Skip to content

    Odds and probabilities are not the same thing

    Odds and probabilities are not the same thing

    If a study reports that one group had “twice the odds” of an outcome, it can sound as though the outcome was twice as likely. Sometimes those numbers are close, but odds and probability aren't the same calculation. The difference becomes especially important when an outcome is common.

    You don't need to become a statistician to read the claim more carefully. A small example shows why it helps to ask for the actual numbers in each group.

    Start with 100 people

    Imagine a fictional weight-care study with 100 people in one group. During a defined period, 20 experience a specified health event and 80 do not.

    The probability, or risk over that period, is 20 out of 100: 20%. The odds compare the 20 people who experienced the event with the 80 who did not: 20 to 80, or one to four.

    Both describe the same group, but they use different denominators. Probability compares the event group with everyone. Odds compare the event group with the people who did not have the event.

    These numbers are invented for illustration. They don't describe a real treatment or forecast any person's outcome.

    Now compare a second group

    Suppose another fictional group receiving a different approach has 100 people, of whom 40 experience the event and 60 do not. Its probability is 40%, and its odds are 40 to 60, or two to three.

    Compare the probabilities: 40% divided by 20% gives a risk ratio of 2. In this example, the event occurred twice as often proportionally in the second group over the stated period.

    Compare the odds instead: 40/60 divided by 20/80 gives an odds ratio of about 2.67. The odds ratio is larger than the risk ratio even though both calculations use the same group counts.

    That's why an odds ratio of 2.67 shouldn't be translated into “2.67 times the risk.” The numbers answer related but different questions.

    Why do researchers use odds ratios?

    Different study designs and statistical methods use different measures. An odds ratio can be useful for analysis and comparison. The issue isn't that the measure is invalid; it's that the wording can be mistaken for a risk ratio or an absolute probability.

    When an outcome is rare, an odds ratio and a risk ratio may be numerically close. As outcomes become more common, the difference can be substantial. You need the underlying risks, when available, to understand the practical size of the finding.

    Some published odds ratios are adjusted for other characteristics. In that case, simply calculating a ratio from the raw group totals may not reproduce the reported adjusted result. Ask how the researchers made the comparison and what the adjustment was intended to address.

    A ratio needs a reference point

    For both an odds ratio and a risk ratio, a value of one represents no difference between the compared groups on that measure. Values above or below one indicate a direction relative to the stated reference group. But whether an increase is favorable depends on the outcome: an increase in recovery is different from an increase in hospitalization.

    Always check which group is the reference and what event was counted. “Higher odds” can sound positive or negative depending on the wording, so translate it back to the actual outcome before drawing a conclusion.

    A ratio also doesn't tell you the absolute number of people affected. A change from one in 1,000 to two in 1,000 and a change from 20 in 100 to 40 in 100 both have a risk ratio of two, but the absolute differences are very different.

    Don't forget uncertainty and design

    A study reports an estimate, not a guaranteed value. Look for its confidence interval and ask how precise the estimate is. If the interval is wide, it may include several effects that would lead to different practical conclusions.

    Also consider the study design. An observed association doesn't automatically prove that a treatment or behavior caused the difference. Participants may differ in other ways, and adjustment cannot always account for every relevant factor. Randomized comparisons and observational studies answer questions differently and each has limitations.

    The follow-up period matters as well. An event rate over six months isn't directly interchangeable with a rate over several years. A headline without a population, comparison and time frame leaves out information you need.

    Ask for the numbers behind the claim

    Try: “How many people had the outcome in each group, and over what period? Is this an odds ratio or a risk ratio? Was the estimate adjusted, and how uncertain is it?” Your clinician can help explain what the study means for your health history.

    The goal isn't to distrust every ratio. It's to keep the denominator in view. Once you know who is being compared and how the numbers were calculated, the headline becomes much easier to understand.