How to: Correlation

Prism can perform correlation analyses either from XY tables or Column tables. The analysis works a bit differently depending on which kind of table you analyze.

Correlation from XY tables

1. Create a data table

From the Welcome or New Table dialog, choose to create XY data table.

If you are just getting started, choose the sample data: Correlation.

If you are entering your own data, choose to enter a single Y value for each point (no replicates, no error values).

2. Enter data

It matters which variable you place in the X column. If you enter Y values for several data sets (column A, B and C), Prism will report correlation results for X vs. YA, for X vs. YB, and for X vs. YC, but not YA vs. YB etc.

3. Analysis choices

Click Analyze and choose Correlation regression from the list of XY analyses.

As explained below, choose whether you want to use a nonparametric test and whether you want to switch to one-tail P values. If not sure, do not choose nonparametric and choose a two-tail P value.

Prism will compute correlation for X vs. YA, for X vs. YB, and for X vs. YC. However, Prism will not report the correlation of YA with YB, or YA with YC, etc.

If the data table contains subcolumns, Prism analyzes only the mean values.

Correlation from Column tables

1. Create a data table

From the Welcome or New Table dialog, choose to create column data table.

If you are just getting started, choose the sample data: Correlation matrix

2. Enter data

It doesn't matter which variable goes in which column. Prism will compute the correlation of each variable (column) with every other variable.

3. Analysis choices

Click Analyze and choose Correlation regression from the list of Column analyses.

Prism gives you two choices:

•	Choose one column to be "X" and another to be "Y". Prism then presents the full correlation results and creates a new XY graph of those two variables.

•

Choose a correlation matrix. Prism computes the correlation coefficient for each pair of variables, and presents the results on two pages. One page presents the matrix of correlation coefficients and the other page presents the matrix of P values. Each P value is computed independently, with no correction for multiple comparisons. When you compute a correlation matrix, Prism does not create a graph.

Correlation choices

Nonparametric?

Prism offers two ways to compute correlation coefficients:

•	Pearson correlation calculations are based on the assumption that both X and Y values are sampled from populations that follow a Gaussian distribution, at least approximately. With large samples, this assumption is not too important.

•	Spearman correlation makes no assumption about the distribution of the values, as the calculations are based on ranks, not the actual values.

One- or two-tailed P values?

Prism can compute either a one-tailed or two-tailed P value. We suggest almost always choosing a two-tailed P value. You should only choose a one-tail P value when you have specified the anticipated sign of the correlation coefficient before collecting any data and are willing to attribute any correlation in the “wrong” direction to chance, no matter how striking that correlation is.