The smallest correlation that counts as statistically significant for a given sample size. Enter your n and α above and the table highlights the matching row. If your |r| is larger than the value shown, the correlation is significant.
How to read the table
- Find your sample size n, the number of pairs, in the first column. The df in brackets is n − 2.
- Choose the column for your α, with Tails set to match your test.
- Compare your correlation, ignoring its sign, with the table value. If |r| is larger, reject H₀: ρ = 0.
A worked example
You measure hours of sleep and reaction time for 20 students and get r = −0.42. Is that significant at α = 0.05, two-tailed?
- n = 20, so df = 18.
- Row 20 (18), column 0.05 gives 0.4438. Show this in the table.
- |−0.42| = 0.42 is smaller than 0.4438, so the correlation is not significant. The exact p-value is 0.065.
That surprises a lot of people. A correlation of 0.42 looks reasonable, but with only 20 pairs, a correlation that size can turn up by chance more than 5% of the time.
Sample size changes everything
Run down the 0.05 column and you can see why “significant” and “strong” are not the same thing:
| n | Smallest significant |r| (two-tailed, 0.05) |
|---|---|
| 10 | 0.632 |
| 20 | 0.444 |
| 30 | 0.361 |
| 100 | 0.197 |
With 100 pairs, r = 0.20 is significant, yet x explains only 4% of the variation in y (r² = 0.04). Always report r itself, not just whether it passed. The correlation coefficient calculator gives r, r², the p-value and a confidence interval for ρ.
Where the values come from
Each value is the r that gives a t statistic exactly equal to the critical t with n − 2 df:
rcrit = tcrit / √(tcrit² + n − 2)
So this table and the t table always give the same decision.
Common mistakes
- Using df = n − 1. A correlation loses two degrees of freedom, so df = n − 2.
- Dropping the sign of a negative r and forgetting about it. Compare |r| with the table, then report the original sign. A significant −0.6 means a negative relationship.
- Treating a significant r as proof of cause. The table tests whether a linear relationship is likely to be real, not why it exists.
Check yourself
With n = 12, r = 0.55. Significant at α = 0.05, two-tailed?
df = 10 and the critical value is 0.5760. 0.55 is smaller, so it is not significant.
Roughly how many pairs do you need before r = 0.30 is significant at 0.05, two-tailed?
Go down the 0.05 column until the value drops below 0.30. The rows jump from n = 42 (0.3044) to n = 47 (0.2876), so the table tells you somewhere between 43 and 47. Type n = 44 into the lookup and you get 0.2973, just under 0.30. So 44 pairs is the minimum.