## 4.2 Transformation techniques

If $$T : \mathcal{Z} \to \mathbb{R}$$ is a map and $$Z \in \mathcal{Z}$$ is a random variable we can simulate, then we can simulate $$X = T(Z).$$

Theorem 4.1 If $$F^{\leftarrow} : (0,1) \mapsto \mathbb{R}$$ is the (generalized) inverse of a distribution function and $$U$$ is uniformly distributed on $$(0, 1)$$ then the distribution of $F^{\leftarrow}(U)$ has distribution function $$F$$.

The proof of Theorem 4.1 can be found in many textbooks and will be skipped. It is easiest to use this theorem if we have an analytic formula for the inverse distribution function. But even in cases where we don’t it might be useful for simulation anyway if we have a very accurate approximation that is fast to evaluate.

The call RNGkind() in the previous section revealed that the default in R for generating samples from $$\mathcal{N}(0,1)$$ is inversion. That is, Theorem 4.1 is used to transform uniform random variables with the inverse distribution function $$\Phi^{-1}$$. This function is, however, non-standard, and R implements a technical approximation of $$\Phi^{-1}$$ via rational functions.

### 4.2.1 Sampling from a $$t$$-distribution

Let $$Z = (Y, W) \in \mathbb{R} \times (0, \infty)$$ with $$Z \sim \mathcal{N}(0, 1)$$ and $$W \sim \chi^2_k$$ independent.

Define $$T : \mathbb{R} \times (0, \infty) \to \mathbb{R}$$ by $T(z,w) = \frac{z}{\sqrt{w/k}},$ then $X = T(Z, W) = \frac{Z}{\sqrt{W/k}} \sim t_k.$

This is how R simulates from a $$t$$-distribution with $$W$$ generated from a gamma distribution with shape parameter $$k / 2$$ and scale parameter $$2$$.