What Is a Correlation Coefficient? Pearson’s r and Psychology Statistics

A correlation coefficient of just 0.3 counts as a strong finding in personality psychology, yet most people misread what Pearson’s r actually proves about cause and effect.

Key Takeaways:

• What does Pearson’s r tell us? It’s a single number between −1 and +1 showing the strength and direction of a relationship between two variables — the closer to −1 or +1, the stronger the link.
• Is a correlation of 0.7 strong? Yes — using Cohen’s (1988) widely used benchmarks, anything above roughly 0.5 counts as a large effect, so 0.7 is a strong, notable relationship.
• Can correlation prove causation? No. Even a very strong correlation coefficient only shows an association — a third variable, reverse causation, or coincidence can always explain the pattern instead.

What Is a Correlation Coefficient?

A correlation coefficient is a single number that tells you how strongly two variables move together. In psychology, you’ll most often meet it as Pearson’s r, named after statistician Karl Pearson, who formalised the product-moment correlation formula in the 1890s while extending earlier work by his mentor Francis Galton on inherited traits (Pearson, 1895). Galton had been trying to quantify how closely traits like height were shared between parents and children, and Pearson’s contribution was to turn that intuition into the standardised, comparable statistic still used today. It is one of the most frequently reported statistics in psychology research papers, textbooks, and news coverage of psychological studies.

The value of r always falls somewhere between −1 and +1:

• +1 means a perfect positive relationship — as one variable increases, the other increases in exact step
• −1 means a perfect negative relationship — as one variable increases, the other decreases in exact step
• 0 means no linear relationship at all

In real psychological data, you almost never see a perfect −1 or +1. Instead, r usually falls somewhere in between, and part of learning to work with correlations is learning to read what the size of that number actually means. One influential set of benchmarks, still widely taught today, comes from Cohen (1988), who proposed that a correlation around 0.1 represents a small effect, around 0.3 a medium effect, and 0.5 or above a large effect. Cohen (1992) later revisited these thresholds in a short, widely cited paper aimed at helping researchers plan adequately powered studies, reinforcing that the same conventions apply whether you’re testing a correlation, a group difference, or a more complex model.

These bands are a widely used rule of thumb rather than a fixed law of statistics — even Cohen himself described his own thresholds as a last resort for use only when no field-specific benchmark exists (Cohen, 1988). A correlation of 0.3 might be considered meaningfully strong in a messy field like personality psychology, but weak in a tightly controlled physics experiment. Context always matters more than the label — a point explored further in the comparison table later in this article. If you want a refresher on how individual scores are standardised before they’re compared like this, our guide to z-scores covers the underlying maths in more depth, and the same logic underpins how IQ scores are distributed around a population mean.

Positive, Negative, and Zero Correlation Explained

Positive correlation means both variables move in the same direction. As one goes up, so does the other. A commonly cited example in psychology is the relationship between hours of sleep and next-day concentration: more sleep tends to go with better concentration scores, producing a positive r. Twin and family studies provide some of the clearest large-scale examples of positive correlation in psychology — for instance, the intelligence scores of identical twins raised in entirely separate households still correlate strongly with one another, a finding covered in more depth later in this article (Bouchard, Lykken, McGue, Segal, & Tellegen, 1990).

Negative correlation means the variables move in opposite directions. As one goes up, the other goes down. A classic example is the relationship between stress levels and short-term memory performance under certain conditions — as reported stress increases, memory test scores tend to decrease, producing a negative r.

Zero (or near-zero) correlation means there’s no consistent linear pattern between the two variables at all. For instance, there’s no meaningful relationship between someone’s shoe size and their score on a general knowledge quiz — knowing one tells you nothing useful about the other.

It’s worth remembering that the sign (positive or negative) tells you the direction of the relationship, while the size of the number (how close it is to 1 or −1, regardless of sign) tells you the strength. A correlation of −0.75 is stronger than one of +0.40, even though the second number is positive. This distinction matters just as much when comparing personality traits — for instance, when comparing Big Five trait scores against other frameworks, the direction and size of any correlation between traits are reported separately for exactly this reason.

Correlation vs Causation: Why They’re Not the Same

This is arguably the single most important thing to understand about correlation coefficients — and one of the most heavily tested ideas in psychology courses. A correlation coefficient, however large, only tells you that two variables are statistically associated. It does not tell you that one causes the other.

The classic illustration used across statistics teaching is the correlation between ice cream sales and drowning incidents, which rise and fall together over the course of a year. Buying ice cream doesn’t cause drowning, and drowning doesn’t cause ice cream sales. Both are driven by a third variable: warm weather, which leads to more swimming and more ice cream purchases simultaneously. This is known as the third variable problem, a concept closely tied to what philosopher of science Hans Reichenbach (1956) formalised as the “common cause principle” — the idea that when two events are correlated without one causing the other, a shared prior cause is usually responsible. It’s one of the main reasons a strong correlation can be seriously misleading if it’s mistaken for evidence of cause and effect.

A second issue is the directionality problem. Even when a real relationship exists between two variables, a correlation coefficient can’t tell you which one is the cause and which is the effect. Take the well-documented association between screen time and depressive symptoms in adolescents. Researchers Twenge, Joiner, Rogers, and Martin (2018) found that US teenagers who spent more time on screen-based activities were more likely to report high depressive symptoms and suicide-related outcomes. It’s tempting to conclude that screen time causes low mood. But the correlational design used in that kind of research can’t rule out the reverse explanation — that teenagers who are already low in mood withdraw into more screen use — or a third factor, such as disrupted sleep, driving both.

Psychologists have documented that people routinely fall into this trap. A widely discussed study (Bleske-Rechek, Morrison, & Heidtke, 2015) gave participants research scenarios on topics like video gaming and aggression, then asked them to draw conclusions. Even when the underlying design was clearly correlational rather than experimental, many participants still drew causal conclusions from it — showing how natural, and how easy, this misinterpretation is.

It’s also worth noting that correlational evidence can, over time and with enough converging studies, build a strong enough case that a causal relationship becomes broadly accepted — even without a true controlled experiment. The historical relationship between smoking and lung cancer is the best-known example. Doll and Hill (1950) published a large case-control study showing a strong statistical association between cigarette smoking and lung cancer diagnoses, at a time when the medical community was deeply divided over whether the link was genuine or explained by some other factor, such as urban air pollution. It took years of additional correlational research — including a landmark prospective cohort study following tens of thousands of British doctors — before the causal link was accepted as established, since randomly assigning people to smoke for decades was never an ethical option. This example shows that correlational research isn’t a weaker substitute for experiments; it’s often the only route available, and repeated, well-controlled correlational findings can accumulate into genuinely strong causal evidence.

Researchers also need to watch for a more subtle statistical trap known as Simpson’s paradox, where a correlation that appears in the data overall can reverse or disappear entirely once the data is broken down by subgroup (for example, by age group or gender). A correlation coefficient calculated on a whole dataset can therefore tell a misleading story if an important subgroup difference is hiding underneath it.

None of this means correlational research is a lesser form of evidence. Correlational studies are often the only ethical or practical way to investigate a question — you can’t randomly assign teenagers to five years of high screen use to test the theory experimentally. Correlational research is genuinely valuable for identifying relationships worth investigating further, generating testable hypotheses, and documenting real-world patterns. The key skill is holding two ideas at once: a correlation coefficient can reveal a real and important pattern, while still being unable, on its own, to prove what’s causing that pattern. For a deeper dive into the reasoning behind this distinction, Scribbr’s guide to correlation and causation (Bhandari, 2022) is a useful companion resource that goes further into research design solutions like longitudinal and experimental studies.

How Strong Is “Strong”? Comparing Correlation Coefficients Across Fields

Students are often asked directly: “what counts as a strong correlation?” There’s no single universal cut-off, but a commonly taught convention, following Cohen (1988), is that anything above roughly 0.5 is considered a large effect, values around 0.3 are moderate, and anything closer to 0.1 is small. What counts as impressive, however, depends enormously on the field of research being discussed.

Personality and social psychology sit at the lower end of this table because human behaviour is shaped by dozens of interacting factors at once, so even a genuine, replicable effect often produces a correlation coefficient no higher than 0.2 or 0.3. Studies examining the MBTI’s relationship compatibility claims, for example, have reported weak correlations (generally below r = .30) between MBTI type-matching and relationship satisfaction outcomes — a finding covered in more depth in our analysis of MBTI test accuracy, which illustrates how a low correlation coefficient is itself meaningful evidence when evaluating whether a popular claim holds up statistically.

At the other end of the scale, published test-retest reliability studies for well-validated personality instruments — such as the Big Five model — commonly report correlation coefficients exceeding 0.80 across repeated testing (McCrae & Costa, 2008), a benchmark that’s much harder to achieve in most areas of psychological research. Our comparison of MBTI and Big Five reliability walks through what a “strong” correlation coefficient looks like in the specific context of personality test consistency, and our wider comparison of MBTI, Big Five, and Enneagram shows how these reliability figures translate into practical differences between the three frameworks.

Behavioural genetics research occupies a middle-to-high position on this scale. The Minnesota Study of Twins Reared Apart found that identical (monozygotic) twins separated at birth and raised in entirely different households still showed IQ correlations of roughly 0.70–0.75 (Bouchard, Lykken, McGue, Segal, & Tellegen, 1990) — remarkably high given that the twins often grew up in very different environments, and a figure researchers point to as strong evidence for a substantial genetic contribution to intelligence differences.

Beyond r itself, r² (the correlation coefficient squared) is often more informative on its own, because it tells you the proportion of variance in one variable that’s statistically associated with the other. A correlation of 0.5 gives an r² of 0.25 — meaning only 25% of the variation is shared between the two variables, with the remaining 75% explained by other factors entirely. As McLeod (2019) notes in a widely used summary of effect size conventions, statistical significance alone says nothing about how large or meaningful an effect actually is — which is exactly why r² is a useful check against overstating what a correlation coefficient, by itself, actually demonstrates.

Pearson’s r vs Spearman’s Rho

Pearson’s r is a parametric statistic, which means it comes with certain assumptions about your data: both variables should be roughly continuous (interval or ratio level), the relationship between them should be linear, the data should be reasonably close to normally distributed, and the spread of one variable’s scores should stay fairly consistent across the range of the other (a property called homoscedasticity). Extreme outliers can distort Pearson’s r considerably even when only one or two data points are affected.

When those assumptions aren’t met — for example, when data is ordinal (ranked categories, like exam grades or a Likert-scale attitude survey), or when the relationship is non-linear, or when there are significant outliers — psychologists typically switch to Spearman’s rho (ρ), a non-parametric alternative first published by Charles Spearman (1904) in his foundational paper on general intelligence. Instead of working directly with the raw scores, Spearman’s rho ranks each set of scores from lowest to highest and calculates the correlation between the ranks. This makes it more robust to outliers and doesn’t require the underlying assumptions that Pearson’s r depends on.

A close relative, Kendall’s tau, offers a third option, also based on ranks but calculated by comparing every possible pair of observations to see whether they’re ranked in the same order on both variables. Kendall’s tau is generally preferred over Spearman’s rho for very small samples or when a dataset contains a lot of tied ranks, though Spearman’s rho remains far more commonly reported in psychology research.

As a practical rule: if you’re working with continuous measurements (reaction time in milliseconds, a validated scale score) and the relationship looks roughly linear on a scatter graph, Pearson’s r is appropriate. If you’re working with ranked or ordinal data, or your scatter graph shows a clear curve rather than a straight-line trend, Spearman’s rho (or Kendall’s tau for smaller samples) is the safer choice. This same logic applies when comparing scores across different MBTI cognitive functions, where preference strength is often ranked rather than measured on a true continuous scale.

A Worked Example: How a Correlation Coefficient Is Calculated

It helps to see, at a conceptual level, where a correlation coefficient actually comes from. Imagine five students’ hours of revision and their exam scores:

Pearson’s r works by measuring how consistently each student’s position relative to the average revision time lines up with their position relative to the average exam score. Student A revised less than average and scored below average; Student E revised more than average and scored above average. Because every student’s relative position on one variable matches their relative position on the other, the pattern produces a correlation coefficient very close to +1 — in this invented example, r would come out at roughly 0.99, reflecting an almost perfectly straight-line relationship between the two columns.

In real psychological data, this kind of clean, near-perfect alignment is rare. Most genuine relationships between two psychological variables — revision time and exam performance included — are muddied by other factors (prior knowledge, sleep, anxiety, exam difficulty), which is exactly why real-world correlation coefficients usually land well below 1, even when there’s a genuine underlying relationship.

Reading a Scatter Graph

A scatter graph (or scatterplot) is the standard way to visualise a correlation before calculating r. Each point on the graph represents one participant or observation, plotted according to their score on both variables — one on the x-axis, one on the y-axis.

To read one:

• Look at the overall direction. If the points trend upward from left to right, that’s a positive correlation. If they trend downward, that’s negative.
• Look at how tightly the points cluster around an imaginary line. The closer the points sit to a straight line, the stronger the correlation — a tight, narrow cloud of points suggests a value of r close to 1 or −1. A wide, scattered spread suggests a value closer to 0.
• Watch for curves. If the points form a clear curve rather than a straight-line trend, Pearson’s r can be misleading, since it only measures linear relationships.
• Watch for outliers. A single extreme data point sitting far away from the rest of the cloud can distort a correlation coefficient considerably, especially with small sample sizes.
• Watch for a “fanning” pattern. If the spread of points widens or narrows as you move across the x-axis (heteroscedasticity), this can also violate the assumptions behind Pearson’s r, even when the general trend still looks roughly linear.

The importance of this step is famously demonstrated by Anscombe’s quartet (Anscombe, 1973) — four datasets that each produce an almost identical Pearson’s r, mean, and variance, yet look completely different when plotted: one shows a genuine linear relationship, one a clear curve, one a near-perfect line thrown off by a single outlier, and one dominated entirely by one extreme point. Anscombe’s original demonstration remains one of the clearest illustrations in statistics of why a correlation coefficient should never be trusted without first looking at the graph behind it.

Statistical Significance of a Correlation (a Light Touch)

A correlation coefficient tells you about the strength and direction of a relationship in your sample. It doesn’t, by itself, tell you whether that relationship is likely to reflect a genuine pattern in the wider population, or whether it could easily have arisen by chance. That’s a separate question, answered using a significance test that produces a p-value, often reported alongside a confidence interval showing the plausible range for the true population correlation.

As a general principle, larger sample sizes make it easier to detect a statistically significant correlation even when the underlying relationship is fairly weak, while small samples can fail to reach significance even for a moderately strong r. Cohen (1990) made this point forcefully in a widely cited retrospective on his own career, arguing that statistical significance is often the least interesting thing about a result, and that researchers should pay closer attention to the actual size of an effect rather than whether it crosses an arbitrary threshold. This is why researchers report both the size of r and its statistical significance, rather than either figure alone. For the full detail on how significance testing works, including common misinterpretations of p-values, see our guide to statistical significance, p-values, and effect sizes.

Real Psychology Research Using Correlational Design

Correlational designs appear constantly in published psychology research, particularly where an experimental design would be impractical, unethical, or simply impossible.

One well-known example is the research into adolescent screen time and psychological wellbeing carried out by Twenge, Joiner, Rogers, and Martin (2018), published in Clinical Psychological Science. Using large-scale survey data, the researchers examined the relationship between time spent on screen-based activities and measures including depressive symptoms and suicide-related outcomes among US teenagers, finding that higher screen time was associated with worse outcomes on both measures, while time spent on non-screen activities showed the opposite pattern. This kind of large sample, correlational survey design is common in developmental and health psychology precisely because researchers cannot ethically or practically assign young people to different levels of screen exposure over years and observe the outcome.

A very different example comes from behavioural genetics. The Minnesota Study of Twins Reared Apart (Bouchard, Lykken, McGue, Segal, & Tellegen, 1990) tracked more than 100 pairs of identical and non-identical twins who had been separated in infancy and raised in different households, later bringing them together as adults for extensive testing. The correlation between the IQ scores of separated identical twins came out at roughly 0.70–0.75 — far higher than the correlation between separated non-identical twins — and researchers have used this gap as one of the central pieces of evidence for a substantial genetic contribution to intelligence differences. This design is a particularly clever use of a correlational approach: since the twins couldn’t ethically be assigned to different households as an experiment, comparing naturally occurring identical and non-identical twin pairs raised apart offered a way to isolate genetic influence from shared environment.

Correlational research also underpins much of the evidence used to evaluate personality assessment tools. Studies examining the MBTI’s relationship compatibility claims, for example, have reported weak correlations (generally below r = .30) between MBTI type-matching and relationship satisfaction outcomes — a finding covered in more depth in our analysis of MBTI test accuracy, which illustrates how a low correlation coefficient is itself meaningful evidence when evaluating whether a popular claim holds up statistically. The same correlational logic underpins broader discussions of personality psychology as a field, where reliability and validity coefficients are constantly used to judge which assessment tools deserve trust.

Common Mistakes and Misuses of Correlation Coefficients

A handful of errors come up repeatedly, both in student work and in media reporting of psychological findings:

• Treating correlation as proof of causation. As covered above, this is the single most common misuse, and one of the most heavily tested concepts in research methods (Bhandari, 2022).
• Ignoring restriction of range. If a sample only includes people with a narrow range of scores on one variable (for example, only high-achieving students), the calculated correlation can appear artificially weak, even if a stronger relationship exists in the wider population. A study of exam stress and performance that only surveys students already predicted to get top grades, for instance, may find little correlation at all, simply because there’s so little variation left to detect.
• Assuming linearity without checking. Pearson’s r only detects straight-line relationships. As Anscombe (1973) demonstrated, a genuinely strong curved relationship between two variables can produce a low or near-zero Pearson’s r, which is why plotting a scatter graph before calculating r is considered good practice.
• Over-relying on statistical significance alone. A correlation can be statistically significant (unlikely to be due to chance) while still being practically weak or unimportant, particularly in large samples (Cohen, 1990). Significance and strength answer two different questions.
• Missing a hidden subgroup effect (Simpson’s paradox). A correlation calculated across an entire dataset can reverse direction, or vanish entirely, once the data is split by a relevant subgroup such as age or experience level — a reminder that an overall correlation coefficient doesn’t always tell the full story.
• Cherry-picking or “spurious” correlations. With enough variables and enough data, some correlations will appear purely by chance. Two entirely unrelated statistical trends can produce a surprisingly high correlation coefficient with no meaningful connection between them at all — a pattern Reichenbach’s (1956) common cause principle helps explain when a shared underlying factor is genuinely at play, and one that’s simply coincidence when it isn’t.

Being alert to these pitfalls is part of what separates a superficial reading of a correlation coefficient from a genuinely critical one — the kind of judgement psychology courses are specifically designed to build, and one that Rodgers and Nicewander (1988) argued is best developed by learning to see a correlation coefficient from multiple angles rather than treating it as a single, self-explanatory number. If you’re building out your understanding of personality testing more broadly, our guide to the free MBTI function stack test and complete guide to the 16 MBTI types both apply these same correlational reasoning skills to evaluate how much confidence a personality result deserves.

Conclusion

A correlation coefficient is one of the most useful — and most misused — statistics in psychology. Pearson’s r tells you the direction and strength of a relationship between two variables on a simple −1 to +1 scale, and Spearman’s rho (or Kendall’s tau) offers a robust alternative when your data is ranked or non-linear. What counts as a “strong” correlation depends heavily on the field — a 0.3 that would be unremarkable in behavioural genetics can be a genuinely important finding in personality psychology. But the size of a correlation, however strong, never proves that one variable causes the other. Reading a scatter graph, checking for outliers and hidden subgroups, and asking whether a third variable could explain the pattern are just as important as the number itself. Used carefully, correlation coefficients remain one of the simplest and most powerful tools for spotting genuine relationships in psychological data.

Frequently Asked Questions

What does Pearson’s r tell us?

Pearson’s r tells you the strength and direction of a linear relationship between two continuous variables. It ranges from −1 (a perfect negative relationship) to +1 (a perfect positive relationship), with 0 meaning no linear relationship at all. The closer the value is to −1 or +1, the stronger the relationship, regardless of whether it’s positive or negative.

Why is a Pearson’s R test used?

Pearson’s r is used because it condenses the relationship between two variables into a single, comparable number. Researchers use it to check whether variables like stress and memory, or sleep and concentration, move together, and to compare the strength of different relationships across a study without inspecting every data point individually.

Is r 0.75 a strong correlation?

Yes. Using Cohen’s (1988) widely taught benchmarks, a correlation around 0.5 or above is generally considered large, so 0.75 falls comfortably into the strong to very strong range. In fields like personality psychology, where correlations are often much smaller, a value this high would be considered an unusually robust relationship — comparable to the IQ correlations found between identical twins raised apart.

What does a Pearson correlation of 0.5 mean?

A Pearson correlation of 0.5 indicates a moderate-to-strong positive relationship: as one variable increases, the other tends to increase too, though the relationship isn’t perfect. Squaring it (r² = 0.25) shows that only 25% of the variation in one variable is statistically associated with the other, leaving 75% explained by other factors.

What is correlation coefficient?

A correlation coefficient is a single statistical value, typically between −1 and +1, that describes how strongly two variables are related and in which direction. The most common version in psychology is Pearson’s r, used for continuous data, alongside Spearman’s rho for ranked or ordinal data.

Can a correlation coefficient prove causation?

No. A correlation coefficient, however large, only shows that two variables are statistically associated — it cannot show that one causes the other. A genuine relationship can still be explained by a third variable, reverse causation, or coincidence, which is why experimental research is needed to establish cause and effect.

What is the difference between Pearson’s r and Spearman’s rho?

Pearson’s r measures the linear relationship between two continuous variables and assumes minimal outliers. Spearman’s rho measures the relationship between the ranks of two variables rather than their raw scores, making it better suited to ordinal data, non-linear but consistently increasing or decreasing relationships, or data with outliers.

What is considered a negative correlation coefficient?

A negative correlation coefficient is any value below zero, down to −1. It means that as one variable increases, the other tends to decrease. For example, as reported stress increases, performance on some memory tasks tends to decrease, producing a negative Pearson’s r.

References

• Anscombe, F. J. (1973). Graphs in statistical analysis. The American Statistician, 27(1), 17–21.
• Bhandari, P. (2022). Correlation vs. causation: Uses and misuses. Scribbr.
• Bleske-Rechek, A., Morrison, K. M., & Heidtke, L. D. (2015). Causal inference from descriptions of experimental and non-experimental research: Public understanding of correlation-versus-causation. The Journal of General Psychology, 142(1), 48–70.
• Bouchard, T. J., Jr., Lykken, D. T., McGue, M., Segal, N. L., & Tellegen, A. (1990). Sources of human psychological differences: The Minnesota Study of Twins Reared Apart. Science, 250(4978), 223–228.
• Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Lawrence Erlbaum Associates.
• Cohen, J. (1990). Things I have learned (so far). American Psychologist, 45(12), 1304–1312.
• Cohen, J. (1992). A power primer. Psychological Bulletin, 112(1), 155–159.
• Doll, R., & Hill, A. B. (1950). Smoking and carcinoma of the lung: Preliminary report. British Medical Journal, 2(4682), 739–748.
• McCrae, R. R., & Costa, P. T. (2008). The five-factor theory of personality. In O. P. John, R. W. Robins, & L. A. Pervin (Eds.), Handbook of personality: Theory and research (3rd ed., pp. 159–181). Guilford Press.
• McLeod, S. (2019). What does effect size tell you? Simply Psychology.
• Pearson, K. (1895). Note on regression and inheritance in the case of two parents. Proceedings of the Royal Society of London, 58, 240–242.
• Reichenbach, H. (1956). The direction of time. University of California Press.
• Rodgers, J. L., & Nicewander, W. A. (1988). Thirteen ways to look at the correlation coefficient. The American Statistician, 42(1), 59–66.
• Spearman, C. (1904). General intelligence, objectively determined and measured. American Journal of Psychology, 15(2), 201–292.
• Twenge, J. M., Joiner, T. E., Rogers, M. L., & Martin, G. N. (2018). Increases in depressive symptoms, suicide-related outcomes, and suicide rates among U.S. adolescents after 2010 and links to increased new media screen time. Clinical Psychological Science, 6(1), 3–17.

Further Reading and Research

Recommended Articles

• Rodgers, J. L., & Nicewander, W. A. (1988). Thirteen ways to look at the correlation coefficient. The American Statistician, 42(1), 59–66.
• Cohen, J. (1992). A power primer. Psychological Bulletin, 112(1), 155–159.
• Aldrich, J. (1995). Correlations genuine and spurious in Pearson and Yule. Statistical Science, 10(4), 364–376.

Suggested Books

• Field, A. (2013). Discovering Statistics Using IBM SPSS Statistics (4th ed.). Sage Publications.
  • A widely used statistics textbook covering correlation, regression, and significance testing with worked examples aimed at psychology students.
• Huff, D. (1954). How to Lie with Statistics. W. W. Norton & Company.
  • A short, accessible classic explaining how statistics — including correlation — can be misused or misread, aimed at general readers.
• Cohen, J. (1988). Statistical Power Analysis for the Behavioral Sciences (2nd ed.). Lawrence Erlbaum Associates.
  • The foundational reference for effect size conventions, including the correlation strength benchmarks used throughout this article.

Recommended Websites

• Scribbr: Pearson Correlation Coefficient (r) — Guide & Examples
  • A step-by-step guide to calculating and interpreting Pearson’s r, with worked examples and assumption checks.
• Simply Psychology — Correlation and Research Methods
  • General-audience explanations of correlation, causation, and related statistical concepts for students and interested readers.
• Khan Academy — Statistics and Probability
  • Free video lessons covering correlation, scatterplots, and coefficient calculation step by step.

Kathy Brodie

Kathy Brodie is an Early Years Professional, Trainer and Author of multiple books on Early Years Education and Child Development. She is the founder of Early Years TV and the Early Years Summit.

Kathy’s Author Profile

To cite this article please use:

Early Years TV What Is a Correlation Coefficient? Pearson’s r and Psychology Statistics. Available at: https://www.earlyyears.tv/what-is-a-correlation-coefficient-pearsons-r-and-psychology-statistics/ (Accessed: 7 August 2026).