An alternative hypothesis is a statement that predicts a real effect or difference exists between groups or variables. It’s the claim a researcher actually wants to support, and it gets accepted only when the data provide enough evidence to reject the opposing statement (the null hypothesis). In notation, it’s written as Ha or H1.
How It Relates to the Null Hypothesis
Every statistical test starts with two competing statements. The null hypothesis (H0) says nothing interesting is happening: there’s no difference between groups, no relationship between variables, no effect of a treatment. The alternative hypothesis says the opposite: something IS going on. These two hypotheses are mutually exclusive, meaning only one can be true.
Here’s the key thing that trips people up: you never directly prove the alternative hypothesis. Instead, you test whether the null hypothesis is plausible given your data. If the evidence is strong enough to reject the null, you accept the alternative by default. Think of it like a courtroom. The null hypothesis is “not guilty.” You don’t prove guilt directly; you show enough evidence that “not guilty” is no longer believable.
One-Tailed vs. Two-Tailed Alternatives
The way you phrase your alternative hypothesis determines what kind of test you run. There are three options, and each one uses a different symbol:
- Greater than (>): You predict one value is higher than another. Example: cholesterol levels after eating oatmeal daily for six weeks will be lower than before. Written as Ha: μpost < μpre.
- Less than (<): You predict one value is lower than another.
- Not equal to (≠): You predict a difference exists but don’t specify which direction.
The first two are called one-tailed tests because you’re only looking for an effect in one direction. The third is a two-tailed test, where the effect could go either way. If you have a strong reason to expect a specific direction (say, a new drug should lower blood pressure, not raise it), use a one-tailed test. If you’re genuinely unsure which direction the effect might go, use a two-tailed test.
This choice matters practically. A two-tailed test splits your attention across both directions, so it’s harder to reach statistical significance. A one-tailed test focuses all the statistical power in one direction, making it easier to detect an effect there, but it completely ignores the possibility of an effect in the opposite direction.
How the Alternative Hypothesis Shapes Your P-Value
Once you’ve stated your alternative hypothesis, it directly controls how you calculate your p-value. The p-value answers a specific question: “If the null hypothesis were true, how likely is it that I’d see data at least this extreme in the direction my alternative hypothesis predicts?”
For a right-tailed test (Ha: μ > 3), you look at the probability of getting a result as large or larger than what you observed. For a left-tailed test (Ha: μ < 3), you look at the probability of getting a result as small or smaller. For a two-tailed test (Ha: μ ≠ 3), you look in both directions, and the p-value is always double what you’d get from a one-tailed test on the same data.
You then compare this p-value to your significance level (commonly set at 0.05). If the p-value is smaller, you reject the null hypothesis in favor of the alternative. If it’s larger, the data aren’t strong enough to reject the null.
Type II Errors and Statistical Power
Choosing the right alternative hypothesis also affects your ability to detect a real effect. A Type II error happens when you fail to reject the null hypothesis even though the alternative is actually true. In plain terms, it means missing a real effect. The probability of this happening is called beta (β).
Statistical power is the flip side: it equals 1 minus β, and it represents your chance of correctly detecting an effect when one exists. If you set β at 0.10, your study has 90% power, meaning that 90 times out of 100, you’d successfully detect an effect of the size you’re looking for. A vague or poorly chosen alternative hypothesis, or one that doesn’t match your study design, can reduce this power and make it more likely you’ll miss something real.
How to Write an Alternative Hypothesis
Start with your research question. If you’re studying whether a new teaching method improves test scores, your research question might be: “Do students taught with Method B score differently than students taught with Method A?” From there, identify the variables (test scores, teaching method) and decide on direction. If you have prior evidence that Method B should produce higher scores, your alternative hypothesis is directional: Ha: μB > μA. If you’re exploring without a strong prediction, go non-directional: Ha: μB ≠ μA.
A few rules always apply. The alternative hypothesis must directly contradict the null. If the null uses an equals sign (=), the alternative uses not-equal (≠), greater than (>), or less than (<). If the null uses greater-than-or-equal-to (≥), the alternative uses less-than (<). They always mirror each other. You should also state your hypothesis before collecting data, not after looking at results, since choosing a direction after seeing your data inflates the risk of a false positive.
Real-World Examples
In clinical trials, alternative hypotheses typically claim that a new treatment outperforms an existing one. A cancer study might compare survival times between two treatments, with the alternative hypothesis stating that patients on the new drug survive longer. The specific statistical test chosen (such as a log-rank test or a comparison of medians) implicitly defines exactly what “better” means, whether that’s a higher average survival time, a greater probability of outliving the comparison group, or a shift in the median outcome.
In paired studies, where the same person is measured twice, the alternative hypothesis focuses on individual differences. For instance, researchers studying tumor response to chemotherapy might measure tumor size before and after treatment in each patient. The alternative hypothesis would state that the median difference in tumor size is greater than zero, meaning tumors shrank. Similarly, eye studies comparing a treated eye to an untreated eye in the same person use paired alternative hypotheses.
In psychiatric research, an alternative hypothesis might predict that patients who attempted suicide report different medication habits than a control group. The null says there’s no difference. If the test rejects the null, the alternative is supported, and the researchers conclude a meaningful association exists between the medication and the outcome.