-.033* -.005 -.006 .200? .321** .323** 450.773 (106) 0.995 0.055 .010 .285** 456.076 (110) 0.994 0.054
(.013) (.014) (.014) (.113) (.104) (.098)
.085 -.011 -.013 .077 .135 .145
-.048* .030 .044 -.047 -.260** -.162** 678.769 (208) 0.991 0.046 .004 -.155** 690.415 (212) 0.990 0.046
(.021) (.030) (.044) (.067) (.067) (.062)
-.072 .043 .059 -.032 -.180 -.120
Note: All analyses control for spouse's education,durationof marriage,numberof children, spouse's employmentstatus, respondentrace, and respondentgender.N = 1,065 (537 women and 528 men). CFI = ComparativeFit Index; RMSEA = Root Mean Square of Approximation. tp < .10. *p < .05. **p < .01.
tween job satisfaction and marital satisfaction were constrainedto be equal at all three time points, changes in job satisfactionwere not sigrelatedto changesin marital satisfaction nificantly over time. The second row of Model 2 corresponds to our second hypothesisregardingspillover from maritalsatisfaction job satisfaction. to This coefficient was positive and significant,insatisfaction were dicatingthatincreasesin marital significantlyrelatedto increasesin job satisfaction. This is consistentwith the notionof positive spillover.This model also fit the datawell, as indicatedby goodness-of-fit statistics. We compared differencein X2valuesto inthe dicate whetherthe fit of the model was signifi-
cantly improvedby allowing the paths to vary across time points. The X2 statistic for Model 1 wherethe pathswereallowedto varywas 450.773 (df = 106), andthe X2statisticfor Model2 where the paths were constrainedwas 456.076 (df = 110), resultingin a differenceof 5.303 (df = 4). This differencewas not statistically insignificant, that little informationwas lost in condicating strainingthe pathsto equality.Thus we can have confidencein the overallpattern positive spillof over from maritalsatisfaction job satisfaction. to To test the stabilityof these findings,we alteredthe model in severalways. Becausejob satisfactionwas an observedindicatorthat was presumedto be measured withouterror, because and
492 marital satisfaction a latentconstruct which was in measurement errorwas taken into account,it is resultsthatdo not possible to arriveat artifactual reflectthe truerelationship betweenthe variables (Lorenzet al., 1995). To investigatethis possibility, we tested the model shown in Figure 1 using a modifiedindicator job satisfaction of with 20% measurement errorassumed.We also reevaluated the modelallowingthe errortermsfor marital satisfactionandjob satisfaction be correlated, to providing a more stringent test of the reciprocal paths.None of these changesto the model altered the substantive findingsshownin Table2, providevidence thatthese relationships were ing further stable. * Maritaldiscordandjob satisfaction. results The of parallel analyses of the relationship between maritaldiscordandjob satisfactionare presented in the bottom panel of Table 2. The first three rows of Model 1 addressour first hypothesisof to spilloverfromjob satisfaction maritaldiscord. Increasesin job satisfactionfrom 1980 to 1983 were significantlyrelated to declines in marital discord by 1983, consistent with the notion of positive spillover.However,changesin job satisfaction and changes in maritaldiscord were not relatedat the othertwo time points. significantly The second three rows in Model 1 provide evidencerelatedto...