This list explains the z score used for an 85% confidence interval, along with the formulas, assumptions, and practical checks that make the value useful. It is for students and analysts who need to calculate or interpret a two-sided interval for a population mean or proportion.
Core z Score Facts
- The central 85% z score is about 1.440. It leaves 7.5% in each tail of the standard normal distribution.
- Use z* = Φ⁻¹(0.925) for a two-sided 85% interval. The cumulative probability includes the lower 7.5% tail plus the central area.
- The critical value is positive, approximately 1.4395. Tables and software commonly round it to 1.44.
- The lower critical value is approximately −1.4395. Symmetry of the standard normal curve gives the matching lower endpoint.
- An 85% interval captures the central 0.85 of a standard normal curve. This is a probability statement about the interval procedure under its assumptions.
- The combined tail probability is 0.15. Dividing it equally gives 0.075 per tail for a two-sided interval.
- In a z table listing area from zero to z, look for 0.425. That is half of the central 85% area.
- In a cumulative-left z table, look for 0.925. This directly identifies the upper critical value.
- Software can calculate the value with an inverse-normal function. For example, use the 0.925 quantile of a standard normal distribution.
- The z score is not 0.85. Confidence level is an area, while z* is the corresponding horizontal distance from the mean.
- The value is smaller than the 90% critical value, 1.645. A lower confidence level produces a narrower interval when other inputs stay fixed.
Building the Interval
- For a mean with known population standard deviation, use x̄ ± 1.4395(σ/√n). The standard error scales the critical value into the measurement units.
- For a population proportion, use p̂ ± 1.4395√[p̂(1−p̂)/n]. This is the conventional large-sample Wald interval.
- The margin of error is z* times the standard error. It is the distance from the sample estimate to either endpoint in a symmetric interval.
- Add the margin of error to get the upper endpoint. Subtract it to get the lower endpoint.
- Keep the estimate and standard error in matching units. For example, a mean measured in kilograms produces endpoints in kilograms.
- Use the sample mean x̄ as the point estimate for a population mean. The interval centers on the observed sample average.
- Use the sample proportion p̂ as the estimate for a population proportion. Express it consistently as a decimal or percentage throughout the calculation.
- For a 85% interval, enter 0.85 as the central probability, not 0.85 on each side. Each tail contains 0.075.
- Round the critical value only after finding it. Keeping 1.4395 during calculation reduces avoidable rounding error.
- Round final endpoints to a precision supported by the data. Excess decimal places imply more measurement accuracy than the sample provides.
- Check that the calculated interval is centered on the estimate when using a symmetric z formula. Unequal distances usually indicate an arithmetic or rounding error.
Choosing the Right Method
- Use a z interval for a mean when σ is known. This is the standard normal-theory method specified by the formula.
- When σ is unknown, a t interval is usually preferred for a mean. It uses the sample standard deviation and degrees of freedom.
- For large samples, the t critical value approaches the z value. The difference may be small, but the method should still match the information available.
- Use a proportion interval only for binary outcomes. Examples include yes/no survey responses or success/failure trials.
- Consider a Wilson interval for a proportion near zero or one. It generally behaves better than the simple Wald formula in those cases.
- Do not use the basic proportion formula with very few successes or failures. Its normal approximation can give poor coverage or endpoints outside zero and one.
- For a difference of independent means with known standard deviations, combine the standard errors in quadrature. The standard error is √(σ₁²/n₁ + σ₂²/n₂).
- For paired measurements, calculate differences within pairs first. The interval then estimates the mean paired difference rather than treating observations as independent.
- For a single mean from a small, strongly skewed sample, a normal-based interval may be unreliable. Inspect the data or use a suitable alternative method.
- For complex survey samples, account for weights and design effects. A simple z interval can understate uncertainty when observations are clustered.
- For a prediction about one future observation, use a prediction interval, not a confidence interval for the mean. Individual outcomes vary more than estimated averages.
Assumptions and Interpretation
- Random or representative sampling supports population inference. Convenience samples may not represent the population, regardless of the z score.
- Independent observations are important for the usual standard error. Repeated or clustered measurements require methods that model their dependence.
- A mean interval needs a suitable sampling distribution. Normal data or a sufficiently large sample often supports the normal approximation.
- The central limit theorem concerns the sampling distribution of the mean. It does not make a skewed population itself normal.
- For a proportion, expected successes and failures should be adequate for a normal approximation. A common check is at least 10 of each, though guidance varies.
- An 85% confidence procedure has 85% long-run coverage under its assumptions. It does not mean the fixed interval has an 85% probability of containing the fixed parameter.
- A confidence interval describes uncertainty in an estimate. It does not measure the probability that individual observations fall within its bounds.
- Confidence level is not the same as statistical significance. Hypothesis-test conclusions depend on the test, null value, and chosen significance level.
- An interval containing a proposed value does not prove that value is correct. It indicates compatibility at the interval’s stated procedure and assumptions.
- Bias is not repaired by a larger sample alone. A precise interval can still be misleading if the sampling or measurement process is biased.
- Report the confidence level alongside the endpoints. Readers need it to interpret the stated uncertainty.
Worked Calculation Checks
- Suppose x̄ = 50, σ = 10, and n = 100. The standard error is 1, so the 85% interval is approximately 48.56 to 51.44.
- For that example, the margin of error is about 1.44. Multiplying 1.4395 by the standard error of 1 gives the result.
- If the same σ is paired with n = 25, the standard error is 2. The margin of error doubles to about 2.88.
- If a proportion estimate is 0.50 with n = 400, its Wald standard error is 0.025. The approximate 85% margin is 0.036.
- For that proportion example, the interval is about 0.464 to 0.536. The endpoints remain within the valid zero-to-one range.
- Check the standard error before applying z*. A mistaken square root of n can make the interval dramatically too wide or narrow.
- For a mean, divide σ by √n, not by n. Averaging reduces standard error at the square-root rate.
- For a proportion, use p̂(1−p̂) under the square root. The product reflects the variance of a binary outcome.
- When the standard error is zero, investigate the data or inputs. A zero-width interval is rarely a meaningful uncertainty summary in ordinary sampling.
- If a proportion interval extends below zero or above one, the Wald approximation is unsuitable for that dataset. Consider Wilson or another appropriate interval.
- Use the same confidence level for comparisons you intend to interpret together. Mixing levels changes the widths and the long-run coverage statement.
Common Calculator and Reporting Issues
- Set the calculator to inverse normal, not normal density. The inverse function returns the quantile for a specified cumulative area.
- For an inverse-normal calculator using mean and standard deviation inputs, enter 0.925, 0, and 1. The result is the upper standard-normal cutoff.
- Some tables report right-tail area rather than left-tail area. In that format, use 0.075 to find the same positive cutoff.
- Check whether a table gives area between zero and z or cumulative area. The target entry differs by table convention.
- State whether the interval is one-sided or two-sided. The 1.4395 value applies to the central, two-sided 85% interval.
- A one-sided 85% bound uses a different quantile, approximately 1.036. Do not substitute it for the two-sided critical value.
- Use a minus sign for the lower z cutoff if reporting both critical values. The interval formula usually uses the positive magnitude symmetrically.
- Do not confuse a z score for an observed value with z*. An observed z score standardizes data; z* is a confidence-interval cutoff.
- Include sample size and variability information when reporting a calculated interval. These inputs allow readers to understand how precision was determined.
- Label estimated quantities clearly, such as “85% CI for the population mean.” This avoids implying the interval describes raw observations.
- Describe the method when it matters, such as “normal approximation” or “known σ.” The label makes assumptions transparent.
- Do not report only the margin of error. The estimate and both interval endpoints provide the full result.
How to Use This List
- Start with the parameter you want to estimate: a mean, proportion, difference, or future observation. That choice determines the interval method.
- Confirm whether the population standard deviation is known before choosing a z interval for a mean. If it is estimated from the sample, consider t.
- Identify whether your interval is two-sided. For the standard central 85% interval, use 1.4395 or its rounded value 1.44.
- Calculate the standard error from the correct formula before finding endpoints. Most practical mistakes begin with the wrong standard error.
- Check sampling and distribution assumptions before interpreting the result. If they fail, use a method suited to the study design or data.
- Use the worked examples as arithmetic checks, not as substitutes for your own sample inputs. Recalculate with the actual estimate, variability, and size.
- For software, verify the quantile convention and confidence level. A cumulative probability of 0.925 should produce approximately 1.44.
- Report the interval with its confidence level, units, and method. This gives readers enough context to interpret the numbers responsibly.
Frequently Asked Questions
- What is the z score for an 85% confidence interval? For a two-sided interval, the positive critical value is approximately 1.4395, usually rounded to 1.44.
- Why do I use 0.925 in an inverse-normal calculator? The 15% outside the central interval splits into 7.5% in each tail, leaving 92.5% to the left of the upper cutoff.
- Is 1.44 appropriate for a one-sided 85% interval? No; a one-sided 85% bound uses approximately 1.036 because all 15% excluded area lies in one tail.
- Can I use z when the population standard deviation is unknown? For a mean, the usual choice is a t interval based on the sample standard deviation, especially with smaller samples.
- Does an 85% confidence interval mean an 85% chance the parameter is inside? In frequentist terms, 85% describes the long-run success rate of the interval method, not a probability assigned to a fixed parameter.
- Why choose 85% rather than 90% or 95%? It may be a specified reporting standard or a deliberate precision-coverage tradeoff; lower confidence gives a narrower interval but less long-run coverage.
Final Thoughts
For a two-sided 85% confidence interval, the standard normal critical value is about 1.44. Apply it to the correct standard error, verify the method’s assumptions, and report the interval with its confidence level and context.
