Rafael Wittek: Resource competition and violent conflict
and more than twice as high in this model than in the preceding one for land-acquisi-
tion. The interaction effects in both models are equally strong.
Statistical power of the models
Statistical power (or, the control of the ß-error) is a long neglected topic in anthropo-
logical research (Schweizer and Lang 1989). The statistical power of a test is defined as
1—f. fi is the beta-error, that is, the probability of rejecting the alternative hypothesis
though it is true. E.g. if the level of the f-error is considered to be 10%, the power of
the test is 1—8 = 90%. In other words: the probability that the hypothesis is true 1s
90%. The power of a log-linear model depends upon four parameters: sample size, de-
grees of freedom, the specified a-level and a critical effect size *w". *W" measures how
Strong the alternative hypothesis deviates from the null-hypothesis. A log-linear
model exactly fits, if w = 0. Cohen (1988: 224) distinguishes small (w = .10), medium
(w = .30) and large (w = .50) w values, and suggests w = .10 for strong tests. As this
would make necessary a sample size of over 1500 cases to attain 90% power at a. = 5%,
more liberal effect sizes have to be accepted when working with the SCCS.
Table 6. Power of six models for W —.30 and a — 596
Models N udi Power
Independent variables 135 2 85%
Landacquisition, Plunder 125 4 75%
Subjugation 125 5 70%
Tribute 125 6 70%
Xternal war 117 6 65%
Table 6 shows the power of the six logit-models for w = .30 and a = 5%, appro-
Ximately determined according to the power tables in Cohen (1988: 259). Thus, with
More than 75% the logit models for the acquisition of land and plunder yield more
Power than the other models for violent conflict. This power value comes close to 8096
With o = 5%, which Cohen suggests as a convention. It can be concluded, that at least
for these two dependent variables as well as for the interaction of the independent
Variables, the statistical results are robust enough to warrant the acceptance of the dis-
Covered models.
Discussion
Some comments are necessary with regard to the role of the variable “population den-
Suy in the respective models. First, the fact that plunder is conducted by low-density
ocleties with food shortages contradicts the assumption of the demographical expla-
ation, This result indicates that population density cannot be taken as a measure of