1 Baby-Boom, Baby-Bust and the Great Depression By Andriana

Baby-Boom, Baby-Bust and the Great Depression
By Andriana Bellou and Emanuela Cardia
Université de Montréal and CIREQ
VERY PRELIMINARY DRAFT April 18, 2014
Abstract
The Baby-Boom and subsequent Baby-Bust of the 50s have shaped much of the history of the second half of the 20th century, yet
it is still largely unclear what caused them. Often cited explanations of the surge of births after WWII, are the war and a catch-up
phenomenon with the return of men from the war and/or the lasting effects of the Great Depression that created low material
aspirations and once the economy revived, larger families. This paper presents a new hypothesis based on the finding of a
persistent and robust pattern of entry in the labor market for women who were 20 to 34 years old in 1930. These women, young
and of working age during the Great Depression, had fewer children and kept entering the labor market throughout the 1940s and
1950s, maintaining a strong significant link with economic conditions during the Great Depression. We show that in the 1960s
these effects changed direction with the retirement of the older cohort, turning the baby-boom into a baby-bust.
1. Introduction
It is still unclear what caused the unprecedented increase in fertility between 1946 and 1964, the
official dates of the baby-boom: from the cohort of women born between 1906 and 1910, to the cohort born
between 1931 and 1935, fertility increased by 40%. What is equally surprising is that the spectacular
increase then evaporated within a decade. At the end of the baby-boom, the cohort born between 1946 and
1950 had an average of only 2.22 children, lower than the average for the cohort born between 1906 and
1910. Due to the long period, a study of the baby-boom and baby-bust would have to encompass several
cohorts of women, during a period that witnessed many dramatic events, from the Great Depression and
WWII, to sweeping improvements in home technologies, decreases in maternal and infant mortality and the
economic-boom of the post WWII, to name the most important. Any of these events could have increased
fertility, although only two provide a story for the subsequent baby-bust, the Great Depression and WWII,
since both were temporary events. To complicate the analysis, this period also witnessed increasing incomes
which increase the opportunity cost of raising children and have been associated with declining fertility.
Studies of the baby-boom usually focus on completed fertility, but one characteristic of the fertility
boom-bust is that births of women in different age brackets spiked at the same in the 1950s and started to
decline in the late 1950s/early 1960s. The cohort born between 1936 and 1940 had more births in 1960 when
they were 20 to 24 years old, than all other baby-boom cohorts, but overall they had fewer kids than the
previous cohort born between 1931 and 1935: 3.02 children versus 3.21 children. Something happened in
between the end of the 1950s and the beginning of the 1960s that led all cohorts, even the one that had the
most births at the end of the 1950s, to decrease births. A story of the baby-boom and baby-bust therefore
needs to explain the increase and decline in completed fertility, be consistent with the timing of the yearly
births, and also explain the brusque change at the end of the 1950s that affected all cohorts. Although there
were two official recessions between 1957 and 1958 and between 1960 and 1961, this last one was a short
recession, and overall the 1960s were a period of growth and economic prosperity. It seems unlikely that a
mild recession in 1960 could have created such a dramatic and permanent reversal. Other possible
explanations such as the introduction of the pill or divorce don’t fit the timing of the reversal as they
occurred later, the use of the pill over the course of the 1960s, divorce laws in the 1970s.1
In this paper we propose a new explanation of the baby-boom and baby-bust that fits the timing of
the increase in the yearly births, the bust and the increase in completed fertility. Our explanation links the
baby-boom and the baby-bust to the Great Depression, but via a different channel than the one proposed by
Easterlin (1961). He argued that young women who grew up during the depression had low material
aspirations and responded to the post-WWII economic recovery with renewed optimism and a desire for
larger families. This hypothesis has received less attention recently, partially because the cohort with most
1
The first pill was released in 1960 (see Bailey, 2006), but it took time between its release and its broad use, as well, in most states unmarried women under the
age of 21 could not obtain it without parental consent. Bailey (2006) shows that greater fertility control contributed to the boom in young women’s market work
from 1970 to 1990.
1
births was born between 1936 and 1940 and was therefore too young to have been directly affected by the
Great Depression. This cohort spent most of its formative years, if not all, in the 1940s. We show instead
that the Great Depression induced young married women, 20 to 34 years old in 1930 (henceforth called the
D-cohort), to enter the labor market in the 1930s, and even more massively in the following decades. Their
entry crowded-out young women in the 1950s, while their retirement in the late 1950s and 1960s, reversed
the trend and produced a bust. This explains why in a period of economic prosperity like the 1950s and
1960s we witnessed both a boom and a bust. In the 1950s, the crowding-out decreased young women’s
employment opportunities and wages, thus weakening the negative impact of the substitution effect
(working rather than staying at home); in the 1960s, the retirement of the D-cohort, 50 to 64 years old in
1960, freed positions and increased the opportunity cost of raising children. Jones and Tertilt (2006)
carefully detail and examine fertility changes over the 19th and 20th century for women born between 1826
and 1960. They find a strong negative relation between income (proxied by occupational status) and fertility,
which would explain most of the fertility decline observed in the late 19th century and early 20th century,
but does not adequately explain the baby-boom that occurred during a period of prosperity.
WWII could seem the most obvious explanation of the baby-boom, as this occurred soon after the reentry of soldiers from the war. From the official entry of the US in the war, December 1941, 16 million men
were drafted and it took 3 and half year for the war to end. This alone could have triggered a catch-up effect
and a baby-boom. But could this have been sufficient to explain the observed baby-boom, which spanned
nearly two decades, and the subsequent baby-bust? More importantly, even if delayed fertility could explain
the baby boom and bust in the 1950s and 1960s, this should not affect the number of children ever born,
while completed fertility rates increased substantially. Another puzzling fact that could make the war an
unlike candidate to be the exclusive story behind the baby boom, is that the cohorts of women that increased
fertility the most were too young during WWII to have had to catch-up their lost fertility years. It is possible
that the war affected completed fertility as well via a different channel. Doepke, Hazan and Maoz (2012)
suggest that women who entered the labor market during WWII acquired work experience and remained
afterwards, thus crowding out younger women with less experience, including the cohorts of women who
were too young to have been directly affected by the war. We show however that the cohort that was 20 to
24 year old in 1960, the one who contributed the most to the baby-boom, actually worked more in states
with higher mobilization, while the older cohort, 40 to 54 years old in 1950, worked less. These results are
consistent with estimates presented in Acemoglu, Autor and Lyle (2004), Olivetti and Goldin (2013) and
Bellou and Cardia (2013), who show that the war increased the labor force participation of young women
with higher education and did not significantly increase the overall participation of women who were 40 to
54 years old in 1950. Even if via a different channel, the Great depression rather than WWII, our results
though confirm the crowding-out hypothesis put forward by Doepke et al. (2012).
Greenwood, Seshadri and Vandenbroucke (2005) credit instead the dramatic transformations in
home production that started in the first part of the 20th century, and the rapid diffusion of modern
appliances over the 1940s and 1950s, for freeing time and increasing the demand for children. The same
mechanism however cannot explain the baby-bust, which is instead attributed to an increase in the cost of
education over the 1960s. Bailey and Collins (2011) use census county data on appliances and fertility and
annual data on the diffusion of electricity across counties to assess their impact on fertility. She shows that
the link between home technology and fertility is either negative or non-significant. Finally, recently,
Albanesi and Olivetti (2013) link the baby boom to improvements in health that significantly decreased
maternal mortality during the early part of the 20th century. Neither of these explanations can easily account
for the sudden change in yearly fertility in the late 1950s that affected all cohorts, and particularly the cohort
born between 1936 and 1940. Similarly, if the reason of the sharp decline was an increase in the cost of
raising children as in Greenwood et al., one would have to show that this increase happened suddenly and
sharply at the end of the 1950s.
This paper is divided in two parts, the first documents the entry of older cohorts of women in the
labor market during the 1940s and 1950s and the link to the Great Depression; the second examines its
impact on births and completed fertility. In the first part, we use several panels of microdata, 1920-1930,
1930-1940, 1940-1950 and 1940-1960, to examine the impact of the Great Depression and of changing
2
current economic conditions on the labor market of several cohorts of women.2 We show that the entry of
the D-cohort can be traced to the late 1920s and early 1930s and is significantly higher in states where the
depression was more severe. Interestingly, only married women increased their presence in the labor market
during this early phase. These women were between the age of 20 and 34 years old in 1930. A potential
explanation for the entry of married women in bad times is what is known as the Added Worker Effect. A
decrease in family income and credit market constraints that limit borrowing will push women not currently
in the labor market to enter the labor market to offset the loss of family income. Finegan and Margo (1994)
using census data calculate that in 1940 the labor market participation of women whose husband was
unemployed but not on work relief , was 50% higher than that of women whose husband was employed in
the private sector or in a non-PEW (Public Emergency Work) job during the census week.
Economic conditions in the 1930s had a lasting impact and decades after it affected not only women
the D-cohort but young cohorts of women, 20 to 29 years old in 1950 or 1960. Using the panels 1940-1950
and the panels 1940-1960, we find an opposite entry/exit response for old/young cohorts, with the D-cohort
entering the labor market and the young cohorts exiting in states with more commercial failures in the early
1930s.3 The negative impact on the young cohorts can only be indirect as these women were not of working
age in the early 1930s, some not even born. We also show that in states which were more severely affected
by the Great Depression female wages were lower decades later, while we find no lasting significant effects
on the employment of men, and little effects on their wages. We find a similar opposite entry/exit pattern
when we examine the impact of improvements in current economic conditions in the 1950s, with fewer
business failures leading to more entry of women in the D-cohort and to fewer young women working.
Figure 1 shows the share of women who are employed between 1930 and 1970, by age and year. As
can be seen, the employment of older cohorts, and in particular of the D-cohort, increases tremendously
throughout the decades. In the 1930s and 1940s, and earlier on, women tended to withdraw from the labor
force after marriage and not to re-enter. This changed drastically between 1940 and 1950, but the increase is
even more remarkable between 1950 and 1960, well after WWII was over. For example, 39% of the women
in the D-cohort are working in 1960, while only 18.6% in that age bracket, were working in 1940. This also
implies that a large number of women was about to retire and to exit the market by 1970. Instead, with the
exception of the very young women (18 to 22 years old), it is only between 1960 and 1970 that the
employment of young cohorts increases.
Figure 1: Share of White Women Working by Age and Year
0,6
0,5
0,4
0,3
0,2
0,1
0
70 68 66 64 62 60 58 56 54 52 50 48 46 44 42 40 38 36 34 32 30 28 26 24 22 20 18 16
work1930
2
work1940
work1950
work1960
work1970
We also use the 1920-1940 and the 1910-1960 to check robustness, the results are reported in the Appendix.
3
Also younger cohorts of married women entered the labor market between 1930 and 1940, but the link with the Great Depression weakens and
is not significant in the 1940-1950 panels.
3
While there is a theory of why married women may enter the labor market in bad times (see Finegan
and Margo, 1994), their persistent entry few decades after is surprising. We can offer some insight of why it
is plausible that for this cohort the impact of the great Depression was long lasting and may have
permanently shifted out their labor supply. This is a cohort that has lived through bad times when they were
married or of age of marriage and had young children. They may have lost their homes not yet fully paid,
their business and their life savings, their husbands may have decreased their earning potential permanently
by having been out of the labor market or unemployed for many years.
To test whether there is a link between the behavior of the D-cohort and the baby-boom and bust, we
construct a measure of the share of older women entering or exiting the labor market in 1950s and 1960s.
Specifically, to predict births in the 1950s we use the share of women in the D-cohort working in 1950
minus share of women in the same cohort working in 1940. An increase means that more women worked in
1950 than in 1940. Similarly, to predict births in the 1960s, we use the share of women in the D-cohort
working in 1970 minus share of women in the same cohort working in 1960. A decrease means that more
women retired. Finally, to predict births in 1940 we use the share of women in the D-cohort working in 1940
minus share of women in the same cohort working in 1930. This measure captures the entire life-cycle of the
D-cohort, from 1930 to 1970. If the crowding-out and crowding-in hypothesis are correct, we expect the
effects on births to be positive and significant in the 1950s and negative and significant in the 1960s. First
we show that this measure predicts a decline in the share of women 20 to 24 and 25 to 29 years old who are
working between 1940 and 1960, and an increase for the women in the same age brackets, between 1940
and 1970. This means that the constructed measures are good predictors of the probability that young
women work in the 1950s and 1960s. Second, we show that the constructed measure also predicts
significantly more births in the 1950s than in 1940, and significantly fewer in the 1960s than in 1940 – in
both panels these effects are quantitatively important and apply both to women 20 to 24 years old and to
women 25 to 29 years old in the 1950s and 1960s. The measure we constructed cannot be easily used to
examine completed fertility, so instead we examine whether it can predict higher (fewer) overall births for
women who were 24, 25 to 29 and 29 years old in 1960 (or in 1970) relatively to 1940. We find the same
results as for yearly births. The crowding-out measure predicts that these women have overall significantly
more children in 1960 than in 1940, while the crowding-in measure predicts a decline in the overall number
of children.
Our empirical strategy relies on using several panels of data for the work patterns to reduce the
possibility of picking up spurious correlation or a common factor that determines both births and work
behavior, which would invalidate our causal hypotheses that the behavior D-cohort led to the baby-boom
and to the bust, and not vice versa. In addition as a measure of economic conditions we use business failures,
which are more likely to reflect shifts in labor demand and unwanted decline in the participation to the labor
market, than a labor supply shift. Since the Great Depression could affect fertility by changing preferences
during formative years, we also examine the impact of economic events during childhood on completed
fertility (the Easterlin hypothesis). This channel would run counter the direction of the causality we propose,
suggesting that the Great Depression affected young cohorts of women in the 1950s by shifting inward their
labor supply. Numerous falsifications are used to assess whether the economic conditions are specific to the
Great Depression and not business cycle trends. There are many cohorts that enter and exit the labor market
at the same time, picking one, even if large, could lead to overestimating their power to crowd-out or crowdin young cohorts. For this reason we also examine the crowding-out and crowding-in effects of cohorts that
are younger and older than the D-cohort. In all cases we find that the D-cohort is the only cohort that
produces significant crowding-out and crowding-in on births.
The paper is divided in 8 sections. In Section 2, we describe the data, the samples and the
characteristics of the different cohorts we examine. Section 3 examines the impact of economic conditions
and the Great Depression on work and wages using several panels of data from 1920 to 1960. Section 4
describes the measure we use to determine the crowding-out and crowding-in. Section 5 contains our main
results about the impact of the crowding-out and crowding-in on the yearly births of different cohorts in
the1950s and 1960s. Section 6 examines their impact on cumulative births in 1960 and 1970 and Section 7
assesses the Easterlin hypothesis. Section 8 concludes.
4
2. Data and Samples
The data used comes from the 1920 to 1990 IPUMS files (Ruggles et al., 2010) and the US Statistical
Abstract. In the per-period analysis of fertility, we use 1940 as the reference year as this preceded the babyboom which started after WWII, as well as significant improvements in economic conditions. To measure
changing economic conditions, we use state-level data on commercial failures and exploit differences in the
extent of commercial failures across states. This data was taken from the US statistical abstract, various
years between 1909 and 1970, and originally comes from Dun and Bradstreet Inc., NY. It is available at
state-level and on a yearly base between 1900 and 1968, while other measures of economic conditions by
state-level such as unemployment are only available every 10 years (census). The aggregate series is
reported in Figure 3, together with aggregate unemployment. As can be seen, commercial failures are less
variable but correlated with changes in unemployment. Although we cannot use unemployment throughout
as a measure of economic conditions, there are reasons to prefer business failures when examining their
effects on labor markets. While unemployment is affected by shifts in labor demand and labor supply,
commercial failures lead to layoffs that are less likely to be attributable to labor supply shifts (see also
Ananat, Gassman-Pines and Gibson-Davis, 2013). A shift towards having more children could lead women
to work less and to fewer women looking for a job. The negative correlation between unemployment and
births in this case could reflect a preference shift. Instead, a negative correlation between births and
commercial failures is more akin to a labor demand shift and reflect layoffs that lead women to drop out of
the labor market.
To examine the impact of the Great Depression on labor participation rates and wages, we use an
average of commercial failures for the years 1929 to 1933.4 To measure the impact of WWII, we follow
Acemoglu et al. (2004) and use aggregate mobilization rate for men 14 to 44 years old, calculated from the
Selective Service Act.5 Our focus will be explaining births in 1950s and 1960s, omitting the years
immediately after the war. Although the war may have impacted fertility, a large part of the fertility increase
occurred during the 1950s.
In the fertility analysis, we compare each year births from 1950 to 1970, to children born in 1940 to
women in the same age brackets. For the 1950s and 1960s we use the date of birth of the child and the
location of the mother in the household to link the child to his/her own mother.6 We therefore compare the
number of children born in 1940 to a woman in a particular cohort, to children born in 1950, 1951, 1952, …,
1960, 1961, 1962,…1970, to a woman in the same cohort. We opted for this set-up rather than for example
comparing births in 1950 to births in 1930, births in 1952 to births in 1932, etc., to keep the comparison year
constant. This has important advantages. First, it allows us to keep the same 1940 covariates as we consider
the impact of various factors on the births of different cohorts of women over the 1950s and 1960s. Second,
it makes it easier to compare our results across years as the only conditions that are changing are the ones
related to the post-1940 period. This is particularly important as using the other approach, the base year
would be moving through the depression years. Not only are these years characterized by dramatic changes
in economic conditions, these are also the economic conditions we are testing the effects of, which could
make the results more difficult to interpret. The chosen set up allows keeping these effects constant across
the births in the base year though, clearly, how many births women of different ages had in 1940 may have
been affected by past historical conditions. For example the Great Depression may have led to postponing
births.
The choice of the base year is important. Women who were 20 to 24 years and 25 to 29 years old in
1940 had the lowest completed fertility since the beginning of the 20th century (with an average of 2.41 and
2.59 children, respectively), while women who were 20 to 24 years old and 25 to 29 years old throughout
the 1950s, contributed the most to the baby boom (from an average of 2.85 children to an average of 3.21
children).
4
For two years the statistics on commercial failures are not available, 1931 and 1944.
In unreported estimates we use as instrument for men’s mobilization rates, as in Acemoglu et al. (1994), the fraction of males aged 13–44 in 1940 that are
German-born, interacted with a 1960 dummy. In all cases the results of the estimates using instrumental variables are similar to the ones we report in the paper.
6
We use momrule and momloc to link the child to the probable mother and set stepmom=0 so that the mother identified in momloc is the biological mother or
there was no mother of this person present in the household.
5
5
Figure 2: Ratio of Commercial Failures and Unemployment
Table 1 below, shows the year of birth of the cohorts of women we include in our analysis and their
ages between 1930 and 1970. It should help linking the cohorts examined in our per-period analysis to their
completed fertility. To maintain some level of aggregation we group women in 8 cohorts, each including
women born 1 to 4 years apart (listed in the left side of Table 1). We report the completed fertility (number
of ever born children by age 40 to 44) of white women on the right side of the Table. The first three cohorts
in the Table are the women in the D-cohort. The shaded green area shows the cohorts in the base year (20 to
24 and 25 to 29 years old in 1940) to which we compare changes in work or births in the 1950s and 1960s.
We highlight in different shades of grey the cohorts that experienced the baby-boom and baby-bust when
they were 20 to 24 or 25 to 29 years old. These are also are the cohorts and ages we use in our fertility
analysis. The darkest grey is for the cohort who had the highest completed and/or the highest yearly fertility.
As can be seen, the cohort with the highest yearly fertility when 20 to 24 years old, did not have the highest
fertility. As the table shows, the cohorts whose fertility we examine and the cohorts that we think affected
their behavior (the D-cohort) are two cohorts removed from each other. As well, the base year cohorts
(green shaded area) are not part of the D-cohort, or part of the baby-boomers. This ensures that the results
are not contaminated by within-cohort overlaps. The two cohort cushion we kept between the D-cohort and
the baby-boomers is justified by our finding no persistent effects of the Great Depression on their entry in
the labor market in the 1950s and 1960s. However, it is clear that this cohort can have contributed to the
baby-boom and baby-bust and we will examine this in Section 6.
Table 1: Cohort Table
Born in:
Completed Fertility
1930
1935
1940
1945
1950
1955
1960
1965
1970
30-34
35-39
40-44
45-49
50-54
55-59
60-64
65-69
70-74
2,83
25-29
30-34
35-39
40-44
45-49
50-54
55-59
60-64
65-69
2,60
1906-1910
20-24
25-29
30-34
35-39
40-44
45-49
50-54
55-59
60-64
2,30
1911-1915
15-19
20-24
25-29
30-34
35-39
40-44
45-49
50-54
55-59
2,41
1916-1920
10-14
15-19
20-24
25-29
30-34
35-39
40-44
45-49
50-54
2,59
1921-1925
5-9
10-14
15-19
20-24
25-29
30-34
35-39
40-44
45-49
2,85
5-9
10-14
15-19
20-24
25-29
30-34
35-39
40-44
3,11
5-9
10-14
15-19
20-24
25-29
30-34
35-39
3,21
5-9
10-14
15-19
20-24
25-29
30-34
3,02
5-9
10-14
15-19
20-24
25-29
2,56
5-9
10-14
15-19
20-24
2,22
1896-1900
1901-1905
1926-1930
1931-1935
1936-1940
1941-1945
1946-1950
D-Cohort
Figure 3 plots completed fertility by cohort (the solid line) and also the births the same cohort had
when they were 20 to 24 years old (the dotted line), 25 to 29 years old (the small dashed line) and 30 to 34
years old (the long dashed line). Below the lines we report the year in which each cohort reached that age.
6
The graph is rescaled: information about completed fertility is on the left and information about births, on
the right. Reading the information vertically one can find the completed fertility of a particular cohort and its
average number of children at different points in time. The dotted line (births to 20 to 24 years old) and the
small dashed line cross over (births to 25 to 29 years old): at the beginning of the 1950s more children were
born to 25 to 29 years old women than to 20 to 24 years old women; later in the 1950s, more children were
instead born to 20 to 24 years old women than to 25 to 29 years old women. The years 1959/1960 marks the
beginning of the decline in births for all women, independently of their age.
Figure 3: Completed Fertility and Yearly Births
3. Economic Conditions, Great Depression, Work and Wages
In this section we use 4 panels of data, 1920-1930, 1930-1940, 1940-1950 and 1940-1960, to
examine the impact of changing economic conditions on labor markets. In the first subsection we examine
the first two panels, in the second, the last two.
3.1. Great Depression, Economic Conditions and Female Labor Supply: (1920-1930, 1930-1940)
To examine the impact of changing economic conditions on the probability that women enter the labor
force, we estimate a simple linear probability model:
yits  o  1Failurests   2 Failurests * married  (a  f s  gt ) * married   its
(1)
where yits is an indicator for whether a woman i in state s is in the labor force in year t, t are the years 1920
and 1930. Failurests are average commercial failures in state s, between year t and the three previous years,
and ϕs, fs and gt are respectively, age and the square of age, state of residence and state of births, year fixed
effects. We restrict our analysis to white nonfarm men and women born in the USA. Failurests are average
commercial failures between 1918 and 1920 for 1920, or between 1928 and 1930 for 1930.
In a second set of estimates we interact failures and all other covariates with a dummy, married, that
takes the value of 1 if a woman is married and 0 if she is not. This will test whether married and unmarried
women behaved differently, but controlling for the impact that commercial failures or differences across
states or cohorts, may have had on the probability of being married. The estimate of  2 indicates the effect of
failures on marriage and the sum ( 1 +  2 ), the overall impact of failures on married women. The coefficient
on failure alone, 1 , indicates the impact of failures on the probability that single women work. Below the
7
estimates of we report whether the sum of this coefficient is significant and positive (y), or not (n). Since we
use commercial failures to size improving or worsening economic conditions, a positive sign in the estimate
means that a worsening in the economic conditions increases labor market participation.
Point-estimates and robust standard errors for the impact commercial failures on the labor force
participation of women are reported in Tables 2 to 4 for the panel 1920-1930 and in Tables 5 to 6 for the
panel 1930-1940. The first row of Table 2 reports results for women independently of their married status.
In this case the dummy married is 0 for all women and we only estimate 1 . For women in the D-cohort, the
coefficient estimate is negative and significant. This means that increased commercial failures significantly
decreased the presence of these women on the labor market. In the estimates reported below we interact
failures (and other covariates) with the dummy marriage. As can be seen the coefficient on the interacted
failure regressor is significant and positive, while the estimate on the uninterated failure regressor is negative
and significant. The sum the two however is significant and positive at 1% or less for the cohort 20 to 34
years old in 1930 (for 25 to 29 years old women, it is only significant at the 10% significance level).
Is the effect of failures on work reflecting the early impact of the Great Depression on married
women or spurious correlations with business cycles or other unobserved trends? As a falsification we run
the same regressions using average failures between 1914 and 1916 for 1920, and between 1924 and 1926,
for 1930. During this period average commercial failures decreased. The new estimates are reported in the
lower panel of Table 2. In the same vein as before but in the opposite direction, the presence of women in
the labor market increases as commercial failures decrease. Now however, differently from before, there are
no significant effects for married women (the sum of the two coefficients is statistically non-significant),
which suggests that the previous results are not picking up unobserved trends that affect married women in
general. 7
When we run the same regressions for men as in the top panel (that is with average failures between
1928 and 1930), see Table 3, we find that both men overall and married men decrease their presence in the
labor market as failures increase. Repeating the falsification we did in Table 2, and using failures in 19141916 for 1920 and in 1924-1926 for 1930, we find no significant effects, nor for men overall, nor for
married men. These results suggest that at the end of the 1920s, economic conditions at the start of the Great
Depression significantly affected the behavior of both men and women, but while men, women and married
men decreased their participation, married women showed a unique pattern of entry in the labor market as
failures increased.
To check the robustness of our results to how we measure changing economic conditions, we rerun
the regressions with total unemployment using IPUMS census data. To measure the early impact of the
Great Depression we use average unemployment between 1929 and 1930, for 1930 and between 1919 and
1920, for 1920. Since IPUMS reports unemployment only for 1910 and 1930 and not for 1920, we use
interpolated unemployment state-level data from these census to proxy unemployment in 1918, 1919 and
1929. The results from these regressions are reported in Table 4. As can be seen, and as before, 20 to 34
years old women and men significantly decrease their presence in the labor market as unemployment
increases, while married women in the same cohort significantly increase their presence. As with
commercial failures, married men respond differently from married women and decrease their participation
to the labor market as commercial failures increase.
Tables 5 and 6, use IPUM 1930-1940 micro data to evaluate the impact of the Great Depression on
the D-cohort in the 1930s and estimate the probability that a woman works:
yits   o  1Failurests   2 Failures _ 1930ds  a  f s  gt   its
(2)
yits is an indicator for whether a woman i in state s is employed in time t, where t is 1930 and 1940.8
In previous estimates we examined labor force participation rates because in the 1920 census, the reference
7
In all cases, the 1920-1940 panels reveal a similar entry pattern for married women in the D-cohort as the 1920-1930 panels but 10 years after,
a pattern that is not found for men.
8
We set work=1 if empstat==1, 0 otherwise. The results presented here do not change when we use labor force participation rate (not reported).
8
year, people were not asked about their employment status. Since we are interested in whether the entry of
the D-cohort in the labor market crowded out younger cohorts, we now shift the attention to the impact of
the Great Depression on work. To capture changes in current economic conditions we use the rate of average
commercial failures between 1929 and 1930 for 1930 and between 1939 and 1940 for 1940, during which
period they decreased. We don’t include the marriage interaction term as the results now apply to all
women, not just to married women. We now have a new term that captures the worst years of the Great
Depression, Failures_1930. These are average failures between 1919 and 1922 for 1930 and between 1929
and 1932 for 1940. During this period failures increased considerably. The results in Table 5 show that the
Great Depression significantly increased the share of women working (both married and unmarried), while
improving contemporaneous economic conditions decreased their employment. When we include average
failures in the early 1920s, these are as for the 1920-1930 panels, non-significant. These results are reported
in Table 5 under the heading Falsification. In addition to Failures_1930 and Failures we include average
commercial failures between 1908 and 1910 for 1930, and between 1918 and 1920 for 1940, as well as
average failures between 1914 and 1916 for 1930 and between 1924 and 1926 for 1940. The only significant
effects come from contemporaneous failures or failure during the Great Depression.
The same entry pattern in response to worse conditions during the Great Depression is significant at
the 1% level, or less, for women in the D-cohort and also younger women. In Table 6, we estimate the same
equation but for men. We find that improving economic conditions increase the probability that men work,
while worse economic conditions during the Great Depression decrease the probability they work. The
estimates confirm the patterns we found for the 1920-1930 samples: significant entry of women in the Dcohort (and also of younger women) in response to economic conditions during the Great Depression, while
no similar pattern is found for men. While when we used the 1920-1930 panels we found that only married
women showed an entry pattern linked to the Great Depression, this pattern now extends to all women,
independently of their marital status. Before we were captuiring the early years of the Great Depression, in
this sample we are capturing the worst years of the Great Depression.
9
Table 5: Female Employment & Commercial Failures: 1930-1940
Dep. Variable = 1 if currently working
Ages:
20-24
25-29
20-29
30-44
45-64
contemporaneous failures
(1939/1940 vs 1929/1930)
(Change: -0.46)
0.056
(0.015)***
0.036
(0.014)**
0.046
(0.014)***
0.006
(0.012)
-0.032
(0.017)*
past failures (Gr. Depression)
(1929/1932 vs 1919/1922)
(Change: 0.52)
0.147
(0.017)***
0.092
(0.017)***
0.122
(0.016)***
0.029
(0.009)***
-0.002
(0.013)
contemporaneous failures
(1939/1940 vs 1929/1930)
(Change: -0.46)
0.057
(0.015)***
0.034
(0.014)***
Falsification
0.046
(0.014)***
0.004
(0.011)
-0.033
(0.017)*
past failures (Gr. Depression)
(1929/1932 vs 1919/1922)
(Change: 0.52)
0.140
(0.016)***
0.099
(0.017)***
0.122
(0.015)***
0.035
(0.009)***
0.002
(0.014)
past failures
(1924/1926 vs 1914/1916)
(Change: -0.17)
0.035
(0.028)
-0.009
(0.014)
0.014
(0.019)
-0.014
(0.008)*
-0.021
(0.010)*
past failures
(1918/1920 vs 1908/1910)
(Change: -0.45)
-0.032
(0.029)
-0.032
(0.028)
-0.033
(0.027)
-0.006
(0.014)
0.021
(0.019)
N
48737
43631
92368
103083
80904
Notes : Coefficients from OLS regression of an indicator of employment on current commercial failures, past failures, age,
current/birth state and year fixed effects. Sample includes white, non-farm women born in the United States. Standard errors
(parentheses) are clustered by state-year. ***. **. * indicate significance at 1%. 5% and 10% respectively.
Table 6: Male Employment & Commercial Failures: 1930-1940
Dep. Variable = 1 if currently working
Ages:
20-24
25-29
20-29
30-44
45-64
contemporaneous failures
(1939/1940 vs 1929/1930)
(Change: -0.46 )
-0.043
(0.011)***
-0.042
(0.006)***
-0.046
(0.008)***
-0.019
(0.008)**
-0.007
(0.014)
past failures (Gr. Depression)
(1929/1932 vs 1919/1922)
(Change: 0.52)
-0.063
(0.014)***
-0.016
(0.008)*
-0.042
(0.010)*
-0.005
(0.005)
0.008
(0.007)
N
44484
42107
86591
102614
78843
Notes : Coefficients from OLS regression of an indicator of employment on current commercial failures, past failures, age,
current/birth state and year fixed effects. Sample includes white, non-farm men born in the United States. Standard errors
(parentheses) are clustered by state-year. ***. **. * indicate significance at 1%. 5% and 10% respectively.
10
3.2. Great Depression, Economic Conditions and Female Labor Supply: 1940-1950, 1940-1960
We now turn to the impact of economic conditions during the Great Depression on the following
decades. Did they have lasting effects for the old cohorts of women? Did they affect the younger cohorts? In
this section we pool data from 1940 and 1950 and 1940 and 1960, to examine whether the pattern of entry in
the labor market of women in the D-cohort persists into the 1950 and 1960. We estimate equations of the
general form:
yits   o  1Mobrates   2 Failurests  3 Failures _ 1930ds   4 Z1940, s  a  f s  gt   its
(3)
yits is an indicator for whether a woman i in state s is employed in time t, where t is 1940 and 1950 or 1940
and 1960, depending on the panel used. In a second set of regressions we use as dependent variable the log
of weekly earnings. ϕs, fs and gt are age, state and year fixed effects. As before, we restrict our analysis to
white nonfarm men and women born in the USA. Mobrates is WWII mobilization rate in state s (see
Acemoglu et al. 2004) interacted with a 1950 or 1960, depending on the panel used, year dummy. Z1940,s is a
vector that contains the 1940 state averages of male education, and the ratios of male farmers and non-white
males. All covariates are interacted with a 1950 or 1960 year dummy (depending on whether we use the
samples 1940-1950 or the samples 1940-1960). To capture the lasting impact of the Great Depression we
include as regressor Failuresds, which are average commercial failures in state s, between 1929 and 1933 for
1940, and between 1920 and 1923 for 1950, when we use the panel 1940-1950 or between 1910 and 1913
for 1960, when we use the panel 1940-1960. To controls for changes in current economic conditions, we
include the regressor Failurests, which are average commercial failures in state s between 1937 and 1940 for
1940, and between 1947 and 1950 for 1950, when we use the panel 1940-1950 or between 1957 and 1960
for 1960, when we use the panel 1940-1960.
In Tables 7 to 8, we report the point-estimates and robust standard errors for the panel 1940-1950
and in Tables 9 to 10, for the panel 1940-1960. We report the estimates of the effects of WWII ( WWII
mobilization = α1 ), of the effects of changes in current economic conditions between the base year and 1950
or 1960 (Failures = α2 ) and of the impact of increased failure in the 1930s (Failure_1930s = α3 ).9 Failures
decreased both in 1950 and 1960 relatively to the base year.
We first discuss the estimates using 1940 and 1950 census data. The estimates presented in Table 7
show that in states with more commercial failures during the Great Depression, 45 to 54 years old women
worked significantly more in 1950 than in 1940 (see the second to last column). Instead the younger group
in the D-cohort, 40 to 44 years old in 1950, was not significantly affected. Improving current economic
conditions also increase the probability that women in the 45 to 54 cohort are in the labor market.10
Interestingly, women 20 to 24 years old are only affected by economic conditions dating back to the Great
Depression, not by current economic conditions. Overall, for the cohort 20 to 34 years old in 1950, however,
both worsening economic conditions between the 1920s and the 1930s and improving current economic
conditions decrease their employment.11
The 30 to 39 years old cohort (in 1950), for whom we found an increased presence in the labor
market between 1920 and 1940, is instead not responsive to changes in economic conditions, current or past,
in 1950. The different response of the D-cohort relatively to this younger cohort could be due their age at the
time of economic recovery, the first was too old to have significantly more children, while the second was
20 to 29 years old in 1940 – all sufficiently young to have children over the next decade. Also the younger
group of the D-cohort (40 to 44 years old in 1950) did not enter significantly more in states with more
failures during the 1930s; they too in 1940 were sufficiently young to have more children. The last panel of
Table 7 reports the results of the estimations for men, they show no significant effects from the Great
9
There are no data for 1931, so for symmetry, we omit both 1931 and 1921 from the averages.
When considering married women (not reported), the employment of all women in the D-cohort is significantly higher in 1950 in states with more business
failures during the Great Depression, while changes in current economic conditions have no significant effects.
11
To understand the robustness of this result, we estimated a specification that does not include the effects of economic conditions during the Great Depression
(not reported). When we take out this channel, we again find no significant effect of current economic conditions, which suggests that these channels are
independent from each other.
10
11
Depression and an increase in the probability that they are employed in states with contemporaneous
improving economic conditions.
We lastly turn our attention to the impact of WWII. As can be seen, the effects are significant and
positive for women 20 to 24 years old in 1950 and for the 30 to 39 years old cohorts (both with a
significance level of 1% or lower), while they are negative and non-significant (significant at the 10% level)
for the D-cohort. These results are not consistent with the hypothesis that WWII led to the crowding out of
young cohorts, which would imply a decrease in the share of young women working.12
Table 8 reports the impact of current and past failures on log weekly earnings (divided by the year
CPI index) in 1950 versus 1940, restricting women to have worked at least 20 weeks in the previous year.13
The most striking result is the negative sign associated with commercial failures in the 1930s for nearly all
cohorts of women. These results are consistent with a labor supply shift of the older cohorts that decreased
wages for women of all ages. Interestingly, the opposite sign old/young extends to wages when we examine
the impact of current economic conditions. The decrease in current commercial failures is associated with
lower wages for the D-cohort, 40 to 54 years old in 1950, which is also consistent with a labor supply shift.
The opposite sign for the 20 to 24 years old cohort suggests that their decreased presence created upward
pressures on wages.
The second panel of Table 8 reports results from regressions that include the share of women in the
D-cohort who were working in 1930 interacted with a 1950 year dummy. If our proposed channel is right,
we would expect that in states with higher participation of women in the labor market in 1930 female wages
are lower. Our results support our hypothesis as for nearly all cohorts of women, wages are significantly
lower in 1950 than in 1940 in states with a higher share of women in the labor force in 1930. The last panel
of Table 8 reports the impact of past failures on men. As can be seen there are some negative effects but
only for men who were 30 to 39 years old in 1950, beside this the Great Depression seems to have had no
lasting effects on their wages.14
12
Though, the coefficient is only significant at the 10% level when we omit commercial failures during the Great Depression (not reported).
The results are similar when we don’t restrict the number of weeks.
14
IPUMS 1920 does not contain information on employment or weekly wages; therefore for 1920 we only report results for labor participation rates.
13
12
Table 7: Employment: 1940-1950
Dependent Variable Female Employment (All Women)
Dep. Variable = 1 if currently employed
Ages:
(Dep. Var.: 1940 mean)
20-24
(0.397)
25-29
(0.304)
30-39
(0.248)
40-54
(0.203)
0.942
(0.189)***
-0.195
(0.181)
0.560
(0.151)***
-0.294
(0.171)*
-0.103
(0.209)
0.148 0.318
(0.188) (0.126)**
Failures
(Change 1950-1940: -0.36)
-0.024
(0.021)
0.053
(0.011)***
0.015
(0.014)
-0.004
(0.011)
-0.042
(0.014)***
-0.014 0.036
(0.016)(0.009)***
Failures_1930s
(Change 1930-1920: 0.37)
-0.045
(0.017)***
-0.017
(0.017)
-0.007
(0.013)
0.028
(0.018)
0.061
0.072 -0.025
(0.023)** (0.026)***
(0.010)***
N
24390
25852
Dependent Variable Male Employment (All Men)
43772
48977
30612
21405
WWII mobilization
0.017
(0.118)
-0.023
(0.093)
-0.132
(0.095)
-0.032 -0.136
(0.092) (0.166)
-0.005
(0.014)
-0.017
(0.014)
-0.033 -0.038
(0.009)***
(0.009)***
-0.018 -0.003
(0.011) (0.009)
WWII mobilization
-0.007
(0.293)
-0.242
(0.139)*
-0.024
-0.05
(0.011)** (0.009)***
45-54 (all)45-54 (married)
20-34
(0.196)
(0.091) (0.326)
Failures
(Change 1950-1940: -0.36)
-0.033
(0.016)**
Failures_1930s
(Change 1930-1920: 0.37)
0.027
(0.018)
-0.015
(0.012)
-0.010
(0.008)
-0.017
(0.010)*
-0.020
(0.011)*
N
20140
23147
41403
44805
27844
22664
73346
65170
Notes: Coefficients from OLS regression of work indicator on contemporaneous failures, failures during the Great Depression years, WWII mobilization rates,
age, share of males in 1940 that are nonwhites, share of males in 1940 that are farmers, 1940 average male education, state and year fixed effects. Sample
includes white, non-farm men born in the United States. Standard errors (parentheses) are clustered by state-year. ***, **, * indicate significance at 1%, 5%
and 10% respectively. Sources: 1940-1950 IPUMS USA, Statistical Abstracts of the United States.
Table 8: Wages: 1940-1950 (Dep. Variable: Log real weekly wage, worked at least 20 weeks in previous year)
Dpendent Variable Female Wages
Ages:
20-24
25-29
30-39
40-54
45-54
20-34
WWII mobilization
-0.663
(0.418)
-0.683
(0.426)
0.386
(0.367)
-0.690
(0.553)
0.337
(0.883)
-0.382
(0.275)
0.005
(0.055)
-0.041
(0.030)
Failures
-0.099
0.074
(0.033)*** (0.026)***
-0.009
(0.039)
0.025
(0.031)**
Failures_1930s
-0.196
-0.110
(0.038)*** (0.037)***
-0.063
(0.035)*
-0.140
-0.197
-0.119
(0.050)*** (0.067)*** (0.020)***
Failures
-0.122
0.081
(0.036)*** (0.034)**
-0.042
(0.039)
Share women 20-34 yrs old in
-0.523
(0.172)***
labor force in 1930*yr1950
N
8956
Dpendent Variable Male Wages
-0.021
(0.178)
-0.026
(0.033)
-0.554
-0.953
(0.198)*** (0.273)***
-0.047
(0.048)
-0.059
(0.031)*
-1.005
(0.383)**
-0.372
(0.142)**
7268
11172
12148
7297
22109
WWII mobilization
-0.293
(0.220)
-0.172
(0.257)
0.111
(0.221)
0.220
(0.417)
0.111
(0.418)
-0.042
(0.181)
Failures
(Change 1950-1940: -0.36)
0.036
(0.027)
0.016
(0.030)
0.068
(0.015)***
0.018
(0.028)
0.003
(0.028)
0.043
(0.018)**
Failures_1930s
(Change 1930-1920: 0.37)
-0.03
(0.025)
0.042
(0.026)
-0.061
(0.029)**
-0.039
(0.033)
-0.044
(0.036)
-0.011
(0.019)
N
16476
20307
35434
35204
21425
55791
Notes: Dependent variable: log weekly wages. For more details see footnote to previous table.
13
Tables 9 and 10, report the same estimates as for Tables 7 and 8, but for the 1940-1960 panels. The
most remarkable finding is the similarity in the exit-entry pattern for young and old cohorts of women we
found for 1940 and 1950, if anything this pattern is stronger, as is also the link to the Great Depression. The
D-cohort is now 50 to 64, working significantly more in 1960 than in 1940 in states which had more
commercial failures during the Great Depression, including the 50 to 54 years old group for which we had
found no significant effects (unless we restricted the sample to married women) using the 1940-1950
dataset. Again, we find no link between male employment and the Great Depression. The D-cohort also
works more in states with improving economic condition (see the estimates for failures). Instead, 20 to 34
years old women work significantly less in states with more commercial failures during the Great
Depression and also significantly less in states with improving current economic conditions.
Table 10 reports results from the wage regressions for the panel 1940-1960. The estimates show a
negative link between wages and the depression years or with women labor participation in 1930, though the
link is weaker for the 20 to 24 and 25 to 29 years old cohorts, and becomes non-significant when we use the
share of women in the D-cohort who were in the labor force in 1930. This is not surprising as the D-cohort is
50 to 64 years old in 1960 and some of them are starting to retire. In both the 1940-1950 and 1940-1960
panels, we see that the 20 to 24 cohorts have higher wages in states with improving economic conditions. In
the 1950-1950 panels these effects were countered by the negative impact of the Great Depression on female
wages, while in the 1940-1960 panels failures in the 1930s have a smaller negative effect (only significant
at the 10% significance level) on the wages of this cohort, suggesting a change in labor market conditions.
In the lower section of Table 10, we examine the impact of failures in the 1930s on men’s wages. We
find a decrease in wages for men in the D-cohort, but only significant at the 10% significance level. Men in
this cohort lost their jobs or interrupted their education during the depression years, at a crucial stage of their
life that may have led to permanently lower wages and also have contributed to the massive entry of women
in the D-cohort in the labor market. We find a positive effect in response to improving contemporaneous
economic conditions for the 25 to 29 years cohort, this is consistent with rising male incomes encouraging
marriage and family formation.
14
Table 9: Employment: 1940-1960
Dependent Variable Female Employment (All Women)
Dep. Variable = 1 if currently employed
Ages:
(Dep. Var.: 1940 mean)
20-24
(0.397)
25-29
(0.304)
30-39
(0.248)
WWII mobilization
0.400
(0.172)**
-0.205
(0.270)
Failures
(Change 1960-1940: -0.097)
0.059
0.047
(0.019)*** (0.021)**
Failures_1930s
(Change 1930-1910: 0.35)
-0.058
-0.064
-0.026
(0.019)*** (0.019)*** (0.011)**
40-49
(0.209)
50-64 (all)
(0.161)
50-64 (married)
(0.066)
20-34
(0.326)
-0.285
(0.250)
-0.121
(0.215)
0.240
(0.170)
-0.045
(0.010)***
-0.055
(0.015)***
-0.086
(0.014)***
0.060
(0.015)***
-0.010
(0.012)
0.036
(0.014)***
0.023
(0.012)*
-0.053
(0.012)***
0.534
-0.289
(0.143)*** (0.089)***
0.008
(0.013)
N
51043
52741
Dependent Variable Male Employment (All Men)
112067
99222
105712
70129
159145
WWII mobilization
0.297
(0.226)
0.021
(0.168)
0.014
(0.133)
0.267
(0.140)*
-0.300
(0.187)
-0.095
(0.166)
0.115
(0.153)
-0.039
(0.017)**
-0.017
(0.016)
-0.021
(0.011)*
0.022
(0.016)
-0.015
(0.017)
-0.019
(0.015)
-0.028
(0.011)**
Failures_1930s
(Change 1930-1910: 0.35)
0.010
(0.016)
0.005
(0.009)
-0.006
(0.008)
-0.003
(0.010)
0.015
(0.019)
-0.001
(0.015)
0.005
(0.008)
N
42528
49887
106537
94512
95851
80104
145612
Failures
(Change 1960-1940: -0.097)
Notes: Coefficients from OLS regression of work indicator on contemporaneous failures, failures during the Great Depression years, WWII mobilization rates,
age, share of males in 1940 that are nonwhites, share of males in 1940 that are farmers, 1940 average male education, state and year fixed effects. Sample
includes white, non-farm men born in the United States. Standard errors (parentheses) are clustered by state-year. ***, **, * indicate significance at 1%, 5%
and 10% respectively. Sources: 1940-1960 IPUMS USA, Statistical Abstracts of the United States.
Table 10: Wages: 1940-1960 (Dep. Variable: Log real weekly wage, worked at least 20 weeks in previous year)
Dpendent Variable Female Wages
Ages:
20-24
25-29
30-39
40-49
50-64
20-34
WWII mobilization
-1.162
(0.353)***
-0.467
(0.391)
-0.968
(0.326)***
-1.718
(0.662)**
-1.380
(0.845)
-0.868
(0.284)
Failures
-0.171
(0.045)***
-0.054
(0.037)
-0.025
(0.043)
-0.004
(0.050)
0.101
(0.091)
-0.117
(0.028)***
-0.082
-0.075
-0.183
(0.042)* (0.027)*** (0.057)***
-0.067
(0.081)
-0.057
(0.029)**
-0.165
(0.049)***
-0.039
(0.041)
-0.037
(0.046)
-0.061
(0.054)
0.029
(0.100)
-0.113
(0.029)***
-0.102
(0.187)
-0.054
(0.213)
-0.421
(0.231)*
-1.384
(0.255)***
-1.280
(0.347)***
-0.122
(0.144)
N
19320
Dpendent Variable Male Wages
14766
30888
33819
31325
48593
Failures_1930s
Failures
Share women 20-34 yrs old in
labor force in 1930*yr1960
-0.063
(0.036)*
WWII mobilization
-0.481
(0.405)
-0.411
(0.228)*
-0.400
(0.168)**
-0.660
(0.461)
-2.179
(0.382)***
-0.402
(0.194)**
Failures
(Change 1960-1940: -0.36)
-0.045
(0.051)
-0.090
(0.041)**
0.008
(0.023)
0.044
(0.030)
0.024
(0.044)
-0.036
(0.032)
Failures_1930s
(Change 1930-1910: 0.35)
-0.000
(0.029)
0.070
(0.028)**
0.001
(0.022)
-0.042
(0.036)
-0.048
(0.025)*
0.025
(0.020)
N
37895
45725
93805
78618
71791
131206
Notes: Dependent variable: log weekly wages. For more details see footnote to previous table.
15
In Tables 11 and 12 we perform a number of falsifications, for the panels 1940-1950 and the panels
1940-1960. For past failures we use failures in 1920-1922 instead of failures in 1929-1933. In another
falsification experiment instead of the labor participation of women in the D-cohort in 1930, we use the
labor force of participation of men of the same age as the women in the D-cohort. In no case, we find a
significant effect. The only significant effects are found in relation to conditions that date back to the Great
Depression.
Table 11: Female Employment - Falsification Checks: 1940-1950
Dep. Variable = 1 if currently employed
Ages:
20-34
45-54
45-54
20-34
45-54
45-54
20-34
45-54
45-54
20-34
45-54
45-54
(all)
(all)
(married)
(all)
(all)
(married)
(all)
(all)
(married)
(all)
(all)
(married)
WWII mobilization
0.293
(0.127)**
-0.094
(0.181)
0.076
(0.215)
0.270
(0.129)**
-0.045
(0.208)
0.139
(0.198)
0.394
(0.120)***
-0.132
(0.222)
0.066
(0.108)
0.244
(0.155)
-0.175
(0.240)
0.225
(0.211)
Failures
0.037
-0.044
(0.008)*** (0.013)***
-0.018
(0.017)
0.035
-0.041
(0.008)*** (0.013)***
-0.015
(0.016)
0.024
(0.009)**
-0.04
(0.015)**
-0.004
(0.018)
-0.203
(0.054)***
0.076
(0.094)
0.233
(0.086)***
Failures_1930s
-0.007
(0.013)
0.004
(0.016)
-0.021
(0.019)
-0.016
(0.013)
0.021
(0.016)
-0.003
(0.018)
Share women 20-34 yrs old in labor force
in 1930*yr1950
Share men 20-34 yrs old in labor force
21405
-0.328
(0.269)
73346
-0.307
(0.394)
30612
in 1930*yr1950
N
-0.019
(0.016)
-0.029
0.066
0.071
(0.011)** (0.023)*** (0.026)***
(1930/1933) vs (1920/1923)
Failures_falsif
(1920/1922 vs 1910/1912)
0.038
-0.043
(0.008)*** (0.014)***
73346
30612
21405
73346
30612
21405
73346
30612
-0.382
(0.382)
21405
Table 12: Female Employment - Falsification Checks: 1940-1960
Dep. Variable = 1 if currently employed
Ages:
20-34
50-64
50-64
20-34
50-64
50-64
20-34
50-64
50-64
20-34
50-64
50-64
(all)
(all)
(married)
(all)
(all)
(married)
(all)
(all)
(married)
(all)
(all)
(married)
WWII mobilization
0.431
(0.196)**
-0.290
(0.210)
-0.206
(0.183)
0.300
(0.178)*
-0.208
(0.219)
-0.137
(0.188)
0.536
(0.132)***
-0.449
(0.231)*
-0.265
(0.190)
Failures
0.075
-0.072
-0.088
0.056
-0.062
-0.082
0.047
-0.054
-0.074
0.071
-0.059
-0.089
(0.017)*** (0.013)*** (0.015)*** (0.018)*** (0.013)*** (0.013)*** (0.014)*** (0.016)*** (0.014)*** (0.014)*** (0.016)*** (0.014)***
Failures_1930s
-0.054
0.032
(0.013)*** (0.013)**
(1930/1933) vs (1910/1913)
Failures_falsif
(1920/1922 vs 1900/1902)
-0.002
(0.010)
0.021
(0.012)*
-0.0003
(0.009)
0.009
(0.010)
0.015
(0.011)
-0.005
(0.009)
-0.406
(0.061)***
in 1930*yr1960
0.117
(0.093)
0.283
(0.077)***
Share men 20-34 yrs old in labor force
158988
105650
70096
158988
105650
-0.232
(0.165)
0.025
(0.012)**
Share women 20-34 yrs old in labor force
in 1930*yr1960
N
0.513
-0.490
(0.174)*** (0.213)**
70096
159145
105712
70129
0.360
(0.325)
159145
-0.311
(0.396)
105712
-0.105
(0.369)
70129
Notes: Coefficients from OLS regression of work indicator on contemporaneous failures, failures during the Great Depression years, WWII mobilization rates, age, share of males in 1940 that are nonwhites, share of males
in 1940 that are farmers, 1940 average male education, state and year fixed effects. Sample includes white, non-farm women born in the United States. Standard errors (parentheses) are clustered by state-year.
***, **, * indicate significance at 1%, 5% and 10% respectively. Sources: 1940-1950 IPUMS USA, Statistical Abstracts of the United States.
Overall these two subsections show that deteriorating economic conditions between 1920 and 1940
led to the entry of married women in the labor market (Added Worker Effect). These women where 20 to 44
years old in 1940 - and a subgroup, the 30 to 44 years old, our D-cohort – showed a pattern of persistent and
significant entry in the labor market over the next decades. We interpret these results as supportive of the
hypothesis of a labor supply shift for this cohort. This entry is not so surprising if one considers their age at
the time of the Great Depression, their husbands may have been unemployed for long periods and their
salary have suffered permanently; they may have lost homes with mortgages, their business or savings. In
addition they had families and children. Sufficiently large wealth and income losses can permanently shift
out labor supply.
16
4. Great Depression and the D-Cohort: A Measure of Crowding-Out and Crowding-In
The empirical evidence presented so far shows a pattern of entry of women who were 20 to 34 years
old in 1930, named the D-cohort, linked to bad economic conditions during the 1930s, that persisted over the
arc of several decades. We also found that the same economic conditions significantly decreased the share of
working young women in 1950 and 1960, women who were either children in the 1930s, or not yet born.
Our hypothesis is that this opposite entry/exit pattern was due to a large outward labor supply shift of a
particular cohort, the D-cohort, which crowded-out young women.
The next step is to show that the outward labor supply shift of the D-cohort can generate a babyboom in the 1950s and later, at their retirement, the baby-bust. To do this, we construct a state-level measure
that summarises their life-cycle labor supply changes and test whether it significantly affected births in both
decades. We create two measures, one for the crowding-out (CO) which describes the entry of the D-cohort
in the labor market, and another for the crowding-in (CI), that summarizes the retirement and exit of the Dcohort from the labor market. To predict work and births in the 1950s, we use the change in the share of
women in the D-cohort who were working between 1950 and 1940, that is:
54
CO1960, (1940  1960) 
 work
54
s , i ,1950 /
i  40
 pop
44
s , i ,1950
-
i  40
 work
44
s , i ,1940 /
i  30
 pop
s , i ,1940
i  30
An increase in CO means that there were more women working in 1950 than in 1940. For the panels
1940-1970, we adopt a similar strategy. In this case we use the actual share of women retiring between 1960
and 1970, rather than the change between 1950 and 1960:
64
CI 1 9 7 0, (1 9 4 0  1 9 7 0) 
64
 work
i ,1960 /
i 50
74
 pop
i ,1960 
i 50
 work
74
i ,1970 /
i 60
 pop
i ,1970
i 60
There is a slight asymmetry between this measure and the first one, as the first uses the increase in the share
of women in the D-cohort working between 1940 and 1950, to predict crowding out and births in each year
in the 1950s, and the second uses actual retirement occurring between 1960 and 1970. The first allows a
separation between the entry of the D-cohort and the births of the young cohorts, to avoid the possibility that
we are picking up reverse causality or simultaneity between the decision of the old and the young. If we
were to use the change in women working between 1950 and 1960 to predict retirement however, we would
not be able to predict retirement as the D-cohort keeps increasing its participation in the labor market
between 1950 and 1960, but given their age we know that they will be retiring in the 1960s. Since women
who are 50 to 64 years old in 1960 are retiring over the 1960s for reasons that are mostly exogenous and
driven by their age, we feel that there is less of a problem of endogeneity or possible reverse causality.
In both cases, to predict work and births in the reference year, 1940, we use the change in the share
of women working between 1950 and 1930:
34
CI 1 9 4 0  CO1 9 4 0 

i  20
34
works ,i ,1930 /

i  20
44
pops , i ,1930 -

44
works ,i ,1940 /
i  30
 pop
s , i ,1940
i  30
The change in the share of women working between decades contains important information about
the extent to which labor markets changed over the course of a decade. The bigger the increase in the share
of women in the D-cohort entering in the 1940s, the more wages decreased and the younger women
perceived that there were fewer market opportunities.
Figure 3 reports the share of women in the D-cohort working between 1930 and 1970. We also report
the same information for four other cohorts, three older and one younger. Figure 4 reports the same share
for married women only. Between 1930 and 1940, overall fewer women in the D- cohort worked while - as
17
the figure below shows - married women worked more. The year 1940 marks the beginning of a period of
constant increase in market work of the D-cohort, up until roughly 1960, and then declines drastically by
1970. This is while at the same time much younger women (20 to 29 years old in the 1950s and 1960s)
instead produce an unprecedented number of children. The decline in births of these younger cohorts,
interestingly, coincides with the retirement of the D-cohort and their gradual exit from the labor market. The
immediately younger cohort, 20 to 29 in 1940 also works more in the 1950s, but its work pattern cannot
explain the decline in births in the 1960s as they retire much later.
Figure 3: Women Employment Shares by Cohort
Figure 4: Women Employment Shares by Cohort (married women)
18
Before using these measures to assess their impact on births, we examine if they can predict changes
in the share of women working between 1940 and 1960 and between 1940 and 1970:
yits  o  1Mobrates   2 Failurests  3CO t   4CI t 
(4)
 5 Z1940, s  a  f s  gt   its
The dependent variable is 1 if a woman in cohort i is working, 0 if she is not. Mobrate is the mobilization
rate of men to WWII interacted with a 1960 or 1970 year dummy, depending on the panel used. Z contains
1940 state observable covariates such as the 1940 male share of farmers, non-white male population and,
average male education. The other variables have been defined previously.
The results are reported in Table 13 below. They show that the entry of the D-cohort in the labor
market between 1940 and 1950 predicts a decrease in the share of young women working, at the 1%
significance level for the 20 to 24 years old cohort and at the 10% significance level for the 25 to 29 year old
cohort. Between 1940 and 1970, instead, the crowding-in term induces the opposite effect: as women in the
D-cohort exit, significantly more young women enter the labor market. For both cohorts the estimates are
significant at the 1% significance level or less. The measures constructed therefore capture the labor market
changes that we have hypothesized, we now examine in the next section whether they can also predict
changes in fertility patterns over the 1950s and 1960s.
Table 13: Crowding-Out, Crowding-In & Labor Supply of Young Women (1940-1960, 1940-1970)
Dep. Variable = 1 if currently employed
1940-1960
1940-1970
Ages:
20-24
25-29
20-24
25-29
"crowding out"
(d-cohort: 20-34 yrs old in 1930)
(change 1960-1940: 0.116 - entry )
"crowding in"
(d-cohort: 20-34 yrs old in 1930)
(change 1970-1940: 0.106 - retirement)
N
-0.445
(0.129)***
51043
-0.196
(0.118)*
52741
0.555
(0.139)***
0.497
(0.134)***
73264
65683
Notes: Coefficients from OLS regression of work indicator on WWII mobilization rates, age, share of males in 1940 that are
nonwhites, share of males in 1940 that are farmers, 1940 average male education, state and year fixed effects. Sample includes
white women born in the United States. Standard errors (parentheses) clustered by state-year. ***, **, * indicate significance
at 1%, 5% and 10% respectively. Sources: 1940-1960 IPUMS USA, Statistical Abstracts of the United States.
To see whether these measures have the potential of explaining birth changes in the 1950s and
1960s, we plot the crowding-out measure against the change in births between 1940 and 1960, see Figure 5,
and the crowding-in measure against changes in births between 1940 and 1970 (see Figures 6), to women
who were at the time 20 to 24 or 25 to 29 years old.15 In Figures 7 and 8 we plot both measures against the
cumulative fertility of women who were 25 to 29 years old in 1960 or 1970.16 Women in this age bracket in
1960 had the highest fertility during the baby-boom (an average of 3.21 children), while women in the same
age bracket in 1970 were responsible for the baby-bust and had an average of only 2.57 children. If the
crowding-out hypothesis is correct, we expect this relation to be positive for changes in births between 1940
and 1960 and if the crowding-in hypothesis is correct, we expect this relation to be negative for changes in
births between 1940 and 1970. In the first case it would confirm that the negative impact of the entry of the
D-cohort on the probability that young cohorts work we just found, has the potential to explain more births,
in the second that the positive impact of the retirement of the D-cohort on the probability that young women
work has the potential to explain a decline in births. As can be seen below, the scatter plots show that the
15
We plot the change in births between 1940 and 1960 against the change in the share of women working in 1950 minus share of women working in 1940 against
the change in births that occurred between 1940 and 1960, the change in births between 1940 and 1970 against the change in the share of women working in 1970
minus share of women working in 1960.
16
We use chborn2 of women 25 to 29 years old in 1960 or 1970.
19
crowding-out measure is associated with more births and more cumulative births in states where the share of
women in the D-cohort who were working increased more, while the crowding-in measure is associated with
fewer births and lower cumulative births in states where the share of women in the D-cohort who were
working decreased the most. In the next section we examine whether these measures of crowding-out and
crowding-in are significant and can explain changes in birth patterns over the 1950s and 1960s.
Figure 5: Change in the Work Behavior of the D-Cohort (entry: 1940-1960)
20
Figure 6: Change in the Work Behavior of the D-Cohort (retirement: 1940-1970)
5. Crowding-out, Crowding-in and Yearly Births
In this section we test whether the work behavior of the D-cohort can explain the baby-boom and the
subsequent baby-bust. To examine births we re-estimate equation (4). The dependent variable is now 1 if a
woman in cohort i had a child born in year t. As described in the introduction, when we use the panels 19401960, we compare births in 1950, 1951, 1952, …1959 to births in 1940 to women in the same age bracket.
When we use the panels 1940-1970, we instead compare births in 1960, 1961, 1962, …1969 to births in
1940 to women in the same age bracket.
Table 14 presents the results for births to women who were 20 to 24 and 25 to 29 years old
throughout the 1950s and in 1940. We report results for births that occurred in each year in the 1950s versus
births to women in the same age bracket in 1940. In addition to the measure for crowding-out, we also
include mobilization rates and current failure (one year lagged to reflect economic conditions at the time of
conception) interacted with a 1960 dummy, to allow for different effects in 1960 relatively to 1940.
21
The results are striking, in each year women have significantly more births than in 1940 when there
was a higher entry of women in the D-cohort. For women who were 20 to 24 years old in the 1950s, in all
years but one, the significance level is 1% or lower and the crowding-out effect is quantitatively important.
Let’s take the example of women 20 to 24 years old in 1959, the estimated coefficient is .466, the increase in
the average number of births between 1940 and 1959 is .14 (from .076 to .22 births) and the change in the
crowding-out measure between 1940 and 1960 is .116 (entry). This means that the crowding-out from
women in the D-cohort explains nearly 40% (.466*.116/.14) of the increase in the births at the peak of the
baby-boom relatively to the previous bottom. If we take 1955, with an average of .2 births in 1955, the
crowding out explains 32% of the increase in births. The effects are significant in most years also for the
cohort that was 25 to 29 years old in the 1950s. For example in 1958 women in this age bracket had an
average of .187 children, and in 1940, an average of .084 children. The increased share of women in Dcohort working in the 1940s, explains 31% of the increase in births between 1958 and 1940, to women who
were 25 to 29 years old in these years. These effects are probably a lower bound as they refer to the entry of
women in the D-cohort in the 1940s and not in the 1950s, period that witnessed an important increase in
their presence in the labor market.
As can be seen, WWII has modest positive effects on births. It only increases births to women 25 to
29 years old in 1950 and 1952. Improvements in contemporaneous economic conditions (Failures) affect
young cohorts differently in 1960 and in 1940, but overall they tend to decrease births and follow the pattern
documented for previous decades by Jones and Tertilt (2006).
We now examine whether the retirement of the D-cohort can explain the decline in births in the
1970s. Table 15 presents the results for births to women who were 20 to 24 and 25 to 29 years old
throughout the 1960s and in 1940. We report results for births that occurred in each year in the 1960s versus
births to women in the same age bracket in 1940. The results are striking, while the crowding-out measure
predicted an increase in births, the crowding-in measure predicts a decline in births in almost every year for
both 20 to 24 and 25 to 29 years old women. Let’s take as example 1965, 20 to 24 years old women had on
average .164 births (in 1940, they had an average of .085 children), the crowding-in measure increases by
.106 (exit), and the retirement of women in the D-cohort predicts a 31% decrease in births between 1965 and
1940. Always taking 1965 and calculating the effects for women 25 to 29 years old, the crowding-in (in
1965 they had an average of .151 children) reduced their births by 47%.
One could argue that isolating the effect of a cohort on births may induce effects that could disappear
if one also includes other cohorts, as several cohorts enter and exit and the net effects may not be significant.
It is plausible that all cohorts induce some crowding-out and crowding-in for younger cohorts as they enter
or exit the labor market, and also affects their births. To check this we did several experiments in which we
include other cohorts that may have entered or exited the labor market in the 1950s and 1960s. In no case the
effect of the D-cohort was diminished. In Table 16 we examine whether the immediately older cohort, the
cohort that was 35 to 49 years old in 1930, and 55 to 69 years old in 1950, could have also induced a
crowding-out effect as they increased their presence in the labor market between 1940 and 1950 (see Figure
3). We calculate the crowding-out measure exactly as we did for the D-cohort and include both measures of
crowding-out (but the results don’t change if we include the cohorts separately). The results are reported
under the heading Specification I. As can be seen the crowding-out effect due to the entry of the D-cohort
remains as before, also quantitatively, while the entry of the older cohort does not significantly affect births
(in any year).17
In Table 17 we use the panels 1940-1970 to examine the impact of the retirement of the older cohort,
35 to 44 years old in 1940. Between 1950 and 1960 this cohort goes from being 55 to 64 to being 65 to 74
years old. We use the difference between their participation in 1960 and 1950, because clearly they retire
during the decade. Does their retirement produce an additional crowding-in effect? We include both the
retirement of the D-cohort and the retirement of this older cohort. The results are reported under the heading
of Specification I. As can be seen the estimates of the crowding-in effect of the exit of the D-cohorts are
similar to what we had found in Table 15, and the retirement of the older cohort produces no significant
17
In a second falsification check, we check whether the immediately younger cohort, 15 to 19 years old in 1930, could have decreased the impact of the crowdingout of the D-cohort. We cannot include the cohort that was 10 to 14 years old in 1930, because this cohort was 20 to 24 in 1940 and their births are part of our
1940-1960 panels, which would contaminate our analysis. Even the 15 to 19 years old are part of our reference year, but only when we examine the births of 25 to
29 years old women, which we therefore will not do. The addition of the younger cohort does not change the results we found for the D-cohort.
22
additional crowding-in effect on births. The results are the same if we examine their impact on the births of
women 25 to 29 years old.
Table 14: Annual Births (white women) & "Crowding-Out": 1940-1960
Dep. Variable = 1 if a birth took place in a given year (base year 1940)
Age group: 20-24 years old
average birth rate of 20-24 year olds in 1940: 0.08
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
mobilization
-0.170
-0.253**
-0.139**
-0.026
-0.07
-0.271** -0.301*** -0.423*** -0.358*** -0.498***
"crowding-out"
0.194**
0.215***
0.213***
0.226***
0.256*** 0.339*** 0.301*** 0.330***
0.358*** 0.466***
failures
0.042***
0.054*** 0.037**
0.068***
0.062*** 0.086*** 0.034**
0.081***
0.056** 0.070***
failures*d1960
-0.011
-0.032*** -0.05***
-0.052***
-0.062*** -0.055*** -0.034*** -0.066*** -0.040*** -0.050***
1960
-0.419***
0.376***
0.024
-0.004
observations
56364
56242
1959
0.114
0.161*
0.007
-0.013
1960
-0.118
0.239**
0.020
-0.031
62841
64232
57926
change:
0.116
62056
Age group: 25-29 years old
1950
mobilization
0.250**
"crowding-out"
0.142*
failures
0.028**
failures*d1960
-0.032*
observations
61183
1951
0.053
0.204***
0.023
-0.025**
64797
65207
"crowding-out"predictor:
60157
1952
0.295**
0.072
0.022
-0.002
58760
58330
57731
57339
56724
56935
average birth rate of 25-29 year olds in 1940: 0.08
1953
1954
1955
1956
1957
1958
0.092
0.130
0.139
0.126
-0.025
-0.010
0.277***
0.245**
0.139*
0.316*** 0.221**
0.274***
0.066***
0.062**
0.033**
0.048** 0.048**
0.050***
-0.038*** -0.029
-0.036*** -0.027*
-0.045*** -0.033**
66216
58132
mean 1940:
0.05
58562
59959
60985
mean 1960:
0.066
61858
Notes : Reported coefficients are OLS estimates from a regression of an indicator of whether a birth took place in a given year (1940-1951, 1940-1952, … , 1940-1960 with 1940 the year of reference)
Table 15: Annual Births (white women) & "Crowding-In": 1940-1970
Dep. Variable = 1 if a birth took place in a given year (base year 1940)
Age group: 20-24 years old
average birth rate of 20-24 year olds in 1940: 0.08
1961
1962
1963
1964
1965
1966
1967
mobilization
-0.041
-0.090
-0.199**
-0.205**
-0.288*** -0.267*** -0.196*
"crowding-in"
-0.170**
-0.064
-0.175**
-0.256*** -0.221***
-0.282*** -0.215**
failures
0.033*
0.002
0.012
0.030**
0.026
0.029**
0.016
failures*d1970
-0.041***
-0.014*
-0.011
-0.035***
-0.031*** -0.011
-0.005
observations
60254
62814
mobilization
"crowding-in"
failures
failures*d1970
1961
0.076
-0.295***
0.025
-0.028*
1962
0.186*
-0.235***
0.018
-0.013
observations
56926
56812
57314
mean 1940:
0.052
Age group: 25-29 years old
"crowding-in/retirement"predictor:
65415
67060
68090
70771
73371
average birth rate of 25-29 year olds in 1940: 0.08
1963
1964
1965
1966
1967
0.049
0.184
0.169*
0.387*** -0.014
-0.310***
-0.187**
-0.293***
-0.137* -0.312***
0.018
0.013
0.019
-0.014
0.028*
-0.014
-0.025*
-0.024*
-0.013
-0.021*
57987
58754
60056
mean 1960:
0.158
62616
1968
1969
-0.181*
-0.203**
-0.243*** -0.229***
0.032**
0.033***
-0.022**
-0.014
74500
75700
1968
0.205*
-0.164**
-0.006
-0.015
1969
0.090
-0.268***
0.027*
-0.003
65217
66862
change:
0.106
Notes : Reported coefficients are OLS estimates from a regression of an indicator of whether a birth took place in a given year (1940-1961, 1940-1962, … , 1940-1969 with 1940
being the year of reference) on WWII mobilization rates, business failures in the year prior to birth year and a measure of the change in work behavior of the "d-cohort"
(named "crowding-out"). See text for a detailed definition of this variable. The "d-cohort" consists of women 20-34 years old in 1930. Other controls: age dymmies, 1940
share of men that are farmers, 1940 share of nonwhite men, average male education in 1940, state of residence and state of birth dummies, year fixed effects. All controls
except the state and year dummies as well as the "crowding-out" variable areinteracted with a year dummy. Sample includes white women born in the United States.
Standard errors are clustered at the state of residence-year level. ***, **, * refer to 1%, 5% and 10% significance level respectively.
These falsifications reinforce our hypotheses that the causality runs from the D-cohort to the younger
cohorts and not vice versa. If the young cohorts were voluntarily leaving the market, one would expect that
other cohorts would also be affected, but we find no such effect. Instead, in all cases we find that only the
life cycle employment measure of the D-cohort produces the effect of significantly increasing births in the
1950s and decreasing them in the 1960s.
23
Table 16: Falsifications - Baby Boom & Labor Supply of Various Cohorts: 1940-1960
Dep. Variable = 1 if a birth took place in a given year (base year 1940)
Age group: 20-24 years old
average birth rate of 20-24 year olds in 1940: 0.08
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
Specification I
"crowding out"
0.186
0.219
0.207
0.235
0.259
0.327
0.297
0.318
0.350
0.463
0.388
(D-cohort: 20-34 yrs old in 1930)
(0.081)** (0.074)*** (0.064)*** (0.069)*** (0.085)*** (0.087)*** (0.071)*** (0.093)*** (0.084)*** (0.093)*** (0.107)***
"false crowding out"
(cohort 35-49 yrs old in 1930)
0.049
(0.103)
-0.021
(0.103)
0.041
(0.107)
1950
1951
1952
Age group: 25-29 years old
"crowding out"
(D-cohort: 20-34 yrs old in 1930)
0.155
0.215
(0.077)** (0.076)***
0.079
(0.101)
-0.056
(0.088)
-0.017
(0.101)
0.084
(0.103)
0.035
(0.091)
0.105
(0.118)
0.057
(0.116)
average birth rate of 25-29 year olds in 1940: 0.08
1953
1954
1955
1956
1957
1958
Specification I
0.288
0.261
0.158
0.331
0.245
0.283
(0.083)*** (0.110)** (0.079)** (0.092)*** (0.108)** (0.083)***
0.024
(0.117)
0.129
(0.132)
1959
1960
0.172
(0.089)*
0.163
(0.082)**
"false crowding out"
(cohort 35-49 yrs old in 1930)
-0.076
(0.113)
-0.064
(0.099)
-0.043
(0.107)
-0.067
(0.099)
-0.101
(0.114)
-0.117
(0.113)
-0.110
(0.115)
-0.201
(0.121)*
-0.061
(0.097)
-0.079
(0.114)
-0.075
(0.128)
observations
66216
65207
64797
64232
62841
61858
60985
59959
58562
58132
57926
Notes : Reported coefficients are OLS estimates from a regression of an indicator of whether a birth took place in a given year (1940-1951, 1940-1952, … , 1940-1960 with 1940 being the year of reference) on WWII
mobilization rates, business failures in the year prior to birth year and a measure of the change in work behavior of the cohorts of women who were 35-49 and 15-19 yrs old in 1930 (the later combined with the D-cohort)
. See text for a detailed definition of this variable. Other controls: age dymmies, 1940 share of men that are farmers, 1940 share of nonwhite men, average male education in 1940, state of residence and state of birth
dummies, year fixed effects. All controls except the state and year dummies as well as the "crowding-out" variable are interacted with a year dummy. Sample includes white women born in the United States. Standard
errors are clustered at the state of residence-year level. ***, **, * refer to 1%, 5% and 10% significance level respectively.
Table 17: Falsifications - Baby Bust & Labor Supply of Various Cohorts: 1940-1970
Dep. Variable = 1 if a birth took place in a given year (base year 1940)
Age group: 20-24 years old
average birth rate of 20-24 year olds in 1940: 0.08
1961
1962
1963
1964
1965
Specification I
"crowding out"
-0.185
-0.072
-0.140
-0.235
-0.161
(D-cohort: 20-34 yrs old in 1930)
(0.088)**
(0.083)
(0.074)* (0.076)*** (0.082)**
"false crowding out"
(cohort 35-44 yrs old in 1930)
Age group: 25-29 years old
0.035
(0.071)
-0.089
(0.072)
-0.043
(0.081)
-0.130
(0.078)*
1967
1968
1969
-0.253
-0.224
-0.266
-0.231
(0.094)*** (0.102)** (0.080)*** (0.092)***
-0.058
(0.084)
0.017
(0.081)
average birth rate of 25-29 year olds in 1940: 0.08
1963
1964
1965
1966
1967
Specification I
-0.278
-0.219
-0.325
-0.214
-0.292
-0.082
-0.297
(0.097)*** (0.099)** (0.105)*** (0.095)** (0.077)***
(0.097) (0.099)***
1961
"crowding out"
(D-cohort: 20-34 yrs old in 1930)
0.019
(0.087)
1966
1962
0.049
(0.083)
0.004
(0.070)
1968
1969
-0.142
(0.086)*
-0.267
(0.083)***
"false crowding out"
(cohort 35-44 yrs old in 1930)
-0.041
(0.109)
-0.038
(0.087)
0.036
(0.118)
0.056
(0.107)
-0.001
(0.093)
-0.106
(0.096)
-0.029
(0.102)
-0.046
(0.089)
-0.0007
(0.093)
observations
56926
56812
57314
57987
58754
60056
62616
65217
66862
Notes : Reported coefficients are OLS estimates from a regression of an indicator of whether a birth took place in a given year (1940-1961, 1940-1962, … , 1940-1969 with 1940 being the year of reference)
on WWII mobilization rates, business failures in the year prior to birth year and a measure of the change in work behavior of the cohorts of women who were 35-44 and 15-19 yrs old in 1930 (the later
combined with the D-cohort). See text for a detailed definition of this variable. Other controls: age dymmies, 1940 share of men that are farmers, 1940 share of nonwhite men, average male education in
1940, state of residence and state of birth dummies, year fixed effects. All controls except the state and year dummies as well as the "crowding-out" variable are interacted with a year dummy. Sample
includes white women born in the United States. Standarderrors are clustered at the state of residence-year level. ***, **, * refer to 1%, 5% and 10% significance level respectively.
6. Crowding-out, Crowding-in and Completed Fertility
It is difficult to produce a measure of crowding-out and crowding-in that can be used to examine
completed fertility. For this reason we examine the impact of the crowding-out and crowding-in on the
number of children ever born in exactly the same fashion as we did for yearly births. We examine how the
change in the labor supply behavior of the D-cohort in the 1940s, 1950s and 1960s impacted on the total
number of children ever born to a woman by the age of 24 and 29 years old respectively, as well as for all
women 25 to 29 years old in 1960 and 1970 (relative to women of the same age in 1940). More births will
occur in the 1970s, but we focus on capturing the cumulative births that occurred when the labor markets
were significantly being affected by the entry and retirement of the D-cohort and when these young women
(up to age 29 in 1940, 1960 and 1970) were in their prime childbearing years. The 1950s and 1960s are also
the periods in which we observe the largest increase in yearly births as well as the largest decline.
The results are presented in Table 18. We consider three alternative definitions of the dependent
variable: the total number of children ever born by a given age, whether a woman had given birth to more
24
than 3 and more than 2 children respectively by a given age. The rational for the last two specifications is
the following. Women born, for instance, between 1931-1935, who were 25-29 years old in 1960, had the
highest average completed fertility of 3.21 children. While completed fertility is typically measured upon the
end of the woman’s fertility horizon (at the age of 40 or later), a finding where by the age of 29 a woman has
surpassed the average of 3 children, would suggest that most of the lifetime fertility has been realized by that
age. Our question is whether the labor market behavior of the D-cohort has significantly induced the
younger cohorts to attain this threshold during the boom or fall short of that during the bust.
Our results on cumulative fertility are in line with the direction of the effects established for yearly
births. The crowding-out significantly increased the average number of children born and the crowding-in,
significantly decreased it. Let’s compare the cumulative births of women who were 29 years old in 1960
versus the cumulative births of women of the same age in 1940. Women in that age bracket had an average
of 2.4 children by 1960, versus an average of 1.23 children by 1940. The entry of the D-cohort in the labor
market increased their cumulative fertility by 20% (0.116*2.072/(2.4-1.23)). This is an underestimate of the
effects as we are not including in the measure the women in the D-cohort who entered in the labor market in
the 1950s. Let’s now examine the quantitative relevance of the crowding-in effect. A 29 year old woman
had an average of 2.19 children by 1970 and an average of 1.23 children by 1940. Doing the same
calculations as before, we find that the retirement of the D-cohort in the 1960s, reduced the average number
of children ever born by 38% (0.106*3.453/(2.19-1.23)).
Table 18: Cumulative fertility (white women) : "Crowding-Out" & "Crowding-In"
Dep. Variable:
Children ever born by age:
Have more than 3 children by age:
1940-1960
1940-1960
Age groups:
24
29
25-29
24
29
25-29
[mean dependent variable]
[1.432]
[2.034]
[1.845]
[0.059]
[0.156]
[0.125]
[change dep. var. 1960-1940]
[0.536]
[0.795]
[0.776]
[0.036]
[0.089]
[0.074]
Have more than 2 children by age:
1940-1960
24
29
25-29
[0.180]
[0.339]
[0.288]
[0.102]
[0.214]
[0.195]
"crowding out"
(cohort 20-34 yrs old in 1930)
0.542
(0.173)***
2.230
(0.593)***
2.072
(0.661)***
2.094
(0.455)***
0.289
(0.104)**
0.493
(0.159)***
0.318
(0.087)***
0.591
0.655
(0.197)*** (0.143)***
[change 19601940: 0.116]
observations
Dep. Variable:
Age groups:
[mean dependent variable]
[change dep. var. 1970-1940]
"crowding out"
(cohort 20-34 yrs old in 1930)
9634
10793
51568
Children ever born by age:
1940-1970
24
29
25-29
[1.243]
[2.016]
[1.723]
[0.174]
[0.748]
[0.506]
9634
10793
51568
Have more than 3 children by age:
1940-1970
24
29
25-29
[0.036]
[0.149]
[0.101]
[-0.005]
[0.074]
[0.026]
9634
10793
51568
Have more than 2 children by age:
1940-1970
24
29
25-29
[0.129]
[0.334]
[0.249]
[0.007]
[0.202]
[0.111]
0.434
(1.523)
-1.799
(1.697)
-1.780
(1.043)*
0.124
(0.253)
0.278
(0.376)
-0.213
(0.208)
0.233
(0.4910
-1.063
(0.909)
-3.453
(1.146)***
-2.715
(0.696)***
-0.037
(0.129)
-0.358
(0.256)
-0.369
(0.137)***
-0.156
(0.311)
10750
11224
59164
10750
11226
59164
10750
-0.551
(0.461)
-0.633
(0.292)**
[change 19601950: 0.091]
"crowding in"
(cohort 20-34 yrs old in 1930)
-0.925
-0.875
(0.287)*** (0.186)***
[change 1970-1940: 0.106]
observations
11226
59164
Notes: Reported coefficients are OLS estimates from a regression of the number of children ever born to a woman of a given age on a measure of the change in work behavior of the "d-cohort" (named
"crowding-out" & "crowding-in") in the 1950s and the 1960s. See text for a definition of these variables. The "d-cohort" consists of women 20-34 years old in 1930. All regressions also control for state of
residence, state of birth, age and calendar year dummies. Age effects are also interacted with year dummies. Sample includes white women born in the United States. Standard errors are clustered by state of
residence and year. ***, **, * refer to 1%, 5% and 10% significance level respectively.
7. Crowding-Out or Easterlin?
In this section we perform another falsification check by examining whether our measure of
crowding-out reflects the impact of the Great Depression via a different channel than the labor market
mechanism outlined so far. Easterlin’s hypothesis (1961) postulates that women who grew up in the Great
Depression years had low material aspirations. However, as they witnessed substantially improved economic
conditions and renewed optimism during their childbearing years, they revised upwards their earning
potential expectations and had overall more children. Following this rational, we compare the average
economic conditions a women experienced when 15-29 years, that is during her prime fertility years, to the
average conditions experienced during childhood. The latter is defined by the period when 5-10 years old.
We proxy the state of the economy with the business failure rate, a measure we consistently observe at the
25
state level between 1900 and 1968. We assume that the higher the failures when women were in their
formative years, the lower their material aspirations. We calculate the ratio between the average failures
when 15 to 29 years old, and the average failures they faced when they were 5 to 10 years old, that is in their
formative years. The lower this ratio, the more important the improvements in their relative economic
situation between when they were children and when they were young women. This, following Easterlin’s
interpretation, should lead to a preference shift towards earlier marriages and larger families.
The results of the estimates for different cohorts of women born between 1901 and 1940 are reported
in Table 19. We report three specifications. In the first one we include only average failures when women
were 15 to 29 years old. As can be seen these are always significant and negative for all cohorts, suggesting
that as economic conditions improved, women were more like to produce more children. Specification II
further controls for the average failures when women were 5 to 10 years old. Failures when 15 to 29 remain
significant while failures when 5 to 10 years old are non-significant except for the cohort born between 1926
and 1930. In Specification III, we include the constructed ratio, while still accounting for the conditions
during the childbearing phase. The ratio is significant only for the 1926-1930 cohort, but the sign is the
opposite than one would expect. An improvement in their relative economic situation actually decreases
their completed fertility. These results provide more support to our hypothesis that the causality runs from
the old to the young, rather than from the young to old via a preference shift.
Table 19: Completed Fertility: The Role of Economic Conditions & the Easterlin Hypothesis
Dep. Variable: children ever born
*d1901_1905
-0.261
(0.120)**
*d1906_1910
-0.179
(0.096)*
*d1901_1905
-0.328
(0.181)*
*d1906_1910
-0.246
(0.135)*
*d1901_1905
0.134
(0.107)
*d1901_1905
-0.137
(0.142)
*d1901_1905
-0.086
(0.071)
Specification I : Average Failures when 15-29 years old
*d1911_1915 *d1916_1920 *d1921_1925 *d1926_1930 *d1931_1935
-0.242
-0.304
-0.439
-0.327
-0.299
(0.071)***
(0.095)***
(0.125)***
(0.141)**
(0.124)**
Specification II : Average Failures when 15-29 years old
*d1911_1915 *d1916_1920 *d1921_1925 *d1926_1930 *d1931_1935
-0.304
-0.386
-0.512
-0.257
-0.299
(0.091)***
(0.108)***
(0.138)***
(0.148)*
(0.139)**
Average Failures when 5-10 years old
*d1906_1910 *d1911_1915 *d1916_1920 *d1921_1925 *d1926_1930 *d1931_1935
0.101
0.079
0.031
-0.017
-0.144
-0.124
(0.089)
(0.063)
(0.056)
(0.054)
(0.067)**
(0.07)*
Specification III : Average Failures when 15-29 years old
*d1906_1910 *d1911_1915 *d1916_1920 *d1921_1925 *d1926_1930 *d1931_1935
-0.072
-0.127
-0.221
-0.419
-0.464
-0.373
(0.119)
(0.103)
(0.157)
(0.219)*
(0.163)***
(0.134)***
Ratio: Average Failures when 15-29 years old/Average Failures when 5-10 years old
*d1906_1910 *d1911_1915 *d1916_1920 *d1921_1925 *d1926_1930 *d1931_1935
-0.136
-0.136
-0.133
-0.089
0.133
0.026
(0.071)*
(0.060)**
(0.142)
(0.258)
(0.056)**
(0.023)
*d1936plus
-0.361
(0.132)***
*d1936plus
-0.449
(0.143)***
*d1936plus
-0.044
(0.197)
*d1936plus
-0.426
(0.139)***
*d1936plus
-0.0009
(0.005)
Observations = 262681
Notes : Reported coefficients are OLS coefficients from a regression of the number of children ever born to a woman on the average economic
conditions experienced when 15-29 years old (baseline-specification I ). Additional covariates: average economic conditions experienced
when 5-10 years old (specification II ), the ratio of average economic conditions when 15-29 years old vs average economic conditions when
5-10 years old (specification III ). Economic conditions refer to the individual's state of birth. Birth year, birth state fixed effects and a statespecific linear time trend are included in all specifications. Sample includes white, ever married, non-farm women born in the United States.
Standard errors are clustered at the state of residence-birth year level. ***, **, * indicate significance at 1%, 5% and 10% significance levels
respectively.
8. Conclusions
This paper revisits the determinants of the baby-boom and baby-bust and proposes a new theory
which attributes their origins to the Great Depression. While the most well-known theory of the Great
Depression links the baby-boom to a shift in the preference of women who grew up in the 1930s towards
lower material aspirations, our story hinges upon the older cohort of women who were of working age
during the Great Depression. These women, whom we name D-cohort, were 20 to 34 years old in 1930 and
in the face of adverse conditions that affected their family income and the well-being of their children,
entered the labor market and kept entering decades after. While this paper does not analyze the mechanisms
26
that fostered and sustained their persistent entry decades after the depression years were over, the Added
Worker Effect explains the initial entry of married women when their husbands were losing their jobs.
Further work will be needed to understand whether their entry decades later is linked to wealth losses during
the Great Depression, or a reduction in their family permanent income or to a preference shift.
We use several panels to document this entry that precedes the entry of the younger cohorts of
women in the labor market in the 1950s and 1960s. We show that neither the entry of this cohort, nor the
decreased presence of the younger cohorts of women in the labor market is linked to WWII. We further
establish that the work behavior of the D-cohort significantly increased births in the 1950s, and that at the
end of the 1950s, it led to a reversal in the trend, and to a rapid decline in births. The effects are
quantitatively important, with an order of magnitude of 20 to 50%. The participation of the D-generation
after 1940 remains striking and seemingly capable of substantially altering the labor market opportunities of
new entrants. The age of the D-cohort in 1960, is just right for the flip over. If such a large cohort could
decrease employment opportunities, their exit could do just the contrary.
We provide several pieces of evidence which would be difficult to reconcile with other possible
explanations that fit all the pieces together. These are: 1) the same and only cohort (D-cohort) stays and
keeps entering the labor market between 1930 and 1960; by 1960 these women start to retire and by 1970
very few are still working; 2) an opposite entry/exit pattern for the old/young cohorts that is robust and
found both in the 1940-1950 and in the 1940-1960 panels; 3) the Great Depression is associated with lower
wages for nearly all cohorts of women, including young cohorts in 1950 and 1960 that were not yet of
working age in the 1930s. These effects are also found when we use the 1930 share of women 20 to 34 years
old (D-cohort) who were working, in lieu of average economic conditions in 1930. This is an alternative
way to address and assess the consistency of our proposed mechanism; 4) these patterns are not found for
men or when we use failures from earlier periods (early 1920s) or other cohorts.
9. Bibliography
Ananat Elisabeth Oltmans, Anna Gassman-Pines and Christina Gibson-Davis, (2013) “Community-Wide
Job Loss and Teenage Fertility: Evidence from North Caroline,’’ Demography, (50, pp. 2151-2171.
Acemoglu, Daron, David H. Autor, and David Lyle (2004). “Women, War and Wages: The Effect of
Female Labor Supply on the Wage Structure at Mid-Century,” Journal of Political Economy, 112 (3), pp.
497-551.
Albanesi, Stefania and Claudia Olivetti (2013). “Maternal Health and the Baby Boom,’’ Quantitative
Economics, Forthcoming
Bellou Andriana and Emanuela Cardia (2013), “Occupations after WWII: The Legacy of Rosie the Riveter”,
Université de Montréal, mimeo,
Bailey Martha J. (2006). “More Power to the Pill: The Impact of Contraceptive Freedom on Women’s Life
Cycle Labor Supply,” The Quarterly Journal of Economics, 121 (1), pp. 289-320.
Bailey Martha J. and William J. Collins (2011). “Did Improvements in Household Technology Cause the
Baby Boom? Evidence from Electrification, Appliance Diffusion, and the Amish,’’ American Economic
Journal: Macroeconomics 3 (2), pp. 189-217.
Doepke Matthias, Hazan M. and Yishay D. Maoz (2012). “The Baby Boom and World War II: A
Macroeconomic Analysis,” mimeo.
Easterlin, Richard A. (1961). “The American Baby Boom in Historical Persepctive,’’ American Economic
Review, 51 (5), pp. 869-911.
27
Finegan Aldrich T. and Robert A. Margo (1994). “Work Relief and the Labor Force Participation of Married
Women in 1940,’’ The Journal of Economic History, 54 (1), pp. 64-84.
Goldin, Claudia (1990). “The Role of World War II in the Rise of Women’s Employment,” American
Economic Review, 81 (4), pp. 741–756.
Goldin, Claudia, and Claudia Olivetti (2013). “Shocking Labor Supply: A Reassessment of the Role of
World War II on Women’s Labor Supply,” American Economic Review P&P, 103(3), pp. 257-262.
Greenwood, Jeremy, Ananth Seshadri, and Guillerme Vanderbroucke (2005). “The Baby Boom and the
Baby Bust,’’ American Economic Review, 95(1), pp.183-207.
Jones, Larry E., and Michele Tertilt. (2008) “An Economic History of Fertility in the United States: 18261960,’’ Chapter 5 of Frontiers of Family Economics, edited by Peter Rupert. Bradford: Emerald.
U. S. Census Bureau, Statistical Abstract of the United States (Washington, DC:various years)
Ruggles, Steven, J. Trent Alexander, Katie Genadek, Ronald Goeken, Matthew B. Schroeder and Matthew
Sobek (2010). Integrated Public Use Microdata Series: Version 5.0. Minneapolis: University of Minnesota.
Selective Service System (1956). Special Monographs of the Selective Service System. Vols. 1–18.
Washington, D.C.: Government Printing Office.
28
29