Point estimate vs interval estimate

Point estimate vs interval estimate: the short answer

Both answer the same question, what is the unknown parameter, and they differ in what they admit about themselves.

A point estimate is a single number: the value your statistic took in the one sample you collected, offered as the best guess. From 96 approvals in a random sample of 150 adults, the point estimate is the sample proportion $\hat{p} = \frac{96}{150} = 0.64$ (read "p-hat").

An interval estimate is a range of plausible values, built as the point estimate plus or minus a margin of error. From that same sample, the 95% interval estimate runs from 0.563 to 0.717.

The real difference is not one number against two. It is that the point estimate says nothing about its own precision and the interval estimate carries its precision with it. The number 0.64 looks equally solid whether it came from 150 people or from 25. The range 0.563 to 0.717 tells you at a glance that these data pin the proportion down to about 15 percentage points and no further.

What a point estimate is

A point estimate is the value of a statistic, used as the estimate of a parameter. Each parameter has a standard partner: the sample mean $\bar{x}$ (read "x-bar") estimates the population mean $\mu$ (mu), the sample proportion $\hat{p}$ estimates the population proportion $p$, and the sample standard deviation $s$ estimates $\sigma$ (sigma).

The word point is the whole content of the term. It means a single location on the number line with no width attached. Nothing inside the number 0.64 records the sample size, the variability of the data, or how far the estimate could reasonably sit from $p$. Two studies can both report 0.64 and mean very different things by it: with $n = 150$ the 95% margin of error is 0.0768, and with $n = 600$ it is 0.0384, half as wide, from the identical point estimate.

None of that is an argument against point estimates. The point estimate is the center of every interval you build and the statistic in every test you run, so you compute it first either way. It is an incomplete answer on its own, which is why the course moves from topic 3.1 Estimators into confidence intervals rather than stopping there.

What an interval estimate is

An interval estimate reports a range of plausible values instead of one. In this course it always takes the form on the formula sheet,

so the range sits symmetrically around the point estimate and the margin of error is exactly half its width. Attaching a stated confidence level is what turns an interval estimate into a confidence interval, and the confidence interval is the only interval estimate the AP course constructs.

Return to the 150 adults. The standard error is $\sqrt{\frac{0.64(0.36)}{150}} = \sqrt{0.001536} = 0.0392$, the 95% critical value is $z^* = 1.96$, and the margin of error is $1.96 \times 0.0392 = 0.0768$. The interval estimate is $0.64 \pm 0.0768$, which runs from 0.563 to 0.717. Step by step, that construction is how to calculate a confidence interval, and the confidence interval calculator does the same arithmetic.

The report now answers a question the single number could not. It separates the values these data leave standing from the values they rule out at that level: 0.55 sits outside the interval, 0.70 sits inside it. The level itself describes the method, not the one range you computed, a distinction worked through in confidence level vs confidence interval.

The differences side by side

Nearly every row is the same point said again: the interval estimate is the point estimate with its precision written down beside it.

A point estimate is almost always wrong

Here is the fact that makes the rest of inference make sense. A point estimate is hardly ever exactly right, and when it is you have no way of knowing. Neither of those is a flaw in the estimator.

Take a mean. For a continuous population the parameter $\mu$ is a real number, and the sample mean $\bar{x}$ is a continuous random variable. The probability that a continuous random variable takes any one particular value is 0, so the probability that $\bar{x}$ lands exactly on $\mu$ is 0. Not small. Zero.

Proportions weaken that claim rather than repeat it, because counting leaves the exact hit possible. With $n = 150$, the sample proportion can only be one of the 151 fractions $\frac{0}{150}, \frac{1}{150}, \ldots, \frac{150}{150}$, spaced about 0.0067 apart. If the true proportion is $p = 0.615$, no sample of 150 can produce it, because $150 \times 0.615 = 92.25$ is not a whole number of successes, and the best any sample can do is land nearby. But when $p$ is one of those 151 values the exact hit is live: at $p = 0.62 = \frac{93}{150}$, about 6.7% of samples land on it, roughly one in fifteen.

So "is the estimate right" is the wrong question to ask a point estimate. For a continuous parameter the answer is no with probability 1; for a discrete one it is usually no, and on the sample where it is yes nothing inside that sample tells you so. The useful question is "how close", and one number cannot answer it. That is the reason interval estimates exist. The interval replaces a claim that is probably false and never checkable with a claim that holds at a stated rate: the method behind 0.563 to 0.717 captures the parameter 95% of the time in repeated random sampling.

What makes an estimator good

Topic 3.1 is titled Estimators, and it asks the question that opens up once you accept that a point estimate almost always misses: if you cannot count on any of them being exactly right, what makes one estimator better than another? Two properties, and they are independent of each other.

Bias is where the estimates center. An estimator is unbiased when the mean of its sampling distribution equals the parameter, so it has no systematic pull high or low. From a random sample, $\bar{x}$ is unbiased for $\mu$ and $\hat{p}$ is unbiased for $p$.

Variability is how tightly the estimates cluster. The variability of an estimator is the standard deviation of its sampling distribution: $\frac{\sigma}{\sqrt{n}}$ for $\bar{x}$ and $\sqrt{\frac{p(1-p)}{n}}$ for $\hat{p}$. Smaller is better, and a larger sample buys it.

A good estimator is unbiased and low in variability, centered on the truth and rarely far from it. Notice that both criteria describe the sampling distribution, never the one estimate in your hand, which is precisely what a point estimate hides and an interval estimate exposes. The width of the interval is built from the standard error, the estimated variability, so the interval is the spread of the sampling distribution made visible on your page. An unbiased estimator with high variability gives an honest point estimate and a uselessly wide interval, and the usual fix is more data, since the AP course fixes the formula for you.

The classic mix-up and how to avoid it

The mistake is not confusing the two objects, which look nothing alike. It is reporting the point estimate as though it were the parameter. "Sixty-four percent of adults approve" computes $\hat{p}$ and writes $p$. What the sample supports is narrower: the sample proportion was 0.64, and the data are consistent with population proportions from about 0.563 to 0.717.

The second error runs the other way, treating the confidence level as a statement about where the parameter sits inside the one interval you computed. Once that interval is a fixed pair of numbers it either contains the parameter or it does not, and the 95% describes how often the procedure succeeds across repeated samples. Keep the confidence attached to the method, not to the range.

A third mistake is quieter and costs points on free-response. A narrow interval is precise, not accurate. Precision is width, and it comes from the sample size, the confidence level, and how variable the data are. Accuracy depends on the sample being random and the conditions holding. Measure only the bottles from the fastest filling head and you get a tight interval around the wrong number, and nothing in the arithmetic will warn you.