CHAPTER 4 Mixed Evidence of the SDT in Taiwan
4.3 Family Value Orientation and Household Type in 2011
4.3.4 Regression Models for Family Value Orientation and Household Type
國
立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
76
woman’s role as caretaker, and women working outside the home affecting the family.
However, there is progress in the public sphere compared to Marsh (2004)’s study regarding men’s attitudes toward women and equal pay in the 1960s and 1990s.
Attitudes toward family role also reflect lingering traditional values. Large proportions of respondents still believe it is important to prioritize parents’ expectations over one’s own desires, to live with parents after marriage, to support parents’
livelihoods, and to have a son to carry on the family name.
For attitude toward marriage, over half of respondents believing children are necessary in a marriage gives support to claims that marriage and childbearing are still tightly linked in Asia. Over half of respondents also disagreed that couples could live together without planning to get married, reflecting the taboo nature of cohabitation.
However, support for divorce or remaining unmarried has clearly increased since over half of respondents disagreed that divorce should wait until children are grown or that a bad marriage is better than no marriage.
4.3.4 Regression Models for Family Value Orientation and Household Type As described in Chapter 3, family survey data was first analyzed using a standard multinomial logistic regression to determine the relationship between overall family value orientation and household type. First, a limited regression model with only the independent and dependent variables was run. Second, age, gender, education level, and income were included as control variables in the full model. The age range for both models was restricted to ages 30 to 59. The reference category of the dependent variable was automatically set as the most common household type, Mar+ (Married with
children). For overall value orientation “neutral” was chosen as the reference category, for gender “female” was set as the focus category. Age was input as a continuous variable. For education and income, the most common categories were chosen. For education that category was “high school,” and for income that category was “30,000-59,000.”
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
77
Results of the limited multinomial logistic analysis indicated that the
one-predictor model provides a statistically significant prediction of household type. The -2 Log Likelihood = -1411.8367, χ2 (15, N = 1074) = 42.74, p = 0.0002. McFadden’s pseudo R2 indicated that the model accounted for about one percent of total variance.
The results of regression model for overall family value orientation in 2011 are reported in Table 4.9.
Results of the full multinomial logistic analysis indicated that the five-predictor model provides a statistically significant prediction of household type. The -2 Log Likelihood = -1260.4205, χ2 (70, N = 1074) = 345.57, p < 0.0001. McFadden’s pseudo R2 indicated that the model accounted for about 12 percent of total variance. For all independent variables posttests for multicollinearity reported variance inflation factors of less than two.
As can be seen by the significant exponentiated coefficients of the full model in Table 4.9, in 2011 respondents were 1.6 times more likely to live in a Mar0 household than Mar+ if they had a moderately nonconformist family value orientation compared to neutral. Women were less likely than men to live in a Mar0 household than Mar+.
Respondents with a moderately nonconformist family value orientation were 3.5 times more likely to live in ForM0 household than Mar+. Respondents were 2.6 times more likely to live in a ForM+ household than Mar+ if their income was under 30,000, a quarter as likely if their income was 90,000-119,999, and much less likely if their income was over 120,000, compared to incomes of 30,000-59,999.
‧
Table 4.9: Multinomial Logistic Regression Results for Overall Family Value Orientation for Family Module Survey in 2011
Exp (B) Junior College 0.821
(0.185) Note: The dependent variable was household type with married with children as the reference category;
the independent variable was family value orientation with neutral as the reference category; for control variables, females were the focus of the gender variable; high school was the reference category for education level, and 30,000-59,999 was the reference category for income.
*p<0.05
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
79
Respondents were 2.6 times more likely to live in a Single household than Mar+
if they had moderately nonconformist family value orientations compared to neutral.
Women were about half as likely to live in a Single household than Mar+ and each year increase in age also reduced the likelihood of living in a Single household. Respondents with a university level education were two times more likely to live in a Single household compared to those with high school educations. Respondents with “other” income were 2.4 times more likely to live in a Single household compared to Mar+ and those with an under 30,000 income level were 3.6 times more likely, compared to those with a 30,000-59,999 income level. Finally, respondents with a moderately nonconformist family value orientation compared to neutral were six times more likely to live in a Coh household than Mar+.
Using Wald tests, each independent variable for the 2011 limited and full models was tested if its set of coefficients differed significantly from zero. For the limited model, results for overall family value orientation were significant (χ2 (15) = 40.80, p = 0.0003). For the full model, results for gender, age, and income were significant (χ2 (5) = 22.64, p = 0.0004; χ2 (5) = 114.00, p < 0.0001; and χ2 (25) = 58.56, p = 0.0002
respectively), but results for education and overall family value orientation were not significant (χ2 (20) = 20.83, p = 0.4074; χ2 (15) = 22.94, p = 0.0855 respectively).
Since in the regression model for overall family value orientation found that overall family values were not significant with control variables added to the model, a second standard multinomial logistic regression was run on family survey data with value orientation split into three independent variables: gender role values, family role values, and attitude toward marriage, with all other variables remaining the same. The purpose of the second regression model was to see if any individual category of family values was significant even though overall family values were not.
In the additional regression model, all three value orientation variables used
“neutral” as the reference category and all other variables were set up the same as in the first model. Results of the multinomial logistic analysis indicated that the five-predictor model provides a statistically significant prediction of household type. The -2 Log Likelihood = -1227.9331, χ2 (115, N = 1074) = 410.55, p < 0.0001. McFadden’s pseudo
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
80
R2 indicated that the second 2011 model accounted for about 14 percent of total variance.
The results of the regression model for family value orientation broken into categories is reported in Table 4.10. For all independent variables in both models, posttests for multicollinearity reported variance inflation factors of less than two.
As can be seen by the exponentiated coefficients in Table 4.10, gender, age, education, and income have a similar effect on household type as in the first 2011 model.
For values, respondents with a nonconformist rather than neutral attitude toward marriage are five times more likely to live in a ForM0 household than Mar+. Respondents with a moderately nonconformist attitude toward marriage were twice as likely to live in a Single household than Mar+, and those with nonconformist attitudes toward marriage were six times more likely, compared to those with neutral attitudes. Those with moderately conformist gender role values were 12 times more likely to live in a Coh household than Mar+, while those with moderately nonconformist gender role values were 7.6 times more likely, compared to those with neutral values. Those with
conformist family role values rather than neutral were 9.7 times more likely to live in a Coh household than Mar+. Finally, those with moderately nonconformist attitudes toward marriage were five times more likely to live in a Coh household than Mar+, while those with nonconformist attitudes toward marriage were twenty-one times more likely, compared to those with neutral values.
Using Wald tests, each independent variable for the second 2011 model was tested if its set of coefficients differed significantly from zero. Results for gender, age, income, and attitude toward marriage were significant (χ2 (5) = 22.96, p = 0.0003; χ2 (5) = 107.49, p < 0.0001; χ2 (25) = 56.79, p = 0.0003, and χ2 (20) = 36.45, p = 0.0136
respectively), but results for education, gender role value orientation, and family role value orientation were not significant (χ2 (20) = 21.31, p = 0.3789; χ2 (20) = 12.44, p = 0.9001; χ2 (20) = 18.73, p = 0.5396 respectively).
‧
Table 4.10: Multinomial Logistic Regression Results for Gender Roles, Family Role, and Attitude toward Marriage Value Orientations for Family Module Survey in 2011 Junior College 0.791
(0.180)
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
82
(0.262) (0.000) (0.154) (0.191) (0.719)
120,000 and Over 1.322 (0.339)
0.572 (0.475)
0.072*
(0.077)
1.036 (0.358)
2.836 (2.599) Note: The dependent variable was household type with married with children as the reference category;
the independent variables were gender role values, family role values, and attitude toward marriage, with neutral as the reference categories; for control variables, females were the focus of the gender variable;
high school was the reference category for education level, and 30,000-59,999 was the reference category for income.
*p<0.05
In summary, when looking at overall family value orientation in 2011, moderately nonconformist value orientations rather than neutral made respondents more likely to live in Mar0, ForM0, Single, or Coh households than Mar+. When looking at these relative likelihoods, nonconformist values make respondents less likely to live in a traditional household in which respondents are married with children. However, in the follow-up test for significance across all equations in the model, overall family value was not significant given control variables, while gender, age, and income level were.
When looking at family values broken into categories, moderately conformist and moderately nonconformist gender role values rather than neutral make respondents more likely to cohabitate rather than be married with children. Conformist family role values rather than neutral also made respondents more likely to cohabitate. For cohabitators with more conformist values, it is possible they are avoiding the obligations and burdens of a marriage but still desire a partnership, as predicted by theories regarding rigid family values. However, in follow-up tests for significance for all equations in the model, neither gender role values nor family role values were significant.
Attitude toward marriage was significant in follow-up tests however. Moderately nonconformist marriage attitudes made respondents more likely to be single or cohabitate while nonconformist marriage attitudes made respondents more likely to be formerly married with no children, be single, or to cohabitate. In this model, gender, age, and income level were also significant.
The next chapter will present the findings of in-depth interviews with Taiwanese women about their life course choices and will give additional insight to the reasons behind their demographic behavior.
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
83