NEWS

---

ef2 2.0.0

• package renamed from ef to ef2 so it can be installed alongside ef 1.x; fitted models now have class efp2
• models are fitted with nlminb() rather than optim() (BFGS), which falsely converged on some platforms; control is passed to nlminb()
• the likelihood is computed on the log scale, so it stays finite when capture probabilities are close to 0 or 1
• efp() is renamed efp2(), so ef and ef2 can be attached together
• remove logit(), invlogit() and transpar(), use qlogis() and plogis() from stats instead
• as.gam() is renamed as.gam2() and ef_data is renamed ef2_data, so no names clash with ef
• fix the choice of coefficients removed for the null space of a smooth, which could give a singular penalty and NaN likelihood on macOS
• when standard errors cannot be computed, efp2() warns which parameters are affected, instead of "NaNs produced"
• new capture_prob() and abundance() give the cumulative capture probability, abundance and density of each sample with confidence intervals, optionally summed over groups of samples with by; the cumulative capture probability is also reported in $P
• abundance(target = "expected") gives confidence intervals for the expected abundance, when the number of fish present is Poisson

• move likelihood computation to RTMB, the package no longer has compiled code
• random effects can be included using s(f, bs = "re") terms in the model formula, for example for between visit and between sample variation
• remove overdispersion() and BICadj(), these are replaced by visit and sample random effects
• remove the sample_re argument of efp()
• smooth terms are now penalised and the amount of smoothing is estimated, by fitting smooths as random effects. Previously smooths were unpenalised regression splines, which are still available using fx = TRUE, for example s(x, k = 3, fx = TRUE)
• the recommended smooths are first and second order random walks, using s(f, bs = "mrf", xt = list(penalty = rw_penalty(levels(f)))). The new function rw_penalty() creates the penalty, and these are fitted using sparse matrices so they are quick to fit even with many levels. Other smooths from mgcv can be used but are slower to fit
• add a predict() method for efp fits, with predictions for new data, standard errors and the option to exclude random effects
• fitted(type = "pi") now returns the probability of capture in each pass
• add day of the year (doy) to the ef_data example dataset, a continuous covariate which can be used to fit smooth terms
• add examples of smooth terms to the help and the README (#1)
• add summary(), vcov(), confint(), logLik() and formula() methods for efp fits. residuals(), anova() and plot() now give an informative error, rather than failing in the methods inherited from glm
• remove the unused init argument of efp()

ef2 1.2.0 (2020-10-18)

• move likelihood computation to TMB
• introduce random effects at the sample level

ef2 1.1.0 (2016-08-01)

• initial package developed using STAN for likelihood computations