Glossary of statistical terms

facebook LinkedIn twitter Google+ Email

Prior and posterior probability (difference):

Consider a population where the proportion of HIV-infected individuals is 0.01. Then, the prior probability that a randomly chosen subject is HIV-infected is Pprior = 0.01 .

Suppose now a subject has been positive for HIV. It is known that specificity of the test is 95%, and sensitivity of the test is 99%. What is the probability that the subject is HIV-infected? In other words, what is the conditional probability that a subject is HIV-infected if he/she has tested positive?

The following table summarizes calculations. For the sake of simplicity you may consider the fractions (probabilities) as proportions of the general population.

. Test results .
HIV-status Positive Negative Total
infected 0.01*0.99 = 0.0099 0.01*(1-0.99)=0.0001 0.01
not infected 0.99*(1-0.95) 0.99*(0.95) 0.99
total: 0.0594 0.9406 1.00

Thus, the average proportion of positive tests overall is 0.0594, and the proportion of actually infected among them is 0.0099/0.0594 or 0.167 = 16.7%.

So, the posterior (i.e. after the test has been carried out and turns out to be positive) probability that the subject is really HIV-infected is 0.167.

The difference between prior and posterior probabilities characterizes the information we have gotten from the experiment or measurement. In this example the probability changed from 0.01 (prior) to 0.167 (posterior)

Note also the surprising result in this case, which, although hypothetical, is typical of many medical screening tests. Although the test is 95% effective in correctly identifying an HIV case, a person who tests positive actually has only a 16.7% chance of having the disease. This is due to the very low proportion of actually-infected people in the population -- most of the positive test results are false positives from the non-infected people who are being tested.

See also: A Priori Probability , Posterior Probability .

Browse Other Glossary Entries

Want to learn more about this topic? offers over 100 courses in statistics from introductory to advanced level. Most are 4 weeks long and take place online in series of weekly lessons and assignments, requiring about 15 hours/week. Participate at your convenience; there are no set times when you must to be online. Ask questions and exchange comments with the instructor and other students on a private discussion board throughout the course.

Introduction to Bayesian Statistics

This course will introduce you to the basic ideas of Bayesian Statistics. You will learn how to perform Bayesian analysis for a binomial proportion, a normal mean, the difference between normal means, the difference between proportions, and for a simple linear regression model.

Back to Main Glossary

Promoting better understanding of statistics throughout the world

To celebrate the International Year of Statistics in 2013, we started a program to provide a statistical term every week, delivered directly to your inbox. The Word of the Week program proved to be quite popular, and continues. The Institute for Statistics Education offers an extensive glossary of statistical terms, available to all for reference and research. Make it your New Year's resolution to improve your own statistical knowledge! Sign up here. Rather not have more email? Simply bookmark our home page and check our “Stats Word of the Week” feature.

Want to be notified of future courses?

Student comments