Thinking in Hypotheses

With a clear question in hand, your supervisor introduces the next tool: the hypothesis (an educated guess your data could actually prove wrong). If nothing you could ever observe would change your mind, she says, it is not a real hypothesis.

For the nitrate question, she has you write two paired statements. The null hypothesis (the boring guess, that nothing has really changed): nitrate at well GW-14 has not changed beyond normal year-to-year variation. The alternative hypothesis (the guess that something did happen): nitrate at well GW-14 has increased beyond normal year-to-year variation. Later missions will show you exactly how to test between them. For now, the goal is just to think this way at all.

It feels backwards at first. Scientists usually start by assuming nothing interesting is happening, and only abandon that assumption if the evidence against it is strong. This protects you from seeing patterns that are really just random noise.

A hypothesis is falsifiable (able to be proven wrong by some possible result) if there is some dataset that would force you to reject it. Nitrate levels might be affected by many factors is not falsifiable, since it fits literally any data. Nitrate has increased by more than natural variability would explain is falsifiable. Five years of stable, flat data would reject it outright.

Imagine you think your friend moved your favorite toy. The null hypothesis is nobody touched it, I just forgot where I put it. You do not accuse your friend until you have checked everywhere else first. That is exactly how scientists treat data: assume nothing happened until the evidence says otherwise.

Researchers often write this pair of claims using the Greek letter mu (a symbol for a true average value we cannot observe directly, only estimate from samples).

Every statistical test you will learn from Mission 5 onward, including t-tests, regression significance, and ANOVA, is really just a formal procedure for choosing between a null and an alternative hypothesis like these. Learning to write the pair clearly now means the statistics later will feel like a natural next step.

Explain why scientists default to assuming the null hypothesis is true until shown otherwise, rather than the other way around.