Measuring gender differences in partner violence: implications from research on other forms of violent and socially undesirable behavior

Sex Roles: A Journal of Research,  June, 2005  by Sherry L. Hamby

• E-mail
• Print
• Link

Demand Characteristics

Demand characteristics of surveys and even interviewer bias are another potential cause of over-reporting (Levine, 1976). Demand characteristics include, for example, the effect of response categories on frequency reports. They also include the investment of researchers in finding data consistent with their hypotheses. Archer (2000) found that female first authors reported greater gender differences than male first authors in his meta-analysis. The NVAWS (Tjaden & Thoennes, 2000) found a much lower rate of partner violence than either of the NFVS surveys (Straus & Gelles, 1990). Rather than promote what might be interpreted as a precipitous drop in partner violence, the NVAWS adopted a less severe definition of abuse and claimed a similar number of battered women--1.5 million in the United States--as was estimated in the first NFVS survey, which limited the definition of "wifebeating" to the severe violence items on the CTS (Straus & Gelles, 1990). There was no mention of the change in definition, although Tjaden and Thoennes did suggest that the original NFVS surveys were compromised by overreporting due to a "more leading" (Tjaden & Thoennes, 2000, p. 23) survey format. Such variations in interpretations of rates help to illustrate the investment people make in the numbers.

TRUE POSITIVES AND TRUE NEGATIVES

Given the contentious nature of some discussion about the nature and extent of partner violence, it is worth noting that existing partner violence measures have several strengths and correctly classify large numbers of individuals. In qualitative studies of partner violence (Hamby & Gray-Little, 1997; Hamby et al., 1996), well more than 90% of reports would be defined as violence by most observers. It is likely that the behavioral questions on the CTS2 and similar measures reduce underreporting and increase the rate of true positives. Many problems with false positives have been identified in samples of youth, who might be expected to struggle with these concepts more than adults. Changing the CTS2 response categories or administration mode (paper-and-pencil versus computer) has little effect on rates (Hamby, Sugarman, Boney-McCoy, & Straus, 2001). Measures of partner violence and other crimes consistently find some expected patterns--for example, severe assaults are reported less frequently than minor assaults across methodologies (Bureau of Justice Statistics, 2003; FBI, 2003; Straus & Gelles, 1990; Straus et al., 1996; Tjaden & Thoennes, 1998). Finally, of course, the voluminous literature on partner violence and its correlates is strong evidence that partner violence measures do accurately identify true violence.

WHERE DOES GENDER FIT INTO QUESTIONS ABOUT FALSE NEGATIVES AND FALSE POSITIVES?

There has been a tendency to accept reports by women or even only victimization reports by women as more valid than reports by men. This perception is probably influenced by societal gender roles. A large literature on self-disclosure from the communication field has generally indicated that women self-disclose somewhat more than men in conversation (Dindia, 2002). So perhaps observed gender patterns in partner violence reporting are due to women disclosing more incidents than men. Other hypotheses include that women are reporting acts that are primarily self-defense, whereas males are typically the primary aggressor (e.g., Dobash & Dobash, 1984). In measurement terms, these and similar hypotheses propose that partner violence questionnaires are differentially valid for men and women. Despite an extensive literature discussing possible differences in men's and women's reports of violence, there are surprisingly few empirical studies on this topic. In part this is because differential validity can be an elusive construct--simply calculating rates by gender provides no insight as to whether the patterns reflect the true characteristics of the group or differential validity. Group differences cannot be accurately assessed when differential validity is present, however.