Statistical Tests and Conclusions

A chi-squared goodness-of-fit test can compare observed category counts with expected counts, for example a proposed genetic ratio. A suitable t-test compares means under the relevant sampling and distribution assumptions; its paired or independent form must match the design. A correlation coefficient addresses association between paired variables, not a difference between two means.

Spearman’s rank is a useful correlation approach for suitable ranked or monotonic data. AQA’s current guidance does not require every correlation question to be labelled only with one memorised coefficient name: an appropriate named coefficient or the general category can be relevant. Understand the purpose and justify the choice.

A null hypothesis should identify the population comparison or association being tested, such as no difference between population means or no association between the stated variables. Under a model, a p-value describes how unusual data at least as extreme as observed would be if that null and the test assumptions held.

At a stated 5% threshold, p≤0.05 supports rejecting the null in this framework. A larger p-value means insufficient evidence to reject it, not proof that the populations are identical. A small p-value is not the probability that the null is true, nor proof that a particular causal mechanism has been established.

For chi-squared, a common supplied expression is Σ(O−E)²/E. In a simple goodness-of-fit example with fixed expected proportions, degrees of freedom can be categories minus one; other designs or estimated parameters can change the rule. Use the information provided and appropriate expected-count conditions.

Practise calculation in laboratory learning, but written-exam preparation should prioritise choosing, justifying and interpreting tests and using the statistics or abridged tables supplied. SD calculation is not required in written papers. Statistical significance, effect size, biological importance and causal inference are different conclusions, all affected by sampling quality and design.

Try each question before opening the worked response. Use precise biological terms, connect each cause to its effect, and check units and assumptions when interpreting data.