Review of Economics of the Household

, Volume 10, Issue 3, pp 319–343

Reinterpreting the economics of extramarital affairs

Authors

    • School of Economics and FinanceUniversity of St Andrews
Article

DOI: 10.1007/s11150-012-9146-9

Cite this article as:
Smith, I. Rev Econ Household (2012) 10: 319. doi:10.1007/s11150-012-9146-9

Abstract

The empirical results for the economic variables presented by Fair (J Political Econ 86(1):45–61, 1978) in his seminal study of extramarital affairs are puzzling within his household allocation of time framework. In particular, the theory is unable to accommodate readily the opposite signs for occupation (positive) and education (negative), assuming the wage rate is directly correlated with both variables. This paper provides a new interpretation of Fair’s estimates that accounts for the unexpected education result in terms of the association between schooling and the discount factor applied to expected future sanctions for sexual cheating. Three data sets from the United States, Germany and the United Kingdom are investigated to check the robustness of the partial correlations between infidelity and economic incentives. Taken together, the results across different countries and infidelity measures are substantially in agreement, especially for men. In a novel contribution, this study distinguishes between one off encounters, and irregular and regular forms of infidelity and finds that these are differentially related to occupation and education, consistent with theoretical predictions.

Keywords

InfidelityOccupationEducationSexual behavior

JEL Classification

D13J12J29

1 Introduction

It is well recognized that sexual behavior responds to incentives. The use of contraception, abortion choices, and the timing of sexual initiation and participation decisions are not immune to the systematic influence of economic variables (Posner 1992; Akerlof et al. 1996). Infidelity, however, is a dimension of sexual behavior that somewhat resists a persuasive economic explanation and has only attracted occasional interest among economists.

The pioneering contribution is that of Fair (1978) who modeled extramarital affairs within a household time allocation framework. Assuming an affair is a normal commodity, his principal prediction is that the effect of wages on marital cheating is ambiguous, depending on the relative magnitudes of the income and substitution effects of a change in the price of time. In the absence of wage rate data, Fair’s empirical investigation proxied time costs using variables on occupation and education, assuming both are positively correlated with the missing wage. The estimates produced conflicting signs for these variables, positive for occupation and negative for education, which could not be readily explained by the time allocation model except in terms of their possible inadequacies as proxies. Subsequent studies and replications have not attempted to resolve the puzzle. Indeed, the anomaly is reinforced by the fact that the negative partial correlation between infidelity and schooling also conflicts with the prediction derived from the literature in evolutionary biology (Elmslie and Tebaldi 2008).

In the econometric analysis that follows, we show that Fair’s results are not simply an artifact of his selected magazine samples or the way his variables are measured. Similar results for occupation and schooling emerge using German, British and American nationally representative survey data with alternative variable definitions and specifications. The challenge is to explain these empirical regularities. To do so, this paper pursues one of the alternative modeling strategies suggested by Fair, namely, treating infidelity as an illicit act subject to sanctions if detected, a theoretical approach also adopted by Liu (2008) using a dynamic optimization model.1 This strategy is appealing given that sexual cheating is intentional and hidden, causes harm to third parties and may incur punishment costs if detected by a wronged spouse. Historically extramarital affairs were treated as a criminal offence in many jurisdictions (Rasmusen 2002). But even in the absence of public prosecution and legal sanction in most countries today, adultery is still widely considered an undesirable act that incurs social opprobrium. In the British NATSAL survey used below of a relatively young cohort aged 18–44, 85 % of respondents expressed the opinion that extramarital sex is always or mostly wrong. Sexual exclusivity within committed partnerships then remains the dominant social expectation even among younger sexually permissive generations. Nevertheless, despite the strength of the fidelity social norm, there is in practice significant deviation from this ideal. In the eleven waves of the US General Social Survey that ask an infidelity question, 23.5 % of male respondents and 14.6 % of women report having engaged in extramarital sex during their lifetime.

Modeling infidelity as an illegitimate activity allows occupation and education to have differently signed effects. The key innovation is to treat occupation as an indicator of an individual’s quality as a sexual partner, providing a measure of the benefits from an extramarital liaison. While not exhaustive of the factors that contribute to sexual desirability, it is clear that occupation is positively correlated with various socioeconomic considerations that matter in the (extradyadic) mating market. By contrast, the association of education with infidelity is ambiguous. While schooling augments social and communication skills that raise the gains from an affair, it also increases the weight placed on the expected costs of marital cheating. Insofar as education reduces the discounting of future costs this provides one channel through which schooling diminishes the demand for infidelity all else equal.

A further novel contribution of the paper disaggregates extramarital relationships into different types according to their degree of regularity. This allows occupation and schooling to be differentially related to the demand for casual and repeated liaisons. In particular, we find that quality has a stronger association with more regular extramarital relations while, holding quality fixed, the negative effect of schooling is greater for casual encounters.

Infidelity matters. It is not a purely private affair without public consequences. Insofar as marriages would have remained stable in the absence of cheating, this behavior is of social concern given the potentially adverse economic and nonpecuniary impacts of separation and household dissolution on children and other family members. Indeed, extramarital affairs play a leading role in the destabilization of marriages and non-marital partnerships with non-negligible harm to third parties, especially children. In the British NATSAL 2000 data used below, 37 % of respondents who had experienced the failure of their first live-in partnership identified infidelity as a contributing factor. However, its occurrence may also be a symptom as well as a cause of marital disintegration. As Cox (2008) notes, in some cases infidelity is a strategy for establishing a new match given that the existing marital relationship has failed and will dissolve independently of whether an affair occurs, though the timing may well be affected. In that instance, infidelity may not itself be economically significant in generating external costs.

Nevertheless, in most countries adultery remains the most frequently cited fault ground for the dissolution of marital relationships. Even for those couples whose relationship survives the disclosure of an affair, the gains from the marriage may be diminished by this negative shock given that infidelity is a cause as well as a consequence of marital dissatisfaction. In addition, extramarital sex has potential health impacts as one of the main sources of the transmission of sexual diseases among heterosexuals, especially in developing countries. Finally it should be noted that undetected female adultery is a form of moral hazard for a male spouse since it potentially leads to cuckoldry, the involuntary investment of resources in a genetically unrelated child.

2 Related literature

The motivation for the model and the estimation that follows is the puzzle generated by Fair’s empirical results. He used Tobit models to investigate two admittedly less than ideal datasets collected through magazine surveys conducted by Psychology Today (PT) and the Red Book (RB), the latter confined exclusively to female respondents. Fair restricted his sample to those employed and in their first marriage. Although in both cases, the occupation variable was positively signed, the results turned out to be mixed in that neither schooling nor occupation was statistically significant for the much smaller PT sample (601 observations). Moreover, while both economic variables were statistically significant for the RB sample, education was negatively signed. This striking result was contrary to prior expectations for the schooling variable and Fair did not attempt a theoretical interpretation of this outcome. Based on the estimates for the occupation variable, Fair tentatively concluded that the positive income effect of a change in the wage rate appears to dominate in the allocation of time to extramarital affairs. But given data constraints, he could not be confident in his chief finding that women in higher occupational groups are more likely to cheat.

Fair’s data are publicly available and have been subsequently exploited by econometricians as an application to illustrate new estimation procedures or diagnostic tests but with little attention paid to the substantive behavioral concerns (Chernozhukov and Hong 2002; Li and Racine 2004; Pagan and Vella 1989; Wang 1997; Wells 2003; Yen 1999). Only Chernozhukov and Hong (2002) shed additional light on Fair’s results for occupation and education. They apply quantile regression methods to the RB sample, and report that both the negative education and positive occupation coefficient estimates are strongest at the higher quantiles, that is, for women of higher occupational status and schooling levels.

Since Fair, there has been very little further modeling of extramarital affairs. Notable exceptions are the recent contributions of Cox (2008) and Elmslie and Tebaldi (2008) who draw heavily upon the literature in evolutionary psychology and biology to explain gender differences in mating preferences and behavior. A central idea is that greater biological investment costs in children for women generate systematic differences in optimal infidelity strategies across the sexes. Women care more about securing investment in child quality and search for relatively wealthy males able and willing to provide resources. Men value fertility and are attracted to healthier, younger females. As the sex with the lower necessary biological investment costs, male sexual opportunities are constrained by female preferences. The authors present some empirical estimates related to economic variables with mixed results. Consistent with evolutionary argument that men trade resources for sexual access, Cox reports that poorer women are more likely to be unfaithful, reflecting their greater gains from securing additional male investment. On the other hand, Elmslie and Tebaldi estimate a positive social class effect on infidelity for women. Neither study finds a statistically significant wealth effect on male cheating. Importantly, Elmslie and Tebaldi also report a negative association between education and extramarital affairs. This holds for both sexes but for women the magnitude of the estimated coefficient is generally half the size of that for men and not well determined. The negative sign contradicts the prediction from the evolutionary model. In addition to own schooling effects, the main empirical result from Elmslie and Tebaldi is a negative correlation of spousal education and women’s propensity to cheat. The interpretation is that the higher a husband’s education level, the more a wife expects to lose from divorce given that schooling proxies male resources.

There is some scanty consideration of extramarital affairs in law and economics research. Liu (2008) argues that the expected penalty for adultery in fault based divorce settlements should increase with the quantity of extramarital affairs in order to achieve the socially efficient level of cheating. Allen and Brinig (1998) use a small number of American divorce records from Fairfax county that cite adultery grounds to show that systematic mismatches in spousal sex drives over the lifecycle can explain age patterns in extramarital affairs. Finally, in a recent contribution, Wax (2011) discusses informally the role of time preferences in explaining the patterns of marital and sexual decision-making across different socio-demographic groups, including the influence of patience on infidelity choices. Her important insight is developed more formally in the model that follows.

3 An infidelity model

Consider a simple two period model in which in at date 1 a married risk neutral individual chooses a quantity of infidelity b and generates a total net return r(b, q, e, v) (individual and time subscripts suppressed), rb > 0 and rbb < 0. No distinction is made at this stage between intensive and extensive margins in the infidelity decision. The payoff to affairs in the first period depends upon a person’s quality q where rq > 0 and rqq < 0. The intuition is that a higher quality person can attract a more desirable paramour given that quality summarizes a vector of characteristics including physical attributes, personality factors and economic resources. An individual’s education e also directly affects the net return insofar as it enhances social and communication skills and decision making in romantic relationships. In addition, schooling may have an indirect effect by augmenting quality through, for example, its positive impact on health, qe >0. The return also depends on an individual’s propensity to cheat and opportunities for infidelity, represented by v, and affected, for example, by moral and religious constraints and search costs. This portmanteau variable captures too the expected costs of unintended pregnancy and sexual infections.

As the social norm of sexual exclusivity in committed romantic relationships remains strong, affairs are generally considered undesirable by the betrayed partner and the adulterer seeks to conceal them. In the second period, infidelity is detected by the spouse with probability p(b), pb > 0, where, for simplicity, the spousal choice of monitoring effort is not modeled. In the event of discovery, the philanderer incurs a cost c > 0 imposed by the spouse either through dissolution of the relationship, non-cooperation, retaliation, a renegotiation of the terms of the marriage, or reputational damage. All else equal, the cost is assumed to be lower for a higher quality person cq < 0 who has more outside options in the singles market. The cost is discounted by a factor β(e)ϵ[0, 1]. Following the arguments of Becker and Mulligan (1997) schooling, e, is permitted to affect time preferences βe > 0, βee < 0. The more highly educated place greater weight on future (second period) expected costs as they are better able to imagine the distant outcomes of current (first period) choices and mentally simulate alternative scenarios. The magnitude of the spousal sanction also depends on the extent of the adultery with greater infidelity punished more heavily, cb > 0, cbb < 0.

The individual’s problem is to choose the quantity of infidelity \( b^{*} \) that maximizes expected utility given by:
$$ u = r(b, q,e,v) - p(b)\beta (e)c(q,b) $$
(1)
The first order condition for an interior solution is:
$$ \frac{\partial r(b,q,e,v)}{\partial b} = p^{\prime } (b)\beta (e)c(q,b) + p(b)\beta (e)\frac{\partial c}{\partial b} \ge 0 $$
(2)
Using this condition, we derive the comparative static effects of the exogenous variables q, e and v on b*. With respect to quality, Eq. (3) suggests two main channels through which infidelity decisions can be affected. First, given the marginal return to infidelity is positively related to quality, rbq > 0, then higher quality individuals are more productive in their extramarital relationships and consume more of them. The second channel relates to expected costs. Additional infidelities increase the risk of detection but the expected marginal cost of discovery is lower for higher quality individuals, cbq < 0. They lose less from exposure because spousal sanctions and their consequences are less severe in the event of detection of a clandestine relationship. Higher quality individuals then require fewer benefits to induce them to commit a given level of adultery than those of lower quality all else equal. The marginal effect of quality on cheating derived in Eq. (3) is positive given assumptions on the signs of the derivatives and a positive denominator.
$$ \frac{{db^{*} }}{dq} = \frac{{r_{bq} - p_{b} \beta (e)c_{q} - p(b)\beta (e)c_{bq} }}{{\beta (e)2p_{b} c_{b} + \beta (e)p(b)c_{bb} - r_{bb} + p_{bb} \beta (e)c(q,b)}} > 0 $$
(3)
Education is related to the infidelity decision through three primary mechanisms of interest described in Eq. (4). First, schooling is likely to augment directly the net benefits in the first period of an extramarital relationship through its impact on relational skills, rbe > 0. Second, education has a positive indirect effect on infidelity through quality qe > 0 with the consequences described earlier. Third, the positive association of schooling and patience increases the weight placed on the expected second period costs of extramarital liaisons, given by the final term in the numerator of Eq. (4).2 As these education channels have opposing effects on the optimal level of adultery, the sign of overall influence of the schooling variable is ambiguous.
$$ \frac{{db^{*} }}{de} = \frac{{r_{be} + q_{e} r_{bq} - p_{b} \beta (e)c_{q} q_{e} - p(b)\beta (e)q_{e} c_{bq} - \beta_{e} (p_{b} c(q,b) + p(b)c_{b} )}}{{\beta (e)2p_{b} c_{b} + \beta (e)p(b)c_{bb} - r_{bb} + p_{bb} \beta (e)c(q,b)}} \frac{ < }{ > } 0 $$
(4)
The model could be extended further to make the probability of detection a decreasing function of education as in the crime research of Friehe (2008) and Lochner and Moretti (2004). This would provide another mechanism through which schooling might be expected to increase the level of sexual cheating.

Given the paucity of spousal information available in data sets that measure infidelity, potentially important considerations such as spouse quality and marital gains are not explicitly modeled. However, these variables are likely to be related to both education and the infidelity decision. In particular, empirical evidence suggests that gains from marriage are increasing in schooling and this is reflected in lower divorce risks, especially given strong assortative mating on education (Gustafsson and Worku 2005; Bruze et al. 2012). Greater marital gains for educated couples can be accounted for by both market and nonmarket (household production) returns to human capital. Additionally, the more educated the partners the greater the cross spousal productivity effects in the household and labor market (Grossbard-Shechtman 1993). Educated spouses then have stronger incentives to invest in the marriage and this augments the marital surplus.3 Given they have more benefits to lose, educated partners also face stronger incentives to avoid risking the stability of the marriage through an affair. The implication is that this mechanism provides a credible alternative (or complementary) explanation to discounting for an inverse relationship between schooling and infidelity. With respect to the model, one simple approach to capture the idea is to modify the cost function to include education, which is assumed to be positively assortatively matched across couples, as an argument, c(q, b, e). This would add the term −β(e)(pbce + p(b)cbe) to the numerator of the comparative static for education in Eq. (4). Assuming that additional schooling increases the cost of infidelity, the term is negatively signed.

According to the comparative statics, the effect of the opportunity and preference factors v on infidelity depends on the sign of rbv. This is positive for opportunity variables since the marginal gain from infidelity is increasing in access to extramarital sexual partners. Infidelity will be negatively related to those preference factors that raise the costs of infidelity such as strong moral or religious convictions.

The model allows for possible gender differences in infidelity choices. Insofar as women on average face higher infidelity costs than men due to pregnancy risks or greater religiosity or other moral constraints or less favorable remarriage market opportunities, they will have a lower \( b^{*} \). As well as gender differences in the scale of demand for infidelity there may also be gender heterogeneity in the parameters describing an infidelity function. For example, gender differences in the effect of education on patience will affect the size of the marginal impact of schooling on cheating. In addition, it is likely that the effect of the economic components of quality on extramarital affairs are more important for men given that women are the choosier sex and value economic credentials relatively more highly in their sexual partner (Schmidt 2005).

4 Data and methods

We investigate data from three nationally representative cross section surveys. These are the American General Social Survey (1991–2010), the first wave of the German Pairfam survey (2008/2009) and the British Natsal survey (1999/2001). Due to the absence of data on frequency of cheating in these surveys, the empirical analysis focuses initially on the extensive margin of the infidelity participation decision though later we exploit information on different cheating intensities. Assuming that the probability of infidelity is determined by the same theoretical mechanisms that affect its quantity, the model suggests that cheating is influenced primarily by individual quality, education, opportunity and preference factors.

To permit the results across surveys to be as comparable as possible, our proxy for quality exploits an internationally standardized socioeconomic index of occupational status designed to capture the earning capacity associated with each occupation. This is described in the data appendix and briefly in the subsections reporting the results for each national data set. To permit individuals to have completed their education and to have chosen an occupation, the samples are restricted to respondents aged at least 25, though in practice the results are not sensitive to this choice of age threshold. The education variable is categorical, coded according to the stage of education completed. The preference and opportunity variables are primarily demographic, comprising age, number of children, religiosity, residence in an urban area, age at marriage and whether legally married. Additionally two of the samples measure nights spent away from home for work related reasons. For the British data, we are also able to construct the gender mix by industrial sector as a crude measure of access to extramarital partners.

Tables 6, 7, 8 collected in the Appendix present descriptive statistics on all the variables across the three data sets (excluding Fair’s sample). For infidelity, the means reflect the standard finding that most people are faithful, but men consistently report more cheating than women. Responses to questions on infidelity are naturally vulnerable to both recall and self-presentation biases. Moreover, given the personally confidential and potentially embarrassing nature of survey questions on sexual behavior, there are clearly issues of willingness to respond, accurate disclosure and data quality. In an attempt to minimize the measurement problems, the surveys facilitated disclosure of sensitive information by the private input of answers using the computer assisted self interview method. Nevertheless there remain methodological concerns (Blow and Hartnett 2005).

We estimate a simple Probit model for infidelity by maximum likelihood. The equations are estimated separately by gender to allow for sex differences in behavior. All equations use a robust variance estimator and sampling probability weights. Given the data are cross section samples, it is clearly not possible to infer causal effects and the results should be interpreted as partial correlations. Note also that as homosexual behavior patterns typically differ significantly from those of heterosexuals, homosexual partnerships are excluded throughout.

4.1 American samples

Fair’s Redbook sample is restricted to employed women in their first marriage in 1974. The dependent variable is the average number of extramarital sexual liaisons per year of marriage and the specification is estimated using the Tobit estimator. For the occupation variable, Fair used the Hollingshead (1957) occupational classification, divided into six categories in ascending order of social position with the highest corresponding to professional with an advanced degree. Education is defined in terms of the number of years associated with the schooling level attained, so High School is codified as 12 years and so on. For comparability with other samples, education is recoded here using a set of binary dummy variables for level of education completed (less than High School, some post High School, and College) where High School is the omitted category.

Column (1) of Table 1 reports Tobit average marginal effect estimates with robust standard errors for the occupation and education variables using Fair’s (1978) Red book sample with p-values in parentheses. Controls comprise age, children and degree of religiosity. Fair also included marital duration and happiness but as these two variables are possibly endogenous to the decision to have an affair, reduced form estimates are reported that exclude them though with little qualitative change in terms of the sign and significance of the remaining covariates.4
Table 1

Infidelity equation marginal effect estimates (p-values in parentheses), fair sample

 

(1)

(2)

(3)

Tobit

OLS

Probit

Occupation

0.25 (0.00)

0.03 (0.00)

0.03 (0.00)

Education

 <Secondary

0.44 (0.31)

0.11 (0.18)

0.11 (0.19)

 >Secondary

0.13 (0.33)

0.01 (0.46)

0.01 (0.45)

 Tertiary

−0.73 (0.00)

−0.11 (0.00)

−0.11 (0.00)

Age

−0.02 (0.05)

0.002 (0.31)

0.002 (0.28)

Kids

0.05 (0.26)

0.03 (0.00)

0.03 (0.00)

Religiosity

−0.73 (0.00)

−0.09 (0.00)

−0.09 (0.00)

N

4,427

4,427

4,427

p-values are in parentheses. All respondents are women. Column (1) uses the Red Book sample and Tobit estimator investigated by Fair (1978). All equations include a constant term and controls for age, number of children and religion. The binary education variables are categorised according to attainment, namely secondary schooling not completed (<Secondary), some post-secondary education (>Secondary) and tertiary. The omitted category is completed secondary education (i.e. graduated from High School). Columns (2) and (3) present linear probability and Probit marginal effects respectively with a binary infidelity variable. The marginal effects are measured at the mean values of the continuous explanatory variables and for the discrete change in categorical dummies from 0 to 1

For comparison with samples that only collect data on participation in infidelity rather than its frequency, the dependent variable is reformulated as a binary variable set to one for those who admitted any extramarital affair and zero otherwise. OLS and Probit average marginal effect estimates for the participation equation are presented in columns (2) and (3) respectively and are virtually identical to each other. The central point is that across all three formulated models the results continue to show the positive and statistically significant relationship between the risk of infidelity and occupation reported by Fair. With respect to schooling, it is notably tertiary level education that generates the negative correlation with extramarital affairs.

Fair’s sample is a self-selected group of female respondents to a magazine survey. To investigate whether the results hold in a nationally representative survey, the specification is estimated using pooled cross-section data from the 11 waves of the US General Social Survey (1991–2010) that include a question on whether the respondent has ever participated in an extramarital affair during their lifetime. Descriptive statistics for the variables for those ever married are listed in Table 6 and include comparable measures to those used by Fair. The differences are the use of a quadratic in age rather than Fair’s grouped age categories and adoption of a continuous measure of occupational status, the International Socio Economic Index (ISEI) constructed by Ganzeboom, De Graaf and Treiman (1992). For comparison with the European samples, this index is preferred to that developed for the GSS by Nakao and Treas (1992), though the results are not sensitive to this selection. The ISEI is designed to capture a respondent’s resources in terms of earnings capacity, and reflects the wages and education typically associated with each occupation, where occupation is defined by the codes from the ILO 1988 International Standard Classification of Occupations. The latest version (ISEI-08: Ganzeboom and Treiman 2011) estimates the index using pooled waves from the International Social Survey Program data over the period 2002–2007 for almost 200,000 male and female respondents in 42 countries. The revised index is preferred to earlier versions which only used data on men to score the socioeconomic status of each occupation. This has the drawback that the characteristics of male workers may not be representative in highly feminized occupations, likely biasing the occupational score upward. Note that the ISEI also differs from occupational prestige measures which are based on a subjective evaluation of occupational attractiveness. Compared to Fair’s specification, the GSS also permits inclusion of a richer set of control variables, including race indicators, spousal education, cohort and a measure of the population size of the area in which the respondent resides. No restriction is imposed on employment status.

Columns (1) and (2) of Table 2 report the Probit average marginal effect estimates for respectively men and women. For comparability to Fair’s model, the sample is initially constrained to respondents in their first marriages. This excludes those in informal cohabiting relationships for whom the infidelity question was not asked unless they had been previously married. The results are generally consistent with Fair’s findings. The association between schooling and infidelity is once again negative with the largest effects at the tertiary level. The magnitude of the partial correlation for college education is almost four times as large for men (−0.063) as for women (−0.016) but only well determined in the former case. If spouses have more schooling this also reduces sexual cheating but the estimates are not statistically significant. In order to test for the effects of complementarities in schooling levels across spouses on the infidelity decision, an interaction dummy is specified for those respondents in which both spouses have tertiary education. The interaction term is designed to capture the extent to which the negative association of college education and marital unfaithfulness reflects the additional gains from marriage for a highly educated couple. However, the sign of the coefficient estimate is only negative in the case of women and not statistically significant. The result is the same if the interaction dummy is defined more widely to be set to one for couples in which both partners had completed at least junior college.
Table 2

Infidelity equations: probit average marginal effect estimates, United States, GSS sample

 

First marriage

Ever married

(1)

(2)

(3)

(4)

Male

Female

Male

Female

Occupation

0.0008 (0.04)

0.0000 (0.99)

0.0008 (0.01)

−0.0001 (0.74)

Education

    

 <Secondary

−0.009 (0.74)

−0.008 (0.65)

−0.035 (0.07)

−0.002 (0.88)

 >Secondary

−0.010 (0.71)

−0.003 (0.84)

−0.048 (0.03)

−0.019 (0.23)

 Tertiary

−0.063 (0.00)

−0.016 (0.28)

−0.078 (0.00)

−0.029 (0.01)

Age

0.008 (0.01)

0.007 (0.01)

0.016 (0.00)

0.015 (0.00)

Age squared

−0.000 (0.01)

−0.000 (0.00)

−0.000 (0.00)

−0.000 (0.00)

Kids

0.009 (0.08)

0.005 (0.11)

0.011 (0.00)

0.001 (0.68)

Religiosity

−0.013 (0.04)

−0.019 (0.00)

−0.028 (0.00)

−0.025 (0.00)

Cohort

−0.002 (0.17)

0.000 (0.70)

−0.002 (0.08)

0.000 (0.64)

Black

0.140 (0.00)

0.016 (0.36)

0.120 (0.00)

0.047 (0.00)

Other non-white

0.058 (0.06)

−0.029 (0.03)

0.028 (0.30)

−0.030 (0.07)

Urban

0.002 (0.54)

0.003 (0.18)

0.007 (0.02)

0.002 (0.33)

Spousal education

 <Secondary

0.025 (0.37)

0.006 (0.72)

  

 >Secondary

0.036 (0.18)

−0.018 (0.26)

  

 Tertiary

−0.021 (0.38)

−0.004 (0.79)

  

 Both tertiary

0.030 (0.37)

−0.011 (0.62)

  

N

3,192

3,506

5,914

6,964

The binary education variables are categorised according to attainment, namely secondary schooling not completed (<Secondary), some post-secondary education (>Secondary) and tertiary. The omitted category is completed secondary education (i.e. graduated from High School. Both Tertiary is a binary dummy variable set to one when both partners are college graduates. For the racial indicators, white is the omitted category. All the variables are defined in the data appendix. N is the sample size and p-values are in parentheses. All equations use a robust variance estimator. The GSS estimates are weighted by sampling probability weights

With respect to the occupation variable, a positive and statistically significant association with extramarital affairs is only apparent for men. In terms of the magnitude of the association, a one standard deviation increase in the occupation index raises the estimated probability of male cheating by 1.8 percentage points relative to an infidelity mean of 14.4 % for those men in their first marriage.

The restriction to those currently married renders the estimates potentially vulnerable to a selection bias. This would arise if non-random unobservable (to the econometrician) factors that affect marital survival also determine the decision to cheat and are correlated with the included explanatory variables. Although marital status is not modeled here, it is possible to perform a simple check for indications of selection bias, by re-estimating the model for all ever married respondents, whether or not they currently have a spouse. The results are reported in columns (3) and (4) of Table 2 and display very little qualitative change in the estimates. Those for occupation are identical while the association of infidelity with college education is larger in absolute value and now statistically significant for both sexes. If anything, then, restricting the sample to first marriages underestimates the size of the negative correlation of cheating and tertiary schooling. Following Elmslie and Tebaldi (2008), who also analyzed GSS data, we specify additional control variables for respondent’s income, social class and labor market status in unreported regressions. The results for occupation and own education remain robust to these extensions. It should be noted that a potential source of omitted variable bias in the case of the GSS sample is the absence of an age at marriage variable (since 1994) in the questionnaire. Given schooling and age at marriage are known to be positively correlated, this omission may bias the estimate for education in the direction of the sign of any relationship between age at marriage and infidelity, which is most likely negative given that a higher age of (first) marriage tends to protect against marital failure.

4.2 The German pairfam sample

We also run the infidelity specification using two European samples from Germany and Britain. The German data are taken from the first wave, dated 2008/9, of the new Panel Analysis of Intimate Relationships and Family Dynamics (pairfam) (Huinink et al. 2011). The data set comprises 12,402 randomly selected respondents across three birth cohorts, aged 15–17, 25–27 and 35–37 at the time of the survey. As members of the initial cohort have not generally completed their education, they are excluded from the study.

Rather than the lifetime measure recorded in the GSS, the infidelity question in the German survey asks whether the respondent cheated on their current partner during the past year or whether they are aware if their partner cheated or whether both did so.5 As with the Redbook data used by Fair, the question only captures cheating by those whose relationships are still intact. Again, it is a selected sample of those marriages that have survived discovery of an affair or in which infidelity remains undetected. The restriction to a 1 year window implies that infidelity is a much rarer event in the pairfam data though recent unfaithfulness is still admitted by 2.27 and 1.6 % of male and female respondents respectively. The dependent variable is coded to one for individuals who reported cheating, irrespective of whether or not they report that their partner was unfaithful. Where sexual partners are unmarried, the sample is restricted to those couples who are coresident as an informative indicator of commitment to sexual exclusivity. A limitation of the ISEI occupation variable in the German sample is that it is only recorded for those in employment or vocational training at the time of the survey. Education is coded according to the same scheme as that applied to the US data.

The results are reported for each sex in columns (1) and (2) of Table 3. Standard controls for a quadratic in age, number of kids and religiosity, the age of the respondent on entry into coresidence with the current partner are specified as well as opportunity variables. These include the proportion of nights away from home in the past 3 months due to work related reasons. As such absences are unlikely to be shared with their live-in partner they increase the risk of secondary sexual relationships. We also specify an indicator variable for whether the respondent lives in an urban area with a population of more than 50,000. This measure crudely proxies for the supply of potential alternative partners. The pattern of the results is close to that obtained with the GSS data. Occupation and college education matter but only for men and with the same signs for the associations as in the American case. By contrast, the equation for women is not at all well determined. For men, the estimated magnitudes of the economic effects are large relative to the mean of the dependent variable. A one standard deviation increase in the occupation index increases the cheating risk by 0.87 percentage points compared to an infidelity mean of 2.27 %. Likewise, the marginal effect of tertiary education (−0.019) is much larger relative to the mean in absolute terms than in the GSS sample.
Table 3

Infidelity equations: probit average marginal effect estimates, German sample

 

(1)

(2)

(3)

(4)

Male

Female

Male

Female

Pairfam (2008/2009)

Occupation

0.0004 (0.07)

−0.0001 (0.58)

0.0003 (0.21)

−0.0000 (0.85)

Education

    

 <Secondary

0.021 (0.36)

−0.007 (0.35)

0.019 (0.36)

−0.006 (0.39)

 >Secondary

−0.012 (0.37)

0.012 (0.29)

−0.012 (0.32)

0.011 (0.30)

 Tertiary

−0.020 (0.05)

0.007 (0.44)

−0.012 (0.25)

0.008 (0.48)

Age

0.015 (0.58)

−0.023 (0.33)

0.018 (0.50)

−0.024 (0.30)

Age squared

−0.000 (0.58)

0.000 (0.33)

−0.000 (0.50)

0.000 (0.31)

Kids

−0.002 (0.72)

0.004 (0.31)

−0.002 (0.68)

0.004 (0.29)

Religiosity

−0.006 (0.25)

−0.004 (0.43)

−0.006 (0.23)

−0.005 (0.42)

Urban

0.023 (0.02)

0.001 (0.91)

0.023 (0.02)

0.001 (0.84)

Nights away

0.011 (0.61)

0.037 (0.18)

0.011 (0.59)

0.038 (0.17)

Age at coresidence

−0.000 (0.94)

0.002 (0.02)

−0.000 (0.90)

0.002 (0.02)

Nonmarital cohabitation

−0.004 (0.68)

−0.010 (0.26)

−0.010 (0.64)

0.006 (0.73)

Occupation × cohabitation

  

0.0003 (0.53)

−0.0003 (0.41)

Tertiary × cohabitation

  

−0.036 (0.20)

−0.001 (0.96)

N

1,673

1,567

1,673

1,567

The infidelity specifications are estimated using the first wave of the German Pairfam sample from 2008/2009 with a binary dependent variable set to one for those respondents indicating they had cheated on their partner during the past year. The education indicator variables are categorised according to attainment, namely secondary schooling not completed (<Secondary), some post-secondary education (>Secondary) and tertiary. The omitted category is completed secondary education (i.e. graduated from High School). Columns (3) and (4) also include interaction terms between living together without marriage and, respectively, occupation and tertiary education. The construction of the variables is defined in the data appendix. N is the sample size and p-values are in parentheses. All equations use a robust variance estimator and sampling probability weights

It is plausible that the effects of economic factors on infidelity differ across respondents according to whether they are married or cohabiting without legal marriage. In particular, as cohabitants have selected a typically less committed relationship form, the effect of occupation may be stronger and the protective influence of education weaker than in the case of marriage. To test for this difference, the equations are reestimated with the addition of interaction terms between cohabiting status and, respectively, the occupation and tertiary education variables. The results are reported in columns (3) and (4) of Table 3 and show no statistically significant dependence of the effects of the economic variables on relationship status. In unreported regressions, partner’s education and interactions between the schooling of each partner were also specified to test for the effects of partner complementarities on infidelity but, as with the GSS sample, these estimates were not well determined.

4.3 The British NATSAL sample

The nationally representative cross sectional British data are taken from the second National Survey of Sexual Attitudes and Lifestyles (NATSAL) 2000 (Erens et al. 2001) which interviewed 12,110 respondents, aged 16–44 on a single occasion between 1999 and 2001.6 For those currently in partnerships, only one person was interviewed. The ethnic minority booster sample is excluded from the analysis but the results are unaffected by this restriction. The British questionnaire did not address infidelity directly but rather included a detailed sexual partnership history module with questions on the timing of the past three sexual relationships. Comparing this sexual history information with dates on the timing of live-in partnerships permits computation of the existence of concurrency in sexual relationships. As with the pairfam data, cheating is defined with respect to coresidential partners. However, in the British sample this is not confined to the current relationship but potentially includes infidelity that occurred when living together with a previous partner. Table 8 in the Appendix presents summary statistics for the British sample by fidelity and gender. The occupation variable is coded for all those who have an occupation whatever their current labor market status. However, a limitation, as with Fair’s sample, is that the occupation measure is only available at a highly aggregated level. Here it is divided into nine categories according to the respondent’s major international standard occupation code and converted to the ISEI-08 scale.

The Probit marginal effect estimates are reported in columns (1) and (2) of Table 4. They show that measures of occupation for both men and women are positively signed and statistically significant. For men (women), a one standard deviation increase in the occupation index increases the cheating risk by 2.8 (1.9) percentage points compared to an infidelity mean of 15.54 % (9.37 %). As with the American and German samples, college education is negatively signed for both sexes but only significantly different from the high school base category for men, with a coefficient estimate of −0.061. The equations are reestimated in columns (3) and (4) to allow for the effects of occupation and higher education on cheating to depend on whether the respondent has ever formally married. As in the German case, the coefficient estimates of the interaction terms between the economic variables and never married status are not significantly different from zero for either gender. This suggests there is no difference in the responsiveness of infidelity choices to college education and occupational status according to whether a respondent has ever entered marriage.
Table 4

Infidelity equations: probit average marginal effect estimates, British sample

 

(1)

(2)

(3)

(4)

Male

Female

Male

Female

NATSAL (1999/2001)

Occupation

0.0019 (0.01)

0.0014 (0.00)

0.0018 (0.03)

0.0014 (0.01)

Education

 <Secondary

−0.053 (0.13)

−0.021 (0.25)

−0.054 (0.12)

−0.021 (0.25)

 >Secondary

−0.012 (0.61)

0.004 (0.79)

−0.013 (0.61)

0.004 (0.80)

 Tertiary

−0.061 (0.02)

−0.009 (0.58)

−0.071 (0.01)

−0.010 (0.58)

Age

0.026 (0.26)

0.011 (0.45)

0.026 (0.25)

0.011 (0.45)

Age squared

−0.000 (0.20)

−0.000 (0.37)

−0.000 (0.20)

−0.000 (0.37)

Kids

0.000 (0.96)

−0.003 (0.58)

0.000 (0.96)

−0.003 (0.60)

Religiosity

−0.036 (0.00)

−0.015 (0.01)

−0.035 (0.00)

−0.015 (0.01)

Nights away

0.182 (0.02)

0.226 (0.02)

0.182 (0.02)

0.224 (0.02)

Urban

0.040 (0.06)

−0.000 (0.99)

0.039 (0.06)

−0.000 (0.99)

Female %

0.023 (0.63)

−0.071 (0.01)

0.021 (0.66)

−0.071 (0.01)

Age at first partnership

−0.007 (0.01)

−0.005 (0.00)

−0.007 (0.01)

−0.005 (0.00)

Never married

0.007 (0.75)

0.010 (0.46)

−0.027 (0.72)

−0.007 (0.88)

Occupation × never married

  

0.0005 (0.76)

0.0004 (0.73)

Tertiary × never married

  

0.037 (0.48)

0.002 (0.95)

N

1,821

2,861

1,821

2,861

The infidelity specifications are estimated using the British Natsal sample from 1999/2001 with a binary dependent variable set to one for those respondents who cheated on their current or a previous partner. All equations include standard controls for age, number of children and religiosity. The education indicator variables are categorised according to attainment, namely secondary schooling not completed (<Secondary), some post-secondary education (>Secondary) and tertiary. The omitted category is completed secondary education (i.e. graduated from High School). Nights away is defined as the proportion of the year spent away from home for work related reasons, urban is an indicator variable for residence in an urban or suburban area and Female % is the proportion of women working in the respondents’ two-digit standard industrial classification. Age at first partnership refers to the age at which the respondent first entered a coresidential sexual partnership. Never married is a dummy variable set to one for those respondents who have never entered into formal marriage. Columns (3) and (4) also include interaction terms between never having legally married and, respectively, occupation and tertiary education. The construction of the variables is defined in the data appendix. N is the sample size and p-values are in parentheses. All equations use a robust variance estimator and sampling probability weights

The British data also allow an additional proxy to control for infidelity opportunities. In addition to measures of urban residence and nights away from home for work reasons, a new opportunity variable is derived, namely the proportion of female workers by employment sector (Female %). The idea is that workplace interaction provides a low search cost context for married individuals to find alternative mates. Theoretically the higher the female to male sex ratio in the workplace, the fewer the cheating opportunities for women due to the greater degree of competition for male attention and vice versa for men. Indeed, Wellings et al. (1994) speculate that the greater opportunities to meet women at work may partly explain evidence that professional men have more affairs than those from lower social classes. And, in related research, McKinnish (2007) shows that the divorce risk is increasing in the fraction of opposite sex coworkers, especially for women. Although respondent workplace data are not collected, a crude sex ratio measure can be constructed using aggregate data from the UK Labor Force Survey across 60 sectors on female employment share for the two digit Standard Industrial Classification that the survey assigns to each respondent. The estimated marginal effects show that the workplace gender balance variable is negatively signed for women as would be expected and statistically significant. Although the estimate is positively signed for men as anticipated it is only one-third of the absolute size of that for women at the variable means. This suggests that there is much less of an infidelity benefit for men from a workplace gender imbalance though this effect is not precisely estimated. In other words, the risk of seduction by a colleague decreases more strongly for women as the workplace becomes more feminized than it increases for men.

In summary, the results from the GSS, pairfam and NATSAL samples, using different questions and variable definitions across three countries taken together tell a reasonably consistent story. First, occupational status has a positive association with infidelity for men while college education is negatively correlated across all three data sets. For women the picture is much more mixed with occupation only statistically significant in Great Britain and college education only in the US.

5 Distinguishing different classes of infidelity

Due to the limited scope of most survey questionnaires, the literature on the economics of extramarital sex does not distinguish different types of liaison, especially variation by the regularity of the relationship. Instead empirical studies typically address the probability of all forms of infidelity combined and define these loosely as affairs, a practice reflected in the discussion so far. However, the socioeconomic factors that determine the probability of indulging in a one off encounter may differ, or have different effects, from those that determine the likelihood of taking an occasional lover or of sustaining a long term parallel relationship. It is unlikely that these forms of sexual cheating are perfect substitutes given that the latter requires much more emotional investment and interaction. Likewise the expected sanction from a wronged spouse may be relatively less severe on average if their partner has engaged in casual extramarital sex compared to a much more meaningful ongoing affair. The aim of this section is to disaggregate infidelity into different classes by regularity and test whether the utility maximizing regularity choice differs by quality and education levels.

Theoretical predictions can be derived through a reinterpretation of the model. Specifically, define b as the regularity of a given illicit liaison rather than the quantity of extramarital sexual activity. The economic incentives can be correspondingly redefined. Using Eq. (3), higher quality individuals are expected to have greater demand for longer attachments insofar as the marginal return to more regular contact in a relationship increases with quality, rbq > 0, and the marginal expected sanction from additional interaction declines with quality, cbq < 0. While higher quality is predicted to raise demand for all classes of infidelity, the effect is expected to be stronger for more sustained relationships between cheating partners.

Holding quality fixed, education has ambiguous effects on b as described by Eq. (4). Additional schooling provides skills to facilitate relationship building and this raises demand for a more regular secondary love interest if the marginal return to regularity is increasing in education, that is, if rbe > 0. But by augmenting patience, education also increases the weight placed on the expected costs of infidelity exposure. The strength of this effect on the regularity of infidelity depends on both the marginal expected sanction cost, p(b)cb, and the effect of longer term cheating on the detection probability, pb. If the sanction cost and probability functions are concave in b, pbb < 0 and cbb < 0, then the discounting effects will be larger at lower levels of b. In other words, the patience benefits of additional schooling for resisting sexual temptation will be larger for casual encounters than for sustained affairs if the discounted expected marginal punishment costs from a greater regularity of cheating are larger when the extramarital relationship is relatively undeveloped.

The interpretation of the effects of the opportunity variables can be derived in an analogous fashion. The prediction is that the marginal return to a more sustained relationship is increasing in the opportunities for forming such partnerships. The lower the moral or religious cost of more regular simultaneous sexual relations or the more nights spent away from home for work related reasons, the greater the demand for longer term infidelity.

In terms of empirical implementation, the British NATSAL questionnaire permits inferences regarding whether the extramarital relationships identified in Sect. 4.3 constitute a regular, occasional or one-off sexual contact. These categories are constructed by combining responses to two partnership history questions. The first question classifies the type of relationship according to whether it is considered a regular or irregular sexual liaison. The second question asks whether the most recent sexual encounter with a partner was also the first occasion. This permits the division of irregular partnerships into one off encounters and occasional flings with the same partner. For the small minority of respondents (16 %) who have engaged in more than one type of cheating, they are allocated to the highest regularity category.7

Table 9 in the Appendix presents variable means by class of infidelity and gender. Conditional on cheating, men are much more likely than women to choose a one night encounter (37.8 vs. 27.6 %) while women are more likely to favor a more regular affair arrangement (21.9 vs. 31.7 %). In the raw data, the mean occupation measure is positively correlated with longer duration forms of cheating, especially for men. With regard to schooling, it is notable that within the one night class the share of college educated men is low compared to any other category. The sample means also show that respondents who have had an extended affair have fewer children on average and, in the case of men, more likely to live in an urban area and work with a higher proportion of female colleagues. Those participating in a one night encounter tend to be younger and less religious.

We use multinomial logistic regressions estimated by maximum likelihood to investigate the relationship between different categories of infidelity and the chief economic variables of interest, occupation and education, especially tertiary level education which has the strongest correlation with infidelity. The variable specification adopted is the same as that reported in columns (1) and (2) of Table 4. Table 5 presents the multinomial logit marginal effect estimates for infidelity by gender, but with the results only reported for occupation and higher education. As usual these represent, for an individual with average characteristics, the change in the predicted probability of engaging in each class of extramarital affair with respect to each regressor. In the case of higher education this change is relative to the base category of secondary schooling.
Table 5

Multinomial logistic regression marginal effect estimates for alternative classes of infidelity using the NATSAL sample

 

Occupation

Tertiary education

Men

Faithful

−0.0017 (0.01)

0.055 (0.01)

One night

0.0006 (0.15)

−0.033 (0.01)

Irregular

0.0005 (0.30)

−0.012 (0.41)

Affair

0.0007 (0.01)

−0.010 (0.22)

N

1,821

 

Women

Faithful

−0.0012 (0.01)

0.003 (0.83)

One night

0.0006 (0.02)

−0.007 (0.25)

Irregular

−0.0000 (0.87)

0.009 (0.43)

Affair

0.0007 (0.00)

−0.005 (0.46)

N

2,861

 

The estimates are the marginal effects of the occupation and tertiary education variables on the probability of each category of cheating (or none). They are computed from an equation that includes the controls listed in columns 1 and 2 of Table 4. p-values are in parentheses. N is the number of observations

For occupation, the estimated marginal effects are positive for each class of infidelity with the single exception of the irregular class for women. The magnitude of the effects is highly symmetrical between men and women with the pattern suggesting that occupations of greater socioeconomic status and resources are positively correlated with more extended unfaithful liaisons. In the case of affairs, these effects are statistically significant at the 1 % level for both sexes. In the case of one night stands, the marginal effect for men is statistically insignificant. Although apparently small in absolute terms, the size of the marginal effects is large relative to the unconditional means of the infidelity classes (as given in the appendix, Table 9). For example, a one standard deviation increase in the occupational index is associated with an increase in the predicted probability of an affair for men of 0.010 relative to a class mean of 0.034.

In the case of tertiary education, the most striking result is the negative marginal effect of −0.033 on the predicted probability of a one off liaison for men relative to someone who has only completed secondary schooling and compared to a mean of 0.059 for this class of infidelity. The marginal effect for a more regular affair is much smaller in magnitude at −0.010 relative to mean of 0.034. Although a χ2 test cannot reject the hypothesis that the effect of college education on one night stands and affairs is the same for men, the association between affairs and higher education is not statistically significant at conventional levels nor is it distinguishable from the effect of college on faithfulness. For a given socioeconomic status, then, while higher education is negatively correlated with infidelity in general, the strongest statistically significant association is with casual encounters for men. In terms of the theory, this is consistent with the prediction that the discounting effects of education on infidelity are greater at lower levels of cheating regularity all else equal. And this association is not offset by the marginal relational benefits of schooling which might also be expected to be larger when regularity is lower. Notably, the schooling results for women are much less well determined; the marginal effect estimates are relatively small and imprecisely estimated.

6 Conclusion

Exploiting four different samples across three countries from different time periods over the past 40 years and with alternative measures of infidelity, we find consistent correlations between extramarital affairs and occupational status and education. Given difficulties interpreting these results using Fair’s (1978) time allocation model, an alternative theory is developed that is driven primarily by time preferences and expected sanction costs.

With regard to occupation, the findings are most striking and robust across national samples in the case of men with more mixed results for women: the results show a positive correlation between socioeconomic status and the demand for extramarital sex both in terms of participation and durability of the relationship. It is those in higher occupational groups who are most able to attract and retain parallel lovers.

Taken together across all samples and both sexes, college education is negatively related to the decision to participate in cheating with typically no association arising at other levels of completed education, holding occupation constant. With respect to the model, this is interpreted in terms of a time preference argument; the educated place a higher weight on expected future sanctions relative to the benefits of infidelity. Investigating different classes of infidelity, the results suggest that schooling has a greater deterrent effect for casual encounters, at least for men, a finding consistent with the argument that additional education is associated with higher discounted marginal costs of infidelity at lower levels of regularity. As a caveat, it should be noted that the results for education need not necessarily be causal. They are equally consistent with a selection argument that patient individuals both invest more in their schooling and commit less infidelity, especially in the form of short term cheating. If that mechanism is correct, then the schooling estimates are capturing a correlation with time preferences and should be interpreted accordingly.

However, there also exist plausible alternative or complementary explanations of the result for education. Evidence that links more schooling to higher marital investment and greater gains to marriage provides a potentially powerful channel through which education can provide another disincentive to marital cheating. Although initial tests for an association between marital quality and infidelity using the interactions between the schooling of both partners could not identify an effect, this is only a preliminary result using a simple test which opens up a research agenda rather than resolves the question.

Inevitably, cross section estimates cannot easily provide a causal account. There may be confounding covariates omitted from the specification that explain the associations. With respect to occupation, for example, unobserved characteristics that drive partners to achieve in the labor market, especially without the benefits of formal education, may also be effective and valued in the mating market and this could generate the patterns observed in the data. Better survey measures of infidelity in household panels are required to provide scope to control for such unobserved heterogeneity and offer a more definitive evaluation of infidelity activity.

Finally, it is notable that, whatever the causes of cheating, the consequences of infidelity remain an open and unexplored empirical question. For example, little is known about the determinants of the decision to dissolve a union, conditional on the discovery of an infidelity. There is likewise little discussion in the literature of the social efficiency of extramarital affairs. This largely turns on the trade-off between the gains from polygamous mating and the reallocation of partners in the event of household dissolution compared to the costs imposed on third parties, especially children.

Footnotes
1

Of course, models of illegitimate activity and sanctions can also be constructed within an allocation of time framework. There is clearly scope to extend Fair’s model itself to accommodate the negative education effect.

 
2

Moreover, insofar as education increases risk aversion, the possibility of future sanctions will have a greater deterrent effect. For simplicity, risk neutrality is assumed throughout.

 
3

Empirical evidence that education is associated with greater gender equality in the household division of labor, a measure of martial investment, is presented by Bloemen et al. (2010) and Iversen and Rosenbluth (2006) and references therein.

 
4

Fair also specified a variable for husband’s occupation which was not statistically significant. For ease of comparison with other data sets that do not record spousal occupation, the variable is also excluded from estimation using Fair’s data.

 
5

An attempt was made to exploit the information on the respondent’s awareness of a partner’s unfaithfulness in the German sample by estimating Probit specifications for partner cheating. Since the question was only asked of those whose relationship remained intact, just 0.6% of men reported that their partner had cheated in the past year and 1% in the case of women. Unfortunately, the estimation results were very imprecisely determined.

 
6

The first NATSAL survey was investigated by Cameron (2002) based on Fair’s time allocation model.

 
7

The NATSAL questionnaire provides duration data on the length of time between the beginning and end of the previous three sexual relationships. However, application of survival analysis to investigate the determinants of the duration of infidelity is frustrated by inherent difficulties of using timing data to define what counts as a spell of infidelity. A brief fling with an old flame encountered again many years later would be wrongly classified as an apparently long duration, though irregular, concurrent relationship in terms of the spell between first and last sex with this former partner. In the absence of explicit frequency data on sexual contact with each partner, the approach is constrained here to consider only a division of infidelity into broadly specified cheating classes.

 

Acknowledgments

The author is grateful to the editor and two referees for helpful comments. The British data used in this paper were made available by the UK Data Archive and Catherine Mercer provided help with the variables. This paper also uses data from the first wave of the German Family Panel (pairfam) which is coordinated by Bernhard Nauck, Johannes Huinink, Josef Brüderl and Sabine Walper. The panel is funded as long-term program by the German Research Foundation (Deutsche Forschungsgemeinschaft, DFG).

Copyright information

© Springer Science+Business Media, LLC 2012