Statistics Glossary → Noncentral t confidence interval
Noncentral t confidence intervalconcept
An exact confidence interval for standardized effect sizes such as Cohen's d, built by searching for the noncentral t distributions that could have produced the observed t value.
When the null hypothesis is true the t statistic follows the familiar (central) t distribution. When there is a real effect, t follows a noncentral t distribution whose noncentrality parameter (ncp) depends on the effect size and the sample size (e.g. ncp = δ·√n for one sample). For the confidence interval the question is: for which ncp values does the observed t fall exactly at the upper 2.5% and the lower 2.5% point of the distribution? These two ncp values are found numerically and rescaled by the sample size to give the interval for d (the Steiger and Fouladi approach).
This is more accurate than approximate intervals of the form d ± 1.96·SE; especially with small samples and large effects the true interval is asymmetric and the approximate formula is misleading. The method assumes normality and, for independent groups, equal variances. The same logic is used for exact confidence intervals of Bland-Altman limits of agreement.
An everyday example
Twelve people using a sleep app sleep on average 0.6 hours longer, giving Cohen's d = 0.80. The approximate formula gives a symmetric interval from 0.15 to 1.45, while the noncentral t method gives roughly 0.13 to 1.44, an interval that is not symmetric around d. The smaller the sample and the larger the effect, the more the two methods diverge.
In MerQur
Available as the 'd CI, noncentral t' checkbox in the one-sample and independent t-tests and as 'd_z CI, noncentral t' in the paired t-test; in the independent case the interval is computed from Student's t even when Welch is selected. In Bland-Altman agreement analysis the 'Exact LoA CI' output uses the noncentral t distribution for the limits of agreement; the effect size calculator also gives the d and d_z intervals with this method.
Analyses where this concept appears
One-Sample t-TestIndependent-Samples t-TestPaired t-TestBland-Altman Agreement AnalysisEffect Size Calculator
Related concepts
Cohen's d (effect size)Confidence level and confidence interval