Version 0.5.0: 2026-09-18
        - Validate data and controls with ordinary R errors. Accept numeric or
          integer 0/1 outcomes, preserve missing teacher identifiers, and allow
          missing values in unused covariates. Reject unsupported formula offsets.
        - Honor verbose=FALSE and the selected binomial link throughout fitting.
          REML=TRUE remains the default and has an explicit regression test.
        - Correct joint-model VP updates and ML/REML covariance profiling. Return
          estimates, pseudo-responses, weights and prediction-error covariances
          from the same final working model.
        - Respect EM iteration limits, report convergence explicitly, and guard
          residual covariance updates against invalid matrices and endless loops.
        - Retain fits when optional Hessian differentiation fails and report its
          status. Use unrounded values for inference and consistently suppress
          ordinary Wald tests and intervals for variance components.
        - Support independent-response covariance Hessians and C++ benchmarking
          together with VP and school effects without overriding user options.
          Apply independence constraints to school covariance components too.
        - Retain known teacher histories and last-attended schools when scores
          are missing. Support the default fixed-effects model for a single year.
          Preserve user predictor columns and numeric teacher-year ordering
          beyond nine years.
        - Check R/C++ covariance-update agreement during benchmarking; protect
          native exception boundaries and qualify registered routine lookups.
        - Correct summary residual selection, print return values, and plotting
          arguments, year labels and restoration of graphics prompting.
          Include school covariance summaries and handle undefined correlations.
        - Move package documentation to roxygen source headers, correct data
          counts and covariance descriptions, and add testthat regression and
          independent numerical-reference tests.

Version 0.4-3: 2019-04-22
	- Added the joint covariance matrix, C, of the estimated fixed effects and
		predicted random effects to the output. See Henderson (1975).
	- Paper reference added to manuals and to CITATION file.

Version 0.4-0: 2017-08-13
        - New arguments added to the RealVAMS function
	-- independent.responses defaults to FALSE. If TRUE,
		this option will model the responses independently
		by fixing the covariances in G at 0 as well as the
		covariances in the last row/column of R. The resulting
		estimates are the same as those that would be obtained by
		modelling the test scores in package GPvam (with REML=FALSE)
		and modelling the binary respones in SAS GLIMMIX (RealVAMS
		has been validated against these programs).
	-- cpp.benchmark defaults to FALSE. If TRUE, this option will
		perform the calculations shown in equation (16) of Karl,
		Yang, Lohr (2013) using both R and the embedded C++ code
		to demonstrate the time savings of using C++. A summary 
		table is printed at the end.
	- The functions in the code of the file vp_cp (the estimation routine)
		have been commented to explain their purpose and to reference
		which equations from Karl, Yang, Lohr (2013) or Wolfinger and
		O'Connell (1993) they represent

Version 0.3-3: 2017-03-02
        - Add generic summary, plot, and print functions.
	- Added detail to documentation for the RealVAMS function
	- Reformatted user manual documentation for simulated data
	- Renamed some of the elements of the RealVAMS class

Version 0.3-2: 2015-07-19
        - Minor updates to package structure to match new CRAN requirements.

Version 0.3-1: 2014-11-01
        - First public release.
