The third research question asked: What effects do strategy, household/gender, length of marriage, and pre-divorce total gross income have on post-divorce income-to-needs ratios? An overall repeated measures Multivariate Analysis of Variance (MANOVA) (Howell, 1997; Stevens, 1986) was used to evaluate the differences in the six levels of the dependent variable, represented by the post-divorce female and male income-to-needs ratios using six strategies for calculating incomes.
The two within unit factors were strategies for calculating income (actual court order plus five strategies based on different strategies), and household/gender, while the between unit factors were pre-divorce gross income level, and length of marriage. The SPSS statistical package was used to conduct the analyses. A repeated measures MANOVA is the appropriate statistical procedure when the research design has two or more independent variables, with both between and within unit factors, and a dependent variable, with several levels, that are at least interval scale.
The repeated measures MANOVA explored the main effects and interactions among the independent variables. The design incorporated repeated measures because each of the six strategies for distributing incomes was applied to each household in the sample. Because the same households are used for each strategy, intercorrelations were expected from one strategy of allocating income to the next (Howell, 1997).
Howell explained that repeated measure designs can partial out effects that cause dependence and reduce overall variability, by using a common subject pool for all treatments, while simultaneously removing subject differences from the error term, which keeps the error components independent from treatment to treatment or cell to cell. The assumption of sphericity is a requirement of the repeated measures MANOVA, that when violated, means that the F ratio is positively biased, creating false rejections of the null hypothesis that occur too often (Stevens, 1986).
In other words, violations of sphericity make it more likely that researchers will find a significant relationship that was not there. Although the repeated measures MANVOA is robust to violations of sphericity (Stevens, 1986), when it is violated, the multivariate approach (Wilks Lambda), or the appropriate nonparametric test (Friedman), or a correction to reduce the degrees of freedom to correct the univariate approach, such as the Greenhouse-Geisser should be used (Morgan & Griego, 1998).
Sphericity was violated in the current analyses, so Wilks’ Lambda, a multivariate test of within subjects effects, will be used to test significance, instead of correcting the degrees of freedom to the univariate approach (Morgan & Griego, 1998). Preliminary analyses revealed two problems with the study design. First, there were so few fathers who were primary custodians that it made the analyses unreliable to include the gender of custodian as a between subjects variable.
Including this variable created many cells with zero or very few fathers who were custodians. For this reason, gender of custodian was dropped from the analyses. Second, it was necessary to drop Singer’s formula from the analyses because there was no variability between the male and female income-to-needs ratios (i. e. , each household was awarded 50% of the combined total gross income).
Although Singer’s formula was dropped from the overall model, it will be discussed later to compare other strategies that did not allocate an equal share of the combined incomes. The overall model for the repeated measures MANOVA had two within subjects variables when accommodating the changes to the research design: strategy, which had five levels, and household/gender, which had two levels (male and female). There were two between subjects factors: pre-divorce gross income, which had five levels, and length of marriage which had four levels.