Partial Kendall correlation test
This is a nonparametric test that can be used for analyses of correlation between two
(continuous, ordinal or binary) variables with elimination (control) of an effect of third
(continuous, ordinal or binary) confounding variable.
As far as we know, this test is not included into any commercial statistical package. The basic
module of SPSS as well as for example Statistica contains the procedure for computation
Kendall XY . We programmed an Excel sheet that computes a value and significance of XY|Z
using values XY, XZ, YZ computed with Kendall correlation tests implemented in any
standard statistical package.
The method of computing of a value XY| Z was described in (Siegel and Castelan, 1988). In
principle, the formula for computation of XY|Z is analogical to formula for computation of
partial Pearson’s r
XY|Z=(XY-XZ*YZ)/(1-(XZ)2)* (1-(XZ)2)
The significance of XY|Z is computed using asymptotic method described in (Sheskin 2003).
Main disadvantage of the present method (in comparison with partial Pearsons r ) is an
impossibility to control effects of more than one confounding variable. As far as we know, no
method of such control has been described.
The way of usage of Excel sheet is rather intuitive. It is just necessary to fill XY, XZ, YZ to
cells D9, D10 and D11 and number of observations (N) to the cell D7. The partial  is
automatically computed in the cell D13 and asymptotic two-tailed p value is computed in the
cell D20. Sample of data is shown in the column C.
Author of the sheet is Aleš A. Kuběna.
Please cite the paper
Kaňková S, Kodym P., Flegr J. 2011: Direct evidence of Toxoplasma-induced changes in
serum testosterone in mice. Experimental Parasitology, 128, 181-183.
References
Sheskin DJ (2003) Handbook of Parametric and Nonparametric Statistical Procedures, 3rd ed.
Chapman and Hall/CRC Press, Boca Raton, FL, USA, pp 1079-1090
Siegel S, Castelan JR NJ (1988) Non parametric statistics, ed. Mc Graw-Hill, New York, pp
254-262