I am reading the book Modelling Survival Data in Medical Research. The book says that it's the Taylor series approximation to the variance of a function of a random variable. I learnt Taylor expansion back in high school and as far as I could recollect from my memory, Taylor expansion basically says that the function could be approximated by a polynomial:
If expanding f(x) at x=0, the Taylor expansion of f(x) would look like:
$$f(x)=f(0)+f'(0)(x-0)+\frac{f''(0)}{2!}(x-0)^2+...+o(1)$$
I don't know exactly what o(1) is because I never got a chance to know more advanced maths, but I know at the limit 0(1) should approach to 0.
However, I am still quite blind in deriving the approximation of variance in this case. Could someone please show me the step? I explain my understanding in Taylor series because I was hoping someone could point out if I have misunderstood the theories or if there are other relevant and important theorems and knowledge in deriving the variance approximation. Many thanks!
A first-order approximation should be enough. $$g(X) \approx g(\bar{X}) + \frac{d g(\bar{X})}{d X} (X-\bar{X})$$ Where $\bar{X}$ - some determined (fixed) value. Therefore, $g(\bar{X}) - \frac{d g(\bar{X})}{d X} \bar{X}$ is fixed term in the above expression and has zero variance.
From the property of variance $\text{Var}\{aX\} = a^2 \text{Var}\{X\}$ for any fixed value $a$ the desired expression follows immediately: $\text{Var}\{g(X)\} \approx [\frac{d g(\bar{X})}{d X}]^2 \text{Var}\{X\}$