factors which might account for the non-significant relationships, among them low
statistical power or inaccurate ratings. Given the number of aggregate coworker and focal
pairs (N=107), weak relationships are not likely to reach statistical significance.
Specifically, the power to detect a correlation of moderate magnitude significantly
different from 0 in the population at p< .05 (e.g., r= .20) with a sample number of
respondents equal to 107 is equal to approximately .55 (Cohen, Cohen, West, & Aiken,
2002). In other words, type I error (the rejection of the null hypothesis when it is false) is
9 times more important than type II error (the failure to reject the null hypothesis/detect
significant effects). For a desired a power of .90 under the same parameter
specifications, 258 subjects would be required.
Moreover, the ratings provided by coworkers may also be biased by social
desirability or inaccuracy in spite of the researcher’s attempts to minimize biases.
Specifically, only cases who reported being familiar with the behaviors of the focal
employee were considered, and the raters were assured of the confidentiality of their
feedback, as well as its developmental rather than evaluative purposes, which could be
expected to result in higher truthfulness and reliability of the ratings (Pollack & Pollack,
1996). However, the focal employees were asked to select their raters, which is likely to
have resulted in their selection of friends who may provide inflated ratings (Murphy &
The relationships between roles and supervisors’ ratings of behaviors were not
significant and, in fact, sometimes were opposite of their predicted direction. For
instance, innovative role had a negative albeit not significant relationship with supervisor
5 The importance of type I error relative to type II error may be judged by calculating the ratio of (1- power)/significance level (in this case, (1-.55)/.05.