- implement different maximum-likelihood based estimation methods (naive ML, weighed ML, corrected ML), with option to profile out Weibull’s scale-parameter
- use some log-transformed parameters internally for more stable and less constrained optimization
`test_diff`

: allow for likelihood-ratio test- include restricted mean survival time function for delayed exponential and Weibull
- [experimental] implement two delay phases for exponential and for Weibull
- update simulation R-scripts (in
`inst/scripts/`

) to also use ML-based estimation and LR-tests

- fix package help: due to changes in roxygen

- check for minimal number of observations and fail early if not enough observations
`test_diff`

:- test-statistic gets lower bound of 0 enforced (a restricted model can not have better fit than unrestricted model)
- deactivate Anderson-Darling (AD) GOF-test as its performance in simulations under two-group setting was unsatisfactory

- include simulation R-scripts in package under
`inst/scripts/`

folder. When the package is installed the scripts are found at in the -directory`scripts/`

within the package. - rename methods:
- ‘MSE’ => ‘MPSE’ (as the term MSE has already other meanings)
- ‘MLE’ => ‘MLE0’ (to indicate that this is the standard MLE that is not appropriate for delay models)

- rework implementation of objective-function: do not rely on attributes but use the environment of the objective function instead. This means fewer copies of information are made.

- New implementation of delay smoothing: use objective function along
candidate delay values to guide in smoothing
- define an interval for delay candidate values
- sample from this interval according to the objective function: the better the value of the objective function the more likely it is to sample a value. For speed of computation, we vary only delay and keep the other parameters at their estimate.
- use
`smd_factor`

to either concentrate or flatten around the best delay estimate

- Better implementation for MLE-fitting:
- Support MLE-fitting also for Weibull (one group setting only)
- Move MLE-objective function into a general objective function factory method

- correct URL for bug reports in DESCRIPTION

- more robust test that passes also on Apple-silicon
- doc: add homepage in DESCRIPTION

- use LGPL-licence

- doc: further improvements (document all return values)

- doc: small improvements

- README: better intro text and more text for example

- small changes for CRAN-warnings
- test: bigger data set for delay estimation

- Fix errors/warnings from check
- document every function parameter
- S3-methods have right parameters
- specify package name for external functions (e.g., from
`stats`

) - global variables (use
`.data`

and`aes_`

)

- add
`cran-comments.md`

file

- confint: use better default values
- confint: when using bootstrap smoothing for delay, make sure the
smoothing keeps limited for tiny estimated rates lambda (see
`SMD_MINRATE`

) - bug fix:
`plot.incubate_fit`

has legible colour legend title

- data: add experimental data
`stankovic`

from a publication Stankovic et al EGFL7 enhances surface expression of integrin doi:10.15252/emmm.201708420 (2018) as example data set - This version of package is also published at https://zenodo.org/record/6462382

`power_diff`

: improve heuristics in iterative search for sample size, given power

`test_diff`

: change name of P-values (within object)`power_diff`

: allow to specify power and do iterative computation

- confint: smooth delay for parametric bootstrap & delayed exponential model via rectified Gaussian (instead of normal Gaussian)

- confint: add option to smooth delay for parametric bootstrap & delayed exponential model

- confint: redefine
`logshift`

to mean what is*added*to the argument of log, before it was subtracted. This meaning seems more natural.

- confint: logshift accepts also
`NA`

and`NULL`

. Useful for simulation studies with optional logshift-setting.

- confint: parameterized logshift for delay, new default is 0.01

- confint: log-transformation revisited: shift mainly for delay (and hardly a shift for the other parameters)
- confint: extend test coverage to confint routines

- confint: allow for log-transformation of bootstrapped coefficients

- confint: separate bootstrap data generation and inference step. This allows for more efficient confint-simulations.

- enhance test-coverage for package
- reorganize tests to better match the R-script

`test_diff`

: do GOF-tests also for unrestricted model (e.g.`gof_pearson1`

) besides for the restricted null model (renamed to`gof_pearson0`

)

- remove dependency on
`dplyr`

- add
`bs_infer = 't0'`

- add +3 to degrees of freedom for t-quantiles in
`bs_infer='t'`

and`='t0'`

- version bump belated: these features were temporarily also released as 0.7.3

- fixed mapping of bootstrap inference names for
`boot`

-package

- allow to use
`boot`

package to calculate confidence interval

- allow to choose which tests to perform in
`test_diff`

(helpful when calculation of AD GOF-tests fails on older R-installations)

- change of S3-classes:
- for delay model fitting the new class is
`incubate_fit`

- for tests the new class is
`incubate_test`

- for delay model fitting the new class is
- confidence intervals supports now simple data generation to draw from the data with replacement (besides parametric bootstrap)
- preliminary support for MLE-fitting: this allows to compare confidence intervals based on MSE vs MLE

- make handling of ties in calculation of spacings within the objective function more robust

- implement AD-GOF test within
`test_GOF`

- add separate function
`test_GOF`

for goodness of fit (GOF) tests based on a fitted model - add Moran’s GOF-test

- bug fix in MSE-criterion
- bug fix in tie-handling
- function
`estimRoundingError`

to estimate rounding error (for tie handling)

- simplified power function (dropped stuff and conventions from
`sscn`

-package)

`ties='equidist'`

is default method to handle ties

- more choice in handling of ties:
`equidist`

and`random`

- rename function
`test_delay_diff`

to`test_diff`

- add confidence intervals (based on basic bootstrap) for model parameters

- Weibull:
`bind=`

is implemented for all possible parameter combinations - internal clean up

- more warnings when MSE-optimization fails
- bug fix for power simulation
- bug fix in testing

- initial start as a move out from package
`sscn`

(git history is not moved along, though) - code robustness: add
`try`

code at lower level, directly at where`optim`

is called.