Procedure

In both studies, participants were mailed a packet of questionnaires to complete at home and bring with them to a laboratory session where they completed a consent form approved by the local human subjects review board and participated in a variety of tasks beyond the scope of the current analyses. The packet contained self-report measures of attachment insecurity, marital satisfaction, frequency of sexual intercourse, and the Big Five personality traits. To ensure that participants felt comfortable disclosing sensitive information, we (a) instructed them to complete their questionnaires independently of one another and not to discuss the questionnaires with one another, (b) included separate sealable envelopes in which they were to place their completed questionnaires so their partners were not able to easily view their responses, and (c) informed them that we would not share their responses with their partners. Every 6 to 8 months subsequent to the initial assessment, participants were again mailed a packet of questionnaires that contained the same measures of sexual frequency and marital satisfaction, as well as a measure of infidelity. We again employed the same tactics to ensure that participants felt comfortable reporting sensitive information. These follow-up procedures were used six times and spanned the first 3.5 years of marriage in Study 1 and the first 4.5 years of marriage in Study 2. Participants in Study 1 were paid $60 and participants in Study 2 were paid $80 for participating in the first phase of data collection; participants in both studies were paid $50 for participating in each of the six subsequent phases, except for the sixth phase in Study 2 for which participants were paid $80 because it resembled the first phase. Fifty-two percent of participants completed six or more waves and 36% of participants completed all waves. Given that people who were less likely to complete all waves of data collection had fewer opportunities to report an infidelity, and given that such people may also be higher or lower in attachment insecurity, we controlled for whether or not people completed all waves of measurement and examined whether or not that variable moderated any key effects. As we report, this variable did not moderate any of the effects.

Does own or partner attachment insecurity predict infidelity?

Given that we were examining the implications of both own and partner attachment insecurity for own perpetration of infidelity, and given that husbands and wives’ reports are not independent, we addressed our hypotheses using a 2-level actor–partner interdependence model (Kenny, Kashy, & Cook, 2006) using the Hierarchical Linear Modeling (HLM) 6.08 computer program to account for the non-independence of husbands and wives’ data. In the first level of the model, we regressed our code of own infidelity (0 = no, 1 = yes) onto own Baseline reports of attachment anxiety, own Baseline reports of attachment avoidance, partner Baseline reports of attachment anxiety, partner Baseline reports of attachment avoidance, own Baseline reports of the Big Five personality traits, the average of own reports of marital satisfaction across all waves, participant sex, and a dummy-code for attrition (0 = completed all waves, 1 = did not complete all waves). Given that the average of each couple’s sexual frequency across all waves and a dummy code for study were couple-level variables, we entered both as controls on the level-2 intercept. Because our hypotheses addressed the implications of absolute-levels of attachment insecurity, rather than variations of attachment insecurity within each couple, all variables were grand-centered around the sample mean. The level-2 intercept was allowed to vary randomly across couples. Because the dependent variable was binary, we specified a Bernoulli outcome distribution.

Results of this analysis are presented in Table 3. As can be seen, consistent with expectations, but in contrast to previous research on dating couples (DeWall et al., 2011), own attachment anxiety was positively associated with own infidelity, indicating that partners with higher levels of attachment anxiety were more likely to engage in an infidelity. Specifically, people who scored one point higher than the mean on the attachment anxiety subscale were more than twice as likely to perpetrate infidelity as people who scored at the mean on the scale. A test of the Attachment Anxiety X Participant Sex interaction indicated this effect did not differ across husbands and wives, B = −0.08, SE = 0.47, t(393) = −0.18, p = .86; a test of the Attachment Anxiety X Study interaction indicated it did not differ across the two studies, B = −0.06, SE = 0.37, t(393) = −0.12, p = .87; and a test of the Attachment Anxiety X Attrition interaction indicated it did not differ across the attrition dummy-code, B = 0.34, SE = 0.42, t(393) = 0.82, p = .41. Also, this effect remained significant when partner attachment anxiety and avoidance were not controlled, B = 0.54, SE = 0.20, t(399) = 2.74, p < .01.

In contrast to expectations, however, and also in contrast to what has been found in research on dating relationships (DeWall et al., 2011), own attachment avoidance was unrelated to own infidelity. This null effect did not differ across husbands and wives, B = −0.46, SE = 0.40, t(393) = −1.14, p = .25, the two studies, B = 0.62, SE = 0.44, t(393) = 1.40, p = .16, or the attrition dummy-code, B = 0.47, SE = 0.53, t(393) = 0.89, p = .37, and remained non-significant when partner attachment anxiety and avoidance were not controlled, B = 0.03, SE = 0.24, t(399) = 0.11, p = .91. Notably, low power cannot explain why attachment avoidance was not positively associated with own infidelity because the direction of the non-significant effect was negative.

Consistent with expectations, partner’s attachment anxiety was positively associated with own infidelity, indicating that individuals with partners who were high in attachment anxiety were more likely to engage in infidelity. Specifically, people who had partners who scored one point higher than the mean on the attachment anxiety subscale were more than one and a half times as likely to perpetrate infidelity as people who had partners that scored at the mean on the scale. This effect did not differ across husbands and wives, B = 0.19, SE = 0.58, t(393) = 0.33, p = .75, the two studies, B = −0.02, SE = 0.40, t(393) = −0.05, p = .96, or the attrition dummy-code, B = −0.48, SE = 0.60, t(393) = −0.80, p = .42.

In contrast to expectations, partner’s attachment avoidance was negatively associated with own infidelity, indicating that spouses with partners who were high in attachment avoidance were less likely to engage in infidelity. Specifically, people who had partners who scored one point higher than the mean on the attachment avoidance subscale were .45 times less likely to perpetrate infidelity as people who had partners that scored at the mean on the scale. This effect did not differ across husbands and wives, B = .57, SE = 0.76, t(393) = 0.75, p = .46 or the attrition dummy-code, B = −0.98, SE = 0.82, t(393) = −1.19, p = .24, but did differ across the two studies, B = −2.00, SE = 0.57, t(393) = −3.50, p < .01. Specifically, partner attachment avoidance was negatively associated with infidelity in Study 1, B = −2.26, SE = 0.44, t(393) = −5.09, p < .01, but not in Study 2, B = −.26, SE= 0.39, t(393) = −0.67, p = .50.

Theoretical and Practical Implications

These findings have important theoretical and practical implications. Regarding theory, they provide a more complete picture of the role of attachment anxiety in predicting marital infidelity. For example, consistent with dyadic models of personality (Shoda et al., 2002; Zayas et al., 2002), these findings demonstrate that the implications of one’s own attachment anxiety for infidelity appear to depend on the partner’s level of attachment anxiety, such that attachment anxiety in either member of the couple increases the likelihood that either spouse will perpetrate an infidelity. It may be that both own and partner attachment anxiety predict infidelity for the same reason, e.g., perhaps anxiety in either partner simply provides enough of a threat to intimacy to motivate spouses to seek out alternative partners. Alternatively, own and partner attachment anxiety may lead to infidelity for different reasons. Future research may benefit by identifying the mechanisms responsible for this interactive association.

Additionally, these findings raise important questions regarding the role of attachment avoidance in predicting marital infidelity. In contrast to research on dating couples in which own attachment avoidance was positively associated with infidelity, own attachment avoidance was unassociated with infidelity in these two studies of marriage. Further, partner attachment avoidance was negatively associated with infidelity in Study 1. Given that both findings were unexpected, and given that the negative association between partner attachment and infidelity did not replicate in Study 2, future research may benefit by attempting to replicate and explain both associations. In line with recent research by Lemay and Dudley (2011), partner avoidant attachment may be negatively associated with infidelity because spouses with an avoidantly-attached partner may be particularly careful not to disrupt their relationships.

Finally, these findings suggest that it may be important to study married spouses, rather than dating partners, to best understand the implications of various psychological traits and processes for marital relationships. As noted previously, one important difference between dating and marital relationships is that married individuals demonstrate higher levels of commitment, and existing research demonstrates that commitment moderates the associations between interpersonal processes (Amodio & Showers, 2005; Frye, McNulty, & Karney, 2008; Miller & Maner, 2010). For example, Miller and Maner (2010) reported that whereas cues to a woman’s fertility were positively associated with ratings of her attractiveness among relatively uncommitted men, those same cues were negatively associated with attractiveness among relatively committed men. Commitment may similarly moderate the effects of attachment insecurity on infidelity. Indeed, whereas DeWall and colleagues (2011) demonstrated that people in dating relationships, who tend to be less committed on average, perpetrate infidelities to avoid intimacy, Allen and Baucom (2004) demonstrated that married individuals, who tend to be more committed on average, perpetrate infidelities to increase intimacy. Accordingly, it makes sense that infidelity in a dating relationship is more likely among those high in attachment avoidance whereas infidelity in marital relationships is more likely among those high in attachment anxiety. In other words, not only may commitment mediate the effects of attachment insecurity on infidelity, as demonstrated by DeWall and colleagues (2011), commitment may moderate the effects of attachment on infidelity.

Regarding practice, the current findings suggest several avenues for improving the effectiveness of interventions to reduce infidelity. First, these results highlight the potential benefits of assuaging intimates’ attachment-related concerns. Indeed, interventions such as attachment-based family therapy (Shpigel, Diamond, & Diamond, 2012) and attachment-focused group intervention (Kilmann et al., 1999) have been effective at reducing attachment anxiety and thus may help prevent infidelity among anxiously-attached intimates. Second, given that partners’ attachment insecurity was associated with engagement in infidelity, practitioners may benefit from teaching their clients to be responsive to their partner’s attachment-related concerns. Indeed, intimates report reduced attachment insecurity when they are with responsive partners than when they are with unresponsive partners (e.g., Whiffen, 2005; see also Little, McNulty, & Russell, 2010) and thus being responsive may help prevent infidelity in these relationships.

Strengths, Limitations, and Directions for Future Research

Our confidence in these results is enhanced by several strengths of the research. First, with the exception of the negative association between partner attachment avoidance and infidelity, the effects reported here did not differ across two independent studies, providing confidence that they were not spurious due to the low number of infidelities in each study. Second, whereas other studies have collapsed across married and dating individuals (e.g., Allen & Baucom, 2004; DeWall et al., 2011), potentially obscuring the different implications of attachment infidelity in each type of relationship, our samples consisted of only married couples. Third, helping to minimize the problems associated with retrospective reports, both studies used a prospective design in which spouses first reported on attachment and then reported on the perpetration of infidelity. Finally, both studies controlled for several potential confounds— the frequency of sexual intercourse in the marriage, marital satisfaction, and personality.

Nevertheless, several factors limit the interpretations and generalizations of these results until they can be replicated and extended in future research. First, although these studies were longitudinal and controlled several individual difference variables, the results are nevertheless correlational and thus causal conclusions should be drawn with caution. Second, given that our samples consisted of newlywed couples married less than five years, caution should be used when generalizing these findings to lengthier marriages. Given the effects of attachment insecurity on infidelity appear to differ across married and unmarried people, future research may benefit from examining whether they also differ across couples in marriages of different lengths as well. Third, the definition of infidelity was open to participants’ interpretation and the items assessing infidelity were worded differently for own compared to partner reports of infidelity. As such, our measure of infidelity may have added error variance that limited our ability to detect some effects, such as any effects of own avoidant attachment. Fourth, although the samples were diverse in some important ways (e.g., education obtained, income), they were homogenous in other regards (e.g., age, ethnicity, religion). Ethnicity, in particular, may play an important role given that previous findings indicate that African Americans and Hispanic Americans report greater marital infidelity (e.g., Allen et al., 2005; Amato & Rogers, 1997; Greeley, 1994). Future research may benefit by examining whether ethnicity, or any differences associated with it, moderates the results that emerged here.

Preparation for this article was supported by the National Science Foundation Graduate Research Fellowship under Grant No. DGE-I246794 awarded to V. Michelle Russell and the National Institute of Child Health and Development Grant RHD058314 awarded to James K. McNulty.