An Evolutionary Interpretation of Fertility Decline:
New Evidence
Stephen K. Sanderson
Indiana University of Pennsylvania
This article presents new evidence that partly reinforces and partly qualifies the
results of a recent article on fertility decline published in this journal by Sanderson
and Dubrow. Eight panel regression analyses were carried out, four for the period
between 1960 and 1990 and four more for the period of the original demographic
transition, that between 1880 and 1940. The analyses for the 1960­1990 period
show that Sanderson and Dubrow's original conclusion that infant mortality decline
was causing fertility decline (rather than the reverse) was correct. On the other
hand, Sanderson and Dubrow's conclusion that enhanced female empowerment
led to fertility decline proved incorrect. The new analyses reported here show that
the reverse was in fact the case: women became more empowered as a result of
declining fertility. The panel analyses carried out for the 1880­1940 period showed
that infant mortality decline seemed to be an important cause of fertility decline
between 1880 and 1910 but not between 1910 and 1940. However, the reverse
hypothesis--that fertility decline caused infant mortality decline during this period--was
falsified. I conclude that the causes of fertility decline in the modern
world may be different, at least to some extent, from those in the original demographic
transition. This is an unsatisfying (because unparsimonious) result that suggests
the need for more research.
KEY WORDS: fertility; mortality; evolutionary; empowerment.
In an article published recently in this journal (Population and Environment
21:511­37, 2000), Sanderson and Dubrow presented a series of multiple
regression analyses in which they tested three theories of fertility decline:
the argument that it is the declining economic value of children's
Please address correspondence to Stephen Sanderson, Department of Sociology, Indiana
University of Pennsylvania, Indiana, PA 15705.
Population and Environment: A Journal of Interdisciplinary Studies
Volume 22, Number 6, July 2001
 2001 Human Sciences Press, Inc. 555
556
POPULATION AND ENVIRONMENT
labor with industrialization that causes people to produce fewer children,
the evolutionary argument that fertility levels are adjusted primarily to levels
of infant mortality, and the notion that female empowerment is the critical
factor in fertility decline. Sanderson and Dubrow conducted two studies.
The first was a study of fertility and fertility decline in the period
between 1960 and 1990, whereas the second looked at the original demographic
transition (the period between 1880 and 1940). Their results
showed that infant mortality was the best predictor of fertility levels, that
female empowerment was an important secondary predictor, and that the
economic value of children's labor (as measured by percentage of the labor
force in agriculture or in manufacturing and the percentage of the population
living in urban areas) was negligibly related to fertility levels. They
concluded that both the evolutionary and the female empowerment theories
were supported, but that the evolutionary argument was empirically
stronger.
However, Sanderson and Dubrow's study contained one serious weakness:
it was unable to show statistically that both infant mortality decline
and female empowerment were causal to fertility and fertility decline rather
than merely correlated with it. This article represents an effort to overcome
this deficiency by carrying out a series of panel regression analyses. I will
first concentrate on the most important predictive variable, infant mortality,
and then turn to assess the causal relationship between female empowerment
and fertility.
Panel analysis involves regressing a dependent variable measured at
one time period on a series of independent variables measured at an earlier
time period, while simultaneously controlling for the earlier value of the
dependent variable. Eight panel analyses were carried out. The first such
analysis (Table 1) regressed 1990 fertility levels on 1960 levels of the independent
variables while controlling for 1960 fertility. The results show
strong support for the argument that the direction of causation is from infant
mortality to fertility. All of the beta coefficients except those for infant mortality
and 1960 fertility were small and statistically nonsignificant. The beta
for infant mortality was the largest, and it was substantially larger than that
for 1960 fertility (.586, p < .002 versus .305, p < .042). This is an extremely
important finding, because we usually expect an earlier level of a variable
to be the strongest predictor of a later level of that variable. That this was
not the case suggests a strong causal role for infant mortality. Pairwise and
mean substitution analyses did not invalidate these results. The outcome
was essentially the same for mean substitution. The beta for infant mortality
was .497 (p < .000) and that for 1960 fertility was .280 (p < .000). The pairwise
results showed an even bigger causal effect for infant mortality (beta
557
STEPHEN K. SANDERSON
TABLE 1
Effects of 1960 Levels of Gross National Product, Percentage of the
Labor Force in Agriculture, Percentage Urban, Female
Empowerment, Infant Mortality, and Fertility on Fertility in 1990
ZeroVariable
order Partial St. Beta t P < VIF
LogGNP 1960 -.794 -.026 -.042 -.191 .849 10.658
% Lab. force agri. 1960 .773 -.050 -.094 -.360 .720 14.850
% Urban 1960 -.742 -.114 -.174 -.825 .413 9.658
Female empowerment 1960 -.620 .128 .111 .931 .356 3.052
Infant mortal. rate 1960 .853 .412 .586 3.265 .002 6.946
Fertility 1960 .782 .278 .305 2.085 .042 4.627
R = .871
R2
= .759
Adj. R2
= .731
N = 59
= .735, p < .000), whereas the effect for 1960 fertility was about the same
as in the listwise and mean substitution analyses (beta = .334, p < .003).
A second panel analysis added considerable support to the results of
the first panel analysis. Here I made 1990 infant mortality the dependent
variable and regressed it on 1960 levels of the independent variables while
controlling for 1960 infant mortality (Table 2). Here we see that infant mortality
in 1960 was overwhelmingly the best predictor of infant mortality in
1990 (beta = .832, p < .000), and that fertility in 1960 did not predict infant
mortality in 1990 (more accurately, did not predict changes in infant mortality
between 1960 and 1990) (beta = -.103, p < .428). In fact, the beta
coefficient for fertility in 1960 was negative, and thus running in the opposite
direction. When the analyses were run pairwise and with mean substitution
they were essentially the same. For pairwise deletion, 1960 infant
mortality had a beta of .705 (p < .000) and 1960 fertility a beta of only
-.153. For mean substitution, the 1960 infant mortality beta was .630 (p <
.000) and the 1960 fertility beta was only -.095.
Now we need to consider the causal relationship between female empowerment
and fertility decline. I carried out two panel analyses to test it.
Table 3 shows the regression of 1990 fertility levels on 1960 levels of GNP,
percentage of the labor force in agriculture, percentage urban, and female
empowerment, while at the same time controlling for 1960 fertility levels.
558
POPULATION AND ENVIRONMENT
TABLE 2
Effects of 1960 Levels of Gross National Product,
Percentage of the Labor Force in Agriculture, Percentage Urban,
Female Empowerment, Infant Mortality, and
Fertility on Infant Mortality in 1990
ZeroVariable
order Partial St. Beta t P < VIF
LogGNP 1960 -.837 -.086 -.121 -.621 .537 10.658
% Lab. force agri. 1960 .806 .019 .032 .138 .891 14.850
% Urban 1960 -.767 -.090 -.122 -.654 .516 9.658
Female empowerment 1960 -.549 .175 .134 1.282 .205 3.052
Fertility 1960 .676 -.110 -.103 -.799 .428 4.627
Infant mortal. rate 1960 .885 .591 .832 5.278 .000 6.946
R = .902
R2
= .814
Adj. R2
= .792
N = 59
TABLE 3
Effects of 1960 Levels of Gross National Product, Percentage of the
Labor Force in Agriculture, Percentage Urban,
Female Empowerment, and Fertility on Fertility in 1990
ZeroVariable
order Partial St. Beta t P < VIF
LogGNP 1960 -.794 -.270 -.421 -2.044 .046 7.752
% Lab. force agri. 1960 .773 .006 .013 .046 .964 14.616
% Urban 1960 -.742 -.047 -.077 -.339 .736 9.467
Female empowerment 1960 -.620 .106 .100 .774 .442 3.050
Fertility 1960 .782 .400 .473 3.178 .002 4.053
R = .843
R2
= .710
Adj. R2
= .68
N = 59
559
STEPHEN K. SANDERSON
The results clearly show that female empowerment had little causal influence
on fertility decline between 1960 and 1990. The beta for female empowerment
was only .100, whereas that for 1960 fertility was .473 (p <
.002). The results were essentially the same when the analysis was run
pairwise (N = 63; female empowerment beta = -.162, 1960 fertility beta =
.398) or by mean substitution (female empowerment beta = -.097, 1960
fertility beta = .362).
An additional panel analysis was run in which 1990 levels of female
empowerment were regressed on 1960 levels of GNP, percentage of the
labor force in agriculture, percentage urban, and fertility, while simultaneously
controlling for levels of female empowerment in 1960. This analysis
was designed to determine whether the causal relationship was running in
the opposite direction, i.e., whether it was fertility decline that caused an
enhancement in female empowerment. The results (Table 4) show that this
was indeed the case. In this analysis, the beta for 1960 fertility was substantial
(-.521, p < .000), whereas that for 1960 female empowerment, which
should have been the best predictive variable, was astonishingly low (.068).
When this regression was run pairwise and with mean substitution, the
results were weaker but they still support the argument that it was fertility
decline that caused female empowerment rather than the other way
around. For pairwise deletion, the beta for 1960 fertility was -.346 (p <
TABLE 4
Effects of 1960 Levels of Gross National Product, Percentage of the
Labor Force in Agriculture, Percentage Urban,
Female Empowerment, and Fertility on
Female Empowerment in 1990
ZeroVariable
order Partial St. Beta t P < VIF
LogGNP 1960 .846 .394 .504 2.640 .012 9.535
% Lab. force agri. 1960 -.799 -.042 -.067 -.259 .797 17.640
% Urban 1960 -.747 -.137 -.183 -.853 .399 12.002
Female empowerment 1960 .785 .098 .068 .604 .549 3.332
Fertility 1960 -.878 -.577 -.521 -4.359 .000 3.732
R = .924
R2
= .855
Adj. R2
= .836
N = 44
560
POPULATION AND ENVIRONMENT
.002) whereas the beta for 1960 female empowerment was .275. In the
case of mean substitution, the beta for 1960 fertility was -.295 (p < .000)
and the beta for 1960 female empowerment was .211.
In sum, the panel regression analyses carried out here show that infant
mortality was not only by far the best predictor of fertility, but also that it
was infant mortality decline that determined fertility decline rather than the
other way around. These results thus support the evolutionary argument
which emphasizes that people are adjusting their fertility levels primarily to
the levels of infant mortality, and to some extent child mortality, that prevail
in their society. Contrary to what was argued in the original article, the
results also undermine the argument that it was enhanced female empowerment
that caused lower fertility. The two variables are clearly related, but
female empowerment seems to have been a result of fertility decline rather
than one of its causes. It appears that, with their infant and child care burdens
lessened, it is easier for women to seek more education and to enter
the labor force, two of the key dimensions of female empowerment. Female
empowerment is thus still an important part of the story, even if not in the
way that Sanderson and Dubrow originally thought.
In their earlier analysis of the demographic transition period, Sanderson
and Dubrow found that infant mortality was easily the best predictive
variable for 1880, 1910, and 1940. But in order to determine which way
the causal arrows were pointing, I performed four panel analyses. In the
first, 1910 fertility levels were regressed on 1880 levels of infant mortality,
percentage of the labor force in agriculture, and percentage of the labor
force in manufacturing, while at the same time controlling for 1880 fertility
levels. Running the analysis with listwise deletion of missing cases (Table
5) suggests that infant mortality was not an important causal variable. Here
the beta coefficient for infant mortality was very low (.134), and most of
the fertility decline between 1880 and 1910 was explained by fertility levels
in 1880 (beta = .744, p < .000). However, when the analysis was run pairwise
and with mean substitution the results were much better. In the pairwise
analysis (N = 17), 1880 infant mortality emerged as the most important
cause of the 1880­1910 fertility decline. Its beta was .503 (p < .009),
whereas the beta for 1880 fertility was only .386. Mean substitution produced
very similar results (infant mortality beta = .518, p < .001; fertility
1880 beta = .422, p < .004). Thus, two of the three panel analyses showed
infant mortality decline to be an important causal variable for the 1880­
1910 period.
A second panel analysis was run for the 1910­1940 period (Table 6).
Here 1940 fertility levels were regressed on 1910 levels of the independent
variables, while simultaneously controlling for 1910 levels of fertility. The
561
STEPHEN K. SANDERSON
TABLE 5
Effects of 1880 Levels of Infant Mortality, Percentage of the
Labor Force in Agriculture, Percentage of the Labor Force
in Manufacturing, and Fertility on 1910 Fertility
Variable Zero-order Partial Stand. Beta t P <
Infant mortal. rate 1880 .594 .334 .134 1.227 .243
% L.F. in agri. 1880 .528 .487 .280 1.930 .078
% L.F. in manufac. 1880 -.298 -.173 -.252 -.608 .554
Fertility 1880 .887 .895 .744 6.933 .000
R = .956
R2
= .914
Adj. R2
= .885
N = 17
results show that infant mortality was a very poor predictor (beta = .011,
p < .964), and they only improved modestly when the analyses were run
pairwise and with mean substitution (pairwise, infant mortality beta = .273,
p < .371; mean substitution, infant mortality beta = .183, p < .388). In all
three regression analyses for this period, fertility in 1910 was easily the best
TABLE 6
Effects of 1910 Levels of Infant Mortality, Percentage of the
Labor Force in Agriculture, Percentage of the Labor Force in
Manufacturing, and Fertility on 1940 Fertility
Variable Zero-order Partial Stand. Beta t P <
Infant mortal. rate 1910 .519 .014 .011 .047 .964
% L.F. in agriculture 1910 .419 -.230 -.252 -.784 .450
% L.F. in manufact. 1910 -.529 -.354 -.401 -1.254 .236
Fertility 1910 .793 .637 .713 2.742 .019
R = .823
R2
= .678
Adj. R2
= .561
N = 16
562
POPULATION AND ENVIRONMENT
predictor of fertility in 1940, and percentage of the labor force in manufacturing
emerged as the second best predictor.
Two final panel analyses were performed in order to test for the possibility
that fertility decline may have been producing infant mortality decline.
Table 7 shows the regression of 1910 infant mortality levels on 1880
levels of the independent variables while controlling for 1880 levels of infant
mortality. It is clear that fertility decline was not causing infant mortality
decline during this period. The beta for 1880 fertility was only .146 (p
< .350), whereas that for 1880 infant mortality was a very large .772 (p <
.000). Extremely similar results were obtained with pairwise deletion and
mean substitution of missing cases (pairwise, 1880 fertility beta = .144, p <
.239, 1880 infant mortality beta = .802, p < .000; mean substitution, 1880
fertility beta = .166, p < .086, 1880 infant mortality beta = .832, p < .000).
Basically the same results were produced for the 1910­1940 period
(Table 8). The best predictor of infant mortality in 1940 was infant mortality
in 1910 (beta = .807, p < .004). Not only was fertility in 1910 not causing
infant mortality decline between 1910 and 1940, but its sign was reversed
(beta = -.377). Much the same pattern was found when the analysis was
performed pairwise and with mean substitution (pairwise, 1910 fertility beta
= -.119, p < .596, 1910 infant mortality beta = .851, p < .001; mean substitution,
1910 fertility beta = .212, p < .267, 1910 infant mortality beta =
.661, p < .001).
All four panel analyses for the original demographic transition period
show mixed and inconclusive results. Infant mortality decline did appear
TABLE 7
Effects of 1880 Levels of Fertility, Percentage of the Labor Force in
Agriculture, Percentage of the Labor Force in Manufacturing,
and Infant Mortality on 1910 Infant Mortality
Variable Zero-order Partial Stand. Beta t P <
Infant mortal. rate 1880 .880 .825 .772 5.062 .000
% L.F. in agri. 1880 .307 .261 .190 .938 .367
% L.F. in manufac. 1880 -.041 .002 .001 .007 .994
Fertility 1880 .646 .270 .146 .973 .350
R = .912
R2
= .832
Adj. R2
= .776
N = 17
563
STEPHEN K. SANDERSON
TABLE 8
Effects of 1910 Levels of Fertility, Percentage of the Labor Force in
Agriculture, Percentage of the Labor Force in Manufacturing,
and Infant Mortality on 1940 Infant Mortality
Variable Zero-order Partial Stand. Beta t P <
Infant mortal. rate 1910 .675 .737 .807 3.616 .004
% L.F. in agri. 1910 .496 .203 .213 .687 .506
% L.F. in manufac. 1910 -.553 -.354 -.387 -1.257 .235
Fertility 1910 .434 -.414 -.377 -1.506 .160
R = .837
R2
= .701
Adj. R2
= .592
N = 16
to be an important cause of fertility decline between 1880 and 1910 but
not between 1910 and 1940. However, I have eliminated the reverse hypothesis
for both periods: that the high correlations between infant mortality
and fertility--.814 in 1910 and .567 in 1940--are due to the impact of
fertility on infant mortality. Also, there is perhaps slightly more support here
for the role of the changing economic value of children's labor than was
the case for the 1960­1990 period in that the percentage of the labor force
in manufacturing seemed to be exerting some effect on fertility decline during
the original demographic transition. This means that, at least to some
extent, the causes of fertility decline in the modern world appear to be
different from those in the original demographic transition. This is a rather
unsatisying result because it does not give us a highly parsimonious explanation.
Is this another instance of a "beautiful theory being killed off by an
ugly fact"? Perhaps. My provisional--and, to this point, only justified--
conclusion is that we have learned something important and trustworthy
about the causes of fertility decline in the modern world, but we have a
muddier picture concerning the original demographic transition. More research
with better data, and perhaps a larger number of time periods, is
clearly needed.
ACKNOWLEDGMENTS
I am grateful to Arthur Alderson, Alex Heckert, and Christopher ChaseDunn
for their assistance with the statistical analyses.