Society uses many methods for social control, including the legal system. In the past, society erected a formidable barrier, fault-based divorce law, to prevent (or at least to hinder) the dissolution of a marriage. However, over time, social reforms have led to fewer restrictions regarding divorce in the United States. Perhaps the most profound social reform has been the switch from fault-based divorce law to no-fault divorce law.
Fault-based divorce law was a restrictive law designed to protect marriage. A divorce was granted under the traditional fault-based law only if one spouse was found at fault or "guilty" (e.g., of adultery or cruelty) and the other spouse was found "innocent" (Weitzman, 1985). Furthermore, the consent of the "innocent" spouse was needed to grant the divorce (Weitzman, 1985). Divorce was denied if both spouses were at fault. In theory, fault law awarded alimony, child support, and property distribution to the "innocent" spouse and the monetary value of the economic settlement was linked, in part, to the income level of the "guilty" spouse. Thus, financial gain resulted from proving fault of the other spouse. However, in practice, fault law treated the two genders differently in divorce in that the husband was generally responsible for alimony and child support and the wife was generally responsible for child custody (Weitzman, 1985). Perhaps the most fundamental drawback of the fault-based system was the requirement that at least one spouse be found guilty and the other spouse be found innocent of marital fault, because this framework of guilt and innocence perpetuated acrimony and conflict in the social-psychological and communication climate of the divorce.
In contrast, no-fault divorce law, now law in all 50 states, is a law designed to make divorce less restrictive. The essence of no-fault divorce law is that it does not attribute fault and thus does not require one of the spouses to be considered "innocent" and the other "guilty" (Weitzman, 1985). Rather, no-fault divorce law recognizes the breakdown of the marriage in that the spouses can no longer function as a married couple. Consent of both spouses is not required; rather, one spouse can decide unilaterally to divorce. No-fault divorce law is gender-neutral in that both spouses are responsible for alimony and child support and both spouses are eligible far child custody. Financial awards in terms of alimony, child support, and property distribution are no longer linked to fault but, rather, to the spouses' current financial needs and resources (Weitzman, 1985). The final but perhaps the most fundamental axiom of the no-fault divorce law is to improve the social-psychological and communication climate of divorce by abolishing the concept of fault (the framework of guilt and innocence) and by tempering the adversarial process surrounding divorce proceedings.
No-fault divorce law might logically lead us to expect an increase in the divorce rates because it has reduced the legal obstacles, the economic costs, and the psychological consequences of divorce. However, previous research (e.g., Allen, 1992; Marvell, 1989; Peters, 1986, 1992; Sepler, 1981; Wright & Stetson, 1978; Zelder, 1993) that has examined the effect of no-fault divorce law on the divorce rate has produced inconclusive results. This is probably due in part to the use of cross-sectional designs (e.g., Marvell, 1989; Peters, 1986, 1992; Wright & Stetson, 1978; Zelder, 1993). Hence, the first focus of the current research was to use longitudinal data to simply test the following hypothesis:
Hypothesis 1: No-fault divorce law results in an increase in the divorce rate.
The reduction of the legal obstacles and the economic costs of divorce that came with no-fault divorce legislation might have had differential effects on families with disparate income levels. That is, we might expect no-fault divorce to be more attractive to low-income families who could not afford divorce under fault-based legislation. Conversely, no-fault divorce might also be more attractive to high-income families (especially the "guilty" spouse) who, under fault-based legislation, would have faced the expensive prospect of substantial alimony payments, substantial child support payments, substantial attorney fees, and substantial loss of property to the "innocent" spouse. Thus, the second focus of the current research was to test the following nondirectional hypothesis:
Hypothesis 2: Median family income has a relationship with the increase in the divorce rate due to no-fault divorce law.
Although the prohibitions of the traditional fault-based divorce law gradually gave way to less restrictive no-fault statutes, religious precepts and the educational attainment of marital partners have long served as negative correlates of divorce. In fact, previous research has demonstrated a general inverse relationship between marital dissolution and educational attainment (e.g., Bumpass & Sweet, 1972; Glenn & Supancic, 1984; Glick, 1984; Goode, 1962; Heaton, 1991; Kurdek, 1943; Morgan & Rindfuss, 1985; Udry, 1966) and between marital dissolution and religiosity (e.g., Bumpass & Sweet, 1972; Filsinger & Wilson, 1984; Glenn & Supancic, 1984). One interpretation of the negative effect of religiosity on divorce may derive from theological and subcultural views on marital dissolution and family values (Bumpass & Sweet, 1972; Filsinger & Wilson, 1984). Further, we speculate that participation in a mainstream, highly visible religious denomination should maximize whatever social control effect occurs through religious affiliation. An interpretation of the negative effect of educational attainment on divorce is that the marriages of persons who are well educated may be at a lower risk for dissolution because such persons may embody greater interpersonal skills, maturity, and resources that benefit a marital relationship (Bumpass & Sweet, 1972; Heaton, 1991). Hence, the next focus of the current research was to test the following hypotheses:
Hypothesis 3: Educational attainment has a negative relationship with the increase in the divorce rate due to no-fault divorce law.
Hypothesis 4: Status in high-profile religions--the Roman Catholic, Southern Baptist, or United Methodist churches--has a negative relationship with the increase in the divorce rate due to no-fault divorce law.
The population and unit of analysis for this study were the 50 States. All 50 states were included in the study. A quasiexperimental pre-post treatment design (Cook & Campbell, 1979) and archival data were used to test the hypotheses of this study. The archival data for each variable were obtained from the following sources. The divorce statistics used to define the dependent variable and the covariate came from Vital Statistics of the United States (National Center for Health Statistics, 1987, 1989; United States Bureau of the Census, 1950-1990). The median family income statistics and the educational attainment statistics came from United States Census of the Population (United States Bureau of the Census, 1950-1990). The religiosity statistics came from the Glenmary Research Center, Churches and Church Membership in the United States (Bradley, Green, Jones, Lynn, & McNeil, 1990; Johnson, Picard, & Quinn, 1971; Quinn, Anderson, Bradley, Goetting, & Shriver, 1980; Whitman & Trimble, 1956). The effective dates of the no-fault divorce laws were obtained from Arkansas Code Revision Commission (1993), Louisiana Code Commission (1992), Marvell (1989), Maryland Code Revision Commission (1991), Sepler (1981), and Wright and Stetson (1978).
DESIGN AND VARIABLES
The treatment in this quasiexperimental study was the enactment of the no-fault divorce law for each of the 50 states, which occurred at different dates for the 50 states. The years of enactment ranged from 1953 to 1987, with the majority of the no-fault divorce laws enacted in the 1970s. Seventy-eight percent of the no-fault divorce laws were enacted in the 1970s, 10% in the 1960s, 8% in the 1980s, and 4% in the 1950s. The year of enactment of the no-fault divorce law for each of the 50 states is reported in Table 1.(Table 1 omitted)
The dependent variable for this study was the state divorce rate, measured as number of divorces (including annulments) per 1,000 individuals after the enactment of the no-fault divorce law for each of the 50 states. The dependent variable was based on an average of the divorce rates for the 3 consecutive years following the effective date of the no-fault divorce law for each state. A 3-year average was used as an attempt to stabilize the dependent variable. The post-no-fault divorce rate for each of the 50 states is reported in Table 1.
The covariate for this study was the state divorce rate, measured as number of divorces (including annulments) per 1,000 individuals before the enactment of the no-fault divorce law for each of the 50 states. The covariate was based on an average of the divorce rates for the 3 consecutive years preceding the effective date of the no-fault divorce law for each state. The covariate was included in the model to account for individual differences in divorce rates across states and to increase precision in determining the effect of the independent variables on the dependent variable. This type of design uses a change score that is called a "residualized gain score" in the testing literature (Cronbach & Furby, 1970). The pre-and the post-no-fault divorce rates, and the raw gain scores, are reported for each of the 50 states in Table 1.
Median family income. Median family income for a state was the median of the total income received by related persons in a family. Median family income for each state was adjusted by the consumer price index so that each measure was in 1967 equivalent dollars. Because income data were not always available for the exact year that each state no-fault divorce law went into effect, median family income was taken to the closest year to the enactment of the state no-fault divorce law. Twelve percent of the state income variables exactly matched the year of enactment. In other cases, the closest year was always within 2 years of the actual state no-fault year. If a no-fault year fell between two equally close years, then an average family income value was computed for those 2 years.
Religiosity. The religiosity variables for each state were measures indicating Roman Catholics, Southern Baptists, and United Methodists as a percentage of total state population. These three religions were included because they represent the principal denominations (in terms of percentage of total state population) in 48 of the 50 states. Each of the state religiosity statistics (i.e., Roman Catholic, Southern Baptist, United Methodist) was taken to the closest year to the enactment of the state no-fault divorce law, which was always within 3 years of the actual state no-fault year. Fourteen percent of Roman Catholics, 14% of Southern Baptists, and 14% of United Methodists exactly matched the year of enactment.
Educational attainment. State educational attainment was the percentage of persons 25 years of age and over completing 4 or more years of college. Because educational attainment data were not always available for the exact year that each state no-fault divorce law went into effect, state educational attainment rate was taken to the closest year to the enactment of the state no-fault divorce law. Ten percent of the state educational attainment variables exactly matched the year of enactment. In other cases, the closest year was always within 2 years of the actual state no-fault year. If a no-fault year fell between 2 equally close years, then an average educational attainment rate was computed for those 2 years.
The hypotheses of this study were tested using the dependent t test, multiple regression, and analysis of covariance (ANCOVA). Estimates of the magnitude of the effect size were also computed for each hypothesis. The effect size estimators that accompanied the dependent t test, multiple regression, and ANCOVA were the mean divided by the standard deviation, the standardized regression coefficient, and eta-square, respectively. The Pearson product-moment correlation and the squared semipartial correlation were also used as effect size estimators in specific regression analyses. To test Hypothesis 1, that no-fault divorce law leads to an increase in divorce rates, the dependent t test was carried out comparing the post-no-fault divorce rate to the pre-no-fault divorce rate for all 50 pairs. To show that the change in the divorce rate was different from what it would have been if the divorce law had not changed, a no-treatment comparison group consisting of a random sample of divorce rates, for each state, not separated by no-fault law, was also used in the testing of Hypothesis 1. That is, a random year (a year different from a no-fault year) was selected for each state, and then the dependent t test was carried out comparing a 3-year average divorce rate after the random year to a 3-year average divorce rate before the random year for all 50 pairs. The range of years for each state in the no-treatment comparison group was constrained to not overlap with the range of years for each state in the treatment group. Multiple regression controlling for the effects of the covariate was used to test Hypotheses 2, 3, and 4 (a "residualized gain" analysis).
ANCOVA, with least squares means, also was used to test Hypotheses 2, 3, and 4. A median split criterion was used to dichotomize each independent variable into two levels--high and low (see Table 2).(Table 2 omitted) The goal of using ANCOVA was to assess the effects of categorical independent variables--median family income, educational attainment, and religiosity-upon the post-no-fault divorce rate (dependent variable) independently of the covariate (the pre-no-fault divorce rate).
A reviewer of this article very properly raised the issue of whether statistical tests are appropriate in this setting. If we view the 50 states as our whole population, then we are observing parameters, and statistical tests are not necessary. According to this view, the estimates of the magnitude of the effect size might be considered more appropriate. We agree with this view, as have others who have done state-level analysis (e.g., Richards, 1978, 1984). Thus, we believe our graphs and summary statistics can be considered population portrayals, and we present effect size estimates. However, there is another sense in which statistical tests are useful and enlightening. If we view the enactments of no-fault divorce law as a sample of 50 observations drawn from a whole population of times and settings in which these laws could have been enacted, then our statistical tests answer a broader set of questions. Only in this context do the issues of period effects, historical confounds, selection factors, and other threats to internal validity become relevant. Thus, our broader set of questions involve how enactment of no-fault divorce law generally affects divorce rates in states like those in the United States in times like those in the 1950s through the 1980s.
Dependent t Test
Treatment group. The results of the dependent t test comparing the post-no-fault divorce rate to the pre-no-fault divorce rate across all 50 states revealed that no-fault divorce law had a significant effect on the divorce rate, t(49) = 6.37, p < .0001. Thus, Hypothesis 1 is supported. The estimate of the magnitude of the effect size associated with Hypothesis 1 was .91. The results of the dependent t test revealed a measurable increase in the divorce rate, M = .80, SD = .88. A needle plot of the difference scores on which this analysis is based is shown in Figure 1 and Figure 2.(Figures 1 and 2 omitted) All states showed a positive gain, with six exceptions: Oklahoma, Maryland, Nevada, Arkansas, Illinois, and Utah. These six exceptions are individually interpretable and will be discussed in the discussion section.
No-treatment comparison group. The results of the dependent t test comparing a 3-year average divorce rate after the random year to a 3-year average divorce rate before the random year for all 50 states revealed that the change in the divorce rate that was not separated by no-fault divorce law was not significant, t(49) = -0.74, p = .46. The estimate of the magnitude of the effect size associated with this test, the no-treatment comparison group, was -0.10. The results of the dependent t test for the no-treatment comparison group revealed a measurable decrease in the divorce rate, M = -0.15, SD = 1.49. The findings of the no-treatment comparison group provide further support for Hypothesis 1, and they begin to rule out a period effect interpretation as a threat to the internal validity of the treatment effect.
Multiple Regression and Correlation
We present a second analysis to address the possibility of a period effect. If divorce rates are going up generally and systematically over time, then this would appear as a treatment effect in the previous analysis. The results of the dependent t test for the no-treatment comparison group are strong enough to rule out a period interpretation of the treatment effect. A simple correlation analysis was also used to test for a systematic period effect on divorce rates. For this analysis, the Pearson product-moment correlation (r) was the effect size estimator. The results of this analysis revealed that the year of enactment of the no-fault divorce law was not significantly related to the pre-no-fault divorce rate, r = .15, t(48) = 1.09, p = .28. Also, the year of enactment of the no-fault divorce law was not significantly related to the post-no-fault divorce rate, r = .12, t(48) = .87, p = .39. Moreover, the year of enactment of the no-fault divorce law was not significantly related to the difference score (i.e., the post-no-fault divorce rate minus the pre-no-fault divorce rate), r = -.14, t(48) = 1.04, p = .31. Furthermore, the year of enactment of the no-fault divorce law was not significantly related to the post-no-fault divorce rate when the effects of the pre-no-fault divorce rate were statistically controlled, beta = -.02, t(47) = -.52, p = .60. Here the standardized regression coefficient (beta) was the effect size estimator. The results of the significance tests and the estimates of the effects size suggest that the date of enactment of the no-fault divorce law did not affect divorce rates.
The results of a multiple regression analysis t revealed that median family income had a significant positive relation to the post-no-fault divorce rate when the effects of the pre-no-fault divorce rate were statistically controlled, 1, 47)= 4.86, p < .05. Thus, Hypothesis 2 is supported. The estimate of the magnitude of the effect size (standardized regression coefficient) associated with Hypothesis 2 was .095. The multiple regression results also revealed that the other independent variables--educational attainment, Catholicism, United Methodism, and Southern Baptist--were not significantly related to the post-no-fault divorce rate when the effects of the pre-no-fault divorce rate were statistically controlled. Thus, Hypotheses 3 and 4 were not supported. The multiple regression results, including the estimates of the magnitude of the effect size, are reported in Table 3.(Table 3 omitted)
To assess the relationship between religiosity as a whole and the post-no-fault divorce rate, a model including all the independent variables and the pre-no-fault divorce rate was compared with a model containing only the pre-no-fault divorce rate, median family income, and educational attainment. The results of this model comparison revealed that, when the effects of the pre-no-fault divorce rate, median family income, and educational attainment were statistically controlled, the incremental contribution of mainstream religiosity as a whole was not significant, F(3, 43) = .57, squared semipartial correlation = .002. Here the squared semipartial correlation was the effect size estimator.
ANCOVA with Binary Independent Variables
The ANCOVA results revealed a significant relationship between the post-no-fault divorce rate and median family income when the post-no-fault divorce rate was adjusted for the pre-no-fault divorce rate, F(1, 47) = 6.53, p < .01. The pattern of the post-no-fault divorce rate least squares means showed that the no-fault divorce law had a greater impact on high-income families (least squares mean = 5.19) than on low-income families (least squares mean = 4.68). Thus, this result provided support for Hypothesis 2. The estimate of the magnitude of the effect size (eta-square) associated with Hypothesis 2 was .12.
The ANCOVA results also revealed that there were no significant differences due to the other independent variables--educational attainment, and status as Roman Catholic, United Methodist, or Southern Baptist--when the post-no-fault divorce rate was adjusted for the pre-no-fault divorce rate (covariate). Thus, Hypotheses 3 and 4 were not supported. The ANCOVA results, including the estimates of the magnitude of the effect size, are reported in Table 4.(Table 4 omitted)
A Pearson product-moment correlation revealed a significant relationship between the dependent variable (the post-no-fault divorce rate) and the covariate (the pre-no-fault divorce rate) in the ANCOVA model used in this study, r = .95, p < .0001. This suggests that the assumption of linearity between X and Y in the ANCOVA model was supported. It also reveals that only 10% of the variance was left to be explained by the independent variables.
Statistical Conclusion Validity
A threat to the statistical conclusion validity of this study pertains to autocorrelation in the regression models. That is, if the error terms in the regression models are positively autocorrelated, then the use of ordinary least squares (OLS) has important consequences: (a) the estimated regression coefficients may no longer have the minimum variance property and may be inefficient, (b) mean squared error may underestimate the variance of the error terms, and (c) the standard deviation of the regression coefficient calculated according to OLS procedures may underestimate the true standard deviation of the estimated regression coefficient (Neter, Wasserman, & Kutner, 1989). m view of the seriousness of the problems created by autocorrelated errors, it is important that their presence be detected. Thus, to test that the autocorrelation parameter, p, is zero for each regression model in this study, we employed the Durbin-Watson d statistic, which assumes the first-order autoregressive error model. The value of the Durbin-Watson d statistic is close to 2 if the error terms are uncorrelated (SAS Institute, 1990). Also, according to the Durbin-Watson test bounds, when n equals 50 and alpha equals .05, if the value of the d statistic is greater than 1.77 (upper bound), then the error terms are uncorrelated (Neter, Wasserman, & Kutner, 1989). The Durbin-Watson test for autocorrelation revealed that the d statistic for each regression model in this study was always greater than 1.90 (range = 1.92 to 2.16). Because each d statistic was greater than 1.90 (or close to 2), there appears to be little evidence to suggest that there is significant autocorrelation present in our regression analyses.
A further threat to statistical conclusion validity pertains to homogeneity of variance of the underlying population distributions of state divorce rates. That is, the internal validity of the study may be threatened if variances in the population of state divorce rates were not equal. Thus, to test for homogeneity of variance, we first divided the state pre-no-fault divorce rate data into five groups with equal sample sizes (n = 10); we labeled the groups lowest, low, middle, high, and highest. Then, we tested for homogeneity of variance by assessing the effects of the categorical pre-no-fault divorce rate upon the post-no-fault divorce rate (dependent variable). The Bartlett-Box F statistic and the Cochran C statistic were applied to this design to test for homogeneity of variance. The results of the Bartlett-Box F test and the Cochran C test revealed that the variances in the population of state divorce rates were not equal, F(4, 45) = 11.68, p < .0001; C(4, 45) = .84, p < .0001; SD lowest group = .85, SD low group = .54, SD middle group = .78, SD high group = .83, SD highest group = 3.56. The difference in variances can be attributed to the highest group, which contained the divorce rates of Nevada. Dropping Nevada from the analysis resulted in nonsignificant differences in variances, F(4, 44) = 1.008, p = .40; C(4, 44) = .34, p = .26, SD highest group = 1.11. Dropping Nevada from the analysis strictly to meet the homogeneity of variance assumption would increase the estimates of the magnitude of the effect size, given that Nevada's divorce rate actually decreased after no-fault divorce law was implemented. Also, we do not regard this as a major threat to the internal validity of the study, given that the F statistic and the t statistic that were used in the data analysis of this study are robust to departures from homogeneity of variance and normality (Boneau, 1960; Feir-Walsh & Toothaker, 1974; Ramsey, 1980).
The results of the dependent t test for the treatment group have shown that the switch from fault divorce law to no-fault divorce law led to a measurable increase in the divorce rate. The magnitude of this increase was an average of .80 divorces per 1,000 individuals per year in the state population, which is equivalent to an effect size of .91--a large effect size as defined by meta-analysis standards (e.g., Cohen, 1988). In contrast, the results of the dependent t test for the notreatment comparison group revealed a small and nonsignificant decrease in the divorce rate that was not separated by no-fault divorce law. Three important conclusions emerge from the dependent t test results. First, it is shown that the change in the divorce rate is different from what it would have been if the divorce law had not changed. Second, the results of the no-treatment comparison group provide support for the generalizability of our results at least within the time frame of enactment of no-fault divorce law. Third, the findings suggest support for the basic goal of no-fault divorce law: to provide a divorce law that makes divorce less restrictive by reducing the legal and the economic obstacles of divorce and by improving the social-psychological and communication climate of divorce by abolishing the concept of fault (the framework of guilt and innocence) (Weitzman, 1985).
A visual inspection of the needle plots (see Figure 1 and Figure 2) of the difference scores reveals that all states had an increase in divorce rate after the enactment of the no-fault divorce law, with six exceptions: Oklahoma, Maryland, Nevada, Arkansas, Illinois, and Utah. One interpretation of the decrease in the divorce rate that occurred in these six states may derive from state variation in enforcing the mechanics of no-fault divorce law. Other interpretations include changes in the family institution, decline in the marriage rate, slowdown in migration, economic factors (e.g., unemployment), and a general downturn in divorce rates (Kemper, 1983; Martin & Bumpass, 1989).
The large decrease in the divorce rate that occurred in Nevada may be attributed to the pervasive adoption of no-fault divorce law across the 50 states. That is, the pervasive adoption of no-fault divorce law made divorce less restrictive and more accessible. The result was a reduced need for couples to travel to Nevada for a quick divorce (during the pre-no-fault law era, Nevada was a divorce mill where an expedient divorce was available to married couples in the United States). The needle plot in Figure 2 shows an interesting pattern in that the first states (Oklahoma and Maryland) and two of the last three states (Illinois and Utah) to adopt no-fault divorce law experienced a decrease in the divorce rate. This pattern may be due, in part, to some latency effect; for example, the novelty of the law for Oklahoma and Maryland and the trepidation of consequences of the law for Illinois and Utah. Also, Utah's Mormon culture may act against the typical operation of no-fault divorce law.
The results discussed in the preceding paragraph pertain to the whole population of states. As discussed earlier, summary statistics computed from these analyses can be interpreted as parameters, and effect sizes were presented to support this interpretation. In this context, it is clear that no-fault divorce law did positively influence the divorce rate in 44 of the 50 U.S. states, with the six exceptions being directly interpretable.
The second interpretation discussed above--involving statistical tests--addressed a broader set of questions: Do no-fault divorce laws generally lead to increased divorce rates in places and times like those in the United States during the period of study, and what other variables are related to this process? When we test models including income, education, and religiosity to try to account for state differences in divorce rates, issues of both model misspecification and identification of the general processes influencing divorce rates become relevant. We hesitate to draw conclusions even more broadly (e.g., to general times or cultures), although empirical analysis could certainly be directed toward testing the breadth of these findings. At some point, the cultural definition of marriage itself can help define the boundaries. For example, in polygamous cultures or in cultures lacking elaborate legal systems, the issue of the effect of no-fault divorce law takes on a different role or can perhaps become irrelevant.
An understanding of the hypothesized effects of the independent variables on the post-no-fault divorce rate, when this rate was adjusted for the pre-no-fault divorce rate, was provided by the multiple regression analysis and by the analysis of covariance. The basic finding was that among the independent variables considered, median family income was the only significant predictor of the post-no-fault divorce rate (albeit the magnitude of the effect size was small). The post-no-fault divorce rate increased significantly as median family income increased. This finding suggests that, with the reduction of the legal obstacles and the economic costs of divorce facilitated by no-fault divorce law, there has occurred an increase in divorce among high-income families who can now expect more equitable settlements of property and spousal support. This basic finding, a positive relationship, is inconsistent with previous research (e.g., Cutright, 1971; Goode, 1962; Kurdek, 1993; Schroeder, 1939; U.S. Bureau of the Census, 1950) that has found a negative relationship between husband's income and divorce rate. However, this previous research considered only husband's income, not family income, and was based on individual-level and not state-level data.
The religiosity variables (i.e., Roman Catholic, Southern Baptist, United Methodist), although not significant in this study, provided an interesting and unexpected pattern of findings. Most surprising here was the finding that the hypothesized negative relationship between Catholicism and the post-no-fault divorce rate was not supported. This finding is surprising because the main protagonist against marriage dissolution, compared with other religious denominations, has been the Roman Catholic church, because of its adherence to the doctrine of indissolubility: "In the eyes of the church, a marriage can only be dissolved by God through death or by the church through annulment" (Parkman, 1992, p. 14).
The general religiosity finding in this study is in line with previous research (e.g., Glenn & Supancic, 1984; Trent & South, 1989) that has found religion to be a nonsignificant correlate of divorce. Thus, the religiosity finding in this study may suggest that belief in the sanctity of marriage and adherence to religious precepts, though perhaps once strong enough to prevent the dissolution of marriages, have weakened (Glenn & Supancic, 1984; Lerner & Lerner, 1981; Phillips, 1988). We note that there are inherent limitations in the measurement of our religiosity variables: Broader measures than our three mainstream denomination proportions would be useful. It may be, for example, that religious social controls operate at an individual level, and thus state-level measures cannot detect the operation of such religious controls. However, previous individual-level religiosity research (e.g., Glenn & Supancic, 1984; Trent & South, 1989) has also failed to find the effect of such religious social controls.
These results are, of course, tempered by the quasiexperimental nature of the study. For example, one threat to the internal validity of the current research is history (Cook & Campbell, 1979). That is, there have been many influences on divorce rates other than no-fault divorce laws. Hence, an initial indication of the impact of no-fault divorce laws is whether the increase in divorce rates was part of demographic and socioeconomic trends. For example, factors that were not included in the current research include trends in the age structure at marriage, the presence or absence of children within a marriage, the number of second and subsequent marriages, military conflicts (e.g., Vietnam War), womens' liberation (e.g., working wives may feel more financially empowered and better able to cope with the economic impacts of divorce), and migration (Heaton, 1991; Martin & Bumpass, 1989). A response to this concern involves how spread out in time the initiation of laws was across states. A second and related threat to internal validity is whether no-fault divorce law was a response to, rather than a cause of, the increase in the divorce rate. That is, new laws can alter human behavior, but the laws themselves can reflect legislators' attempts to respond to changes in demographic and socioeconomic forces (Parkman, 1992).
An additional threat to internal validity--more specifically, a threat to statistical conclusion validity--pertains to the independence of observations. That is, the internal validity of the study may be threatened if the state divorce rates were not independent of each other. If, for example, states with earlier enactment stimulated couples from surrounding states to move across the border to divorce, then the results of this analysis may not be considered to be 50 independent replications. The patterns in Figure 1 and Figure 2 are strong enough to rule out this argument for evaluation of the basic treatment effect, albeit we suspect that some of the aforementioned contamination with respect to independence of observations did indeed occur.
Finally, we believe it is beyond the scope of this article to broadly generalize our results to societies outside the United States. The breadth of our results are limited to the effect of the enactment of no-fault divorce law within the United States and during the time frame of the enactment of no-fault divorce law. It is this context in which our results should be interpreted, although our statistical analyses may support generalizing to other settings that appear similar to those in the current study.
Two important results emerge from the current study. First, the enactment of no-fault divorce law had a clear positive influence on divorce rates. Exceptions to this pattern are individually interpretable. Second, median family income had a small but significant positive relation to the post-no-fault divorce rate when the effects of the pre-no-fault divorce rate were statistically controlled.
The authors are grateful to Charles F. Gettys and to two anonymous reviewers for their insightful and helpful comments on earlier drafts of this article.
Allen, D. W. (1992). Marriage and divorce: Comment. American Economic Review, 82, 679-685.
Arkansas Code Revision Commission (1993). Arkansas code of 1987 annotated (Section 9-12-301). Charlottesville, VA: Michie.
Boneau, C. A. (1960). The effects of violations of assumptions underlying the t test. Psychological Bulletin, 57, 356-370.
Bradley, M., Green, N. M., Jones, D. E., Lynn, M.. & McNeil, L. (1990). Churches and church membership in the United Stales. Atlanta, GA: Glenmary Research Center.
Bumpass, L. L., & Sweet, J. A. (1972). Differentials in marital instability: 1970. American Sociological Review, 37, 754-766.
Cohen, J. (1988). Statistical power analysis for the behavioral sciences. Hillsdale. NJ: Lawrence Erlbaum.
Cook, T. D., & Campbell, D. T. (1979). Quasi-experimentation: Design and analysis issues for field settings. Boston, MA: Houghton Mifflin.
Cronbach, L. J., Furby, L. (1970). How we should measure change--or should we? Psychological Bulletin, 74, 68-80.
Cutright, P. (1971). Income and family events: Marital stability. Journal of Marriage and the Family, 33, 291-302.
Feir-Walsh, B. J., & Toothaker, L. E. (1974). An empirical comparison of the ANOVA F-test, normal scores test and Kruskal-Wallis test under violation of assumptions. Educational and Psychological Measurement, 34, 789-799.
Filsinger, E. E., & Wilson, M. R. (1984). Religiosity, socioeconomic rewards, and family development: Predictors of marital adjustment. Journal of Marriage and the Family, 46, 663-669.
Glenn, N. D., & Supancic, M. (1984). The social and demographic correlates of divorce and separation in the United States: An update and reconsideration. Journal of Marriage and the Family, 46, 563-578.
Glick, P. C. (1984). Marriage, divorce, and living arrangements: Prospective changes. Journal of Family Issues, 5, 7-26.
Goode, W. J. (1962). Marital satisfaction and instability: A cross-cultural class analysis of divorce rates. International Social Science Journal, 14. 507-526.
Heaton. T. B. (1991). Time-related determinants of marital dissolution. Journal of Marriage and the Family. 53, 285-295.
Johnson, D. W., Picard, P. R., & Quinn, B. (1971). Churches and church membership in the United States. Atlanta. GA: Glenmary Research Center.
Kemper, T. D. (1983). Predicting the divorce rate--down? Journal of Family Issues. 4, 507-524.
Kurdek. L. A. (1993). Predicting marital dissolution: A 5-year prospective longitudinal study of newlywed couples. Journal of Personality and Social Psychology, 64, 221-242.
Lerner, M. J., & Lerner, S. C. (1981). The justice motive in social behavior. New York: Plenum Press.
Louisiana Code Commission (1992). West's Louisiana statutes annotated (Article 112). St. Paul, MN: West.
Martin, T. C., & Bumpass L. L. (1989). Recent trends in marital disruption. Demography, 26, 37-51.
Marvell, T. B. (1989). Divorce rates and the fault requirement. Law and Society Review, 23, 543-567.
Maryland Code Revision Commission (1991). The annotated code of the public general laws of Maryland (Section 7-103). Charlottesville, VA: Michie.
Morgan, S. P., & Rindfuss, R. R. (1985). Marital disruption: Structural and temporal dimensions. American Journal of Sociology. 90, 1055-1077.
National Center for Health Statistics (1987). Vital statistics of the United States (DHHS Publication No. PHS 91-1103). Washington, DC: U.S. Government Printing Office.
National Center for Health Statistics (1989). Vital statistics of the United States (DHHS Publication No. PHS 93-1103). Washington, DC: U.S. Government Printing Office.
Neter, J., Wasserman, W., & Kutner, M. H. (1989). Applied linear regression models. Homewood, IL: Irwin.
Parkman, A. M. (1992). No-fault divorce. Boulder. CO: Westview Press.
Peters, H. E. (1986). Marriage and divorce: Informational constraints and private contracting. The American Economic Review, 76, 437-454.
Peters, H. E. (1992). Marriage and divorce: Reply. The American Economic Review, 82, 686-693.
Phillips, R. (1988). Putting asunder: A history of divorce in Western society. New York: Cambridge University Press.
Quinn, B., Anderson, H., Bradley, M., Goetting, P., & Shriver, P. (1980). Churches and church membership in the United States. Atlanta, GA: Glenmary Research Center.
Ramsey, P. H. (1980). Exact type 1 error rates for robustness of Student's t test with unequal variances. Journal of Educational Statistics, 5, 337-349.
Richards. J. M., Jr. (1978). An ecological analysis of the impact of the U.S. Supreme Court's 1973 abortion decision. Journal of Applied Social Psychology, 8, 15-28.
Richards, J. M.. Jr. (1984). Ecology of abortions in the United States since 1973: A replication and extension. Population and Environment, 7, 137-151.
SAS Institute. (1990). SAS/STAT user's guide: GLM-VARCOMP (Version 6, Vol. 2, 4th ed.). Cary, NC: Author.
Schroeder, C. W. (1939). Divorce in a city of 100,000 population. Peoria, IL: Bradley Polytechnic Institute Library.
Sepler, H. J. (1981). Measuring the effects of no-fault divorce laws across fifty states: Quantifying a zeitgeist. Family Law Quarterly, 15, 65-101.
Trent, K., & South, S. J. (1989). Structural determinants of the divorce rate: A cross-societal analysis. Journal of Marriage and the Family, 51, 391-404.
Udry, J. R. (1966). Marital instability by race, education, and occupation using 1960 census data. American Journal of Sociology, 72, 203-209.
United States Bureau of the Census. (1950). U.S. Census of the Population (Vol. IV, Special Reports, Part 2, Chapter D, Marital Status). Washington, DC: U.S. Government Printing Office.
United States Bureau of the Census (1950-1990). Statistical abstract of the United States. Washington, DC: U.S. Government Printing Office.
Weitzman, L. J. (1985). The divorce revolution. New York: Free Press.
Whitman, L. B., & Trimble, G. W. (1956). Churches and church membership in the United States. Atlanta, GA: Glenmary Research Center.
Wright G. C., & Stetson, D. M. (1978). The impact of no-fault divorce law reform on divorce in American states. Journal of Marriage and the Family, 40, 575-580.
Zelder, M. (1993). The economic analysis of the effect of no-fault divorce law on the divorce rate. Harvard Journal of Law and Public Policy, 16, 241-267.
Paul A. Nakonezny, The University of Oklahoma, Department of Communication, 610 Elm Avenue, Room 101, The University of Oklahoma, Norman, OK 73019-0335.
Robert D. Shull, The University of Oklahoma, Department of Psychology, 455 West Lindsey Street, The University of Oklahoma, Norman, OK 773019-0535.
Joseph Lee Rodgers, The University of Oklahoma, Department of Psychology, 455 West Lindsey Street, The University of Oklahoma, Norman, OK 773019-0535.…