Below are the first 10 and last 10 pages of uncorrected machineread text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapterrepresentative searchable text on the opening pages of each chapter.
Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 64
64
The categorical treatment of micro socioeconomic variables imposes
constraints on the macr~level analysis. For one thing, it yields more
than one coeff icient for micro variables consisting of more than two
categories (RESC, WED, HED, HOCC, RES). For exploratory purposes, we
focus on a particular contrast of each micro socioeconomic effect: that
between the extreme categories. Thus we compare nonagricultural employers
with nonagricultural employees;8 the most urban residents with the most
rural; the highest educational category with the lowest. We expect
betweenrcountry variability in these coefficients because of differences
in the number and meaning of the categories used. Differences in the
extent to which the contrast between extreme categories captures the
maximum empirical contrast between categories of a particular explanatory
dimension may also introduce variability. However, neither our theory
nor the analyses presented here systematically address such variability.
The interaction terms in the LF structural equation require special
treatment. There is no single effect of respondent's education (WED) or
husband's occupation (HOCC) on LF. The partial derivatives indicate that
the effects of WED and HOCC depend on EF and son sufficiency (SCB):
(2.7a) a(1~Di, = 82,
16 + 517, 16 (EF) + 619, 16 (SCB);
a Cocci 612,16 + 818,16 (EF) + 820,16 (SCB)
To obtain single WED and HOCC coefficients for use in macrolevel
analysis, we not only take extreme categories of each, but also evaluate
the partial derivative at EF = 4 and SCB = 1. These values of EF and SCB
should magnify the potential effect of WED and HOCC on LF. EF = 4
suggests a relatively large family early in the reproductive cycle, while
SCB = 1 indicates at least two sons in that early family.
2 . 5 ILLUSTRATION OF OPE^TION=I ZATION
The purpose of this section is to illustrate and clarify the preceding
description of the micro and macro variables, and the operationalization
of the micro model. To this end, we present a comparison, based on just
two countries, that will clarify the hypotheses to be tested using the 15
WAS countries for which micro results are currently available. The
countries chosen for this purpose are Peru and Korea. The WFS data sets
for these countries are available for unrestricted scholarly use.
Moreover, the two different cultural areas involved should provide an
interesting contrast. The discussion begins with the macro comparison,
and then turns to the micro results.
Macro Comparison
Peru and Korea are both in the high category of the Mauldin et al.,
(1978) social setting index, in which the higher the rank, the more
OCR for page 64
65
advantaged the social setting: of 94 lessdeveloped countries, Peru
ranks 73 and Korea 81. Of the 15 WAS countries studied here, Peru ranks
7 and Korea 11. This indicates, among other things, that this sample of
15 countries is skewed toward the upper end of the social setting index,
and toward the upper end of socioeconomic development, a fact that will
be important in the later analysis of all 15 countries.
It is doubtful whether much importance should be attached to the
difference between the ranking of Peru and Korea on the social setting
index. Indeed, the relative position of these two countries depends to
some extent on which indicators are used. Table 2.4 reproduces the
figures used by Mauldin et al. (1978) to determine the social setting
index scores for the two countries. As the table shows, Korea had the
advantage in education, health conditions, and urbanization, while Peru
had the advantage in income and relative size of the nonagricultural
labor force. Data for Peru and Korea collected in a separate effort
(Entwisle, 1979), paint a similar picture. Table 2.5 presents these
macro data .
Although social setting differences between Peru and Korea are
relatively small, there is a major difference between the two counts ie s
in family planning effort scores. Peru did not have a program in 1972,
and thus is assigned a score of zero on the Mauldin et al. (1978) index.
Korea, on the other hand, had a very strong program at that time. Only
Singapore and China rank higher on the family planning effort scale. m e
Korean government announced the need for a family planning program in
TABLE 2.4 Macro Indicators Reported by Mauldin et al. (1978)
Korea Peru
Indicator
1960 1970 1960 1970
Percent adult literacy 71 88 61 72
Percent of 1519 yearolds in
primary or secondary 65 76 58 75
Life expectancy at birth 54 59 50 55
Infant mortality rate per 1,000 91 60 92 135
Percent of males 1564 in
nonagricultural occupat ions 31 45 37 41
GNP per capita (in 1974 U.S.$)
189 344 521 659
Percent in c ities 100, 000+ 23 38 15
OCR for page 64
66
TABLE 2.5 Socioeconomic Characteristics of Korea and Peru, circa 1965
Indicator
Korea Peru Source
GDP per capita (in 1960 U.S.S) 99 372
Percent in cities 100,000+ 27 19
Hag en and Hawrylyshyn
{1969)
Davis (1969)
Percent labor force in
agriculture 59 47 ILO Yearbook of Labour
Statistics
Life expectancy at birth 56 53
Total school enrollment 74 70
(per 100 population aged 617)
U.N. (1975)
UNESCO Statistical
Yearbook
1961, and initiated it in 1962 as part of a newly formulated national
population policy. The aim was to reduce the population growth rate from
2.9 percent in 1960 to 2.5 percent in 1966 and 2.0 percent in 1971 (Han,
1970~. Selection and training of staff occurred dur ing the 196264
period (Kim et al., 1972~. The first major push to distribute contracep
tives began in 1964 with the IUD program. The program also offered
vasectomies and condoms. In 1968, the pill was added. Abortion, though
illegal, was widely practiced. Information and education received
emphasis almost from the start. In the mid to late 1960s, these efforts
included mass media messages, home visits, Mothers' Clubs, and formal
instruction. Throughout this time, the program benefited from the
interest and support of the government: in the 196575 decade, roughly
onefifth of the nation's health budget was expended on the family
planning program (ESCAP, 1975~.
To complete this sketch of macro differences between Peru and Korea,
it is noteworthy that for the cohort studied here, mean children ever
born is about 7.4 for Peru and 5.4 for Korea. This might be expected.
For two societies of about equal socioeconomic development, the one with
the stronger family planning program should have lower fertility.
Further, if the theory outlined above is correct, Peru and Korea should
be closer in level of early fertility (EF) than in level of later
fertility (LF). In fact, the difference of .5 in mean EF is quite a bit
less than the difference of 1.4 in mean LF. The macro differences
between Peru and Korea suggest that the effects of the socioeconomic
determinants of LF should be smaller for Korea than for Peru, mainly
because of family planning program strength. mere should not be major
differences between the two countries in the socioeconomic determinants
of age at first birth (~FB) and EF. Of course, since the comparison here
involves only two countries, results inconsistent with these expectations
cannot be taken as decisive. Nevertheless, it is helpful to use the main
OCR for page 64
67
hypotheses of the theory to organize this illustrative micro specification
for Peru and Korea.
Finally, before turning to the micro results of the comparison, it
should be noted that we are not interested in ad hoc explanations of
differences between Peru and Korea, or any other pair of countries. The
micro parameters for each country will vary for historical, cultural,
geographic and other reasons having little or nothing to do with level of
socioeconomic development or strength of family planning program effort.
Some of these factors, such as breastfeeding, kinship, and cohabitation
patterns, will directly influence fertility outcomes, and we plan to
incorporate a number of these in future analyses. Even if such factors
were nonexistent, there would still be differences in what we can observe
since we are necessarily working with parameter estimates. Furthermore,
even if we envisioned enormous sample sizes, we would still be sampling
cohorts and countries. For this reason alone, we do not expect to find
strong confirmation of the theory on the basis of a comparison between
two, or even fifteen countries.
Micro Results for Peru and Korea
Tables 2.6 to 2.8 present the AFB, EF, and LF regressions, respectively,
for Peru and Korea. Our purpose in examining these regressions is to
clarify how the various socioeconomic concepts have been operationalized;
to illustrate how the quantities to be analyzed in the subsequent
analysis of the 15 WFS countries are derived; and finally, to illustrate
the implications of the theory in a simple comparison between two
countries.9
Age at First Birth (AFB)
Turning first to the AFB regressions, Table 2.6 indicates that the effect
of childhood residence (RESC) on age at first birth (AFB) is negative but
insignificant for both Peru and Korea. Our theory hypothesizes that this
effect should be negative in traditional settings and positive in tran
sitional settings. Given the positions of Peru and Korea on the social
setting index, a strong and significant negative RESC effect would thus
have been incompatible with the theory. No effect for RESC is compatible
with the theory, although we would have expected a slight positive effect
In the macro analysis, the RESC effect Is taken to be .88 for Peru
and .48 for Korea; these are net mean differences ~n LAB between child
hood residence in a city and in a rural (Peru) or village (Korea) setting
Since the question will undoubtedly arise in the course of this discus
s ion, it is perhaps worth notch here that the RESC effects for Peru and
Korea are not set to zero in the following macro regressions. Instead,
we incorporate the estimated variances of the coefficients. For example,
the city coefficient for Peru has an estimated standard error of .65 and
an estimated variance of about .40. We use the estimated variances of
the coefficients as rec. iprocal we ights in weighted regressions.
OCR for page 64
68
TABLE 2.6 Age at First Birth (AFB) Structural Equation,
Currently Married Women Aged 4044 in 1974, Peru and Korea
World Fertility Surveys
Dependent Variable is AFB
Explanatory Variables
and Other Regression
Information Peru Korea
Childhood Residence (RESC) [.0033*
[~003]*
Rural/Villagea ~ ~~
Town .11 .07
City .88 .48
Respondent's Education (WED) [.051*** [.051***
0 years/0 years ~
12/1S .62 .33
34/68 1.18 .65
56/911 1.95 1.37
7+/12+ 3 47
Religionb [b] i.0051*
Intercept 20.81 21.03C
R2 .06 .06
N 581 700
Note: The asterisks indicate the statistical significance of
selected coefficients (twotailed test):
*p >
**.01 < p ~ .1
***p ~ .01.
The absence of an asterisk indicates that a significance test was
not applied. When the explanatory dimension is a set of
categories, summary information for tests of significance
pertains to the global test that all coefficients within a
classification are simultaneously zero. Bracketed entries are
the squared multiple partial correlations associated with the
dummy variable classifications.
aWhen two category descriptions are provided, the first
pertains to Peru, the second to Korea. Dummy variables omitted
from the regressions for estimation (normalization) purposes are
indicated by a dash.
bThe coefficients for the religion categories are not presented
here. This variable is unavailable for Peru.
CThe effect of religion has been removed from this intercept so
that crossequation comparison of the intercepts is valid.
OCR for page 64
69
Considering next the effects of wife's education (WED) on AFB, we
observe that for both Peru and Korea, this effect is positive and
monotonic with increasing years of schooling. Given that both Peru and
Korea are in the high category of the social setting index, this is
consistent with the theory. Notice that the implied education slope is
much greater for Peru than for Korea. m e highest education category for
Peru is seven or more years, whereas for Korea it is twelve or more
years. Nevertheless, the coefficient for the top education category for
Peru (3.47) is very nearly the equal of that for Korea (3.77), and it is
these coefficients that we use in the macro analysis. What does this
mean? Polytomizing education, as we do here, suggests a slightly stronger
effect for Korea than for Peru. On the other hand, if we were to treat
education as a scaled variable and compute slopes for Peru and Korea, the
slope for Peru would be greater. Given the similar social setting scores
for these two countries, little difference in the slopes of micro
education effects would be expected. For reasons noted at the end of the
previous section, this apparent discrepancy will not be explored here.
It does reaffirm, however, that there is no single allpurpose coding of
education, just as there is no single allpurpose macro measure of
culture, family planning program effort, or socioeconomic development.
Hermalin and Mason (1980} consider at length the problem of, and need
for, different education scales. Our research will not proceed blindly
in the face of this problem.
The WFS collected data on neither ethnic nor religious differen
tiation in Peru, but did so for Korea. m us, the specification for Per u
does not include an ethnic or religious classif ication as one of its
explanatory dimensions, while that for Korea does. me additive effects
for religion in Korea prove to be insignificant for the determination of
AFB. In any case, regardless of their statistical significance, we do
not include ethnic or religious coefficients in the macro analysis. The
role of ethnicity/religion in this analysis is to adjust the other
coefficients in the regression. That is, we obtain a single set of
parameters to be studied at the macro level, controlling for ethnicity or
religion.
Since all of the discrete regressors are normalized using dummy
coding (i.e., zeroone coding), it should be noted that the intercepts
which are studied at the macro levelare always adjusted for ethnicity
when this dimension is present in the source data. To understand this
adjustment, note that the intercept for Peru is completed as Y  b2RESC2
 b3RESC3  g2WED2  g3WED3  g4WED4  g5WED5, where terms for
RESC1 and WED1 do not appear since RESC1 and WED1 are the omitted
categories in the use Of dummy coding for RESC and WED. The intercept
for Korea as derived in the regression computations includes not only the
terms indicated above in the intercept for Peru, but also the terms
{c2REL2  c3REL3), where REL1 is the omitted category in the
religious classification used for Korea. To purge this "rawer intercept
of its religion component, we simply remove the additional terms for the
religion classification. This yields the intercept for Korea presented
in Table 2.6, and makes the intercepts for Peru and Korea comparable
while still allowing for the additive effect of religion in Korea.l0
We apply this treatment to intercepts whenever we use an additive
OCR for page 64
70
religion or ethnicity classification in the AFB, EF, and LF regressions
whose coeff icients provide the quantities studied in the macro analysis.
We do this for Fiji, Guyana, Jamaica, Korea, Lesotho, Malaysia, and Sri
Lanka.
Given that the AFB intercepts for Peru and Korea are comparable,
what do they tell us? For the AFE regression, they tell us that
estimated AFB is virtually the same for those who grew up in a rural or
village environment and received no formal educationabout 21 years of
agefor both Peru and Korea. Recall that, when we regress the
intercepts on the macro variables using all 15 countries, the results
will provide the remains effects of the macro variables on AFB. Thus, the
intercepts perform double duty, and do not merely provide predicted
values for ~ particular concatenation of categories of the explanatory
variables. In terms of this second role, we hypothesize that increases
in socioeconomic development should lead to increases in the intercept of
the AFB equation. As long as both countries are placed in the high
category of the social setting index, the essential equality of the AFB
intercepts for Peru and Korea is compatible with this hypothesis.
Early Fertility (EF)
In turning next to the EF regressions, recall that the theory indicates
that EF should be least affected by socioeconomic position, and LF most
affected. Examining Table 2.7, which presents the EF regressions for
Peru and Korea, note that, as expected, none of the socioeconomic vari
ables (RESC, WED, and WBM) contributes significantly to the determination
of EF. The only included variable that affects EF is AFBan adjustment
variablewhose effect is negative, as hypothesized. The later the AFB,
the less the "exposure" between then and age 30, and hence the lower the
EF. Since we have no expectations for variability in the effect of AFB
on EF by setting, we do not attempt to interpret the difference between
the AFB coefficients for Peru and Korea.
For purposes of the macro analysis, the coefficients for the city,
highest education, and WBM dummy variables are used. As noted earlier,
we also employ the reciprocal of the estimated variance of each coeffici
ent in weighted regression. Of course, since we hypothesize that these
parameters do not depend on socioeconomic development or family planning
program strength, we hypothesize no relationship between the coefficients
and setting variability as characterized here. The only parameter in the
EF equation for which there is a macro hypothesis is the intercept, which
is expected to increase with level of socioeconomic development. That
the EF intercept is larger for Peru than for Korea is inconsistent with
our hypothesis of a positive relationship between the social setting
index and this intercept. Given that there is not much of a difference
in the social setting index for these two countries, there should be
little difference in the intercepts.ll
OCR for page 64
71
TABLE ~.7 Early Fertility (EF) Structural Equation,
Currently Married Women Aged 4044 in 1974, Peru and Rorea
World Fertility Surveys
Dependent Variable is EF
Explanatory Variables
and Other Regression
Information Peru Rorea
Childhood Residence (RESC) [.007]* [.004]*
Rural/Villagea  ~~
Town .22 .01
City .41 .19
Respondent's Education (WED) [.02]*
[.006]*
0 years/0 years  
12/15 .31 .11
34/68 .14 .04
56/911 .003 .11
7+/12+ .45 .22
AFB .36*** .28***
WBM .28* .21*
Religionb [b] [.004]*
Intercept 12.08 9.54c
R2 .55 .52
Note: The asterisks indicate the statistical significance of
selected coefficients (twotailed test):
*P _ .1
**.01 < p < .1
***p ~ .01.
The absence of an asterisk indicates that a significance test was
not applied. When the explanatory dimension is a set of
categories, summary information for tests of significance
pertains to the alohal tact chop al 1 I;; .,::_ _
classification are simultaneously zero. Bracketed entries are
the squared multiple partial correlations associated with the
dummy variable classifications.
~ ~ ~ ~ _ ~ · I_ .L = ~ ~ ~ O ~ ~ ~ 1 1 1 1 1 ~
aWhen two category descriptions are provided, the first
pertains to Peru r the second to Rorea. Dummy variables omitted
from the regressions for estimation (normalizations n''rn^~^c are
ind icated hv ~ {1A ¢:h
i  ~ z'—— A__ _ an. ~
The coefficients for the religion categories are not presented
here. This variable is unavailable for Peru.
CThe effect of religion has been removed from this intercept so
that crossequation comparison of the intercepts is valid.
OCR for page 64
72
Later Fertility (LF)
This brings us, finally, to the LF regressions for Peru and Korea,
presented in Table 2.8. Since the LF specification is complex, we first
indicate the coefficients used in the macro analysis, and then discuss
the results for Korea and Peru. The macro analysis employs the RESC city
dummy, the RES Lima dummy for Peru and city dummy for Korea, the BED 7+
dummy for Peru and 13+ dummy for Korea, the intercepts for each country,
and derived coefficients for WED and HOCC. Expressions (2.7a) and (2.7b),
given earlier, indicate how the derived coefficients are to be computed.
For WED, the highest education category is used (the dummy for 7+ years
in Peru, and 12+ years in Korea), with EF = 4 and SOB = 1. For Peru,
this gives .63  1.24 + .18 = 1.69; for Korea the corresponding
coefficient is .47 ~ .20  .25 = .52. Thus, for women who have borne
four children before age 30, of which at least two were boys, those with
the highest level of education have 1.69 fewer children in Peru and .52
f ewe r in Korea than those with no education, controlling for the other
variables in the regression. Similarly, for husband's occupation (HOCC),
we contrast employers with employees who are primarily in the nonagri
cultural sector, again with EF = 4 and SCB = 1. For Peru, this gives .84
 .76  .49 = .41, while for Korea, the corresponding value is .64  .8
 .03 = .19. Thus, for example, in Peru, women whose husbands are
employers and who have had four children before age 30, of which at least
two were boys, have .41 fewer children than employees, controlling for
the other variables in the regression. We use these derived coefficients
in the macro regressions, together with their derived estimated variances.
We may next consider how observed directions of relationships between
the socioeconomic variables and LF correspond with what the theory
hypothesizes for transitional settings, beginning with those effects not
involved in the interactions. For both Peru and Korea, the effect of
RESC on LF is negative, as expected, though not significant in the case
of Korea. The WBM effect has the correct negative sign for Peru, and is
insignificant for both countries. Work since marriage (WSM) has a
negative effect on LF in Korea. This variable is not available in the
Peruvian data. The effect of HOCC is not significant for either country.
The signs of AFB and AFB2 are precisely the opposite of expecta
tions. The effect of AFB on LF, with all other variables controlled, is
positive for AFB between 15 and 30, and is more pronounced for Peru than
for Koreae Since the effect of AEB on LF is not a direct concern of the
macro analysis (although it is present in the LF intercept), we will not
be able to explore this interesting result further here, although it will
surely 12e the subject of study as we proceed with the evaluation of the
theory.
The effect of ECM is negative instead of positive for Peru, while
positive for Korea. There are several lines of speculation that might
help to clarify the unexpected result for Peru, assuming it does not
result from a Type I statistical error. The effect of selfreported
fecundity is positive as expected and significant. This is true also for
number of marriages in Peru, but not in Korea . The ef feet of marital
duration is positive and signif icant for both countries . Me SCG dummy
variable is not signif icant. Turning now to the interactions, we have
OCR for page 64
73
TABLE 2.8 Late Fertility (LF) Structural Equation, Currently
Married Women Aged 4044 in 1974, Peru and Korea World
Fertility Surveys
Dependent Variable is LF
Explanatory Variables
and Other Regression
Information Peru Korea
Childhood Residence (RESC) [.0091** [.003]*

Rural/Villagea ~~ ~~
Town 004 ~ 04
City .61 .20
Respondent's Education (WED) [d] [d]
O years/0 years  
12/15 1.45 .18
34/68 1.74 .30
56/911 1.80 .26
7+/12+ .63 .47
Age at First Birth (AFB) [~079]*** [.009]**
Linear Component .596 .226
Quadractic Component .009 .004
WBM .37* .25*
ECM .18** .21***
EF .48d .lld
FEC .64*** .30***
NMAR .42** .21*
OUR .01** .01***
Current Residence (RES) [.0191** (.04)***
Rural/Village ~~ ~~
Town .36 .05
City .24 .63
Lima .50 (e)
WSM [e] .52**
Husband's Education (HED) [.0031* [.006]*
0 Years/0 Years ~~ ~~
12/15 .~8 .11
34/68 .34 .17
56/911 .13 .08
7+/12 .09 .25
/13+ [e] .33
OCR for page 64
74
TABLE 2.8 (continued)
Dependent Variable is LF
Explanatory Variables
and Other Regression
Information Peru Korea
Husband's Occupation (COCCI [d] [d]
Employee ~~
Selfemployed, agriculture .83 .05
Selfemployed, nonagriculture 1.79 .72
Employer .84 .64
SCB _.13d .32d
SCG .02* .002*
EF*WED Interaction [.0253*** [.0013*
EF*WED 1 ~~
EF*WED 2 .23 .01
EF*WED 3 .28 .04
EF*WED 4 .22 .04
EF*WED 5 .31 .05
EF*HOCC Interaction [.006]* [.012]**
EF*HOCC 1 ~~ ~~
EF*HOCC 2 .22 .11
EF*HOCC 3 .37 .20
EF*HOCC 4 .19 .20
SCB*WED Interaction [.002]* [.002]*
.
SCB*WED 1  
SCB*WED 2 .28 .35
SCB*WED 3 .17 .18
SCB*WED 4 .31 .12
SCB*WED 5 .18 .25
SCB*HOCC Interaction [.018]** [.000]*
SCB*HOCC 1   
SCB*HOCC 2 .33 .02
SCB*HOCC 3 .S9 .01
SCB*HOCC 4  . 49 . 03
R1 iaic~nb [b] ~ . 001] *
Intercept 8.729 2.303C
R2 .32 .32
OCR for page 64
75
TABLE 2.8 (continued)
Note: The asterisks indicate the statistical significance of
selected coefficients (twotailed test):
*P > .1
**.01 ~ p < .1
***p ~ .01.
The absence of an asterisk indicates that a significance test was
not applied. When the explanatory dimension is a set of
categories, summary information for tests of significance
pertains to the global test that all coefficients within a
classification are simultaneously zero. Bracketed entries are
the squared multiple partial correlations associated with the
dummy variable classifications.
aWhen two category descriptions are provided, the first
pertains to Peru' the second to Korea. Dummy variables omitted
from the regressions for estimation (normalization) purposes are
indicated by a dash.
he coefficients for the religion categories are not presented
here. This variable is unavailable for Peru.
CThe effect of religion has been removed from this intercept so
that crossequation comparison of the intercepts is valid.
dWED, HOCC, EF, and SCB play an interactive as well as additive
role in the LF equation.
eNot applicable.
already derived the interaction contrasts for WED and HOCC that will be
used in the macro analysis. Hey are negative as expected, with the
numerical values for Peru being twice as large as those for Korea.
However, these selected interaction contrasts are only pieces of a larger
pattern. Moreover, it is unclear what they tell us, given that for Peru,
the EF*HOCC and SCB*WED interactions are globally insignificant, while
for Korea, the EF*WED, SCB*WED, and SCB*HOCC interactions are globally
insignificant. If obtaining the bestfitting parsimonious model for Peru
and Korea were our goal' we could reestimate the model, dropping insig
nificant variables and otherwise tinkering with the spec if ication unti 1
we were satisf led. For present purposes, this will not be done. However,
instead of attempting to read off the patterns of effects directly from
Table 2.8, we will use the device of standardizing conditional coeffici
ents to remove insignificant interactions; this will yield reasonable
approximations of the effects that would be obtained if the regressions
were computed with the insignificant terms or classifications omitted.l3
To examine the EF effects on LF in Peru, we standardize over HOCC
(using sample composition for this classification) to obtain EF effects
conditional on WEDthe only significant interaction involving EF in the
Peruvian data. The estimated conditional EF coefficients are as follows:
OCR for page 64
76
WED1:
WED2:
WED3:
WED4:
WED5
.33
.56
.61
.55
.02
where WED5 signifies the highest educational level. This pattern
indicates that in Peru, the effect of EF on LF is positive over all but
the highest level of education, where it is essentially zero. This is
compatible with our hypothesis, although the effect is not as strong as
it could be. A negative EF effect conditional on WED5 would be more
compatible with expectations about the presence and implications of
familysize targets. Nevertheless, given the relatively low level of
educational attainment in Peru, this result is quite acceptable.
Considering next the effects of WED on LF, there are both EF*WED and
SCB*WED interactions, but only the former are globally significant. We
therefore standardize over SCB (assuming a 5050 split between families
with two or more boys and those with fewer), and evaluate the WED effect
at EF = 4:
WED1
WED2 =
WEDS
WED4
WED5
=
0.38
0.72
1.07
1.79
where WED5 is the highest educational category and WED1 is the omitted
category, so that the coefficients for the other categories are deviations
from zero. For EF = 4, the effect of education on LF is perfectly
monotonic and negative, as expected. m is pattern does not emerge until
EF = 3. For lower values of EF, the education effect is quadratic and
concave upward. Thus the WED5WED1 contrast conditional on EF = 4 and
SCB = 1 derived earlier, a value which for Peru is 1.69, appears to be
informative: because the effect of education is monotonic for the
conditioning categories selected, the extreme contrast between WEDS and
WED1 is meaningful.
m e effect of HOCC depends only on boy sufficiency (SCB), and not
EF, in Peru. m at the SCB*HOCC interaction is consistent with expecta
tions is clear from inspection of the terms in this interaction in Table
2.8. For those with at least two boys, nonagricultural employers and
selfemployed workers outside of agriculture have lower LF than employees,
but most importantly, lower LF than farmers. Even though the EF*HOCC
interaction is insignificant, the pattern just described is most clearly
seen for those with average or greater EF (for Peru, EF = 4.1 for the
cohort under examination). With EF = 4, the HOCC effects on LF are as
follows:
OCR for page 64
77
employee
selfemployed, agriculture
selfemployed, nonagriculture
employer
SCB = 0 SCB = 1
.06  .26
.32 .27
.09 .40
These results make it clear that for SCB s 0, there is very little
difference between occupational groups in EF, while there is a maximum
contrast of .66 between farmers and nonfarm employers among those with at
least two boys.
This same point can be made by examining the effects of SCB
conditional on HOCC and standardizing on WED, which yields the following
conditional SCB effects:
employee
selfemployed, agriculture
selfemployed, nonagriculture
employer
.13
.20
.71
.61
m us, the effect of SCB is positive among farmers in Peru and negative
outside of agriculture. Again, this is consistent with the expectations
developed in Chapter 1.
Considering next the interactive results for Korea, it is clear from
Table 2.8 that the SCB variable does not interact significantly with WED
or HOCC, using a global significance test. Therefore, we derive a net
SCB coefficient standardized for both WED and HOCC. This net value is
.48, which is in the hypothesized direction, and confirms the results of
other analyses that report on the importance of sons in Korea (Johnson
Acsadi and Weinberger, 1982). In sum, couples with a sufficiency of sons
had lower LF and, for this cohort at least, this effect did not depend on
WED or HOCC.
WED also fails to ~ nteract significantly with either EF or SCB,
using a global signif icance test. Thus, the effect of WED can be
determined net of EF and SCB. Since EF and SCB are logically related (EF
< 2 implies SCB = 0), we evaluate the effect of WED for SCB = EF = 0
and for EF = 4' but standardized on SCB (assuming a 5050 split). The
results are as follows:
EF = 4;
Standardized on SCB SCB = EF = 0
WED1
WED2
WEDS
WED4
WEDS
e 05
~ ~25
_ dg
~ 38
.18
~ ~ 3 0
~ e26
~ ~ 4 7
These results indicate that although the effect of WED on LF would be
negative if education were scaled by years of schooling completed, it
would be fairly weak. At most, there would be a difference in LF of no
OCR for page 64
78
more than Half a child, and this contrast is between vast extremes in
social positionbetween those with no formal education and those at the
upper extremes of the education distribution. Compared to the effect of
WED on fertility in Peru, the effect for Korea is weaker. This result is
consistent with expectations, given Korea's strong family planning
program.
This leaves for consideration the effects of EF and HOCC on LF.
According to the result. of the global signif icance tests for the Korean
data, EF interacts only with HOCC, and HOCC interacts only with EF;
therefore, examination of the EF effects conditional on HOCC will tell
the same story as examination of the HOCC effects conditional on EF.
Nevertheless, since HOCC is not a continuous variable, it is difficult to
infer the nature of its effect from examination of the conditional EF
effects. For this reason, we present the interaction both ways.
Since the EF*WED interaction is insignificant, we present the EF
effects conditional on HOCC and standardized for WED:
employee
selfemployed, agricultural
selfemployed, nonagricultural
employer
.10
.01
.11
.30
These results indicate a marked inverse EF effect on LF among families in
which the husband is an employer in the nonagricultural sector, but fairly
modest effects of EF for other occupational categories. This result is
moderately consistent with our hypothesis, since employers are presumably
in the more transitional sector of societies undergoing change. Moreover,
there is also a negative EF effect for the employee categorywhich is a
largely nonagricultural group. This, too, is consistent with our
hypothesis.
Whether the EF effects for the remaining occupational categories are
consistent with expectations is unclear. However, we can gain insight
into this pattern by examining its complementthe HOCC effect conditional
on EF. To do this, we standardize over SCB (assuming a 5050 split) where
sensible. We present the HOCC coefficients at two values of EF: four
children and none. Since EF = 0 implies SCB = 0, the HOCC coefficients
given these conditioning values are not standardized for SCB. When EF =
4, however, the HOCC coefficients are standardized for SCB:
EF = 4;
Standardized on SCB
employee
selfemployed, agriculture .39
selfemployed, nonagriculture .09
employer .18
EF = SCB = 0
~
~ e05
~e72
e 64
If our hypothesis about the interaction between EF and HOCC is correct,
the results must certain conform when EF is high. Since EF = 3.5 for
this cohort in Korea, the choice of EF = 4 to evaluate the conditional
HOCC effect should be adequate to determine whether the results are as
OCR for page 64
79
expected. For the two critical occupationalcategories, they are. It
appears that when EF = 4, employers have the lowest LF, and farmers have
the highest. This effect is not large, however, considering the extremes
in position. It is at most not much more than "half a child (.39 
(.18) = .57), which is the same order of magnitude as the extreme in LF
observed for WED. With EF = 0, there is a much larger difference in LF
between the categories of HOCC, but these numbers are probably not inter
pretable, given the very small cell sizes on which they are based.l4
Also, these coefficients are in fact observed regression coefficients:
as can be seen in Table 2.8, they are the coefficients of the "main
effect" of HOCC. None of them is as much as twice its standard error.
Thus, the effect of HOCC on LF emerges in the predicted pattern as EF
increases. Symmetrically, there is an inverse EF effect on LF for
occupations in the nonagricultural sector. Either way of stating the
result is compatible with our hypothesis.
The only coefficients not yet discussed in the LF structural
equations for Peru and Korea are to intercepts. Although both are
negative, the intercept for Peru is markedly larger in absolute value.
Unless we stress the importance of gross domestic product (GDP) per
capita and the relative size of the nonagricultural labor force over
other macro indicators, this result is inconsistent with expectations.
Given Korea's position relative to that of Peru on the social setting
index, and given Korea's strong family planning program effort score
compared to that of Peru, Korea should have the intercept with the larger
absolute value, given that both are negative. There are a number of
possible explanations for this result which we cannot pursue here.
TO summarize the results for the LF structural equation, for both
Peru and Korea there is a mixture of outcomessome consistent and some
inconsistent with our hypotheses. The effects of some adjustment
variables have unexpected signs, while others are as predicted. m e
effects of the socioeconomic factors are perhaps more consistent with our
hypotheses, given that departures from expectations are largely in the
failure of effects to materialize rather than associations that have the
unexpected sign. Even more strongly, recall that one of our macro
hypotheses is that the effect of a strong family planning program effort
should be to attenuate socioeconomic effects on LF. The socioeconomic
effects for Korea are on the whole weaker than those for Peru, given that
there are fewer significant interactions involving socioeconomic status
for the Korean data and that those effects present are somewhat smaller.
Thus, at least with respect to the socioeconomic effects, the results for
Peru and Korea appear to be essentially consistent with the theory
developed in Chapter I.
On the whole, the results for the AFB, EF, and LF regressions are
roughly consistent with our theory. With just two countries being
studied' any pattern of results could be found without necessarily
undermining the theory, whose test must hinge on data for numerous
countries. By the same token, findings for these two countries
compatible with the theory do not necessarily confirm it since a larger
sample of countries could produce quite contrary results. The chief
value of this exercise, then, is that it illustrates the operationaliza
tion used in this initial round of investigation. Among other things, it
OCR for page 64
80
has shown, for example, that the decision to contrast employers with
employees is not as felicitous as expected. It would be preferable to
contrast employers with farmers, since the difference in LF between these
two categories is by far the largest for both Peru and Korea. It has
also shown that AFB does not affect LF in the way expected in Peru and
Korea. Since these are disparate countries, there may be a need to
review our thinking about AFB, and to closely examine the data for an
explanation for this apparent anomaly. We have also gained some insight
into potential problems with the macro tests. m e tests we will present
assume that the hypotheses about the nonsoc ioeconomic variables are
valid. We now have evidence from two countries that this is not always
the case for the LF equation. Therefore, our tests may not yield the
expected resultsnot because our reasoning about socioeconomic effects
is wrong, but because our reasoning about the adjustment and other
variables needs to be revised.
2.6 MACRO ANALYSIS
Our two fundamental hypotheses suggest a particular ordering of steps
within the macro analysis. The first basic hypothesis is that the
socioeconomic determinants of fertility affect the components of the
reproductive process differently. m e second is that the components of
the reproductive process are both additively and interactively affected
by macro phenomena. Translating this second hypothesis into the language
of structural equations and multilevel models, we hypothesize that the
intercepts, and the effects of the socioeconomic determinants, of the
components of the reproductive process vary across societies in meaning
ful and predictable ways. m e first hypothesis directs attention to
withincountry differences in the relationships between the micro socio
economic variables and each component of the fertility process. The
second directs attention to betweencountry variability in these micro
relationships.
Before undertaking macro analysis of the second hypothesis, it is
worth testing the first. If the socioeconomic determinants of the
reproductive process do not differentially affect the components of the
reproductive process, there is little point in testing the second
hypothesis.
WithinCountry Differences
To test the first major hypothesis, we summarize the contrasts, described
earlier, to be used in the macro analysis. Using the criterion that the
ratio of each coefficient to its estimated standard error must be at
least as large as 1.96, that is, using a p = 0.05 level of statistical
significance (twotailed test), we have tabulated the number of positive
and negative significant effects for each socioeconomic variable present
in the micro equations over the 15 WAS countries for age at first birth
(AFB), early fertility (EF), and later fertility (LF). Table 2.9
presents these counts and shows that the significant effects of childhood