Jump to: Page Content, Site Navigation, Site Search,
You are seeing this message because your web browser does not support basic web standards. Find out more about why this message is appearing and what you can do to make your experience on this site better.
Published 10 September 2008, doi:10.1136/bmj.a197
Cite this as: BMJ 2008;337:a197
John Fletcher, clinical epidemiologist
1 BMJ, London WC1H 9JR
jfletcher{at}bmj.com
How would you best assess the degree of linear association between two continuous variables?
d—Correlation measures the degree to which two variables are linearly related.
A scatter plot would display graphically the association between two continuous variables. The x axis displays values for one variable, and the y axis values for the other. Each point represents one paired measurement of each variable. This is a useful way to display the data and would show U shaped and other non-linear associations as well as linear associations.
However, a scatter plot does not measure an association. To put a number to it requires an assessment of correlation. Pearsons correlation coefficient or Spearmans rank correlation coefficient may be used to assess linear correlation. The correlation coefficient may take any value between 1 and –1: 0 represents no linear association, 1 represents a perfect straight line with y values increasing with increasing x values, –1 represents a perfect straight line with y values decreasing with increasing x values. Like other parametric techniques, Pearsons correlation coefficient is sensitive to outlying values and may generate misleading P values when the y axis values are not normally distributed. Spearmans rank correlation coefficient, like other non-parametric techniques, is less sensitive and more robust.
Odds ratios can be used to compare only dichotomous values. Odds express the probability of something happening divided by the probability of it not happening.
A bar chart is used to plot the frequency of a nominal variable (such as eye colour) or an ordinal variable (such as agreement with a survey question: "disagree strongly, disagree, agree, agree strongly").
Cite this as: BMJ 2008;337:a197
Read all Rapid Responses