval. Computation failed for models that had no indirect effects.
List element “reliability” of
assess() is changed to “Reliability” to be consistent with the naming scheme of the other list elements.
Add checks to plot methods for
Several documentation updates and typo corrections.
Implement importance-performance matrix analysis via
doIPMA(). A corresponding plot method is also available.
calculateHTMT(). 1.) Add inference; 2) fix wrong handling of single-indicator constructs (#351); 3) Remove warning produced by
calculateHTMT() when the estimated model contains less than 2 common factors. (#325)
Breaking: Rename argument in
Implement degrees of freedom calculation for second-order constructs.
Fix bug in
doFloodlightAnalysis(). There was an internal bug. Earlier versions returned the wrong direct effect. If you have used
doFloodlightAnalysis() from cSEM v. 0.1.0 results are likely wrong.
Export plot method for
Allow users to specify a lavaan model without a structural model. Now, users can specify a model with several measurement equations (via
=~) but no strucutral equations. Instead the correlations between all! constructs must be given. Failing to do so causes an error.
Update documentation and vignettes
calculateHTMT()(does not support models containing second-order constructs)
assess() now also computes and prints the total and indirect effects for each variable as they are often used for model assessment and may thus be considered a quality criteria. In addition, the variance accounted for (VAF) is computed and printed as well. (#335)
Add the Chi_square statistic and the Chi_square statistic divided by its degrees of freedom to the list of fit indices. See:
Fix bug related to dotdotdot arguments incorrectly passed to functions supplied to
.user_funs when resampling. Add additional example to
assess() illustrating the use of the
.user_funs arguments when given multiple functions. (#334)
Remove warning produced when printing a
cSEMAssess object based on a model containing only constructs modeled as composites.
Update documentation for
.only_common_factors for postestimation function
predict() retruns predictions for composite models as well. This will break existing code that uses
predict(..., .only_common_factors = ...). You will get an
unused argument (.only_common_factors = FALSE) error. Simply remove the argument to fix it. (#330)
testOMF(). Now other fit measures such as the RMSEA or the GFI can be used as the test statistic. This is a rather experimental feature and may be removed in future versions.
.approach_weights = "GSCA" for models containing nonlinear terms gives a more meaningful error message. (#342)
If the weighting scheme is
.disattenuate = TRUE, dissatenuation is longer applied to constructs using modes other than “modeA”" or “modeB”. (#352)
Model-implied indicator correlation matrix for non-recursive models should now be calculated correctly. (#264)
When calculating the HTMT via
assess() the geometric mean of the average monotrait−heteromethod correlation construct eta_i with the average monotrait−heteromethod correlation of other constructs can be negative. NaNs produced are produced in this case and the HTMT was not printed. Added a warning and forced printing the NaNs as well. (#346)
Add CITATION file (#331)
Add informative error message if
.data contains missing values.