Spearman's Rank Correlation

• The Ranking Logic: The fundamental principle is that if two variables are perfectly correlated, the observation with the 1st highest value in variable X should also have the 1st highest value in variable Y. By focusing on the rank rather than the value, the calculation ignores the 'distance' between points and focuses entirely on their relative position.

• The Difference of Ranks ($d_i$): The core of the formula involves calculating the difference between the rank of an item in the first set and its rank in the second set. If the ranks are identical for all items, the sum of squared differences will be zero, leading to a perfect correlation of +1.

• Mathematical Foundation: The formula is derived from the Pearson correlation formula applied to the ranks of the data. When there are no tied ranks, the simplified formula is used: $r_s = 1 - \frac{6 \sum d_i^2}{n(n^2 - 1)}$ where $d_i$ is the difference between ranks and $n$ is the number of pairs.

• Step 1: Assign Ranks: For each variable independently, assign a rank to every data point from smallest to largest (or largest to smallest, as long as you are consistent). If two or more values are identical, they are 'tied' and must be handled specifically.

• Step 2: Handle Tied Ranks: When ties occur, assign each tied value the average of the ranks they would have occupied. For example, if two values are tied for 2nd and 3rd place, both are assigned a rank of 2.5 ($(2+3)/2$).

• Step 3: Calculate Differences ($d_i$): Subtract the rank of the second variable from the rank of the first variable for each pair of data. A useful check is that the sum of these differences ($\sum d_i$) must always equal zero.

• Step 4: Square and Sum: Square each difference ($d_i^2$) to eliminate negative signs and then sum them up to get $\sum d_i^2$. This value represents the total 'disorder' between the two sets of rankings.

• Step 5: Apply the Formula: Plug the sum of squared differences and the sample size ($n$) into the Spearman formula to find the coefficient. The resulting value will always fall between -1 and +1.

Spearman vs. Pearson Correlation

• When to choose Spearman: Use Spearman when your data is ordinal (like survey rankings), when the relationship looks curved on a scatter plot, or when your data fails the normality tests required for Pearson. It is the safer choice when you suspect outliers might be skewing your results.

• When to choose Pearson: Use Pearson when you have a clear linear relationship and your data is normally distributed. Pearson is more 'powerful' in a statistical sense when its assumptions are met because it uses the actual magnitude of the data rather than just the order.

• The Zero-Sum Check: Always verify that $\sum d_i = 0$ before squaring the differences. If the sum of your rank differences is not zero, you have made a mistake in assigning ranks or in simple subtraction, and proceeding will lead to an incorrect coefficient.

• Handling Ties Correctly: Examiners frequently include tied values to test your understanding of the 'average rank' rule. Ensure you don't just skip a rank; if two items are tied for 4th, the next item must be ranked 6th, not 5th.

• Interpreting the Sign: A positive $r_s$ means as one rank increases, the other tends to increase. A negative $r_s$ means as one rank increases, the other tends to decrease. A value of 0 indicates no monotonic relationship exists between the rankings.

• Sanity Check the Result: If your calculated $r_s$ is outside the range of -1 to +1, you have likely made an algebraic error. Additionally, if the data points in a scatter plot move generally in one direction, your $r_s$ should reflect that direction (positive or negative).

• Forgetting to Rank: A common error is calculating $d_i$ using the raw data values instead of their ranks. Spearman's method only works on the relative positions; using raw values will result in a meaningless number.

• Incorrect $n$ Value: Ensure $n$ represents the number of pairs of data, not the total number of individual data points. If you have 10 people with two scores each, $n = 10$.

• Misinterpreting 'No Correlation': A Spearman's coefficient of 0 only means there is no monotonic relationship. The variables could still have a complex non-monotonic relationship (like a U-shape), which Spearman's is not designed to detect.