As described in the previous sections, the delta method can be used to produce analytic expressions for standard errors of functions of estimated parameters that are asymptotically normally distributed. The formulas are specific to each function and are approximations that rely on large samples and the validity of the Central Limit Theorem to be accurate.
Alternatives to the analytic asymptotic distributions of estimators and the delta method are computer intensive resampling methods such as the jackknife and the bootstrap. These resampling methods can be used to produce numerical estimated standard errors for estimators and functions of estimators without the use of mathematical formulas. Most importantly, these jackknife and bootstrap standard errors are: (1) easy to compute, and (2) often numerically very close to the analytic standard errors based on CLT approximations. In some cases the bootstrap standard errors can even be more reliable than asymptotic approximations.
There are many advantages of the jackknife and bootstrap over analytic formulas. The most important are the following:
- Fewer assumptions. The jackknife and boostrap procedures typically requires fewer assumptions than are needed to derive analytic formulas. In particular, they do not require the data to be normally distributed or the sample size to be large enough so that the Central Limit Theorem holds.
- Greater accuracy. In small samples the bootstrap is often more reliable than analytic approximations based on the Central Limit Theorem. In large samples when the Central Limit Theorem holds the bootstrap can be even be more accurate.
- Generality. In many situations, the jackknife and bootstrap procedures are applied in the same way regardless of the estimator under consideration. That is, you don’t need a different jackknife or bootstrap procedure for each estimator.