Inferential statistics are the tools psychologists use to draw conclusions about large groups of people based on data collected from a smaller group. If a researcher studies 200 college students to learn something about anxiety, inferential statistics are what allow them to say their findings likely apply beyond just those 200 people. This stands in contrast to descriptive statistics, which simply summarize the data you already have (averages, percentages, ranges) without making any broader claims.
Why Psychologists Can’t Study Everyone
Psychology aims to understand human behavior and mental processes in general, not just in the specific people who show up to a lab. But studying an entire population is almost never possible. You can’t measure the working memory of every 10-year-old on Earth or survey every person with depression. Instead, researchers study a sample, a manageable subset of the population they care about, and use inferential statistics to bridge the gap between what they observed in that sample and what’s likely true for the broader group.
This leap from sample to population is the defining feature of inferential statistics. It’s also where things can go wrong. The quality of the inference depends heavily on how well the sample represents the population. A study on stress conducted entirely with undergraduate students at one university may not generalize to older adults, people in different cultures, or people outside of academic settings. Random sampling, where every member of the target population has an equal chance of being selected, is the gold standard for making that generalization trustworthy. Random samples are the means, but representativeness is the real goal.
How Hypothesis Testing Works
Most inferential statistics in psychology revolve around hypothesis testing, a structured way of asking: “Is the pattern I see in my data real, or could it just be random noise?” The process starts with two competing statements. The null hypothesis proposes that there’s no real effect or no real difference between groups. The alternative hypothesis proposes that there is one.
Say a researcher wants to know whether a new therapy reduces symptoms of social anxiety more than a placebo. The null hypothesis would be: “There’s no difference in symptom reduction between the therapy group and the placebo group.” The alternative hypothesis would be: “There is a difference.” Statistical tests then calculate the probability that the observed results would occur if the null hypothesis were true. If that probability is very low, the researcher rejects the null hypothesis and accepts the alternative.
The alternative hypothesis can’t be tested directly. It’s accepted only by ruling out the null, which is why this process sometimes feels backward. You don’t prove your theory is correct. You show that the “nothing is happening” explanation is unlikely enough to dismiss.
P-Values and Statistical Significance
The probability calculated during hypothesis testing is called the p-value. It reflects how compatible your data are with the null hypothesis. A small p-value means your results would be very unlikely if there were truly no effect. Psychology has traditionally used a threshold of 0.05: if the p-value falls below 0.05, the result is considered “statistically significant.”
That threshold, though, has come under serious scrutiny. The American Statistical Association published a statement in 2016 challenging the research community to reconsider how it uses the term “statistically significant.” Some researchers have proposed lowering the threshold to 0.005 to reduce false positives, while prominent journals like the New England Journal of Medicine and Nature have supported reducing or abandoning reliance on p-values altogether. The core concern is that a p-value just below 0.05 doesn’t mean the effect is large, important, or even replicable. It only means the data were unlikely under one specific assumption.
Effect Sizes and Confidence Intervals
Because p-values alone don’t tell you how meaningful a finding is, modern reporting standards in psychology emphasize two additional tools: effect sizes and confidence intervals.
An effect size tells you how big the difference or relationship actually is. A therapy might produce a statistically significant reduction in anxiety, but if the improvement is tiny, it may not matter in practice. Common effect size measures include Cohen’s d (which quantifies the difference between two group averages in standardized terms) and R² (which tells you how much of the variation in one variable is explained by another). A large effect size paired with a significant p-value is a much stronger finding than a significant p-value alone.
A confidence interval gives you a range of plausible values for the true effect, typically reported as a 95% CI. Instead of a single yes-or-no verdict, it shows you the precision of the estimate. A narrow confidence interval means the study pinpointed the effect fairly well. A wide one means there’s a lot of uncertainty. Researchers and clinicians are increasingly encouraged to look at confidence intervals rather than simply asking whether a result crossed the 0.05 line.
Common Inferential Tests in Psychology
Different research questions call for different statistical tests. The choice depends on how many groups you’re comparing, what kind of data you have, and whether the same people are being measured more than once.
- T-test: Compares the averages of two groups. An independent samples t-test is used when the groups are made up of different people (therapy group vs. placebo group). A paired samples t-test is used when the same people are measured twice (before and after treatment).
- ANOVA (analysis of variance): An extension of the t-test for three or more groups. If a researcher wants to compare the effectiveness of cognitive behavioral therapy, medication, and a placebo, ANOVA handles that in a single analysis rather than running multiple t-tests.
- Correlation: Measures the strength and direction of the relationship between two variables, producing a coefficient (r) that ranges from -1 to 1. A correlation of 0.8 between sleep quality and mood suggests a strong positive relationship. Importantly, correlation does not establish cause and effect.
- Regression: Goes a step further than correlation by identifying one variable as a predictor and the other as an outcome. It also allows researchers to evaluate multiple predictors at once, such as examining how sleep, exercise, and social support each contribute to mood.
Parametric vs. Non-Parametric Tests
The tests listed above (t-tests, ANOVA, Pearson correlation) are parametric, meaning they assume the data follow a roughly bell-shaped distribution. When that assumption holds, these tests are powerful and precise. But when data are skewed, ranked, or come from very small samples (generally under 30 participants), those assumptions break down.
In those situations, psychologists use non-parametric alternatives that don’t require normal distribution. The Wilcoxon rank-sum test replaces the independent samples t-test, the Kruskal-Wallis test replaces ANOVA, and Spearman’s rank correlation replaces Pearson’s correlation. These tests work by ranking the data rather than relying on means and standard deviations, making them more robust when the data don’t cooperate with parametric assumptions.
Type I and Type II Errors
No statistical test gives you certainty. Two kinds of mistakes are always possible. A Type I error happens when you reject the null hypothesis even though it’s actually true. In practical terms, this means concluding that a treatment works when it doesn’t. The 0.05 significance threshold directly controls this risk: it means you’re accepting a 5% chance of a Type I error.
A Type II error is the opposite. You fail to reject the null hypothesis when there really is an effect. The treatment actually works, but your study didn’t detect it, often because the sample was too small or the effect was subtle. Researchers manage this risk through statistical power, which increases with larger sample sizes and stronger effects. A well-designed psychology study balances both types of error, aiming for enough participants and a sensitive enough design to detect real effects without jumping at noise.
Why This Matters for Psychology
Inferential statistics are what separate anecdotal observation from scientific evidence in psychology. When a clinical trial reports that a treatment significantly reduces PTSD symptoms, or a developmental study finds that bilingual children outperform monolingual children on certain cognitive tasks, those claims rest on inferential methods. The statistics quantify how confident we can be that the pattern is real and not a fluke of the particular group studied.
Understanding these tools also helps you evaluate research you encounter in everyday life. A finding with a tiny effect size and a barely significant p-value from 30 participants tells a very different story than one with a large effect size, a tight confidence interval, and 2,000 participants. The numbers behind the headline determine whether a finding is worth paying attention to.