CAN WE TRUST OUR FINDINGS?
by Raoul Naroll
THE MAIN OBJECT of this pilot study has been to test the feasibility of a cross-historical survey of factors attending War Frequency and Territorial Gain. Our chief focus has been on military deterrence. Here we found no reason to believe that military preparations tended to make wars less likely. How confident can we be of this finding? In effect, what is the lower limit we can reasonably expect to find in the universe of the correlations between War Frequency (Months of War) on one hand, and our seven deterrence variables (Defensive Stance, Strength of Armed Forces, Mobility of Armed Forces, Quality of Armed Forces, Prestige of Armed Forces, Defensive Fortifications and Defensive Alliances) on the other hand?
For this purpose we can use the method of confidence intervals ( Blalock 1960:305-9). The user of this method states the risk of error he is willing to assume. Given such a level of confidence, the method designates the limits or boundaries wherein the true correlation can be expected to lie. The mathematics of the method assumes random sampling; when samples depart from randomness they must also reckon with the possibility that sampling bias alters these limits.
We are content for our present purpose to accept a risk of five percent. Table 24-1 shows the lower limits of the seven deterrence correlations. Column C shows these limits as computed from the present sample. But the length of the confidence interval between the observed correlation and the limits is a simple function of the sample size. Column D shows the lower limits at the 95% level of confidence which the correlations we observed would enjoy