For a two-sided 95% confidence interval, the standard normal critical z score is approximately 1.96 (using the lower and upper values, −1.96 and 1.96). For a one-sided 95% confidence bound, the critical value is approximately 1.645; which value to use depends on the question and the method.
What a z score means
A z score expresses a value in standard deviation units relative to a normal distribution’s mean. A z score of 1.96 is 1.96 standard deviations above the mean; −1.96 is the same distance below it. In confidence-interval calculations, a critical z score marks a cutoff on the standard normal curve.
The familiar 95% figure describes the long-run coverage of a confidence-interval procedure: if the same sampling process were repeated many times and an interval calculated each time, about 95% of those intervals would contain the fixed population parameter, under the method’s assumptions. It does not mean there is a 95% probability that a particular already-calculated interval contains the parameter.
Why 1.96 is used for a two-sided interval
A two-sided 95% interval leaves 5% of the standard normal distribution outside its central region. That remaining probability is split equally between the two tails: 2.5% below the lower cutoff and 2.5% above the upper cutoff. The standard normal quantiles at those boundaries are approximately −1.96 and 1.96.
In symbols, the positive critical value is the 97.5th percentile of the standard normal distribution: z0.975 ≈ 1.96. The central area between −1.96 and 1.96 is approximately 0.95. The value is rounded; software or a table may show more digits, such as 1.95996.
Two-sided versus one-sided 95% confidence
The word “95%” alone does not identify the right critical value. First determine whether the interval or test is one-sided or two-sided.
- Two-sided 95% interval: use 1.96 as the positive critical value, with ±1.96 in the formula. The interval allows for a parameter being either above or below the estimate.
- One-sided 95% bound: use approximately 1.645. All of the 5% excluded probability is in one tail, leaving the 95th percentile as the cutoff.
- Two-sided 90% interval: use approximately 1.645 on each side. This illustrates why 1.645 is not the usual cutoff for a two-sided 95% interval.
Choose the tail structure based on the research question and analysis plan, not simply to obtain a preferred result. A one-sided analysis is appropriate only when the question genuinely concerns one direction and the opposite direction is not being treated as evidence of the same kind of effect.
How to calculate a z-based confidence interval
For an estimate whose sampling distribution is appropriately modeled by a normal distribution, a two-sided confidence interval has the general form:
estimate ± (critical z score × standard error)
The standard error measures the estimate’s sampling variability. The critical value sets how many standard errors to go out from the estimate. For a 95% interval, the margin of error is therefore approximately 1.96 times the standard error.
Example: Suppose a sample proportion is 0.60 and its standard error is 0.02. A simple z-based 95% interval is 0.60 ± (1.96 × 0.02), or about 0.561 to 0.639. Expressed as percentages, that is approximately 56.1% to 63.9%. This calculation is illustrative: a proportion interval also needs an appropriate method and assumptions, and other methods can perform better, especially near 0 or 1 or with small samples.
When a z score is appropriate—and when it is not
A z critical value is commonly used when the standard error is known or when a normal approximation is justified. For example, many large-sample procedures use normal approximations, but sample size alone does not guarantee that a particular approximation is sound. The data structure, distribution, estimator, and method all matter.
For a confidence interval for a population mean when the population standard deviation is unknown and estimated from the sample, a t critical value is generally used instead. The t distribution has heavier tails than the standard normal distribution, particularly for small samples. Its critical value depends on the degrees of freedom and approaches 1.96 for a two-sided 95% interval as the degrees of freedom become large.
Other settings—such as small samples, skewed data, clustered observations, or complex survey designs—may require a different interval method or a standard error that accounts for the design. Do not select a z value without checking what the statistical procedure assumes.
Finding the value in a table or calculator
For the two-sided 95% critical value, look for the standard normal quantile with cumulative probability 0.975: the area to its left is 97.5%. Some z tables instead show the area between zero and z; in that format, look for 0.4750, which corresponds to about 1.96. For a one-sided 95% cutoff, use cumulative probability 0.95, giving about 1.645.
Statistical software may ask for a confidence level, a tail probability, or a quantile. Confirm which input it expects. In a two-sided 95% calculation, the probability in each tail is 0.025, not 0.05; entering 0.95 as a single-tail probability when the software expects a quantile can produce the wrong cutoff.
Frequently Asked Questions
Is the z score for 95% confidence always 1.96?
It is 1.96 for a two-sided interval based on the standard normal distribution. A one-sided 95% bound uses about 1.645, and procedures based on a t distribution or another method may use different critical values.
Why do some sources say 1.645?
That value is the 95th percentile of the standard normal distribution. It is used for a one-sided 95% bound or a two-sided 90% interval—not for the usual two-sided 95% interval.
Does 95% confidence mean a 95% chance the true value is in my interval?
In the standard frequentist interpretation, the parameter is fixed and the interval varies across repeated samples. The method produces intervals containing the parameter about 95% of the time under its assumptions; it does not assign a 95% probability to one particular computed interval.
What is the margin of error for a 95% confidence interval?
For a two-sided z-based interval, it is approximately 1.96 multiplied by the standard error. The margin of error is not 1.96 by itself; its units and size depend on the standard error.
Should I use z or t for a confidence interval for a mean?
If the population standard deviation is unknown and estimated from the sample, the t distribution is generally appropriate. A z method is justified when its assumptions apply, including cases where the relevant standard deviation is known or a suitable normal approximation is used.
