The value returned by nobs.kgaps() was incorrect in cases where there are censored K-gaps that are equal to zero. These K-gaps should not contribute to the number of observations. This has been corrected.
In cases where the data used in kgaps are split into separate sequences, the threshold exceedance probability is estimated using all the data rather than locally within each sequence.
A logLik method for objects inheriting from class "kgaps" has been added.
In the (unexported, internal) function kgaps_conf_int() the limits of the confidence intervals for the extremal index based on the K-gaps model are constrained manually to (0, 1) to avoid problems in calculating likelihood-based confidence intervals in cases where the the log-likelihood is greater than the interval cutoff when theta = 1.
In the documentation of the argument k to kgaps() it is noted that in practice k should be no smaller than 1.
The function kgaps() also return standard errors based on the expected information.
In the package manual related functions have been arranged in sections for easier reading.
Activated 3rd edition of the testthat package
kgaps(), kgaps_imt() and choose_uk() can now accept a data argument that
NAs.cheeseboro is included, which is a matrix containing some missing values.kgaps(), the functions kgaps_imt() and choose_uk() now have an extra argument inc_cens, which allows contributions from censored K-gaps to be included in the log-likelihood for the extremal index.inc_cens in kgaps() (and in kgaps_imt() and choose_uk()) and is now inc_cens = TRUE."confint_gaps" returned from confint.kgaps().confint.spm() and confint.kgaps() the input confidence level is included in the output object.