Null Hypothesis

Null Hypothesis: The null hypothesis is a foundational concept in statistical hypothesis testing. It represents the assumption of no effect, no difference, or no relationship between variables. It serves as a starting point or baseline for statistical comparison. Research is conducted with the aim of either refuting (rejecting) or failing to refute the null hypothesis. This is determined based on whether the observed data demonstrates a significant difference from "no effect."

Clarification: The null hypothesis is not the same as an opposing position or alternative theory. It is specifically the assumption that any observed effect is due to random variation or chance. It is the "no effect" hypothesis, not "the opposite effect" hypothesis.

Purpose and Function: The null hypothesis provides a statistical anchor. It provides the default assumption (no relationship) that can be tested with data. For example, in a study comparing two treatments, the null hypothesis would state that there is no difference in their effects.

Once data is collected, statistical tests determine the probability of observing the study's findings assuming the null hypothesis is true. If this probability (p-value) is sufficiently small (e.g., p < 0.05), the null hypothesis is rejected. The null hypothesis is rejected because the difference in the data was so "extreme" that it would be highly unlikely to occur by chance if the intervention had actually had no effect.

How to Write a Null Hypothesis

To write a null hypothesis, first start by asking a question. Rephrase that question in a form that assumes no relationship between the variables. In other words, assume a treatment has no effect. Write your hypothesis in a way that reflects

Why Not Test the Hypothesis Directly?

A hypothesis is a proposed explanation for a phenomenon that can be tested through research and experimentation. However, research can only demonstrate the likelihood of a relationship between variables and outcome measures. It does not prove why a relationship exists, nor does it determine the precise nature of that relationship. Theoretically, there may be an infinite number of contributing factors to why a relationship was observed (or not). Any hypothesis about why a relationship exists competes with all other plausible explanations, both known and unknown. In short, we cannot test the validity of a hypothesis against an infinite set of alternatives, but we can test the probability of a relationship between a variable and outcome measures.

When we test against the null hypothesis, we arrive at one of two outcomes:

1. We fail to reject the null hypothesis, implying that there is no relationship between the tested variables that cannot be explained by chance.
2. We reject the null hypothesis, indicating that a statistically significant difference from the null exists and that the observed relationship is unlikely to be due to random variation alone.

Determining causality requires methodically designed studies that aim to both demonstrate relationships and refute incorrect hypotheses. Note that a scientific approach, or evidence-based approach , to any topic involves accepting the hypothesis that is most consistently supported by evidence and has demonstrated the best reliability and accuracy for predicting outcomes in repeated experiments. For more reading, and a beautiful quote for a title, check out the article -

• “Science: the slow march of accumulating evidence ” - Katherine Picho, Lauren A. Maggio, and Anthony R. Artino Jr.

A few reasons why starting with the null hypothesis is more rigorous for a few key reasons:

• It Encourages Objectivity and Reduces Bias: The null hypothesis begins from a position of skepticism. The burden of proof lies with the data to demonstrate that the observed results are unlikely to be due to chance. In other words, if the research fails to demonstrate an effect, the default assumption is not the hypothesis generated by the researcher. Furthermore, evidence refuting the null still yields a conservative conclusion, implying only that it is likely an effect occurred. Rejecting the null hypothesis does not automatically validate any specific alternative hypothesis.
• You Can’t Prove a Hypothesis; Only Disprove One: One of the key tenets of the scientific method is that a hypothesis must be falsifiable (it must be possible to prove it false with testing). In science, we don’t “prove” anything with certainty. Determining causation involves demonstrating the low probability of alternative hypotheses. The null hypothesis serves as a clear starting point that aligns with the concept of falsification.
• Probability Models Require Assumptions: Many statistical models are mathematically built on the assumption that the null hypothesis is true. Although this presents a chicken-or-the-egg dilemma—did the null hypothesis result from these models, or did these models lead to the refinement of the null? History suggests they co-evolved.
  • In 1933, Jerzy Neyman and Egon Pearson introduced the formal structure of hypothesis testing, including the distinction between the null (H₀) and alternative (H₁) hypotheses, and the concepts of Type I and Type II errors. They framed hypothesis testing as a decision-making process with predefined error rates.
  • In 1935, Ronald Fisher popularized the null hypothesis and p-values as a way to assess how surprising a result would be, assuming no effect. His focus was on evaluating evidence rather than making strict binary decisions.
  • Together, these contributions laid the foundation for most of the statistical tools used in frequentist inference—t-tests, ANOVA, confidence intervals, etc.—all of which rely on assuming the null hypothesis is true to estimate sampling distributions and make valid probabilistic conclusions.

Related Articles

• Meta-analysis Problems: Why do so many imply that nothing works?
• Is There a Single Best Approach to Physical Rehabilitation?
• NCCA-accredited Certified Personal Trainer (CPT) Exams: Meta-analysis of Topics (Domains)

Related Terms

• Systematic Review
• Evidence-based practice
• Levels of evidence
• Meta-analysis
• P-value
• Null Hypothesis
• Vote-Counting
• Regression to the Mean
• Randomized Controlled Trial
• Experimental Research
• Prospective Study
• Longitudinal Study
• Case Control Study
• Cross-sectional Study
• Retrospective Study

Brookbush Institute Perspectives

Common Misconceptions and Fallacies:

• Misuse #1: Failing to reject the null does not equal proof of the opposite.
  For example, a study that finds no correlation between an anterior pelvic tilt and hip pain does not mean that posture and pain are not correlated. It simply means that the tested postural variable was not strongly correlated with hip pain in the studied population.
• Misuse #2: Absence of evidence is not evidence of absence.
  A non-significant result in a study (a failure to refute the null) does not imply that a relationship does not exist. It only implies that this study failed to demonstrate a relationship, which may be due to the absence of a relationship, or potentially a lack of sensitivity in the outcome measure, the study being underpowered, an outlier in the participant group, etc. All relevant research should be considered in developing a conclusion.
• Misuse #3: "Bait and switch" arguments.
  Too often, gurus and professionals incorrectly treat a failure to refute the null as evidence that their preferred alternative is correct. This is a fallacy, known as the "unsupported default position fallacy". Unless there are only two possible hypotheses (this is rarely the case), then a failure to support one hypothesis does not automatically support the other. Each hypothesis must be considered individually, and the most well-supported hypothesis should be adopted.
• Misuse #4: Overreliance on Statistical Significance to Dismiss Nuance.
  Testing against the null hypothesis often leads to binary interpretations: reject or fail to reject. However, this binary mindset ignores effect size, variance, sample size, and context. A small but consistent trend may be missed due to underpowering, and a single statistically significant finding may be practically meaningless. Evaluating evidence requires a synthesis all relevant research

Frequently Asked Questions (FAQ)

Can a hypothesis be proven?

No. Scientific hypotheses cannot be definitively proven. Even if a hypothesis generates predictions that align with observed results, there may be alternative hypotheses that produce the same predictions. Since it is impossible to test every conceivable explanation, science relies on the accumulation of evidence supporting the most plausible and predictive hypothesis — not proof.

What does the null hypothesis actually say?

The null hypothesis states that there is no effect, no difference, or no association between the variables being tested. It represents the assumption that any observed differences are due to random variation.

If a study fails to reject the null, does that mean the intervention doesn’t work?

Not necessarily. A failure to reject the null only means the study did not provide sufficient evidence to demonstrate an effect. This may be due to a lack of statistical power, poor sensitivity of outcome measures, sampling error, or the possibility that the intervention is truly ineffective under the tested conditions. It does not prove the intervention is universally ineffective.

Does rejecting the null prove my hypothesis is true?

No. Rejecting the null hypothesis provides evidence that an effect likely occurred, but it does not prove why it occurred or confirm that a specific alternative hypothesis is correct. Unless your hypothesis is the only reasonable explanation for the observed effect, additional testing is needed to rule out competing explanations.

Why not just test the hypothesis we believe in?

Testing by attempting to reject the null hypothesis minimizes confirmation bias and improves objectivity. This approach forces researchers to ask: “Are the observed results unlikely under the assumption of no effect?” — rather than simply searching for data that supports a favored idea.

Is the null always the “wrong” hypothesis?

No. Sometimes the null hypothesis accurately reflects reality. Many high-quality studies fail to reject the null, and this is a valid result. When interpreted properly, a retained null helps clarify where effects do not exist or where current methods cannot reliably detect them. The null hypothesis plays a critical role in refining future research.