Point Estimate Calculator

What the point estimate calculator does

A point estimate is a single best guess at a population value from sample data. For a proportion, the obvious guess is the successes divided by the trials, but that simple figure can mislead in some situations. This calculator reports it along with three adjusted estimates, and picks the one best suited to your numbers.

It answers a question that sounds simple but has more to it than it seems: given so many successes out of so many tries, what is the true underlying rate. Below is how it works and why more than one answer is offered.

How to use it

1. Enter the number of successes and the number of trials.
2. Set the confidence level and z-score if you want, or leave the defaults for 95 percent.
3. Press Calculate for the best point estimate and all four methods, or Reset to clear it.

How the estimates are worked out

The starting point is the simplest estimate, the observed proportion, which is just the successes divided by the trials. The calculator then computes three adjusted versions that each nudge the estimate in a way that behaves better at the extremes.

Finally it chooses a best estimate based on where the simple proportion falls. For a proportion up to one half it uses the Wilson estimate, for a middling proportion it keeps the simple one, and for a proportion near certainty it uses the smaller of two adjusted values. The idea is to match the method to the situation rather than trusting one formula everywhere.

Why the simple estimate is not always best

The plain proportion runs into trouble at the edges. If you see 0 successes in 20 trials, it says the true rate is exactly zero, and if you see 20 out of 20, it says exactly 100 percent. Both are almost certainly too strong. A run of twenty without a success is good evidence the rate is low, but not proof it is impossible, and a handful more trials could easily turn one up.

The adjusted estimates fix this by gently pulling the guess away from 0 and 1 and toward the middle, as if adding a few imaginary observations on each side. The effect is largest exactly where the plain proportion is weakest, at the extremes and for small samples, and fades to nothing as the data grows. So the adjustments are a way of not overcommitting on thin evidence.

The four estimates it reports

The maximum likelihood estimate is the plain observed proportion, successes over trials. It is the most natural estimate and the best choice when the proportion is not near 0 or 1 and the sample is reasonable.

The Laplace and Jeffreys estimates both add a small amount to the successes and the trials, shrinking the guess toward one half by different amounts. The Wilson estimate, drawn from the well-regarded Wilson confidence interval, does the same in a way that holds up especially well for proportions near 0 or 1 and for small samples. The calculator's best estimate simply picks whichever of these fits your numbers.

A worked example

Suppose you run 20 trials and see 0 successes. The simple estimate, the maximum likelihood estimate, is 0 ÷ 20 = 0, which claims the true rate is exactly zero.

That is too strong a claim from 20 trials, so the calculator turns to the Wilson estimate, which comes out to about 0.08. It says the rate is low, around 8 percent, but not impossible, which is a far more honest reading of 0 in 20 than a flat zero. The adjustment is the difference between "never happens" and "seems rare, but we have only looked 20 times."

Entering your values

Enter the number of successes and trials, with successes no greater than trials. The confidence level and z-score default to 95 percent, which the Wilson estimate uses, and you can change them if you need a different level. All four estimates and the chosen best one are shown to four decimal places with trailing zeros trimmed.

Questions people ask

What is a point estimate?

A single best guess at a population value from sample data. For a proportion, the basic point estimate is the number of successes divided by the number of trials.

Why does it give four different estimates?

Because the simple proportion can mislead at the extremes and for small samples. The three adjusted estimates handle those cases better, and the calculator picks the one that suits your numbers.

Why is 0 out of 20 not an estimate of zero?

Because twenty trials without a success is evidence the rate is low, not proof it is zero. The adjusted estimates give a small positive value instead, which is more realistic.

Which estimate should I trust?

The best estimate the calculator selects is a sound default. The maximum likelihood estimate is fine away from the extremes, while the Wilson estimate is the safer choice near 0 or 1 and for small samples.

References

A quick note on where the methods here come from. Point estimation of a proportion, and the Wilson interval behind one of these estimates, are set out in the NIST/SEMATECH e-Handbook of Statistical Methods, the US government's public statistics reference. OpenStax Introductory Statistics is a free, widely used textbook covering point estimates.

1. NIST/SEMATECH e-Handbook of Statistical Methods (estimation and proportions). https://www.itl.nist.gov/div898/handbook/
2. OpenStax, Introductory Statistics (point estimation and proportions). https://openstax.org/details/books/introductory-statistics-2e