What the Critical Value Calculator Does
This calculator finds the critical value you need for a statistical hypothesis test or confidence interval. A critical value is the cut-off point on a probability distribution that separates the region where you reject the null hypothesis from the region where you do not. Instead of looking up dusty statistical tables, you enter a few details and get an exact figure for the normal (Z), Student's t, chi-square (\(\chi^2\)), or F distribution.
The Inputs Explained
- Confidence Level (%) – e.g. 95. The tool converts this to the significance level alpha using \(\alpha = \frac{100 - \text{confidence level}}{100}\). A 95% level gives \(\alpha = 0.05\).
- Distribution Type – choose Normal (Z), Student's t, Chi-square (\(\chi^2\)), or F.
- Degrees of Freedom – required for t, chi-square, and F. For F this is the numerator degrees of freedom.
- Degrees of Freedom (denominator) – only used by the F distribution.
The Formula Behind It
The calculator uses the inverse cumulative distribution function (the quantile function) of the chosen distribution:
- Normal & t (two-tailed): critical value = \(\left|\text{inverseCDF}\!\left(\frac{\alpha}{2}\right)\right|\). The alpha is split between two tails.
- Chi-square & F (right-tailed): critical value = \(\text{inverseCDF}(1 - \alpha)\), all of alpha placed in the upper tail.
Worked Example
Suppose you run a two-tailed t-test at a 95% confidence level with 20 degrees of freedom. \(\alpha = \frac{100 - 95}{100} = 0.05\), so \(\frac{\alpha}{2} = 0.025\). The calculator evaluates the inverse CDF of the t-distribution with 20 df at 0.025 and returns the absolute value, giving a critical value of approximately 2.086. If your test statistic exceeds 2.086 in magnitude, you reject the null hypothesis.
For a chi-square test at 95% with 10 df, it computes \(\text{inverseCDF}(0.95) \approx\) 18.31.
Interpreting Your Critical Value
The critical value is the boundary of the rejection region — the point on the distribution's horizontal axis beyond which a test statistic is considered too extreme to be attributed to chance alone at your chosen significance level \(\alpha\).
The decision rules
- Two-tailed Z or t test. Reject the null hypothesis \(H_0\) when \(|\text{test statistic}| \ge\) critical value; otherwise fail to reject. The rejection region is both tails, each containing area \(\alpha/2\).
- One-tailed Z or t test. For a right-tailed alternative, reject when the statistic \(\ge +\) critical value; for a left-tailed alternative, reject when the statistic \(\le -\) critical value. The single tail contains the full area \(\alpha\).
- Chi-square and F tests. These are used as upper-tail tests: reject \(H_0\) when the statistic exceeds the upper critical value, otherwise fail to reject. Both statistics are non-negative, and larger values mean greater departure from the null.
The equivalent p-value rule
Comparing a statistic to a critical value and comparing a p-value to \(\alpha\) are two descriptions of the same decision, and they always agree:
$$\text{reject } H_0 \iff |T| \ge c_{\alpha} \iff p \le \alpha$$This holds because the critical value \(c_{\alpha}\) is defined as the statistic whose tail probability is exactly \(\alpha\). At \(T = c_{\alpha}\) the p-value equals \(\alpha\) exactly; a statistic further into the tail has a smaller p-value, and one closer to the centre has a larger one. For example, with a two-tailed z test at 95% confidence, \(z = 2.30\) exceeds 1.960 and its two-tailed p-value is about 0.021, which is below 0.05 — the same rejection either way.
What the size of the critical value tells you
- A larger critical value means a stricter test. Raising the confidence level lowers \(\alpha\) and pushes the boundary outward: 2.576 at 99% versus 1.960 at 95%. A statistic of 2.10 rejects at 95% but not at 99%, so the significance level must be fixed before looking at the data.
- t critical values shrink toward Z as df increases. With few degrees of freedom the standard error is itself poorly estimated, so the t distribution has heavier tails and demands more evidence: 2.571 at df = 5 against 2.042 at df = 30 and 1.960 in the limit.
- Chi-square and F cutoffs depend on df in the opposite direction for chi-square — they rise with df because the statistic sums more terms, so a chi-square of 20 is extreme at df = 5 (cutoff 11.070) but not at df = 20 (cutoff 31.410).
What a non-rejection does and does not mean
Failing to reject \(H_0\) is not proof that \(H_0\) is true. It means the data did not provide enough evidence against it at the chosen \(\alpha\) — an outcome equally consistent with a true null, with a real effect too small to detect, or with a sample too small to have adequate power. A test can only reject or fail to reject; it never confirms.
Link to confidence intervals
The same critical value that defines the rejection region also sets the width of the matching confidence interval:
$$\text{estimate} \pm c_{\alpha} \times \text{standard error}$$A two-tailed test at level \(\alpha\) rejects \(H_0: \mu = \mu_0\) exactly when the \(100(1-\alpha)\%\) confidence interval excludes \(\mu_0\). With a sample mean of 52.0, a standard error of 1.5 and \(z_{\alpha/2} = 1.960\), the 95% interval is \(52.0 \pm 1.960 \times 1.5 = 52.0 \pm 2.94\), i.e. 49.06 to 54.94 — so a null value of 50 would not be rejected, while a null value of 48 would be.
Frequently Asked Questions
Why is the Z/t result two-tailed? The tool divides alpha by two for the normal and t distributions, reflecting the standard two-sided test. For a one-tailed test, enter a confidence level adjusted accordingly (e.g. use 90% to mimic a one-tailed 95% bound).
Do I need degrees of freedom for the normal distribution? No. The Z critical value depends only on the confidence level. Degrees of freedom matter for t, chi-square, and F.
When do I fill in the denominator degrees of freedom? Only for the F distribution, which needs both numerator and denominator degrees of freedom (common in ANOVA and variance comparisons).