Quick Reference: Noncentral t Parameter Definitions
| Symbol | Meaning |
|---|---|
| t | Observed noncentral t statistic value |
| ν (or df) | Degrees of freedom (must be > 0) |
| λ (lambda) | Noncentrality parameter — any real value, positive or negative |
| F(t; ν, λ) | Cumulative distribution function — P(T ≤ t) |
| P(T > t) | Upper-tail probability = 1 − F(t; ν, λ) |
| P(T < t) | Lower-tail probability = F(t; ν, λ) |
| λ = 0 | Reduces to the ordinary central Student's t distribution |
Both ν and λ must be specified to identify any noncentral t probability. A t value alone — without df and λ — is not sufficient.
Noncentral t Distribution Calculator
What Is the Noncentral t Distribution?
The noncentral t distribution generalizes the ordinary Student's t by allowing the numerator of the t statistic to have a nonzero mean. It arises naturally in power calculations for t-tests and in confidence interval construction for standardized effect sizes.
Formally, a random variable T follows a noncentral t distribution when:
where Z ~ N(0,1), V ~ χ²(ν), Z and V are independent, ν is the degrees of freedom, and λ is the noncentrality parameter.
When λ = 0, the formula reduces to the ordinary Student's t distribution. When λ ≠ 0, the distribution shifts and becomes asymmetric. The further λ is from zero, the more the distribution departs from the central t.
Noncentral t Distribution Table
Each table below gives a specific statistical quantity for selected degrees of freedom and noncentrality parameter values. Select the tab for the quantity you need. Values are computed using the noncentral t CDF via the regularized incomplete beta function, accurate to 4 decimal places. Methodology: equivalent to R's pt(q, df, ncp) function.
Table A: P(T ≥ 1.96) for the noncentral t distribution at selected df and λ. λ = 0 column gives the central t upper-tail probability. Computed using the regularized incomplete beta function; values rounded to 4 decimal places. Source: equivalent to R's 1 − pt(1.96, df, ncp=λ).
Why the Noncentral t Distribution Needs Two Parameters
The central t distribution is indexed by one parameter: df. Its shape is fully determined by degrees of freedom alone. The noncentral t distribution requires two parameters: df and λ.
df controls spread
Degrees of freedom govern the shape associated with sampling variability. Higher df produces heavier concentration around the center; lower df produces heavier tails.
λ controls location
The noncentrality parameter shifts the entire distribution. Positive λ moves probability mass to the right; negative λ moves it left. Larger |λ| produces greater asymmetry.
These two roles are distinct: you cannot recover information about λ from df alone, nor the reverse. A table that omits λ is incomplete for any calculation where λ ≠ 0.
Central vs Noncentral t: Side by Side
| Feature | Central t | Noncentral t |
|---|---|---|
| Degrees of freedom | Required | Required |
| Noncentrality parameter λ | λ = 0 (fixed) | Any real value |
| Symmetry around zero | Yes | Only when λ = 0 |
| Standard t table sufficient? | Yes | No |
| Typical use | Null hypothesis testing | Power analysis, effect-size CI |
| Arises under | H₀ (null hypothesis) | H₁ (alternative hypothesis) |
| Two-sided critical value symmetric? | Yes: ±t_crit | No: asymmetric when λ ≠ 0 |
| Software parameter name | No ncp argument needed | R: ncp=λ · Python: nc=λ |
The central t distribution is the noncentral t with λ = 0. They share the same mathematical family. Using a central t table for a noncentral calculation, or vice versa, produces incorrect results whenever λ ≠ 0.
The Noncentrality Parameter λ
The noncentrality parameter represents the signal relative to the noise in the test statistic. It quantifies how far the distribution is from the null case λ = 0. In most applications, λ is written as:
where δ is the true departure from H₀ and SE is the standard error of the test statistic. The exact formula depends on the design.
| Test Design | λ Formula | df |
|---|---|---|
| One-sample t-test | λ = d√n where d = (μ − μ₀)/σ |
n − 1 |
| Two-sample t-test (equal n) | λ = d√(n/2) where d = (μ₁−μ₂)/σ_pooled |
n₁ + n₂ − 2 |
| Paired t-test | λ = d_pair × √n standardized by σ_diff |
n − 1 |
| Regression slope (simple) | λ = β / SE(β) under H₁ |
n − 2 |
Never apply the one-sample formula to a two-sample design. The exact λ formula depends on the test's structure, sample allocation, and how effect size is defined.
Degrees of Freedom in the Noncentral t Distribution
Degrees of freedom ν control the shape of the noncentral t distribution in the same way they control the central t. Common values by test type:
One-sample t-test: df = n − 1. With n = 20 observations, df = 19.
Independent two-sample (equal variance): df = n₁ + n₂ − 2. With two groups of 15, df = 28.
Welch's t-test: df from the Welch-Satterthwaite equation — generally non-integer and computed by software.
Paired t-test: df = n − 1, where n is the number of pairs.
Simple regression slope: df = n − 2.
Note on non-integer df: Printed noncentral t tables tabulate integer df only. Welch's test frequently produces non-integer df such as 23.7 or 41.3. Statistical software (R's pt(), Python's scipy.stats.nct) evaluates the noncentral t at any positive real df without rounding. When using a table with Welch df, round down to the nearest integer for a conservative result.
How to Read a Noncentral t Distribution Table
A single noncentral t table is always for one specific quantity (upper-tail probability, CDF, critical value, or quantile). Confirm what the table contains before reading it.
Step 1 — Identify degrees of freedom ν
Locate the row for your df. If your exact df is between tabulated values, use the row with the next smaller df (conservative) or interpolate carefully.
Step 2 — Identify your noncentrality parameter λ
Locate the column for your λ. If your λ falls between columns, linear interpolation gives an approximation; for precision, use software.
Step 3 — Confirm what quantity the table provides
P(T ≥ t), P(T ≤ t), CDF, and critical values are four different quantities. Table A above gives P(T ≥ 1.96); Table B gives critical t at given α. Never substitute one for another.
Step 4 — Read the cell at your (ν, λ) intersection
That cell gives the tabulated value for the combination of parameters you selected.
Step 5 — Verify tail direction
Confirm whether you need an upper-tail or lower-tail probability. For a one-tailed power calculation, the direction must match your alternative hypothesis.
Never use a central t table for a noncentral calculation
A standard t table gives probabilities only for λ = 0. Using it when λ ≠ 0 produces incorrect probabilities and wrong power estimates. The error grows with |λ|.
Noncentral t Critical Values
Noncentral t critical values depend on three quantities: df, λ, and α. They differ from central t critical values whenever λ ≠ 0. Common significance levels:
| Significance Level α | Common context |
|---|---|
| 0.10 | Exploratory research; less stringent threshold |
| 0.05 | Conventional level in most social and biological sciences |
| 0.01 | More stringent; common in medical and psychological research |
| 0.001 | Very stringent; genetics and large-scale studies |
The upper-tail critical value t_crit at level α satisfies:
Two-sided noncentral t critical values require careful handling: because the distribution is asymmetric when λ ≠ 0, the two critical values are not simply ±t_crit from the central t. Power for a two-sided test must account for both tails separately.
CDF and Tail Probabilities
Four related quantities describe the noncentral t distribution at a given t value. They are distinct and must not be confused:
CDF — F(t; ν, λ)
Probability of observing T at or below t. In R: pt(t, df, ncp=λ)
Upper-tail P(T > t)
Probability of exceeding t. In R: 1-pt(t,df,ncp=λ)
PDF — f(t; ν, λ)
Density at t. Does not give a probability directly. In R: dt(t, df, ncp=λ)
Quantile function
Inverse CDF. In R: qt(p, df, ncp=λ)
Visualizing the Noncentral t Distribution
Adjust df and λ below to see how the distribution shifts. The blue curve is the noncentral t; the dashed curve is the central t (λ = 0) for comparison. The shaded area to the right of the dashed vertical line represents power at α = 0.05 (one-tailed).
Noncentral t Distribution Properties
| Property | Value / Description |
|---|---|
| Support | All real values (−∞, +∞) |
| Parameters | ν > 0 (degrees of freedom), λ ∈ ℝ (noncentrality) |
| λ = 0 case | Central Student's t distribution |
| Symmetry | Symmetric around 0 only when λ = 0; asymmetric otherwise |
| Mean | λ × √(ν/2) × Γ((ν−1)/2) / Γ(ν/2) when ν > 1; undefined when ν = 1 |
| Variance | Exists when ν > 2; involves ν and λ² in a closed form |
| Skewness | Generally nonzero when λ ≠ 0; sign follows sign of λ |
| Large-ν behavior | Approaches N(λ, 1) as ν → ∞ for fixed λ |
| Relationship to F | [T(ν, λ)]² ~ F(1, ν, λ²) — noncentral F with 1 numerator df |
| Tail behavior | Heavier than normal; exact shape depends on both ν and λ |
The mean and variance formulas require ν > 1 and ν > 2 respectively. Do not use unconditional formulas for ν = 1.
Noncentral t Distribution and Statistical Power
Statistical power is the probability of correctly rejecting H₀ when a true effect exists in the population:
Under H₁, the t statistic follows a noncentral t distribution with noncentrality parameter λ determined by the true effect size and sample size. Power equals the probability that this noncentral t statistic exceeds the critical value from the central t:
| Design change | Effect on λ | Effect on power |
|---|---|---|
| Increase sample size n | λ increases (∝ √n) | Power increases |
| Larger effect size |d| | |λ| increases | Power increases |
| Raise α (less stringent) | No effect on λ | Power increases (lower t_crit) |
| Reduce measurement error σ | |λ| increases (d = δ/σ) | Power increases |
| Switch two-tailed → one-tailed | No effect on λ | Power increases (lower t_crit) |
Noncentral t and Cohen's d
Cohen's d is a standardized effect size measuring the difference between means relative to the standard deviation. It connects directly to λ in specific test designs.
One-sample t-test
With d = 0.5 and n = 25: λ = 0.5 × √25 = 0.5 × 5 = 2.5
Two-sample t-test (equal group sizes n₁ = n₂ = n)
With d = 0.5 and n = 25 per group: λ = 0.5 × √(25/2) = 0.5 × 3.536 ≈ 1.77
Confidence intervals for Cohen's d are constructed by inverting the noncentral t distribution — finding the values of λ such that the observed t statistic equals specific quantiles of the noncentral t. The R package MBESS automates this for several designs.
Worked Example: Power Calculation for a One-Sample t-Test
Setup: A researcher plans a one-sample t-test with n = 25 participants. The expected Cohen's d = 0.50, α = 0.05 (one-tailed). Compute the expected power.
Step-by-Step Solution
| Step | Action | Result |
|---|---|---|
| 1 | Compute degrees of freedom | df = n − 1 = 25 − 1 = 24 |
| 2 | Compute noncentrality parameter | λ = d√n = 0.50 × √25 = 0.50 × 5 = 2.50 |
| 3 | Find central t critical value | t_crit = qt(0.95, df=24) = 1.711 |
| 4 | Compute power (upper-tail of NCT) | Power = 1 − pt(1.711, df=24, ncp=2.50) |
| 5 | Result | Power ≈ 0.800 (80.0%) |
Interpretation
With n = 25, d = 0.50, α = 0.05 (one-tailed), the expected power is approximately 0.80 — exactly the conventional target. This confirms that a sample of 25 is adequate for detecting a medium effect in a one-sample design at this α level. Reducing n to 20 drops power to roughly 0.70; increasing to 30 raises it to about 0.86.
R verification: 1 - pt(qt(0.95, 24), 24, ncp=2.5) = 0.7995
Noncentral t in R and Python
Statistical software evaluates the noncentral t distribution exactly without rounding and handles non-integer df. Use software when precision matters or when parameters fall between tabulated values.
R (base)
Python (scipy)
Other software: SAS uses CDF('T', t, df, lambda) for the noncentral t CDF. MATLAB uses nctcdf(t, nu, delta). Mathematica uses CDF[NoncentralStudentTDistribution[nu, delta], t]. All produce equivalent results. Verify the parameterization convention for the noncentrality parameter in your software's documentation, as notation can differ between packages.
Table Accuracy and Interpolation
The values in this table are rounded to 4 decimal places. They are computed using the regularized incomplete beta function representation of the noncentral t CDF, the same algorithm used in R's pt() function.
Rounding precision
Table values are rounded to 4 decimal places. The true value may differ by up to ±0.00005. For calculations requiring 6+ decimal place accuracy, use software directly.
Interpolating between λ values
Linear interpolation between tabulated λ values introduces error because the CDF is nonlinear in λ. Interpolation is acceptable for small gaps (λ steps of 0.5 or less). Do not interpolate across gaps of 1.0 or more.
Interpolating between df values
When df falls between rows (e.g., Welch df = 23.7), use the next smaller integer df for a conservative result, or software for the exact value.
Common Noncentral t Table Mistakes
Frequently Asked Questions
Key Takeaways
The noncentral t distribution requires both df (ν) and the noncentrality parameter λ. A central t table covers only λ = 0.
Setting λ = 0 reduces the noncentral t to the ordinary Student's t distribution. All central t results are special cases.
Every noncentral t table must clearly state what quantity it tabulates: CDF, upper-tail probability, or critical value. These are not interchangeable.
The noncentral t distribution is asymmetric when λ ≠ 0. Two-sided calculations cannot simply double a one-sided value.
Power analysis for t-tests is mathematically defined through the noncentral t distribution. Larger |λ| — from bigger n or larger effect size — gives higher power.
The λ formula for Cohen's d depends on the test design. One-sample: λ = d√n. Two-sample (equal n): λ = d√(n/2). Do not apply one formula universally.
For exact values at any df and λ — including non-integer df from Welch's test — use software: R's pt() or Python's scipy.stats.nct.