Topic 5.4 Biased and Unbiased Point Estimates: define an unbiased estimator (its sampling distribution centers on the parameter), and distinguish the bias of an estimator from its variability.

What this topic is asking

The College Board (Topic 5.4) wants you to define an unbiased estimator, a statistic whose sampling distribution centers on the parameter, and to distinguish the bias of an estimator from its variability.

Unbiased estimators

Unbiasedness is a statement about the center of the sampling distribution, not about any single estimate. No individual sample mean need equal $\mu$ (and usually none does), but if the sample means from all possible samples average to $\mu$, then $\bar{x}$ is unbiased. This is why the previous topics established that $\mu_{\bar{x}} = \mu$ and $\mu_{\hat{p}} = p$: those equalities are precisely the statements that $\bar{x}$ and $\hat{p}$ are unbiased estimators.

Bias versus variability

The classic image is a target. Bias is whether your shots cluster around the bullseye (unbiased) or around a point off to the side (biased). Variability is how tightly your shots group, regardless of where they center. You want shots that are both on target (unbiased) and tightly grouped (low variability). The two problems, wrong aim versus inconsistent aim, are fixed by different things, and conflating them is the error this topic guards against.

Why both properties matter

A good estimator needs both low bias and low variability, and recognizing the trade-off between them is the heart of Topic 5.4. An unbiased estimator that is wildly variable can be far from the parameter on any given sample, even though it is right on average, so unbiasedness alone does not guarantee a useful single estimate. A very precise estimator that is biased will be reliably wrong, consistently missing the parameter in the same direction, so low variability alone is no comfort either. The exam often presents two estimators, one unbiased but imprecise, one precise but slightly biased, and asks you to compare them, which requires you to evaluate both properties rather than declaring a winner on one. The practical resolution depends on context and on the size of the bias: a tiny bias with much lower variability can beat a large variability with no bias. Crucially, because bias cannot be reduced by collecting more data (only the variability shrinks with $n$), an unbiased estimator has a real advantage: you can drive its variability down with a larger sample and trust that it is converging on the right value. This is exactly why $\bar{x}$ and $\hat{p}$, being unbiased, are the estimators used throughout inference.

Connecting to sampling and inference

Topic 5.4 ties the descriptive idea of bias from Unit 3 to the formal sampling-distribution view. In Unit 3, a biased sampling method systematically missed the truth; here, a biased estimator systematically centers off the parameter, and again, more data does not cure it. The link to inference is direct: confidence intervals and significance tests assume the statistic is an unbiased estimator, so that the interval is centered on a value that, on average, equals the parameter. The standard error, the estimated standard deviation of the sampling distribution, that drives the width of every interval is a measure of the estimator's variability, and the $\dfrac{1}{\sqrt{n}}$ behavior from the previous topics is why larger samples give narrower, more precise intervals around an unbiased center. So understanding bias and variability separately is essential preparation: it tells you that inference works because the estimators are unbiased (right on average) and that precision improves with sample size (variability shrinks), the two facts every later procedure depends on.

Try this

Q1. Define an unbiased estimator in terms of its sampling distribution. [2 points]

• Cue. An estimator whose sampling distribution is centered on the parameter, that is, the mean of the sampling distribution equals the parameter, so it does not systematically over- or under-estimate.

Q2. Does increasing the sample size reduce an estimator's bias? Explain. [1 point]

• Cue. No; a larger sample reduces variability (the spread of the sampling distribution) but not bias, which is a systematic centering problem unaffected by sample size.