CHAPTER 6 A COMPARATIVE ANALYSIS OF INTER BIRTH
6.11 Multivariate analysis 13*
The analysis in this section is based on intervals that were begun in the 10 years ending
two years before the survey: 1968 - 1978 in GFS and 1977-1987 in GDHS. The
resultant set of intervals is intended to minimize selection bias. In addition, the series
provided is long enough to allow examination of trends and the sample size large
enough to permit stable estimates of coefficients (see Bumpass et al., 1986). The
analysis covers the second to the sixth birth intervals. The first birth interval was
examined in Chapter 5. Intervals for births of very high orders are excluded because
they are highly biased towards intervals that were begun close to the time of the survey.
The independent variables considered show significant effects for different birth
intervals. The results for the GFS show that all the five intervals initiated in the six
years before the survey are less likely to be closed shortly than those that have been
initiated 7- 11 years before the survey. The data also show that while ethnicity appears
to be significantly related to the pace of closing the second and the fourth birth
intervals, religion is significantly associated with only the sixth.
Also, place of
residence and age at initiation of the interval are marginally significant for the fifth
birth interval while years of education is associated with the fourth. The period effect
is quite similar for the GDHS too, except for the second birth interval for which the
differential is not statistically significant. Another finding is the reduced impact of the
period effect for all the intervals relative to that found in the GFS. As with the GFS
data, different covariates are marginally related to different birth intervals.
For
example, ethnicity is associated with the second, fourth and fifth intervals; years of
education for the second and fourth intervals; while age is associated with the sixth
interval.
Table 6.4.3 gives estimates of relative risks for the second to the sixth intervals pooled
together within each survey. The data indicate that there is no difference in the risk of
having the second to the sixth births for religious and place of residence groups in both
surveys. In the GFS, the risk of closing the second to the sixth birth net of the effect of
education and period, is significantly lower for the Ewe and the Mole-Dagbani than for
the Akan group. For example, the Ewes have about 16 per cent while the Mole-
Dagbani have about 22 per cent lower risk than the Akans. However, in the GDHS,
only the Mole-Dagbani are significantly different from the Akans with about 28 per
cent lower risk.
Table 6.4.3 Estimated regression co-efficient from discrete time Cox proportional hazard model:
S u r v e y s GFS G D HS RR Z -stat RR Z -stat Covariates Ethnicity A kan 1 1 G a-A dangbe 1.017 -0.257 .997 -0.038 Ew e 0.838 -3.222 .936 -1.167 M ole-D agbani Education 0.778 -5.025 .724 -4.687 N one 1 1 1-10 years 0.884 -3 .0 7 4 .934 -1.493 11 + years Period 0.920 -.820 .736 -3.096 1 st 5 years 1 1 2nd 5 years 0.483 19.50 .743 -7.26
The effect of education on child-spacing is as expected in the GDHS. The relative risk
of having a birth shortly after the previous one is inversely related to the years of the
woman's education but the difference in the risk is only significant for women with
higher education. The GFS appears to show a curvilinear relationship with women
having no education and those having higher education being more likely to have
subsequent births soon after a previous birth compared to women having up to ten
years of education. A similar finding about the effect of education for fertility was
reported by Shah and Singh (1985) and was noted in Chapter 1.
Education is generally negatively associated with fertility. Cochrane et al. (1980)
mentioned three mechanisms by which higher educational achievement may result in
lower fertility. They noted that higher education is mostly associated with later age at
marriage which in turn leads to lower fertility. Also, more educated women tend to be
more efficient users of modem contraception, and finally education acts to change the
focus of women with respect to desired family size through changes in value systems.
This negative association appears to be true in the Ghanaian situation for the more
recent data in this analysis. This suggests that more female education, if encouraged,
can become a promising path to lower fertility in Ghana.
The period variable seems to be the most important one, influencing child-spacing net
of the effect of ethnicity and education in the 11 years preceding either survey although
its impact appeared to have lessened considerably in the GDHS. For example intervals
initiated in the six years before the GFS have about 52 per cent lower risk of being
closed early compared to those started 7 - 1 1 years before the survey. The intervals in
the six years before the GDHS have only 26 per cent lower risk of being closed shortly
after their inception.
6.12 Conclusion
This chapter presented various analyses of the birth interval data in the GFS and GDHS
data. One main objective was to compare the distributions of the lengths of the
intervals obtained in the two surveys. The analysis also sought to determine whether
the average birth interval length has been increasing over time and whether selected
background variables influence the time taken to close specific intervals. One finding
is that the average length of intervals actually closed by year of closure appears to be
increasing with time. While this may represent a genuine trend, it could also be an
artefact of the data.
The distributions of the time to closing the second to the sixth birth intervals are
generally similar in both surveys. The risk of closing these intervals is either the same
or slightly lower in the GDHS. The evidence in Table 6.1 shows that the differences in
the length of birth intervals for the younger age groups are only marginally different
between the two surveys. Finally, there is some evidence of differentials for period,
ethnic and education categories. These findings suggest that there is some increase in
child spacing in Ghana although it does not appear to be a substantial change.
In document
Reproductive change in Ghana : evidence from two national surveys
(Page 153-157)