Testing the Statistical Significance of Pearson’s Correlation Coefficient

You’ve calculated Pearson’s r and it shows a moderate positive correlation between two variables. But here’s the critical question: does that number actually mean something, or could it simply be the result of random chance in your sample? This is exactly what significance testing of r is designed to answer. Without this step, even a seemingly impressive correlation coefficient can be statistically meaningless – a statistical artifact rather than a real-world relationship. Understanding how to test the significance of Pearson’s r is not just a procedural formality; it is the foundation of drawing trustworthy, scientifically sound conclusions from your data.

Why significance testing matters 🔗

The correlation coefficient r tells us about the strength and direction of the linear relationship between two variables, but the reliability of any observed correlation also depends heavily on how many data points are in the sample. A correlation of r = 0.60 calculated from 8 participants is a very different claim from the same r calculated from 800 participants. Any statistical relationship in a sample can be interpreted in two ways: either there is a genuine relationship in the population, or the relationship in the sample is simply a reflection of sampling error. Significance testing helps us decide which interpretation is more defensible.

This is why Pearson’s correlation coefficient functions as both a descriptive and an inferential statistic – it not only summarizes the relationship in your sample but, once tested, allows you to make inferences about whether that relationship likely exists in the broader population.

Setting up the hypotheses 🔗

Before running the test, you must formally state what you are testing. This is done through two competing hypotheses.

The null hypothesis (H₀) 🔗

The null hypothesis states that there is no linear relationship between the two variables in the population. In formal notation: H₀: ρ = 0, where ρ (rho) is the Greek letter used to denote the population correlation coefficient (as distinct from r, which is the sample estimate). This hypothesis proposes that any correlation you observed in your sample is purely due to chance – a product of sampling variability rather than a real association.

The hypothesis test lets us decide whether the value of the population correlation coefficient ρ is close to zero or significantly different from zero, based on the sample r and the sample size n.

The alternative hypothesis (H₁) 🔗

The alternative hypothesis asserts that a genuine, non-zero correlation exists in the population: H₁: ρ ≠ 0. This is a two-tailed alternative because you are allowing for the possibility that the true relationship could be either positive or negative. In some research contexts where the direction of the relationship is predicted in advance, a one-tailed alternative (H₁: ρ > 0 or H₁: ρ < 0) may be appropriate, but the two-tailed test is the standard default in psychological research.

Choosing the significance level (α) 🔗

Before collecting or analysing data, researchers specify a significance level (α) – the threshold probability below which the null hypothesis will be rejected. In 1925, Ronald Fisher suggested a probability of one in twenty (0.05) as a convenient cutoff level to reject the null hypothesis, and Jerzy Neyman and Egon Pearson later formally named this the significance level.

In psychological research, α = 0.05 is the most widely used threshold, meaning you accept a 5% risk of incorrectly rejecting a true null hypothesis (a Type I error). For higher-stakes research such as clinical trials or mental health interventions, a stricter alpha of 0.01 or 0.001 is often used, where the consequences of a false positive are more serious. The significance level must be set before looking at the data – not after – to preserve the integrity of the test.

Calculating degrees of freedom 🔗

Degrees of freedom (df) represent the number of independent pieces of information available to estimate the population parameter. For the significance test of Pearson’s r, degrees of freedom are calculated as:

df = n − 2

where n is the total number of data pairs in the sample. Two degrees of freedom are subtracted because both variables are being estimated simultaneously – one for each variable in the pair. So if you have 30 participants, df = 28; with 50 participants, df = 48.

As degrees of freedom decrease, the t-distribution develops thicker tails, which accounts for the greater uncertainty associated with smaller sample sizes. This is why the same r-value can be significant in a large sample but not in a small one – the test becomes more demanding as sample size shrinks.

Converting r to a t-statistic 🔗

Once the hypotheses are set and degrees of freedom calculated, the next step is to convert the observed r into a t-statistic using the following formula:

t = r√(n − 2) / √(1 − r²)

This t-test can be used when x and y are linearly related, the variables are random variables, and when the population of the variable y is normally distributed. The formula has two key components: r√(n − 2) captures how large the correlation is relative to the sample, while √(1 − r²) reflects the proportion of variance that the linear relationship does not explain. A larger t-statistic (positive or negative) provides stronger evidence against the null hypothesis.

The fact that a correlation coefficient can be directly transformed into a t-statistic has significant theoretical implications – it means the same inferential framework used across many other statistical tests applies here too, making significance testing of r conceptually coherent with the broader hypothesis-testing tradition in psychology.

Using the t-distribution to make a decision 🔗

With the t-statistic and degrees of freedom in hand, the next step is to determine whether the result is statistically significant. There are two equivalent approaches.

The critical value approach 🔗

Consult a t-distribution table using your chosen α level and df. The table provides a critical value – the threshold t-value that the test statistic must exceed for the result to be deemed significant. If |t calculated| > t critical, you reject H₀. For example, at α = 0.05 with df = 28 (a sample of 30), the critical value for a two-tailed test is approximately ±2.048. If your calculated t-statistic exceeds this boundary in either direction, the correlation is statistically significant.

The p-value approach 🔗

The p-value is the probability of obtaining a t-statistic as extreme as (or more extreme than) the one calculated, assuming the null hypothesis is true. If the p-value is less than or equal to α, reject the null hypothesis in favour of the alternative hypothesis. Statistical software such as SPSS, R, or Python automatically calculates p-values, which is why the p-value approach has become the standard in modern psychological research reporting.

A worked example 🔗

Suppose a researcher studies the relationship between hours of sleep and memory test performance in a sample of 30 university students. After calculating Pearson’s r, they find r = 0.45. Is this significant?

The steps are as follows. The hypotheses are H₀: ρ = 0 and H₁: ρ ≠ 0. The degrees of freedom are df = 30 − 2 = 28. Plugging into the t-formula: t = 0.45 × √28 / √(1 − 0.45²) = 0.45 × 5.292 / √(1 − 0.2025) = 2.381 / √0.7975 ≈ 2.381 / 0.893 ≈ 2.67. With df = 28 and α = 0.05 (two-tailed), the critical t-value is ±2.048. Since 2.67 > 2.048, the null hypothesis is rejected. The conclusion: the correlation between sleep and memory performance is statistically significant at the 0.05 level – it is unlikely to reflect sampling error alone.

In APA style, this would be reported as: r(28) = .45, p < .05. APA style requires reporting r alongside its degrees of freedom and p-value, with degrees of freedom noted in parentheses beside r.

What statistical significance does – and does not – tell you 🔗

A statistically significant r confirms that the observed correlation is unlikely to have arisen by chance. However, significance is not the same as importance. Statistical significance does not imply practical, theoretical, or clinical significance – a correlation can be highly significant in a large sample but reflect a relationship so weak that it has little real-world relevance. This is why researchers are increasingly encouraged to report effect sizes (r itself serves as an effect size) alongside p-values.

Equally important: a statistically significant correlation does not imply a causal relationship. Two variables can be significantly correlated because one causes the other, because both are influenced by a third variable, or due to coincidence over time – often called a spurious correlation. Significance testing establishes that a relationship is real; it cannot tell you why it exists.

The role of sample size 🔗

One of the most practically important things to understand about significance testing of r is the influence of sample size. The p-value is a function of sample size – when the sample size is large, the p-value is likely to be small or “significant.” This means that in very large samples, even a trivially small r (say, 0.05 or 0.08) can reach statistical significance, even though the relationship is practically negligible. Conversely, in small samples, a moderate or even strong correlation may fail to reach significance simply because the test lacks sufficient statistical power to detect it reliably.

This is why both the value of the correlation coefficient r and the sample size n must be considered together when interpreting results. A significant result in a small sample (e.g., n = 15) with a large r is often more compelling than a significant result in a massive dataset with a tiny r. Researchers planning studies should conduct power analyses in advance to ensure their sample size is adequate to detect a correlation of the expected size.

Assumptions underlying the test 🔗

For the t-based significance test of r to be valid, several assumptions must be met. Both variables should be measured on a continuous (interval or ratio) scale. The two variables should be approximately normally distributed in the population. The relationship between the variables should be linear – if the true relationship is curved, Pearson’s r and its significance test are inappropriate tools. Like any parametric technique, Pearson’s correlation test makes assumptions that need to be met for the statistical test to be reliable. Checking these assumptions – through scatterplots, normality tests, and residual inspection – should always precede significance testing, not follow it.

Reporting the results correctly 🔗

Once you have completed significance testing, results should be communicated transparently. In formal psychology writing, this means reporting: the value of r, the degrees of freedom, and the p-value. For example: There was a significant positive correlation between sleep duration and memory performance, r(28) = .45, p = .012. If the result was not significant, it would be reported as: r(28) = .20, p = .29. The APA Publication Manual recommends reporting exact p-values to two or three decimal places, and reporting p-values smaller than .001 as p < .001. This level of transparency allows readers and other researchers to evaluate and replicate your findings accurately.

What do you think? Consider a study where researchers find a statistically significant correlation between social media use and anxiety in a sample of 500 adolescents, with r = 0.12 – would you interpret this as a meaningful finding, and what additional information would you want before drawing conclusions? If the same r = 0.12 were found in a sample of just 20 participants and was not statistically significant, would that change how you view the underlying relationship between the variables?

References

1. https://courses.lumenlearning.com/introstats1/chapter/testing-the-significance-of-the-correlation-coefficient/
2. https://opentextbc.ca/researchmethods/chapter/understanding-null-hypothesis-testing/
3. https://www.scribbr.com/statistics/pearson-correlation-coefficient/
4. https://stats.libretexts.org/Bookshelves/Introductory_Statistics/Introductory_Statistics_1e_(OpenStax)/12:_Linear_Regression_and_Correlation/12.05:_Testing_the_Significance_of_the_Correlation_Coefficient
5. https://en.wikipedia.org/wiki/Statistical_significance
6. https://www.simplypsychology.org/p-value.html
7. https://psychology.nottingham.ac.uk/staff/lpzjd/psgy1001-21/pearson-correlation-test.html
8. https://statisticsbyjim.com/hypothesis-testing/degrees-freedom-statistics/
9. https://stats.libretexts.org/Bookshelves/Introductory_Statistics/Mostly_Harmless_Statistics_(Webb)/12:_Correlation_and_Regression/12.01:_Correlation/12.1.02:_Hypothesis_Test_for_a_Correlation
10. http://www2.psychology.uiowa.edu/faculty/mordkoff/GradStats/part%201/I.12%20corrltns.pdf
11. https://online.stat.psu.edu/statprogram/reviews/statistical-concepts/hypothesis-testing/p-value-approach
12. https://resources.nu.edu/statsresources/Pearsonsr
13. https://pmc.ncbi.nlm.nih.gov/articles/PMC6859766/
14. https://tuos-bio-data-skills.github.io/intro-stats-book/correlation-tests.html
15. https://pmc.ncbi.nlm.nih.gov/articles/PMC2816758/