Definition
A standard error is the sampling standard deviation; an estimated standard error estimates it. It is defined under a specified data-generating and estimation procedure and measures sampling uncertainty, not every source of model, measurement or selection uncertainty. The calculation must match dependence, heteroskedasticity and the estimand.
How it is calculated
Standard error of theta-hat = square root of the sampling variance of theta-hat.
| Symbol | Meaning | Unit |
|---|---|---|
| SE | standard error | theta units |
| theta | estimated parameter | parameter units |
| Var | sampling variance | squared parameter units |
Interpretation
If we could repeat the study many times, the estimate would move around. The standard error is the typical scale of that movement under the stated experiment. It is not a meter for all uncertainty surrounding a research claim.
Assumptions and conventions
-
The variance estimator reflects the data dependence and sampling design.
-
The target parameter and estimator are specified.
Same units as the estimator; variance has squared units.
Limitations and common misreadings
-
A small standard error under a misspecified model can be misleading.
-
Standard error is not standard deviation of the raw observations.
References
- Larry Wasserman. (2004). All of Statistics: A Concise Course in Statistical Inference. Springer Texts in Statistics. Chapter 6, Models, Statistical Inference and Learning — point estimation and standard error. doi:10.1007/978-0-387-21736-9. Source

