I treat both types of models here because they are also basically invoked by the same command, rate ( );, even though they may have some different subcommands. The basic structure for some simple models is like this:
| rate( | |
| xa(j,k) = Vara, Varb,..., | |
| )= model_number; |
xa refers to the covariates in the so called a-term of the different models, which are given here as Vara, Varb and so on; I will say a little bit more on this shortly. The indices j and k in the parentheses refers to the transition from origin j to destination k; if you have several destinations and/or origins, you may repeat the xa subcommand with different indices (and the same applies to the xb, xc and xd subcommands wherever appropriate). Finally, the model_number refers to a number that indicates the type of model to be estimated. The different models available are as follows:
| rate = | 1 | Cox model (Partial Likelihood) |
| 2 | Exponential model | |
| 3 | Piecewise constant exponential model | |
| 4 | Polynomial rates, I | |
| 5 | Polynomial rates, II | |
| 6 | Gompertz-Makeham models | |
| 7 | Weibull model | |
| 8 | Sickle model | |
| 9 | Log-logistic model, type I | |
| 10 | Log-logistic model, type II | |
| 12 | Log-normal model | |
| 13 | Generalized gamma model | |
| 14 | Inverse Gaussian model | |
| 16 | Piecewise constant exponential model with period-specific effects |
|
| 20 | Logistic regression model | |
| 21 | Complementary log-log model |
Definition of covariates depends somewhat on the model to be estimated, since some models consist of several terms, called "a", "b" and/or "c" term, respectively. Usually, the "a" term of the model refers to those variables and their associated parameters that directly affect the baseline hazard, while the "b" and "c" terms refer to other parameters that influence the shape of the hazard, as for instance in Weibull or log-logistic models. But in some cases, the "a" term has a different meaning, for instance, as the "Makeham" component in a Gompertz-Makeham model. In addition, some models need extra commands. For instance, to estimate a piecewise constant exponential model, you have to provide information about the time periods for the different constants. Therefore, you have to study carefully chapter 16.7 of the TDA manual and the examples provided therein when you proceed to the estimation of more complex models.
In our simple case, we have only one transition from state "0" (attending college) to state "1" (dropping out of college), and we want to estimate a model containing the variables sex, grade, part-time student, lag (time elapsed between finishing high school and starting college), and the part-time * lag interaction. Therefore, we shall write, for instance:
rate(
xa(0,1) = V4,V5,V6,V7,V10,
)=1;
to estimate a Cox regression model.
Last update: 28 Jan 2000