standard error
/ STAN-derd ERR-er /
Every estimate is a little shaky — measure a thing twice with different data and you get slightly different answers. The standard error is the number that says exactly how shaky: it is the typical amount by which an estimate would bounce around if you could repeat the whole study many times. It converts the vague worry 'is this number reliable?' into a concrete figure you can attach a plus-or-minus to.
Formally, the standard error is the standard deviation of an estimator's sampling distribution. The most famous case is the sample mean: if individual observations have standard deviation s, then the mean of n of them has standard error s / sqrt(n). Note carefully that this shrinks as the sample grows — four times the data halves the standard error — which is the precise mathematical reason that more data buys more confidence. In practice we do not know s, so we plug in the sample standard deviation; the resulting estimated standard error is what software reports next to every coefficient and rate.
The standard error is the actuary's universal yardstick of estimation precision. It is the engine inside a confidence interval (estimate plus-or-minus about two standard errors gives a rough 95% interval) and inside a hypothesis test (the test statistic is the estimate divided by its standard error). When a regression prints a tiny standard error beside a fitted relativity, you can lean on that factor; when the standard error is large relative to the estimate, the apparent effect may be noise. Crucially, the standard error captures only random sampling error — it says nothing about a biased sample, a wrong model, or bad data.
A mortality study estimates a rate of 2.0% from claims with a standard deviation that yields a standard error of 0.15%. The actuary reports 2.0% give-or-take about 0.3% (two standard errors), signaling the rate is pinned down fairly tightly.
Standard error turns an estimate into estimate-plus-or-minus, the basis of every confidence statement.
Standard deviation describes the spread of the data; standard error describes the spread of an estimate. They are different things, and the second is almost always smaller because of the sqrt(n) in the denominator.