Hypothesis Testing for Correlation

Population Correlation Coefficient ($\rho$): This represents the true linear relationship between two variables across an entire population. It is a fixed but usually unknown parameter that we seek to make inferences about.

Sample Correlation Coefficient ($r$): Also known as the Product Moment Correlation Coefficient (PMCC), this is calculated from a specific subset of data. Because of sampling variation, $r$ may be non-zero even if the population has no correlation.

The Null Hypothesis ($H_0$): In correlation testing, the null hypothesis almost always states that there is no linear correlation in the population, expressed as $H_0: \rho = 0$.

The Alternative Hypothesis ($H_1$): This reflects the researcher's claim. It can be one-tailed (predicting a specific direction: $\rho > 0$ or $\rho < 0$) or two-tailed (predicting any correlation: $\rho \neq 0$).

Step 1: State Hypotheses: Define $H_0: \rho = 0$ and choose $H_1$ based on the wording of the problem (e.g., 'test for positive correlation' implies $H_1: \rho > 0$).

Step 2: Identify Parameters: Determine the sample size ($n$) and the significance level ($\alpha$). These are used to find the critical value from a statistical table.

Step 3: Compare and Decide: Compare the calculated sample PMCC ($r$) to the critical value. If $|r| > |\text{critical value}|$, the result is significant.

Step 4: Contextual Conclusion: State whether there is 'sufficient evidence' or 'insufficient evidence' to support the original claim, always referencing the specific variables involved.

Check the Parameter: Always use the Greek letter $\rho$ in your hypotheses. Using the sample statistic $r$ in $H_0$ or $H_1$ is a common error that results in lost marks.

Absolute Values: When dealing with negative correlations, compare the magnitude. For example, if $r = -0.6$ and the critical value is $-0.5$, the result is significant because $|-0.6| > |-0.5|$.

Sample Size Matters: Ensure you use the correct $n$ when looking up critical values. Small samples require much higher $r$ values to be considered significant compared to large samples.

Standard Phrasing: Use the phrase 'There is sufficient evidence at the $x\%$ level of significance to suggest...' to ensure your conclusion meets examiner expectations.

Correlation vs. Causation: A significant result in a hypothesis test proves a linear relationship exists, but it does not prove that one variable causes the other.

Non-Linear Relationships: A non-significant result ($r \approx 0$) only means there is no linear correlation. The variables could still have a strong non-linear (e.g., quadratic) relationship.

p-value Confusion: If provided with a p-value, remember the rule: 'If the p is low, the null must go.' If $p < \alpha$, reject $H_0$.