When a confidence interval includes 0, it means the data are compatible with no real effect or no real difference. In statistical terms, the result is not statistically significant at the corresponding threshold (for a 95% confidence interval, that threshold is a p-value of 0.05). But “not significant” doesn’t always mean “no effect,” and the width of that interval matters more than most people realize.
Why Zero Is the Key Number
A confidence interval gives you a range of plausible values for an effect. If you’re comparing two groups, say a drug versus a placebo, the interval tells you how big the difference between them might realistically be, given the data you collected. Zero represents the null hypothesis: no difference between the groups.
If the entire interval sits above zero, the data suggest a real positive effect. If it sits entirely below zero, the data suggest a real negative effect. But if the interval stretches from a negative number through zero to a positive number, the data can’t rule out the possibility that there’s no difference at all. That’s what it means to “include zero.”
This applies specifically to additive measures like mean differences and risk differences. For ratio-based measures like odds ratios or relative risks, the null value isn’t 0. It’s 1, because a ratio of 1 means equal risk in both groups. So for those measures, the equivalent question is whether the confidence interval includes 1.
The Direct Link to P-Values
There’s a clean mathematical relationship here. If a 95% confidence interval includes the null value, the p-value for that comparison will be greater than 0.05. If the interval excludes the null value, the p-value will be less than 0.05. They are two ways of expressing the same underlying calculation.
The confidence interval, though, gives you something the p-value doesn’t: a sense of direction and magnitude. A p-value of 0.30 tells you the result isn’t significant. A confidence interval of -2.5 to 8.0 tells you the same thing, but also shows that the effect, if real, could plausibly be as large as 8.0 or could go in the opposite direction. That extra information is why many statisticians, including the American Statistical Association, recommend reporting confidence intervals alongside or even instead of p-values. A p-value alone doesn’t measure the size of an effect or its practical importance.
Not Significant Doesn’t Mean No Effect
This is where most people go wrong. A confidence interval that includes zero does not prove the effect is zero. It means the study couldn’t distinguish the effect from zero with enough certainty. Those are very different statements.
The width of the interval is what separates “we found no meaningful effect” from “we simply couldn’t tell.” Consider two hypothetical study results, both comparing a treatment to a placebo for pain reduction on a 10-point scale:
- Study A: 95% CI of -0.3 to 0.4. The interval includes zero, but it’s narrow. Even the upper end of the range is small. This study had enough precision to conclude that any real effect is probably too small to matter.
- Study B: 95% CI of -3.0 to 5.0. The interval also includes zero, but it’s wide. The data are compatible with a large beneficial effect, no effect, or even a harmful one. This result is genuinely inconclusive and should not be interpreted as evidence that the treatment doesn’t work.
Study B’s wide interval typically signals a small sample size. With more participants, the interval would narrow, and the result would become more informative one way or the other. Study A’s narrow interval around zero, on the other hand, is meaningful evidence that the treatment has little to no real-world impact.
What Determines Whether the Interval Includes Zero
Three factors control the width of a confidence interval and whether it crosses zero: the size of the observed effect, the variability in the data, and the sample size.
A large observed effect pushes the interval away from zero. If your treatment group improved by an average of 15 points and the placebo group improved by 2, the center of your interval is already far from zero, making it harder for the lower bound to dip below it. High variability (lots of individual differences within each group) widens the interval, making it more likely to include zero. And a larger sample size shrinks the interval because it reduces the uncertainty in your estimate. This is why studies that are “underpowered,” meaning they enrolled too few participants, frequently produce wide intervals that span zero even when the treatment genuinely works.
How to Read the Interval, Not Just the Verdict
Rather than treating the confidence interval as a pass/fail test (does it include zero or not?), look at what values it contains. Ask yourself three questions:
- Where is the center? The point estimate (usually the middle of the interval) is still your best single guess of the true effect, even when the interval includes zero.
- How wide is the interval? A narrow interval means the study was precise. A wide one means there’s a lot of uncertainty, and the study may need replication with more data.
- Does the interval include values that would matter in practice? If you’re testing a new painkiller and the upper bound of the interval reaches a clinically meaningful level of pain reduction, you can’t rule out that the drug works, even though the result is technically non-significant. A clinically important effect remains plausible.
Conversely, a statistically significant result (interval entirely above zero) can still be clinically irrelevant if the entire interval falls below a threshold that anyone would care about. A blood pressure drug that lowers systolic pressure by somewhere between 0.5 and 1.5 mmHg has a “significant” effect that no doctor would consider meaningful.
Choosing a Different Confidence Level
The 95% confidence interval is the most common, but it’s not the only option. A 90% interval is narrower and less likely to include zero for the same data. A 99% interval is wider and more likely to include zero. The choice of confidence level is essentially a decision about how much uncertainty you’re willing to tolerate.
If a 95% interval barely includes zero (say, -0.1 to 4.5), a 90% interval for the same data might exclude it entirely. That doesn’t mean the effect magically became real. It means you’re using a less strict threshold, accepting a higher chance of being wrong. This is why researchers should choose their confidence level before looking at the data, not after.
Practical Takeaway
A confidence interval that includes zero tells you the observed effect is not statistically distinguishable from no effect at your chosen confidence level. But the story doesn’t end there. A narrow interval around zero is strong evidence of no meaningful effect. A wide interval that happens to cross zero is inconclusive, meaning the study lacked the precision to give a clear answer. Reading the full interval, not just checking whether it crosses zero, is what separates a useful interpretation from a misleading one.