Event History and Survival Analysis

Göran Broström



This vignette is not written yet! That is, it is under construction. With each new version of eha, this vignette will (hopefully) be more complete.

What is eha in relation to survival?

The eha package can be seen as a complement to the recommended package survival: In fact, eha depends on survival, and for standard Cox regression, eha::coxreg() simply calls survival::agreg.fit() or survival::coxph.fit(), functions exported by survival. The simple reason for this is that the underlying code in these survival functions is very fast and efficient. However, eha::coxreg() has some unique features: Sampling of risk sets, The “weird” bootstrap, and discrete time modeling via maximum likelihood.

I have also put effort in producing nice and relevant printouts of regression results, both on screen and to \(\LaTeX\) documents (HTML output is on the TODO list). By relevant output I basically mean avoiding misleading p-values, show all factor levels, and use the likelihood ratio test in front of the Wald test where possible.

Parametric survival models in eha

There is a special vignette describing the theory and implementation of the parametric failure time models. It is not very useful as a user’s manual.

Accelerated Failure Time models

The parametric accelerated failure time (AFT) models are present via eha::aftreg(), which is corresponding to survival::survreg(). An important difference is that eha::aftreg() allows for left truncated data.

Parametric proportional hazards models

Parametric proportional hazards (PH) modeling is available through the functions eha::phreg() and eha::weibreg(), the latter still in the package for historical reasons. It will eventually be removed, since the Weibull distribution is also available in eha::phreg().


The primary applications in mind for eha were demography and epidemiology. There are some functions in eha that makes certain common tasks in that context easy to perform, for instance rectangular cuts in the Lexis diagram, creating period and cohort statistics, etc.