When the degrees of freedom (df) or probability you need is not printed in a statistics table, estimate the value by linear interpolation between the two nearest rows or columns:
value = v₁ + (x − x₁) / (x₂ − x₁) × (v₂ − v₁)
For t-tables there is a better option: interpolate in 1/df (often written 120/df) instead of df. It is almost exact. For example, the two-tailed 5% critical t for 45 df is 2.0158 by plain interpolation but 2.0140 by the 1/df method, against an exact value of 2.0141.
You can do the arithmetic with our interpolation calculator. The examples below show exactly what to enter.
Why tables skip rows
Printed t-tables usually list every df from 1 to 30, then jump: 40, 60, 120 and infinity. The same goes for chi-square tables, which often step by 10 past 30 df. z-tables list z to two decimals. If your study has 45 df, or your z-score is 1.234, the exact entry is not there.
You have three options:
- Use the more conservative row. For a t critical value, the smaller df gives a larger t, so using 40 df for 45 df is safe but slightly strict. Many textbooks recommend this.
- Interpolate for a closer value.
- Use software for the exact value. More on this at the end.
t-table: degrees of freedom between rows
Here are two-tailed 5% (upper-tail 0.025) critical values from a typical table:
| df | 40 | 60 | 120 |
|---|---|---|---|
| t (0.025) | 2.021 | 2.000 | 1.980 |
Method 1: linear in df
For 45 df, between 40 and 60:
- Fraction = (45 − 40) / (60 − 40) = 5 / 20 = 0.25
- t ≈ 2.021 + 0.25 × (2.000 − 2.021) = 2.021 − 0.00525 = 2.01575, or 2.0158
Method 2: linear in 120/df (harmonic interpolation)
Critical values of t change roughly in proportion to 1/df, not df. So convert each df to 120/df and interpolate on that scale:
| df | 40 | 45 | 60 |
|---|---|---|---|
| 120/df | 3 | 2.6667 | 2 |
| t (0.025) | 2.021 | ? | 2.000 |
- Fraction = (3 − 2.6667) / (3 − 2) = 0.3333
- t ≈ 2.021 − 0.3333 × 0.021 = 2.0140
How close are they?
| df | Linear in df | Linear in 120/df | Exact |
|---|---|---|---|
| 45 | 2.0158 | 2.0140 | 2.0141 |
| 80 | 1.9933 | 1.9900 | 1.9901 |
The 120/df method is within 0.0001 of the exact value in both cases. Plain interpolation is a little high, more so over the wider 60 to 120 gap. Both are far better than guessing, and either is fine for a hypothesis test unless your test statistic sits right on the boundary.
The number 120 is used only because it makes the table’s df values into tidy numbers (120/120 = 1, 120/60 = 2, 120/40 = 3, and 120/infinity = 0). Any constant gives the same answer.
z-table: finding z for a probability
A standard normal table lists cumulative probabilities for z in steps of 0.01. Two common jobs need interpolation.
Finding z for a given probability (inverse look-up)
What z has 95% of the distribution below it? The table shows:
| z | 1.64 | 1.65 |
|---|---|---|
| Cumulative probability | 0.9495 | 0.9505 |
0.9500 is exactly halfway, so z ≈ 1.64 + 0.5 × 0.01 = 1.645. The exact value is 1.6449, which is why 1.645 is the critical value you see quoted for a one-tailed 5% test. This is an example of inverse interpolation: you know y (the probability) and want x (the z-score).
Finding a probability for a z with three decimals
What is P(Z < 1.234)?
| z | 1.23 | 1.24 |
|---|---|---|
| Cumulative probability | 0.8907 | 0.8925 |
- Fraction = (1.234 − 1.23) / 0.01 = 0.4
- P ≈ 0.8907 + 0.4 × 0.0018 = 0.89142
The exact value is 0.8914. The normal curve is close to straight over a 0.01 step, so linear interpolation works very well on z-tables.
Chi-square table: df between rows
Upper 5% critical values of chi-square:
| df | 30 | 40 |
|---|---|---|
| χ² (0.05) | 43.773 | 55.758 |
For 35 df:
- χ² ≈ 43.773 + 0.5 × (55.758 − 43.773) = 49.766
- Exact value: 49.802
The estimate is 0.036 low. Chi-square critical values grow a little faster than linearly with df, so straight-line interpolation slightly underestimates. In a test, that means you might reject when you should not if your statistic falls between 49.766 and 49.802. If your value is that close, use software.
Estimating a p-value between two columns
Tables also let you bracket a p-value. Suppose t = 2.30 with 20 df. The 20 df row shows t = 2.086 at one-tailed 0.025 and t = 2.528 at 0.01. Your t is between them, so 0.01 < p < 0.025 (one-tailed).
Interpolating linearly gives p ≈ 0.0177, while the exact one-tailed p-value is 0.0162. Tail probabilities shrink faster than linearly, so this estimate is rougher than the critical-value examples. For reporting, state the bracket (“p between 0.01 and 0.025”) or compute the exact p-value.
Check it with the calculator
The calculator rounds to 2 decimals, which is too coarse for table values like 2.0158. Scale the table values by 1,000 first:
t for 45 df, linear in df:
| Field | Value |
|---|---|
| X1 | 40 |
| Y1 | 2021 |
| X2 | 60 |
| Y2 | 2000 |
| X3 | 45 |
| Y3 | leave empty |
Y3 shows 2015.75, so t ≈ 2.0158 (2.01575 before rounding).
t for 45 df, linear in 120/df: enter X1 = 3, Y1 = 2021, X2 = 2, Y2 = 2000, X3 = 2.6667. Y3 shows 2014.00, so t ≈ 2.0140.
When to skip the table
Interpolation is useful in exams and when only a printed table is at hand. Otherwise, software gives exact values in one line:
- Excel or Google Sheets:
=T.INV.2T(0.05,45)for the two-tailed t,=NORM.S.INV(0.95)for z,=CHISQ.INV.RT(0.05,35)for chi-square. - Python (SciPy):
scipy.stats.t.ppf(0.975, 45),scipy.stats.norm.ppf(0.95),scipy.stats.chi2.ppf(0.95, 35).
The NIST/SEMATECH e-Handbook of Statistical Methods has clear reference tables for the t distribution, the standard normal distribution and the chi-square distribution.
For the general method, see linear interpolation. For tables in other fields, see how to interpolate steam tables and lookup table interpolation.
FAQ
What do I do if my degrees of freedom are not in the t-table? Use the next lower df for a conservative answer, or interpolate. Interpolating on 120/df instead of df gives values within about 0.0001 of exact for common tables.
Is it OK to interpolate a z-table? Yes. Over a 0.01 step in z the normal curve is nearly straight, so linear interpolation is accurate to about four decimals.
Why is interpolating chi-square less accurate? The gaps between printed df are usually wider (often 10), and critical values curve upward with df. Linear interpolation lands slightly low.
Should I interpolate p-values? Only as a rough guide. Tail probabilities fall off quickly, so linear interpolation between table columns can be noticeably off. Report the range from the table or compute the exact p-value.