Gordon R. Stavig

Group Size and Societal Conflict

A number of studies dealing with subunits and total organizations have confirmed the relationship between large group size and increase in conflict and disputes together with decrease in morale, cohesiveness, and consensus. The question arises: Are the theoretical implications, empirical findings, and hypotheses for subgroups and total organizations generalizable to larger social systems? The size-conflict hypothesis is examined in relationship to the population size of nations and the magnitude of intranation and internation conflict. The results of the study show that the propositions formed at one level of analysis can be employed successfully at another level.

Coefficients of Association Analogous to Pearson's r for nonparametric Statistics

The rz and rp coefficients of association are discussed. Both coefficients, like Pearsons r, are based on a z/z max framework. They yield coefficients directly comparable for all levels of measurement being based on an obtained/maximum departure from independence in z units interpretation. The r z coefficient can be applied to any nonparametric test statistic in which a normal approximation equation is appropriate. The rp coefficient is applicable to any nonparametric test statistic in which exact probabilities are known.

Alternatives to Goodman and Kruskal's Lambda

Lambda and kappa coefficients of nominal scale association are developed for research hypotheses that involve predictions of modality, agreement, or some theoretically specified configuration. The κλ coefficient is offered as an alternative to Goodman and Kruskals λ.

The Median Controversy

Disagreement exists in the literature concerning whether the median should be computed using the ungrouped or grouped equation. For the original unaltered data the median equation for ungrouped data satisfies the criterion that Σ|x — Mdn|min. The median equation for grouped data satisfies this criterion for the linearly interpolated underlying continuous distribution. An equation is given to estimate the underlying distribution. The so-called ungrouped equation can be applied to grouped data. The grouped equation is applied to discrete data for comparative purposes. If the data ate not rectangularly distributed in the class interval of interest, a logarithmic adjustment can be made.

Measures of Sampling Error Applied to Statistical Inference

The classical theory of errors is applied to sample statistics. Two measures of sampling error, |ē| and kσ̅ are presented that can be used in conjunction with conventional methods of statistical inference. The measures have two interpretations. The |ē| measures the amount of a sample result that is expected by chance. The kσ̅ estimates both the proportion of a sample result that is expected by chance and the proportional deviation that is expected between an observed sample statistic and its population parameter.