Correlation

Further information on this topic can be found in Chapter 11 of:

Maths from Scratch for Biologists

Numerical ability is an essential skill for everyone studying the biological sciences but many students are frightened by the 'perceived' difficulty of mathematics, and are nervous about applying mathematical skills in their chosen field of study. Maths from Scratch for Biologists is a highly instructive, informal text that explains step by step how and why you need to tackle maths within the biological sciences. (Amazon.co.UK)

Correlation, the relationship between two variables, is closely related to prediction.

The greater the association between variables, the more accurately we can predict the outcome of events. There is rarely an exact correlation of observed results with a mathematical function - the points never fit exactly on the line. The question is therefore whether an association between two variables could have occurred by chance.

There are numerous methods for calculating correlation, e.g:

• The parametric Pearson, or "r value", correlation
• The nonparametric Spearman correlation
• Many others

Pearson correlation calculations are based on the assumption that both X and Y values are sampled from populations that follow a normal (Gaussian) distribution, at least approximately, although with large samples, this assumption is not too important.
Alternatively, the nonparametric Spearman correlation is based on ranking the two variables, and so makes no assumption about the distribution of the values.

A correlation analysis is performed in the same as any other statistical test of significance:

1. Formulate the null hypothesis. A simpler hypothesis has priority over a more complex theory, so the null hypothesis (H₀) is therefore that "There is no correlation between the datasets". You also need to set the significance level (a) before performing the test (e.g. 0.05).

2. Calculate the correlation coefficient (r) for the test data. The Pearson formula is rather cumbersome:

so don't learn this equation - use software to perform the calculation, e.g. MSExcel (see below).

3. Determine whether the value of r is equal to or less than the critical value required to accept the null hypothesis. To do this you need to calculate the number of degrees of freedom for the test - this takes into account the number of independent observations used in the calculation of the test statistic and is needed to find the true value in a probability table. For a one-tailed test, df = n-1, and for a two-tailed test (most usual) df=n-2. Look up the calculated value of r in a table of critical values of the correlation coefficient.

4. When you look up the r value, ignore the sign (+/-). If you did not do this, you would need two tables of r values, one positive and one negative, but otherwise identical. The sign of the r value tells you whether the correlation is positive or negative, but can be ignored when deciding whether a result is significant or not.

5. If the calculated value of r is less than or equal to the critical value of r, accept the null hypothesis - there is no proof of significant correlation between the variables. (Note: not "there is no correlation", simply, "there is no proof of correlation".)
   If the calculated value of r is greater than the critical value of r (from the table), reject the null hypothesis and accept the alternative hypothesis.

Values of r range from +1 (perfect correlation), through 0 (no correlation), to -1 (perfect negative correlation):

Warning:
Correlation tests are in some ways the most misused of all statistical procedures!
They are able to show whether two variables could be connected. However, they are not able to show that the variables are not connected! If one variable depends on another, i.e. there is a causal relationship, then it is always possible to find some kind of correlation between the two variables. However, if both variables depend on a third, they can show a correlation without any causal dependency between them. Take care!

Example: There is a direct correlation between the number of mobile phone masts and the decline in the numbers of house sparrows, Passer domesticus. But do mobile phone masts harm sparrows, or are both effects caused by something else? Or are they both completely independent observations which just happen to correlate? We don't know because correlation tests do not reveal this information - further investigation is necessary.

Example:
In patients undergoing renal (kidney) dialysis, is there any association between heart rate and blood pressure?

This can be seen visually by plotting a scatter graph of this data and drawing a trendline through it:

Warning: You cannot accurately assess whether a significant correlation between variables exists by visual examination alone!

Nonparametric Correlation: The Spearman Test

While desirable, it is not always possible to use a parametric test such as the Pearson method. Fortunately, there are also nonparametric correlation tests, the most frequently-used of which is the Spearman test. Unfortunately, there is no built-in Spearman test in MSExcel, so you're going to have to do some work!

Calculation of the Spearman rank order correlation coefficient (r_s) is used when the data consists of ordinal variables (i.e. variables with an ordered series where numbers indicate rank order only). Although this is a nonparametric statistic, it may be a better indicator than the Pearson coefficient of a non-linear relationship between two variables.

To perform the Spearman test, the data must first be converted into rank order. When converting to rank order, the smallest value on X becomes a rank of 1, etc, e.g:

e.g:

The equation for the Spearman calculation is:

where:
N is the number of pairs (XY)
D is the difference between each pair (X - Y)

There is no built-in formula for the Spearman calculation in MSExcel, but you can easily perform a Spearman calculation as follows. Start by converting the data to rank order if this has not already been done, then calculate the differences between the pairs (D), the squares of the differences (D²) and the sum of the squares (SD²):

Convert the Spearman formula (above) into an MSExcel formula:

r_s = 1-(6*SD^2/(N(N^2 -1)))

After calculating the value of r_s, this is compared with the critical value of r in deciding whether to accept or reject the null hypothesis. For a one-tailed test df = n-1 and for a two-tailed test (most usual) df=n-2.

Values of r range from +1 (perfect correlation), through 0 (no correlation), to -1 (perfect negative correlation). In general terms, correlation coefficients:

• up to 0.33 are considered to indicate weak relationships
• between 0.34 and 0.66 indicate medium strength relationships
• over 0.67 indicate strong relationships (i.e. likely to be regarded as significant, assuming that r_calc is greater than r_crit)

© MicrobiologyBytes 2007.