On this page
A null hypothesis is the baseline assumption in statistical testing that no effect or relationship exists between variables. Researchers collect data and calculate whether those data would be rare if the null were true. If the data are sufficiently unlikely under that assumption, the null is rejected, providing evidence for an alternative claim. Rejecting the null does not prove the alternative is true; it only shows the data are inconsistent with no effect.
- The null hypothesis is the default position of no effect; the alternative hypothesis is the claim a researcher wants to test.
- A p-value answers: if the null were true, how rare would this data be? A small p-value suggests the null assumption is unlikely given what was observed.
- Researchers choose a significance level (commonly 0.05 ) before testing; they reject the null only if the p-value falls below that threshold.
- Rejecting the null means the data are inconsistent with no effect, not that the alternative is definitely true or that the effect is large or clinically important.
- Type I error is rejecting a true null hypothesis (false positive); Type II error is failing to reject a false null hypothesis (false negative).
A null hypothesis is the default assumption in statistical testing that no effect or relationship exists. It is the baseline claim a researcher assumes to be true, then tests against the data collected. Understanding what a null hypothesis is and how it functions is essential for reading research reports and evaluating whether evidence supports a claimed finding.
The null hypothesis is not the researcher's goal or prediction. It is the opposite: the claim that nothing is happening, no difference exists, no relationship holds. The researcher's actual prediction is called the alternative hypothesis. By starting with the assumption of no effect, statistical testing protects against the human tendency to see patterns in random noise. This framework forces researchers to ask: would I see data like this by chance alone if nothing were really going on?
What a null hypothesis actually is
A null hypothesis is a statement that no effect, no difference, or no relationship exists between the variables being studied. It is the default assumption: the claim that the status quo holds, that an intervention does nothing, that two groups are identical, or that two measurements are unrelated.
The null hypothesis is written as a formal statement so that it can be tested mathematically. For example, if a researcher wants to know whether a new training program improves employee productivity, the null hypothesis would be: the training program has no effect on productivity. If a researcher wants to know whether two populations have different average heights, the null hypothesis would be: the two populations have the same average height.
The null hypothesis is not a guess about what the researcher expects to find. It is the opposite of what the researcher often hopes to demonstrate. By starting with the assumption that nothing is happening, the framework ensures that a researcher cannot claim an effect exists simply because the data look suggestive. The data must be unusual enough under the null assumption to justify rejecting it.
How null hypothesis testing works
Null hypothesis testing follows a structured logical sequence. The researcher assumes the null hypothesis is true, collects data, calculates a test statistic, and determines how rare that statistic would be if the null were actually true. If the data are sufficiently rare under the null assumption, the researcher rejects the null. If not, the researcher fails to reject it.
The process does not prove the null is true or false. Instead, it asks: given that the null is true, would I expect to see data like this? A small probability of observing such data suggests the null assumption is questionable. A large probability suggests the data are consistent with the null.
Setting up the null and alternative hypotheses
Before collecting any data, a researcher must state both the null hypothesis and the alternative hypothesis. The null hypothesis claims no effect exists. The alternative hypothesis claims an effect does exist, or that a relationship holds, or that a difference is real.
These two statements must be mutually exclusive and exhaustive: exactly one of them is true, and the data will determine which one is more consistent with what was observed. The alternative hypothesis can be one-sided (claiming an effect in a specific direction) or two-sided (claiming an effect exists in either direction). A two-sided alternative is more conservative and is the standard choice when the researcher has no strong prior reason to expect the effect to go in one direction rather than the other.
Calculating the test statistic and p-value
Once data are collected, the researcher calculates a test statistic. The test statistic is a number that summarizes how far the observed data deviate from what the null hypothesis predicts. Different types of research questions use different test statistics (a t-statistic for comparing means, a chi-square statistic for comparing proportions, and so on), but the logic is the same.
From the test statistic, the researcher calculates a p-value. The p-value answers a specific question: if the null hypothesis were true, what is the probability of observing a test statistic as extreme as (or more extreme than) the one calculated from the data? A small p-value means the observed data would be rare if the null were true. A large p-value means the observed data are consistent with what the null hypothesis predicts.
Making a decision: reject or fail to reject
Before running the test, the researcher chooses a significance level, often called alpha. Common choices are 0.05, 0.01, and 0.10. The significance level is the threshold: if the p-value is less than or equal to alpha, the researcher rejects the null hypothesis. If the p-value is greater than alpha, the researcher fails to reject the null.
Failing to reject the null does not mean the null is true. It means the data are not unusual enough under the null assumption to justify abandoning that assumption. The data could be consistent with the null, or the sample size could be too small to detect a real effect, or random variation could have masked the true pattern. Rejection means the data are unlikely under the null assumption; it does not mean the alternative is proven.
A worked example: testing a new drug's effect
Suppose a pharmaceutical company has developed a new drug and wants to know whether it reduces blood pressure. The null hypothesis is: the drug has no effect on blood pressure. The alternative hypothesis is: the drug does reduce blood pressure.
The researchers recruit a group of patients, give half of them the drug and half a placebo, and measure blood pressure in both groups after eight weeks. They calculate the average blood pressure in each group and find that the drug group has an average of 5 millimeters of mercury lower blood pressure than the placebo group.
Now the researchers calculate a test statistic based on the difference between groups, the variability within each group, and the sample size. Suppose the test statistic is large enough that the p-value comes out to 0.03. This means: if the drug truly had no effect, there would be only a 3% chance of observing a difference this large (or larger) by random variation alone.
The researchers chose a significance level of 0.05 before running the study. Since 0.03 is less than 0.05, they reject the null hypothesis. They conclude that the data provide evidence the drug does reduce blood pressure. But this conclusion is provisional: it is based on the assumption that their test was conducted correctly and that the observed difference is not a rare coincidence. It does not mean the drug definitely works or that the effect is clinically important.
Why null hypothesis testing matters in research
Null hypothesis testing protects against a common human bias: the tendency to find patterns in random data. If a researcher simply looked at two groups and saw that one had a higher average than the other, that observation alone would not prove a real difference exists. Random variation could easily produce such a difference by chance.
By formalizing the question as a hypothesis test, researchers force themselves to ask: how unlikely would this difference be if nothing were really going on? This framework prevents researchers from claiming success whenever the data point in the desired direction. It requires that the data be sufficiently inconsistent with the null assumption before a claim of an effect is made.
The framework also makes the reasoning transparent and reproducible. Another researcher can read the hypothesis, the significance level chosen, the test statistic, and the p-value, and can verify whether the conclusion follows logically from the evidence. This standardization is one reason null hypothesis testing has become the default approach in most scientific fields.
Common misreadings and limitations
Null hypothesis testing is widely used but frequently misunderstood. Several misconceptions are common enough that they appear regularly in research reports and news coverage of research.
What rejecting the null actually tells you
When a researcher rejects the null hypothesis, it means the data are unlikely under the null assumption. It does not mean the alternative hypothesis is true, that the effect is large, or that the effect is clinically or practically important.
Rejection says only that the data are inconsistent with no effect. A very large sample size can produce a small p-value even when the true effect is tiny and irrelevant in practice. For example, a drug might lower blood pressure by 0.5 millimeters of mercury on average. With a large enough sample, this minuscule difference could be statistically significant (p-value below 0.05 ), and the null would be rejected. But a 0.5 millimeter difference is not clinically meaningful.
Conversely, rejecting the null does not prove the alternative is definitely true. The data are inconsistent with the null, but other explanations might exist: measurement error, confounding variables, or bias in how the study was conducted. Statistical significance means the data are unlikely under the null assumption. It does not mean the finding is correct or that the mechanism proposed by the researcher is real.
Why p-value does not equal probability the null is true
A common misreading is that a p-value of 0.03 means there is a 3% chance the null hypothesis is true. This is incorrect. The p-value is the probability of observing data as extreme as (or more extreme than) what was observed, given that the null hypothesis is true. It is not the probability that the null is true given the data.
This asymmetry matters. The p-value is calculated under the assumption that the null is true. It answers: how rare would this data be if I assume the null is true? It does not answer: how likely is the null to be true given the data I observed?
To calculate the probability that the null is true given the data, a researcher would need to use Bayesian methods, which incorporate prior beliefs about the likelihood of the null before seeing the data. Null hypothesis testing is a frequentist approach and does not provide this probability. Over the past decades, reliance on the conventional cut-off of p = 0.05 has fostered misconceptions about what p-values mean, including the belief that statistical significance implies practical importance.
How null hypothesis testing connects to confidence intervals and effect size
A p-value alone tells only whether the data are inconsistent with the null assumption. It does not tell how large the effect is or how precisely it was estimated. Two additional pieces of information are essential for a complete picture: a confidence interval and an effect size estimate.
A confidence interval provides a range of plausible values for the true effect. If a study estimates that a drug reduces blood pressure by 5 millimeters of mercury, a 95% confidence interval might be 2 to 8 millimeters. This interval tells the reader that the true effect could plausibly be anywhere in that range, given the data and the sample size. A narrow confidence interval suggests a precise estimate; a wide one suggests uncertainty.
An effect size is a standardized measure of how large the effect is, independent of sample size. Two studies might both reject the null hypothesis with p-value below 0.05, but one might show a large effect while the other shows a tiny one. Effect sizes allow readers to compare the magnitude of effects across studies and to judge whether an effect, even if statistically significant, is practically meaningful.
Together, the p-value, confidence interval, and effect size provide a more complete picture than any one of them alone. Researchers increasingly report all three, and readers of research should look for all three when evaluating whether a claimed finding is credible and important.
Do this automatically
A researcher manually setting up null and alternative hypotheses, calculating test statistics, determining p-values, and documenting the decision threshold for each study can spend hours on these computational steps alone, multiplied across dozens of experiments per year. Software can automate the formulation of competing hypotheses, statistical calculations, and significance level comparisons, reducing the time spent on mechanical tasks to seconds. Academia research runs this kind of analysis end to end and cites every figure back to its original source, across as many subjects as you need, with Mimir attaching a citation to every figure. It will not write your paper for you. It gathers and cites the evidence; the argument stays yours, and deliberately so. Every plan includes it, including the free one.
Start your first report and run it on something you already know well.
This article is general information and research commentary. It is not professional advice, and it does not account for your particular circumstances.
