A scientific hypothesis is a specific, falsifiable prediction about the relationship between variables. Good hypotheses in psychology follow an 'If–then' or directional format and must be testable with observable data. The null hypothesis (H₀) assumes no effect or relationship, while the alternative hypothesis (H₁) predicts one. Falsifiability — the possibility of being proven wrong — is what makes a claim scientific rather than metaphysical.
Practice converting vague research questions ('Does stress affect memory?') into precise hypotheses with named variables and directions. Evaluate famous claims for falsifiability.
From the scientific method, you know that science advances through a cycle of observation, theory, and empirical test. A hypothesis is the bridge between theory and test — it is the specific, concrete prediction that links the abstract idea to observable data. The challenge is that most interesting questions ("Does stress affect memory?") are too vague to test directly. To move from question to hypothesis, you must specify exactly what you mean by "stress," exactly what you mean by "memory," and exactly what relationship you expect to find. This process of specification is what turns a research idea into something falsifiable.
A good hypothesis has three properties. First, it is specific: it names the variables and predicts a direction or relationship ("Participants exposed to a time-pressure stressor will recall fewer words from a studied list than participants in a control condition"). Second, it is falsifiable: you can describe what data would count as evidence against it. If the stressed group recalled *more* words, the hypothesis would be wrong — and that possibility must be real, not hypothetical. Third, it is grounded: it connects to existing theory or prior findings, which is what distinguishes a scientific prediction from a random guess. You are predicting stress impairs memory *because* stress responses compete for attentional resources, or because cortisol disrupts hippocampal encoding — the mechanism matters.
The null hypothesis (H₀) is the formal machinery that operationalizes "nothing is going on." It typically states that there is no relationship, no difference, or no effect: the two groups have the same mean; the correlation is zero; the intervention has no effect. The alternative hypothesis (H₁) is your prediction — the effect you expect to find. The logic of significance testing is that you temporarily assume H₀ is true and ask: if there were truly no effect, how likely is it that I would observe data at least this extreme by chance? If the answer is "very unlikely" (p < 0.05 by convention), you reject H₀ in favor of H₁. Crucially, you never confirm H₁ directly — you only reject or fail to reject H₀. This asymmetry is not arbitrary; it follows from the logical structure of falsification.
The distinction between a directional (one-tailed) hypothesis and a non-directional (two-tailed) hypothesis matters for both statistical testing and scientific honesty. A directional hypothesis — "stressed participants will perform *worse*" — specifies where you expect the effect to land. A non-directional hypothesis — "stressed participants will perform *differently*" — allows for either direction. Directional hypotheses are appropriate when prior evidence strongly suggests a direction and are slightly more powerful statistically; non-directional hypotheses are more honest when the direction is genuinely uncertain. Pre-registering your hypothesis before collecting data — committing in advance to what you predict and how you will test it — is the discipline that prevents you from mining the data for any pattern and then claiming you predicted it.