v0.3.17

Miscellaneous

• Added Rcpp specific flag to avoid warnings when using the upcoming new version of Rcpp.

What's Changed

• Prevent Rcpp from masking valid uses of Rf_error by @Enchufa2 in #556
• New minor release by @danielinteractive in #557

New Contributors

• @Enchufa2 made their first contribution in #556

Full Changelog: v0.3.16...v0.3.17

v0.3.16

New Features

• Anova.mmrm() now has a test.statistic argument allowing the user to choose the Chi-squared test as opposed to the default F-test.
• mmrm_control() now has a disable_theta_vcov argument allowing the user to disable computation of the variance-covariance matrix of the covariance parameters. In that case, the nlminb optimizer cannot be used and other optimizers should be selected. Avoiding this computation can be useful in cases where the number of covariance parameters is large and computation of this very large matrix is slow and memory-intensive.
• model.frame() has a new argument exclude which allows to exclude specific model variables from the returned data frame, after creating the initial model.frame. This is useful when the same rows should be kept as in the original model fit (e.g. after na.action has been applied), but some variables are not needed in the returned data frame. The terms attribute of the returned data frame is updated accordingly.

Bug Fixes

• Previously, the compound symmetry correlation models (cs and csh) could have divergence issues, especially when the correlation was negative. This was because the range of the correlation parameter was from -1 to 1, whereas the correct range is between -1/(m-1) and 1, where m is the number of observations per subject. This is fixed now by using another mapping from the real line to the correct range, and therefore compound symmetry correlation models will work more robustly.
• Previously, the predict() method failed when requesting an unconditional prediction interval. This is fixed now.
• Previously, the emp_start() starting values could only work with variables included directly in data. Now they can also work when variables are not included in data but are available in the parent environment: The variables will be correctly included in the design matrix. This is achieved by an internal update to h_mmrm_tmb_data().
• Previously, the model.frame() method failed when the model formula included variables from the environment. This is fixed now. In addition, whereas model.frame() used to apply na.action before processing the include argument, na.action is now applied afterwards. As a result, any variables excluded from include are no longer considered when na.action is applied. Therefore, in the default case when na.action = "na.omit", this can lead to more rows being kept in the returned data frame compared to previous versions.
• Previously, the model.frame() method did not pass on additional arguments to the model.frame.default() method, e.g. subset. This is fixed now.
• Previously, the vcov() method returned the asymptotic covariance matrix, even if an adjusted one has been used. This is now fixed and the requested adjusted covariance matrix will be returned. (Please note this was not an issue for downstream summary() or emmeans() calls, where the adjusted covariance matrix was correctly used.)
• Previously, type 3 tests in the Anova.mmrm() method could give incorrect results which were not compatible with results from lme4::lmer() or other software (e.g. when using other than treatment contrasts). This is fixed now. In particular, any combination of contrasts can be used and will yield consistent results.
• Anova.mmrm() can now handle intercept-free models in type 2 tests, even when the first effect containing a categorical variable has an aliased parameter.
• Previously, the naming conventions used in the Anova.mmrm() method were not recognized by broom::tidy(). This is fixed now by using more standard names for the resulting data frame columns (e.g. F instead of F Statistic etc.)
• logLik.mmrm_tmb() now includes an nobs attribute containing the number of subjects in the provided model (i.e., component(mmrm_object, "n_subjects")). As a result, BIC(logLik(mmrm_object)) can now be successfully calculated.

What's Changed

• 517: fix cs, csh correlation range by @danielinteractive in #518
• Adding anova.mmrm by @NikKrieger in #519
• Augment logLik with attributes by @NikKrieger in #521
• Fit data utilities and diagnostics. by @NikKrieger in #522
• helpers for determining LRT suitability by @NikKrieger in #526
• Add h_refit_mmrm() and h_generate_new_name() by @NikKrieger in #528
• 523: add antedependence tests by @danielinteractive in #530
• 524: vcov() now returns adjusted covariance matrix by @danielinteractive in #531
• Added anova.mmrm(). Moved tests to test-mmrm-methods by @NikKrieger in #529
• 525: make predict work when response is missing by @danielinteractive in #527
• Extend the mmrm method for car::Anova() by @NikKrieger in #536
• Remove duplicate tests from test-tmb-methods by @NikKrieger in #535
• make h_get_minimal_fit_data() extract names from calls by @NikKrieger in #537
• fix nesting checks for interaction terms by @NikKrieger in #534
• 543: add disable_theta_vcov option by @danielinteractive in #544
• 502: refactor Anova type 3 calculations by @danielinteractive in #545
• 546: model.frame() supports environment variables in formulas by @danielinteractive in #547
• 548: Fix model.frame() behavior for missing values by @danielinteractive in #549
• Attempt to fix type-2 0-row contrast matrices by @NikKrieger in #552
• 550: Clarify error message when factor is required by @danielinteractive in #553
• New release by @danielinteractive in #554

New Contributors

• @NikKrieger made their first contribution in #519

Full Changelog: v0.3.15...v0.3.16

v0.3.15

New Features

• mmrm now returns score per subject in empirical covariance. It can be accessed by component(obj, name = "score_per_subject").

Bug Fixes

• Previously, when fitting a model with empirical covariance matrix estimation to a data set with a large number of subjects and/or a large number of coefficients, the model fitting could take very long and exhaust the memory in the worst case. This was due to an inefficient implementation of a matrix needed only in case of Satterthwaite degrees of freedom adjustment. This is fixed now, by returning the matrix empirical_g_mat in the mmrm object, instead of the previous empirical_df_mat matrix. The model fit is now much faster and does not exhaust the memory anymore. If old model fit objects are used, the empirical_df_mat will still be used correctly, however a deprecation warning will be issued. Please consider re-fitting the model to get the new empirical_g_mat matrix.
• Previously, when compiling mmrm from source using a TMB version below 1.9.15, and installing a newer TMB of version 1.9.15 or above, would render the mmrm package unusable. This is fixed now, by checking in the dynamic library of mmrm whether the version of TMB has been sufficient.

What's Changed

• 476: be more robust against TMB version/order by @danielinteractive in #478
• return meat per subject when computing empirical covariance matrix of coefficients by @zhangh12 in #495
• Update DESCRIPTION by @clarkliming in #498
• 506: optimize empirical vcov / Satterthwaite d.f. calculations by @danielinteractive in #507
• Fix AR correlation structure in gls calls by @KonstantinLang in #512
• 515: release 0.3.15 by @danielinteractive in #516

New Contributors

• @zhangh12 made their first contribution in #495
• @KonstantinLang made their first contribution in #512

Full Changelog: v0.3.14...v0.3.15

v0.3.14

Bug Fixes

• In version 0.3.13, when the tape optimizer from TMB was switched on, a warning would be given by fit_mmrm(), instructing users to turn off the tape optimizer. However, this is not necessary for reproducible results. Instead, it is now checked whether the deterministic hash for the TMB tape optimizer is used, and a warning is issued otherwise.
• In version 0.3.13, the above described warning by fit_mmrm() was not visible to the user when calling mmrm() because it was caught internally, causing the first fit in each session to fail for the first tried optimizer and falling back to the other optimizers. The warning is now issued directly by mmrm(). This change ensures that the first model fit is consistent regarding the chosen optimizer (and thus numeric results) with subsequent model fits, avoiding discrepancies observed in version 0.3.13.

What's Changed

• 472: fix TMB option warning by @danielinteractive in #473
• add Lukas as contributor, bump up version, polish news by @danielinteractive in #475

Full Changelog: v0.3.13...v0.3.14

v0.3.13

Bug Fixes

• When running with TMB package versions below 1.9.15, MMRM fit results are not completely reproducible. While this may not be relevant for most applications, because the numerical differences are very small, we now issue a warning to the user if this is the case. We advise users to upgrade their TMB package versions to 1.9.15 or higher to ensure reproducibility.
• Previously, mmrm ignored contrasts defined for covariates in the input data set. This is fixed now.
• Previously, predict always required the response to be valid, even for unconditional predictions. This is fixed now and unconditional prediction does not require the response to be valid or present any longer.
• model.frame has been updated to ensure that the na.action works correctly.
• Previously emmeans::emmeans returned NA for spatial covariance structures. This is fixed now.
• Previously car::Anova gave incorrect results if an interaction term is included and the covariate of interest was not the first categorical variable. This is fixed now.
• Previously car::Anova failed if the model did not contain an intercept. This is fixed now.

Miscellaneous

• Upon fitting an MMRM, it is checked whether a not reproducible optimization feature of TMB is turned on. If so, a warning is issued to the user once per session.
• mmrm now checks on the positive definiteness of the covariance matrix theta_vcov. If it is not positive definite, non-convergence is messaged appropriately.
• model.matrix has been updated to ensure that the NA values are dropped. Additionally, an argument use_response is added to decide whether records with NA values in the response should be discarded.
• predict has been updated to allow duplicated subject IDs for unconditional prediction.

What's Changed

• fix emmeans on spatial covariance by @clarkliming in #450
• clean up simulations by @clarkliming in #456
• keep contrast using droplevels by @clarkliming in #455
• fix anova type 3 by @clarkliming in #460
• 464: TMB warnings by @danielinteractive in #466
• make unconditional prediction default by @clarkliming in #463
• signal non-convergence if theta_vcov is not positive definite by @danielinteractive in #468
• Create FUNDING.yml by @danielinteractive in #471
• 469 cran by @danielinteractive in #470

Full Changelog: v0.3.12...v0.3.13

v0.3.12

New Features

• Add parameter conditional for predict method to control whether the prediction is conditional on the observation or not.

Bug Fixes

• Previously if the left hand side of a model formula is an expression, predict and simulate will fail. This is fixed now.

What's Changed

• add rhub workflow by @danielinteractive in #436
• add conditional arg for predict by @clarkliming in #438
• Adapt missing data sims by @danielinteractive in #443
• docs: Link the coverage badge to the HTML report by @cicdguy in #444
• change email, bump version by @danielinteractive in #447

Full Changelog: v0.3.11...v0.3.12

v0.3.11

Bug Fixes

• Previously if a secondary optimizer fails mmrm will fail. This is fixed now.
• Previously character covariate variables will make Anova fail. This is fixed now.

What's Changed

• fix issue of h_record_all_outputs by @clarkliming in #424
• fix anova error by @clarkliming in #427
• remove biocVer by @clarkliming in #429
• update docs by @clarkliming in #428
• Missing data simulations by @danielinteractive in #430
• fix eigen deprecated warning by @clarkliming in #432
• add tests for residual.R, bump up version by @danielinteractive in #433

Full Changelog: v0.3.10...v0.3.11

v0.3.10

Miscellaneous

• Fix internal test skipping functions for MacOS R.

What's Changed

• 419: make skipping code more general to also run on old R versions by @danielinteractive in #420

Full Changelog: v0.3.8...v0.3.10

v0.3.7

Miscellaneous

• In documentation of mmrm_control(), the allowed vcov definition is corrected to "Empirical-Jackknife" (CR3), and "Empirical-Bias-Reduced" (CR2).
• Fix a compiler warning related to missing format specification in error message function call.

Full Changelog: v0.3.6...v0.3.7

v0.3.6

New Features

• The argument method of mmrm() now only specifies the method used for the
  degrees of freedom adjustment.
• Add empirical, empirical Jackknife and empirical bias-reduced adjusted coefficients
  covariance estimates, which can be specified via the new vcov argument of mmrm().
• Add residual and between-within degrees of freedom methods.
• Add Kenward-Roger support for spatial covariance structures.
• Add model.matrix() and terms() methods to assist in post-processing.
• Add predict() method to obtain conditional mean estimates and prediction intervals.
• Add simulate() method to simulate observations from the predictive distribution.
• Add residuals() method to obtain raw, Pearson or normalized residuals.
• Add tidy(), glance() and augment() methods to tidy the fit results into summary tables.
• Add tidymodels framework support via a parsnip interface.
• Add argument covariance to mmrm() to allow for easier programmatic access
  to specifying the model's covariance structure and to expose covariance
  customization through the tidymodels interface.

Bug Fixes

• Previously mmrm() follows the global option na.action and if it is set
  other than "na.omit" an assertion would fail. This is now fixed and hence NA
  values are always removed prior to model fitting, independent of the global
  na.action option.
• Previously a model.frame() call on an mmrm object with transformed terms, or new
  data, e.g. model.frame(mmrm(Y ~ log(X) + ar1(VISIT|ID), data = new_data),
  would fail. This is now fixed.
• Previously mmrm() always required a data argument. Now fitting mmrm can also use
  environment variables instead of requiring data argument. (Note that
  fit_mmrm is not affected.)
• Previously emmeans() failed when using transformed terms or not including the visit
  variable in the model formula. This is now fixed.
• Previously mmrm() might provide non-finite values in the Jacobian calculations,
  leading to errors in the Satterthwaite degrees of freedom calculations. This will raise
  an error now and thus alert the user that the model fit was not successful.

Miscellaneous

• Use automatic differentiation to calculate Satterthwaite adjusted degrees of freedom,
  resulting in 10-fold speed-up of the Satterthwaite calculations after the initial model fit.
• Add an interactive confirmation step if the number of visit levels is too large
  for non-spatial covariance structures. Use options(mmrm.max_visits = ) to specify the
  maximum number of visits allowed in non-interactive mode.
• Removed free_cores() in favor of parallelly::availableCores(omit = 1).
• The model.frame() method has been updated: The full argument is deprecated and
  the include argument can be used instead; by default all relevant variables are
  returned. Furthermore, it returns a data.frame the size of the number of observations
  utilized in the model for all combinations of the include argument
  when na.action= "na.omit".
• Overall, seven vignettes have been added to the package. All vignettes have a slightly
  different look now to reduce the size of the overall R package on CRAN.
• The used optimizer is now available via component(., "optimizer") instead of previously
  attr(., "optimizer").

Full Changelog: v0.2.2...v0.3.6