In some sources 1 one reads that the frailty model expands on the cox proportional hazards model $$h_i(t|x_i)=h_0(t)\exp(\beta x_i)$$ by adding a frailty term $z$ like so $$h_i(t|x_i,z_i)=h_0(t)\exp(\beta x_i+z_i)$$

In other sources I see frailty figuring as a factor “acting multiplicatively” 2 on the baseline hazard function $h_0(t)$ $$h_i(t|x_i,z_i)=z_ih_0(t)\exp(\beta x_i)$$

How are individual frailty and shared frailty correctly expressed?

1: Predicting Horse Race Winners Using Advanced Statistical Methods

2: Frailty Models


1 Answer 1


These are essentially equivalent, except for a different interpretation of $z_i$ in the two representations. For the exponential function,

$$\exp(x+y)=\exp(x) \exp(y) .$$

If you take $\exp(z_i)$ from the first representation

$$h_i(t|x_i,z_i)=h_0(t)\exp(\beta x_i+z_i)$$

you get $z_i$ in the second representation

$$h_i(t|x_i,z_i)=z_ih_0(t)\exp(\beta x_i).$$

The first representation is used for modeling in common R packages. In the R survival package, the coxph() function can model $z_i$ in $\exp(\beta x_i + z_i)$ with either a Gaussian, gamma, or t distribution. That's explained in Chapter 9 of Therneau and Grambsch. The coxme package extends what you show as $z_i$ in the first representation to more complicated random effects with Gaussian distributions.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.