← Back to Statistics

Point estimation

A point estimate is a single best guess for a parameter. Good estimators are unbiased, precise, and grounded in a sound method.

Estimators and estimates

An estimator is a rule for computing a guess from data, such as 'take the sample mean'. An estimate is the number that rule produces for a particular sample. \(\bar{x}\) estimates \(\mu\), \(\hat{p}\) estimates \(p\), and \(s\) estimates \(\sigma\).

Unbiasedness

An estimator is unbiased if its sampling distribution is centred on the true parameter. \(\bar{x}\) and \(\hat{p}\) are unbiased. The sample variance with denominator \(n - 1\) is unbiased for \(\sigma^2\); with \(n\) it would systematically underestimate. That is why \(n - 1\) appears in the formula.

Efficiency

Among unbiased estimators, prefer the one with the smallest standard error. For normal data, the sample mean is more efficient than the sample median: it gives more precise estimates from the same data. For heavy-tailed data, the median can be better because it resists outliers.

Consistency

An estimator is consistent if it gets closer and closer to the true value as the sample size grows. The law of large numbers guarantees this for \(\bar{x}\). Consistency is a minimum requirement: an estimator that does not improve with more data is not worth using.

Maximum likelihood, briefly

A general method chooses the parameter value that makes the observed data most probable. If 7 of 10 flips are heads, the maximum likelihood estimate of \(p\) is \(0.7\). Maximum likelihood is used throughout modern statistics and machine learning because it works for complicated models.

Point estimates need uncertainty

A single number hides how uncertain it is. Reporting '42% support' without a margin of error invites overconfidence. The next units attach uncertainty to estimates through confidence intervals and test claims about parameters with hypothesis tests.

Worked example

A sample of 50 battery lifetimes has \(\bar{x} = 412\) hours and \(s = 35\) hours. The point estimate of the mean lifetime is 412 hours, and its estimated standard error is \(\frac{35}{\sqrt{50}} \approx 4.9\) hours. That standard error tells us the estimate is probably within about 10 hours of the true mean.

Common mistakes

Confusing an estimator (the method) with an estimate (the number). Reporting a point estimate without its standard error. Assuming an unbiased estimator is always close to the truth; it can still be imprecise with small samples.

Key takeaways

Estimators should be unbiased, efficient, and consistent. \(\bar{x}\), \(\hat{p}\), and \(s^2\) (with \(n - 1\)) are standard choices. Maximum likelihood is a general estimation method. Always pair a point estimate with its standard error.

Flashcard

Estimator

A rule for estimating a parameter

Such as the sample mean.

Check your understanding

In 40 coin flips there are 26 heads. What is the maximum likelihood estimate of p?

← Previous lesson Next lesson →