To find the confidence level from a confidence interval, you work backwards: extract the margin of error from the interval’s bounds, divide it by the standard error to get the critical value, then look up that critical value in a z-table or t-table to find the corresponding confidence level percentage. The process requires knowing (or being able to calculate) the standard error, which means you need the sample size and either the standard deviation or proportion.
The Core Formula, Reversed
A confidence interval follows this structure:
CI = Point estimate ± (Critical value × Standard error)
The critical value is the piece that determines the confidence level. A critical value of 1.96 corresponds to 95% confidence, 2.575 corresponds to 99%, and so on. When you already have the interval but not the confidence level, your job is to isolate that critical value. The rearranged formula is:
Critical value = Margin of error ÷ Standard error
Once you have the critical value, you convert it to a percentage using a statistical table. That percentage is your confidence level.
Step-by-Step Process
Step 1: Find the Point Estimate
The point estimate sits at the center of the interval. Add the upper and lower bounds together and divide by two. For example, if your confidence interval is (4.2, 5.8), the point estimate is (4.2 + 5.8) ÷ 2 = 5.0.
Step 2: Find the Margin of Error
The margin of error is the distance from the point estimate to either bound. Subtract the lower bound from the upper bound and divide by two: (5.8 − 4.2) ÷ 2 = 0.8. Alternatively, subtract the point estimate from the upper bound: 5.8 − 5.0 = 0.8.
Step 3: Calculate the Standard Error
This is where you need additional information beyond just the interval. The standard error for a mean equals the sample standard deviation divided by the square root of the sample size. If the standard deviation is 2.5 and the sample size is 40, the standard error is 2.5 ÷ √40 = 0.395. For a proportion, the standard error uses a slightly different formula based on the sample proportion and sample size, but the logic of the remaining steps stays the same.
Step 4: Solve for the Critical Value
Divide the margin of error by the standard error. Using the numbers above: 0.8 ÷ 0.395 = 2.025. This is your critical value.
Step 5: Look Up the Confidence Level
Take that critical value to a z-table or t-table (more on which one below). For a z-table, find the area to the left of your critical value. If the area to the left of 2.025 is about 0.9786, that means 97.86% of the distribution falls below that point, leaving 2.14% in the upper tail. Since confidence intervals are two-sided, you double the tail area: 2 × 2.14% = 4.28% total outside the interval, which means the confidence level is approximately 95.7%.
The shortcut version: look up the area to the left of your critical value, subtract 0.5, double it, and convert to a percentage. So (0.9786 − 0.5) × 2 = 0.9572, or about 95.7%.
Common Critical Values as a Quick Reference
If your calculated critical value matches one of these standard values, you can skip the table lookup entirely:
- 1.645 → 90% confidence level
- 1.96 → 95% confidence level
- 2.33 → 98% confidence level
- 2.575 → 99% confidence level
In practice, most confidence intervals in published research use 95% confidence, so if your critical value lands near 1.96, that’s almost certainly what was used. The 90% and 99% levels are the next most common. If your value falls between entries on this list, you’ll need the full table lookup described in Step 5.
Z-Table vs. T-Table
Which table you use depends on what you know about the data. If the population standard deviation is known and the sample is large, the z-table (standard normal distribution) applies. This is the simpler case, and the critical values listed above all come from the z-distribution.
If the population standard deviation is unknown and was estimated from the sample (which is the more common real-world scenario), the t-distribution is technically more accurate, especially with smaller samples. The t-distribution requires one extra piece of information: degrees of freedom, calculated as the sample size minus one. For a sample of 25, you’d use 24 degrees of freedom when looking up your critical value in the t-table.
The practical difference between the two shrinks as sample size grows. With 30 or more observations, the t-distribution and the normal distribution give nearly identical results. With very large samples, it barely matters which table you use. With small samples (under 30), the t-distribution produces wider intervals for the same confidence level, so using the z-table would give you a slightly inflated confidence level estimate.
A Worked Example
Suppose a study reports a mean of 72 with a 95% confidence interval but you want to verify that claim. The interval is (68.1, 75.9), the sample standard deviation is 12, and the sample size is 36.
Point estimate: (68.1 + 75.9) ÷ 2 = 72. That checks out.
Margin of error: 75.9 − 72 = 3.9.
Standard error: 12 ÷ √36 = 12 ÷ 6 = 2.0.
Critical value: 3.9 ÷ 2.0 = 1.95.
Looking up 1.95 in a z-table gives a cumulative area of about 0.9744. The tail area is 1 − 0.9744 = 0.0256. Double it for both tails: 0.0512. Subtract from 1: the confidence level is 1 − 0.0512 = 0.9488, or about 94.9%. That’s consistent with the 95% claim, with the tiny discrepancy coming from rounding in the reported bounds.
When You Don’t Have the Standard Error
If you only have the confidence interval bounds and nothing else (no sample size, no standard deviation), you cannot determine the confidence level. The interval alone doesn’t contain enough information because many combinations of confidence level and standard error can produce the same width. You need at least the sample size and either the standard deviation or proportion to calculate the standard error. Without those, the problem is unsolvable.
In a classroom setting, these values are typically given in the problem. In a research paper, they’re usually reported alongside the interval or can be derived from other reported statistics like the sample size and standard deviation.