Null hypothesis

In scientific research, the null hypothesis serves as a fundamental starting point for investigation. Often denoted by the symbol $H_0$, the null hypothesis is a formal claim that no effect or relationship exists between the variables being studied. It acts as a default position, suggesting that any observed patterns or differences in data are merely the result of random chance. Scientists use this concept to separate genuine discoveries from statistical noise. By assuming there is no effect, researchers create a high bar for proof. This ensures that a new theory or medicine is only accepted when the evidence is truly strong.

To test this idea, researchers follow a specific mathematical mechanism. The process begins by collecting a random sample from a larger population. This sample must be representative to ensure the results are valid. Scientists then construct a statistical model of what the data would look like if chance alone were responsible for the results. This model is known as the distribution under the null hypothesis. The researchers then compare their actual observed results to this distribution. They use a tool called a test statistic to measure how much the observed data departs from what the null hypothesis predicts.

If the observed data is very unlikely to occur under the null hypothesis, the researcher rejects it. This rejection leads to the support of an alternative hypothesis, denoted as $H_1$ or $H_a$. The alternative hypothesis claims that a real relationship or effect does exist. However, if the data is consistent with the null hypothesis, the researcher does not reject it. It is important to note that failing to reject the null hypothesis does not prove it is true. It simply means there is insufficient evidence to claim otherwise. In many cases, this result is interpreted as a "don't know" rather than a confirmation of no effect.

There are several distinct types of hypotheses used in these tests. A simple hypothesis specifies the population distribution completely. In contrast, a composite hypothesis does not specify the distribution entirely. Researchers also distinguish between exact and inexact hypotheses. An exact hypothesis, or point hypothesis, specifies one precise parameter value. An inexact hypothesis specifies a range or an interval of values. Additionally, a one-tailed hypothesis has directionality. This means the researcher is looking for a change in only one direction, such as a value being specifically higher or lower than a certain point.

History shows how these methods evolved through different scientific approaches. Ronald Fisher developed the significance testing approach. In Fisher's method, a null hypothesis is rejected if the observed data is significantly unlikely to have occurred by chance. He famously used the "lady tasting tea" example to illustrate these concepts. Later, Jerzy Neyman and Egon Pearson introduced a different approach. They focused on contrasting the null hypothesis with an alternative hypothesis using specific error rates. Their method helped researchers make more structured decisions about which model best fits the data.

Statistical significance is often measured using specific thresholds. A researcher might decide that a result is significant if the probability of it occurring by chance is less than 5% or 1%. This threshold helps define the class of data-sets that are considered rare. For example, if a scientist tests whether a new drug reduces heart attack risks, the null hypothesis would state the drug has no effect. If the data shows a change that is statistically significant, the null hypothesis is rejected. This process provides a mathematical way to decide if a treatment is actually working.

Null hypotheses serve many different goals across various scientific fields. Technical null hypotheses are used to verify that statistical models are accurate. Scientific null assumptions can advance major theories, such as testing if the angular momentum of the universe is zero. Researchers also use null hypotheses of homogeneity to ensure that different experiments produce consistent results. Finally, null hypotheses of equality are used to compare different treatments, such as a drug versus a placebo. By testing these different claims, science maintains a rigorous and reliable standard of truth.