When we use a sample group to gain insight into an entire population, whether we’re talking about people or a product built in a factory, we risk the sample group not completely reflecting the whole population. Therefore, we need confidence intervals.
Further, we use confidence intervals to calculate a degree of certainty that the sample group accurately represents the entire population from which they were drawn. Another way of thinking of it: if you drew a same-sized sample hundreds of times and built an interval from each one, a set percentage of those intervals would contain the population mean. That percentage is the confidence level.
A confidence interval is a range of values built from one sample. It gives a plausible range for the population mean, at a stated confidence level.
This page covers every kind of confidence interval you’ll need for your exam. Jump straight to the one you need:
- Confidence interval for a mean (σ known, Z-score)
- Confidence interval for a mean (σ unknown, t-score)
- Confidence interval comparing two population means
- Confidence interval for variation (standard deviation/variance)
- Confidence interval for a proportion
New to the concept of estimating a population from a sample? Start with Point and Interval Estimation first, then come back here for the worked examples of each type.
Confidence Interval vs. Confidence Level
A confidence interval is a range of values that probably contain the population mean.
A confidence level is the percentage of intervals that would contain the population mean if you repeated the sampling many times. It equals 1 – α. A 95% level means about 95 of every 100 intervals built this way capture the true mean.
Confidence Interval vs. Prediction Interval
A prediction interval gives the range where the next single data point will probably fall. They tell you about the distribution of data values, whereas confidence intervals tell you about a probable population mean.
Estimating Population Mean and Standard Deviation
In most cases, we won’t know the mean and standard deviation for an entire population. However, we can calculate the mean and standard deviation for each of our samples. So we can use the sample figures to estimate those of the whole population. This is what we call a point estimate.
Once we find the point estimate, we also need to know how accurate it is. The Central Limit Theorem helps here. With a large enough sample (30 or more is the usual rule of thumb), the sample mean follows an approximately normal distribution. That distribution centers on the population mean. Its standard deviation, called the standard error, is σ/√n. That is what lets us put a margin of error around a sample mean.
Margin of Error
The margin of error is the maximum expected difference between the actual population parameter and a sample estimate of the parameter. In other words, it is the range of values above and below the sample statistic.
Surveys use the margin of error widely. It tells you how much uncertainty the results carry.
Margin of error ≈ 1/√n (a quick rule of thumb for a 95% interval on a proportion near 50%; the exact formula is below)
Where n= sample size
The margin of error is the critical value times the standard error. For the same sample size and spread, a higher confidence level needs a bigger critical value, so the margin of error grows. A bigger sample shrinks the standard error, so the margin of error shrinks.
For example, to estimate the confidence interval, we may use X̅±E
E is the margin of error.
Confidence interval = X̅±Zα/2*σ/√n (use this when the population standard deviation σ is known)
- Where σ = population standard deviation
- Zα/2 = normal distribution value for the desired confidence level
- n = sample size
- X̅ = sample mean
When σ is unknown and you only have the sample standard deviation s, swap Z for t with n – 1 degrees of freedom: X̅±tα/2*s/√n. Sample size does not decide between Z and t. Knowing σ does.
Z Scores
This page uses a cumulative z table: each value is the area to the left of z, so a 95% confidence level (0.025 in each tail) means finding 0.975 in the table, which gives 1.96. If your handbook’s table shows the area from the mean (0 to z), look for 0.475 instead; if it shows the upper tail, look for 0.025.
A Z score is the number of standard deviations between a data point and its mean. Thus, you can use a Z-score table to find the corresponding Z-score for common confidence levels or calculate the α value using this formula:
α = 1 - confidence level
If your desired confidence level is 95%, then your calculation would look like this:
α = (1 - 0.95) α/2 = (1 - 0.95) / 2 = 0.025 Zα/2 = Z0.025 =1.96
Common confidence levels and corresponding Z scores
| Confidence Level | Z Score |
| 90% | 1.645 |
| 95% | 1.96 |
| 99% | 2.576 |

Confidence Interval Question Using Z-Score
Question: We conduct a random survey of 500 newly-enrolled university students. We know that the standard deviation for university enrollment age is eight years. The mean age of our sample is 24. Calculate the confidence interval for all first-year university students with a 99% confidence level to 3 decimal places.
Calculation: We know the population standard deviation σ, so use a Z-score. A 99% confidence level requires a Z-score of 2.576.
Zα/2 = 2.576, X̅ = 24, σ = 8 and n = 500
Margin of error E = Zα/2*σ/√n = 2.576*8/√500 = 0.922
Confidence interval = X̅±Zα/2*σ/√n = 24±0.922
Answer: Hence, the confidence interval for the age of first-year university students is 23.078 to 24.922, with a 99% confidence level.
Confidence Interval Question Using T-Score
Question: A factory produces tennis balls. A sample of 19 balls is taken from one factory production day. The mean weight of the sample balls is 58.2g. The standard deviation for the sample balls is 0.4g. Calculate the confidence interval with a confidence level of 95%.
Calculation: The population standard deviation is unknown. All we have is the sample standard deviation s, so use a T-score, which uses a t-distribution. Sample size does not decide between Z and t. Knowing σ does. Finding a confidence interval for a mean is a two-tailed interval.
You’ll need the alpha value. To calculate it, use this simple equation:
α = (100% - confidence level%) α = (100% - 95%) α = 5%
You also need the degrees of freedom (df), which is the sample size minus one. Or in equation form:
df = n - 1
df for this question is 18.
So, use a T-table to look up the t-score for a two-tailed interval with an α of 5% (0.025 in each tail) and a df of 18: the answer is 2.101.
tα/2=2.101, X̅=58.2, s=0.4 and n=19
Margin of error E = tα/2*s/√n = 2.101*0.4/√19 = 0.193
Confidence interval = X̅±tα/2*s/√n = 58.2±0.193
Answer: Hence, a 95% confidence interval for tennis balls produced in the factory is 58.007 to 58.393 g.
Calculating the Confidence Interval Comparing Two Population Means
Using samples from each population, you can also use confidence intervals to compare two population means. Use this method to compare two manufacturing methods, or to look for a difference between two groups of people (for example, smokers and nonsmokers).
The confidence interval for comparing two means is a range of values in which the difference between those two means probably lies. If the interval includes 0, the data do not show a real difference. If the whole interval sits above or below 0, they do.
We almost never know the population standard deviations. We only have the sample standard deviations, s1 and s2. So use a t interval, not Z.
Default: unequal variances (Welch’s t interval). Use this unless the question tells you to assume equal variances.
Confidence Interval = (x̄1 - x̄2) ± tα/2, df * √((s12 / n1) + (s22 / n2))
The degrees of freedom come from the Welch-Satterthwaite equation. Here are both, in a slightly easier-to-read format:

Round df down to a whole number before you use a t-table. Software uses the exact value.
Option: equal variances (pooled t interval). Use this only when the question says to assume the two populations have equal variances.
sp = √(((n1 - 1)s12 + (n2 - 1)s22) / (n1 + n2 - 2)) Confidence Interval = (x̄1 - x̄2) ± tα/2, n1+n2-2 * sp * √(1/n1 + 1/n2)
Example
High blood pressure has been causally linked to smoking tobacco products. To test this, you want to compare systolic blood pressure between smokers and nonsmokers. You’ll use a confidence level of 95%.
- You have 45 smokers and 56 nonsmokers. The two groups are made up of comparable people (similar age, gender and health). That makes the comparison fair. It is not the same as assuming equal variances.
- In the sample group of smokers, the mean systolic rate is 138.
- In the sample group of nonsmokers, the mean systolic rate is 135.
- The sample standard deviation for smokers is 16.5.
- The sample standard deviation for nonsmokers is 14.9.
- These are sample standard deviations (s), not population values (σ). So use t, not Z.
n1 = 45, n2 = 56 x̄1 = 138, x̄2 = 135 s1 = 16.5, s2 = 14.9 95% confidence, so α = 0.05 and α/2 = 0.025
Step 1: the standard error.
s12 / n1 = 16.52 / 45 = 272.25 / 45 = 6.050 s22 / n2 = 14.92 / 56 = 222.01 / 56 = 3.964 SE = √(6.050 + 3.964) = √10.014 = 3.16
Step 2: the degrees of freedom (Welch).
df = 10.0142 / ((6.0502 / 44) + (3.9642 / 55)) = 100.29 / (0.8319 + 0.2858) = 100.29 / 1.1177 = 89.7, round down to 89 t0.025, 89 = 1.987
Step 3: the margin of error and the interval.
margin of error = t * SE = 1.987 * 3.16 = 6.29
Confidence Interval = (138 - 135) ± 6.29
= 3 ± 6.29
= (-3.29, 9.29)
If the question says to assume equal variances, use the pooled interval instead:
sp2 = ((45 - 1)(272.25) + (56 - 1)(222.01)) / (45 + 56 - 2)
= (11,979 + 12,210.55) / 99
= 244.34
sp = √244.34 = 15.63
SE = 15.63 * √(1/45 + 1/56) = 15.63 * 0.2002 = 3.13
df = 45 + 56 - 2 = 99
t0.025, 99 = 1.984
margin of error = 1.984 * 3.13 = 6.21
Confidence Interval = 3 ± 6.21 = (-3.21, 9.21)
Answer: With 95% confidence, the true difference in mean systolic pressure (smokers minus nonsmokers) is between -3.29 and 9.29 (Welch). The pooled version gives -3.21 to 9.21. Both intervals include 0. So these samples do not show a real difference in systolic pressure between smokers and nonsmokers at the 95% level.
The margin of error is the half-width of the interval: 6.29 for Welch, 6.21 for pooled. We are 95% confident the true difference lies within that distance of the sample difference of 3. With samples this large, a Z of 1.96 gives almost the same answer (±6.20). On the exam, use t whenever you only have s.
Additional Confidence Intervals Videos
Confidence intervals for Variation
The point estimate for σ is s. s2 is an unbiased estimate of σ2.
Use the chi-square distribution to construct a confidence interval for the variance and standard deviation.
If the random variable x has the normal distribution, then this statistic follows a chi-square distribution with n – 1 degrees of freedom:

For sample size n > 1
Confidence intervals for variation:

Where n= sample size
s2 = point estimate of variance (the sample variance)

Example: XYZ pharmaceutical company randomly selected 25 samples of flu medicines. The sample variance is 6 mg2. Assume the weights are normally distributed and construct a 99% confidence interval for the population variance.
n=25
Degrees of freedom = n -1 = 24
s2 = 6
χ2α/2 = 45.559 (area of 0.005 to the right, 24 df)
χ21-α/2 = 9.886 (area of 0.995 to the right, 24 df)

Confidence intervals for variation:

Hence, with 99% confidence, you can say that the population variance is between 3.161 and 14.57 mg2. Take the square root of each end for the standard deviation: 1.778 to 3.817 mg.
Confidence intervals for Proportion
The Central Limit Theorem says that with large samples, the sample proportion follows an approximately normal distribution. When np and n(1 – p) are both at least 5, you can use the normal (Z) distribution to build a confidence interval for a proportion.
The confidence interval for a proportion is:

Where:
n=sample size
p = sample proportion (your estimate of the population proportion)
Zα/2 = Z value for your confidence level, from the Z table
Example: In a bulb manufacturing unit, 24 defective bulbs were identified in a sample size of 400 bulbs. Calculate the 95% confidence interval for the proportion.
n=400
p = 24/400 = 0.06
1-p = 0.94
Check: np = 24 and n(1-p) = 376. Both are at least 5, so Z is fine.
Zα/2 = 1.96 (95% confidence)

0.06±0.0233
0.037≤p≤0.083
So, with 95% confidence, the population proportion is between 0.037 and 0.083.
What is the Difference Between Control Limits and Confidence Intervals?
These are two entirely different concepts. One is used in the Analysis of a process, and the other in the Control of a process.
Control limits come from your process data. They usually sit at the average plus or minus 3 standard deviations of the plotted statistic. They show what the process actually does. A process that stays inside its control limits with no unusual patterns is stable, or in statistical control.
These are different from Specification limits. Specification limits come from the customer or the business. They say what is acceptable. They do not belong on a control chart. A colleague once described them as if the process goes above this level, we all must update our resumes. So, if it goes below this other limit, don’t worry about updating resumes; no one will ever hire us again!
Confidence Interval
Confidence intervals are a device of statistics for when you do not have perfect knowledge of all of the data. For example, imagine you are trying to infer the chance of some event happening by sampling from a population. Let’s take US voting polls. CNN can’t possibly get all of the voting data, but they can sample the population through exit polls. From that sample, they can predict the winner. So, the question is, how sure and certain are they of how accurate their answer is? Are they 90% certain? Are they 95% certain? 99.5%? It all depends on the confidence level required. Thus, you get answers like, “Candidate X has 52% of the vote, plus or minus 3 points, at a 95% confidence level.”
The confidence level is how sure you are that your interval captures the true population value. The confidence interval is that range. Neither one is a control limit. Control limits describe how a process behaves over time.
Also, see:
How to Calculate a Sample Size Given Standard Deviation, Confidence Interval, and Margin of Error
Other Confidence Intervals Problems
For more practice, work through Point and Interval Estimation and Student’s t-Distribution.
ASQ Six Sigma Black Belt Exam Confidence Intervals Questions
Question: Which of the following describes the 95% confidence interval of a 20% absentee rate in a department with 30 people?
(A) 6% to 34%
(B) 8% to 32%
(C) 13% to 27%
(D) 17% to 23%
Answer:
When you’re ready, there are a few ways I can help:
First, join 12,000+ other Six Sigma professionals by subscribing to my email newsletter. A short read every Monday to start your work week off correctly. Always free.
* * *
If you’re looking to pass your Six Sigma Yellow Belt, Green Belt or Black Belt exam, I’d recommend starting with my affordable study guide:
1)→ 🟡Pass Your Six Sigma Yellow Belt: built for ASQ, IASSC and ILSSI, 1,700+ practice questions with explanations, and six full-length mock exams for your exam
2)→ 🟢Pass Your Six Sigma Green Belt: a study guide organized by the exam Body of Knowledge, 1,300+ practice questions with detailed answer explanations, and 20 full-length mock exams
3)→ ⚫Pass Your Six Sigma Black Belt: the full Black Belt Body of Knowledge, 2,200+ practice questions with detailed answer explanations, and 20 full-length mock exams
You’ve spent so much effort learning Lean Six Sigma. Why leave passing your certification exam up to chance? This comprehensive study guide offers 1,700+ exam-like questions for Yellow Belts, 1,300+ for Green Belts and 2,200+ for Black Belts with full answer walkthroughs, access to instructors, detailed study material, and more.
Thousands have used this system to prepare. Join them here.

Comments (14)
I don’t see a formula displayed for margin of error. In the sample workthroughs, its just assumed that a person working through it would know what value was the margin of error and how to appropriately apply it.
Alex,
I’ll see what I can add here. Remember that these materials should be supplementary to previous Six Sigma training.
Best, Ted
1.96 corresponds alpha =0.025 , not 0.05, the calculation has a typo.
Also the formula here is very different from ASQ CSSGB handbook P 287.
Z= (xbar -miu) sqrt n/ sigma
xbar=sample mean
miu=population mean
n-sample size
sigma=population standard deviation
This is getting very confusing
Hello Ming,
We have updated the article with the correct alpha value, and also the confidence interval formula.
Thanks
I posted a question, but disappear?
Hi Ming,
Non-members need to have their comments approved by an administrator. So, you may see some delay on comment approval until you join the members program.
Best, Ted.
Are the confidence intervals for variation correct? You use (a/2) for the lower, and (1-a/2) for the upper limit.
According to the Handbook Of Parametric And Nonparametric Statistical Procedures by David J. Sheskin (5th edition, page 217) it should be the other way around: (1-a/2) for the lower, and (a/2) for the upper limit.
Best regards, Daniel
Kind of weird to answer my own question, but here we go 🙂
I found out that Sheskin meant the right-tailed probability of the chi-squared distribution when he wrote “X²”. Is there any naming convention in statistics, which defines which tailed probability one assumes when writing “X²” for Chi-squared?
My confusion was actually caused by Excel, which has an old CHIINV and a new CHISQ.INV function. The old CHIINV function (up to Excel 2007) only returns the right-tailed probability. Using this CHIINV function, one must indeed use (a/2) for the lower, and (1-a/2) for the upper limit.
When I used the formulas on this webpage in Excel 365, utilising the CHISQ.INV function, I got unexpected results. To get correct results with the CHISQ.INV function, I had to use (1-a/2) for the lower, and (a/2) for the upper limit, as described in Sheskin’s book. The reason behind this switch lies in the definition of the CHISQ.INV function as the inverse of the left-tailed probability. The right-tailed probability can be calculated by adding “.RT” to the CHISQ.INV function (CHISQ.INV.RT).
I hope this explains my confusion and helps other to avoid mistakes when doing statistics with Excel 🙂
I appreciate you sharing both your original question and what you found out! Much appreciated!
Good morning Ted. Looks like the math addition is incorrect for the upper end of interval . the lower end figure is correct at 58.008 but the upper end needs to be the mean plus .192 which is 58.332. thanks Ted.
Confidence interval = X̅±tα/2*σ/√n = 58.2±0.192
Answer: Hence, 95% confidence interval for tennis balls produced in the factory is 58.008–58.192g.
The above comment was for the section called “Confidence Interval Question Using T-Score” with the 19 tennis balls.
Thank you, Barbara! I updated the typo.
Hey Ted,
I apologize if I have misunderstood, but in the example for Calculating the Confidence Interval Comparing Two Population Means, shouldn’t we use the difference in sample means as opposed to the proposed difference in population means when calculating the confidence interval?
The numbers used:
confidence interval = difference in means ± margin of error
confidence interval = -3.2025 ± 6.2025
confidence interval = -9.405–3
While:
x̄1 = 138
x̄2 = 135
Which equates to 3, so the confidence interval should be 3 ± 6.2025.
Best Regards
Hi valentin grozdanov,
Thanks for your feedback. For better clarity, we have updated the article.
Thanks