When the net effect of each of fourteen factors upon the profits for the year 1921 on 231 Warren County farms is measured by coefficients of net correlation, it is found that the most important factors, in order of importance, are the production per animal unit, efficiency in the use of man labor, value of the real estate per acre because it influences the deduction made for the use of land, crop yields, and the amount of pasture used to carry one animal unit. All other coefficients are too small to carry definite significance.

The results, 'while not as conclusive as were expected, confirm in general the value of hypotheses tentatively held in organizing the analysis and demonstrate that care must be exercised in analyzing farm management data in order to determine whether apparent results are in fact due to the causes to which they are imputed. With data collected when farming conditions are more stable than were conditions in 1921, much better results can be expected. When economic conditions are in such a turbulent state, there are many additional and unusual factors introduced which cause wide variations in profits. In the absence of these unusual influences under more normal conditions, it is much easier to select a group of variables which will account for most of the significant influences on profits on individual farms.

Much careful study is needed preliminary to launching upon the correlation problem to make sure that the variables included are the most important factors affecting the results studied and that they are in their simplest form. Each variable should afford an answer to a specific and definite question. It is very necessary to avoid including any phase of the same factor in more than one variable. The fact that relationships within economic data are seldom of a linear nature must be borne in mind in making a strict interpretation of results obtained by assuming straight-line relationships.

If the precautions are taken necessary to secure satisfactory results, the method of partial or net correlation offers a much needed statistical tool for singling out the effect of anyone of many concomitant variables.