We categorize the associations between reading and violent and nonviolent offending as tentatively favorable to the differential etiology of violence hypothesis. In four studies examining the association between reading ability and violence, the findings were divided. As can be seen in Table 5.1, two studies had findings indicating an association between reading problems and violence (Andrew, 1979; Harmon-Jones et al., 1997), and two had findings that were ambiguous (Brownlie et al., 2004; Koda, 1999). It should be pointed out that the reason for ambiguity in those two studies was that the data indicated an inverse association between reading and violence for males, but the opposite for females. There was only one study with findings on nonviolent-only offending, and the findings for that indicate an inverse association for males but not females (Brownlie et al., 2004).
There were seven studies directly comparing violent to nonviolent offenders, and the evidence supports reading as a potential factor in the differential etiology of violence, though this conclusion is tentative. In all cases, we have to be concerned about sampling issues; it is unclear in most cases that any effort was made to match the subjects on important characteristics or use a random sample of violent and nonviolent offenders to reduce the chances that systematic differences, not due to differential etiology, existed between groups. The studies had remarkably similar sample sizes (n ~ 100). Four of the seven studies reported a PoC reflecting significantly lower reading scores among violent offenders than nonviolent offenders in direct tests (Bryant et al., 1984; Kennedy, 2006; Lewis et al., 1979; Marcus & Gray, 1998). Another found that violent offenders had lower reading scores, but the difference between them and nonviolent offenders was not statistically significant (Hill-Smith et al., 2002). Two studies, however, reported a PoC that was in the opposite direction to that predicted (Hollin & Wheeler, 1982; Tarter et al., 1983). In the Hollin and Wheeler study, the sample size was extremely small (n = 20), so selection problems must be considered a serious threat.