the rank (ordinal) variables in the two populations defined by the nominal variable ( Siegel, pp. 116-127). Consider the statement: "Those companies starting part-time transfer a greater amount of technology (.036)". This means that all the companies in the sample are divided into two populations on the basis of the nominal variable, "start part time? yes or no". The average amount of technology transferred in the sample comprised of those firms that did start on a part-time basis is larger, at the .036 level of statistical significance.
The Kendall tau test, referred to in Table A-6, simply measures the degree to which two rank variables are correlated ( Siegel, pp. 213-223). If two rank variables are significantly positively related, when one variable increases (decreases) the other variable is also expected to increase (decrease). (The converse is true when the variables are found to be negatively correlated.) Note that these variables are not said to be linearly correlated, or even that a specific change in one variable predicts the magnitude of the change in the other. It simply says that they "move" together. (In calculating the levels of significance, the "tau not corrected for ties" is used consistently. As such, the level of significance stated is always conservative.)
The reader is cautioned against associating relationship with causality. Frequent mention is made that a strong relationship or correlation exists between two variables. This does not, however, imply that one is caused by the other. Occasionally this inference is drawn based on reasonable assumptions, but it is usually impossible to prove causality mathematically.
Finally, I wish to point out a fact of life in the science of statistics. All other things being equal, the strength of the statement that can be made concerning a relationship increases approximately as the square root of the sample size. While this is so for very legitimate mathematical reasons, it does not detract from the fact that due to relatively small samples in individual studies throughout the research program, the strength of the conclusions that can be stated is much less than would be the case were the samples larger.
Bank of Boston. MIT: Growing Businesses for the Future ( Boston: Economics Department, Bank of Boston, 1989).
S. Siegel. Non-parametric Statistics for the Behavioral Sciences ( New York: McGraw-Hill Book Company, 1956).