teenage kicks: new (old) evidence on the pill and teenage childbearing
TRANSCRIPT
Teenage Kicks: New (Old) Evidence on the Pill andTeenage Childbearing
Kelly S. Ragan∗
Stockholm School of Economics
March 4, 2015
Abstract
How did the introduction of oral contraception (‘the Pill’) alter teenage childbearingin Sweden? Teen fertility was halved in the decade following the Pill’s introduction.New data on oral contraceptive sales reveals that the largest declines occurred in com-munities with high take-up of the Pill. Differences-in-differences-in-differences (DDD)comparisons, exploiting time variation across localities and age groups, point to a strongnegative relationship between Pill use and fertility. Illegitimacy patterns from a centuryearlier are used as instruments to isolate that part of Pill use which is predetermined. IVestimates imply that the Pill’s diffusion could account for the entire decline in teenagechildbearing observed in the data. The estimated effect on non-marital childbearing islarge and negative; the data do not support the predictions of Akerlof et al (1996) butare consistent with the female empowerment model of Chiappori and Orrefice (2008).JEL Codes:Keywords: contraception, teenage childbearing, out-of-wedlock birthPreliminary and Incomplete
∗Address: Stockholm School of Economics, Box 6501, 113 83 Stockholm, Sweden. [email protected] work was supported by the Swedish Science Council (Grant 2012-643) and by the Swedish Royal Acad-emy of Science. I appreciate helpful comments from seminar participants at the NBER Summer Instituteand SOFI, Stockholm University.
1
1 Introduction
This article presents new Swedish data on the use of oral contraceptives (’the Pill’) and
develops an empirical approach to isolate a causal channel between Pill use and teen fertility.
I establish a strong first stage relationship between illegitimacy in 1860 and Pill take-up a
century later. I use these predetermined differences in Pill adoption to isolate the causal
effect of the Pill on teenage childbearing. The identifying assumption behind my choice of
instrument is consistent with the teen fertility data along many dimensions including placebo
policy reforms, the time pattern of reduced form effects, as well as overidentification tests
using alternative instruments. The Pill had a statistically and quantitatively significant
effect on teenage childbearing that could explain the halving of the teen birth rate after
the Pill was introduced. By identifying the effect of Pill use on teen fertility this paper
quantifies an empirical relationship that previous reduced form estimates have not been able
to characterize.
There are few more compelling examples of technology as liberator than the Pill. Yet,
the literature is divided regarding its importance. Declines in fertility were underway well
before the Pill was introduced.1 These long run trends, common across many countries, led
Becker (1991) and others to conclude that the Pill may have only represented a shift along
the technological frontier rather than a contraceptive innovation. Even if the Pill was an
improvement relative to existing contraceptive methods, the theoretical literature is ambigu-
ous regarding the Pill’s probable effects on nonmarital childbearing. Akerlof, Yellen, and
Katz (1996) provide a theoretical model of contraceptive innovation and female immisera-
tion, and argue that the Pill could increase the rate of out-of-wedlock childbearing among
young women.2 Chiappori and Oreffi ce (2008) develop a matching model with an explicit role
for contraception which clearly predicts that universal access to an inexpensive and highly
effective mode of contraception, such as the Pill, should reduce non-marital childbearing. My
IV approach, unlike previous quasi-experiments, is not limited to shifting Pill take-up among
either married or unmarried teens; hence I can decompose fertility effects by marital status.
1A prominent study of the Swedish case can be found in Schultz (1985).2Akerlof et al (1996) emphasize how the pill and liberalized abortion access altered ’shotgun marriage’
customs and reduced the bargaining power of women with preferences against contraception, inducing someto engage in premarital sex whom otherwise would not, a force leading to increased non-marital fertility.
2
Theory is ambiguous, but the data is clear. The Pill substantially reduced nonmarital child-
bearing among teens in both absolute and relative terms. The data do not support Akerlof
et al (1996), but are consistent with the predictions of Chiappori and Oreffi ce (2008).
Goldin and Katz (2002) use legal reforms that altered contraceptive access for unmarried
women between the ages of 18 and 21 as a source of exogenous variation in Pill use to
estimate the effect of Pill access on marital delay and career choice among young college
educated women in the U.S. Much of the ’power of the Pill’literature has focused on early
legal access to the Pill, not Pill use per se. Bailey (2010) uses the repeal of Comstock laws
which banned contraceptive sales to estimate the effect of Pill access on marital fertility.
Ananat and Hungerman (2012), Bailey (2006), and Bailey, Hershbein and Miller (2012)
are prominent examples which use family planning policy and legal reforms that altered
contraceptive access to estimate how Pill access affected women and the well-being of their
children. These studies generally find negative fertility effects. Yet, recent work by Myers
(2012) argues that once abortion access is taken into account, confidential access to the Pill
had little effect on women’s propensity to marry or have a child at a young age. Focusing on
teenagers does not resolve this ambiguity. Using a quasi-experimental design Guldi (2008)
finds small negative fertility effects of confidential early access to the Pill among white teens
but no effect for other teenagers.3
This paper empirically establishes a negative causal relationship between Pill use and teen
fertility in total, as well as among subpopulations of teens, using a new empirical strategy
which does not rely on variation in contemporary legal reforms or family planning policy.
The Swedish institutional setting mitigates confounding factors such as abortion legalization
emphasized by Myers (2012) and Joyce (2013).4 Concerns related to policy endogeneity in
the quasi-experimental literature and potential bias when relying on legal reforms as a source
of identification are moot since the Pill was available to all women over the age of fifteen
with no restrictions regarding marital status or parental consent. The instrumental variables
approach I use relies on a very different source of variation in Pill take-up than any previous
3This is not inconsistent with the literature on teen pregnancy prevention interventions. DiCenso et al(2002) survey 22 randomized studies and present a meta-analysis of interventions to reduce teen pregnancy.They find little effect of these interventions on contraceptive use or pregnancy among teens.
4Abortion was illegal in Sweden until 1975.
3
study. The reduced form estimates I report point to a sizable fertility reduction among teens
related to confidential Pill access. Moreover, I estimate first stage equations that clearly
establish the channel through which exogenous variation in Pill use operates, and complement
reduced form analysis with a full model where the causal chain is made explicit.5
Decomposing fertility effects by marital status not only allows us to discern which theory
match the data but also presents new puzzles with respect to the role the Pill played in the
secular declines in marital childbearing among teens observed in the data. Total teen fertility
declines match up well with the diffusion of the Pill. Consistent with Bailey (2010), I find
that marital fertility among married teens declines significantly. Yet, my estimates overstate
declines in non-marital fertility and predict an increase in marital fertility, consistent with
the positive correlation in the cross section data between the diffusion of the Pill and marital
fertility among teens. To better understand the observed and predicted increase in marital
fertility with respect to Pill diffusion I examine how the Pill altered marital behavior. My
estimates suggest that the Pill increased marriage among teens, consistent with increased
marital childbearing, but at odds with the marital delay mechanism emphasized in Goldin
and Katz (2002).
The empirical approach developed here could be used to evaluate the impact of the Pill on
a host of other behaviors. Establishing the impact of the Pill on fertility is a necessary first
step before one can establish any causal relationship between the Pill and other downstream
outcomes such as earnings or child outcomes. The paper proceeds with an overview of the
introduction of the Pill in Sweden with particular attention to the data used in this study
as well as important aspects of the institutional environment. A first look at the data on
the Pill and teenage childbearing is then presented. Using variation across time, across
localities, and across age groups with differential take-up rates I establish a strong negative
relationship between the Pill and teen fertility. Section 4 presents the empirical model and
the identifying assumptions behind the instrumental variables (IV) approach. Section 5
makes a case for the instrument, showing how illegitimacy from 1860 is highly correlated
with Pill use, but uncorrelated with changes in teen fertility before the Pill is introduced.
5Bailey, Hershbein and Miller (2012) provide evidence that early legal access increased pill use, but theirmain results, consistent with this literature focus on estimating the effect of early legal access on behavioraloutcomes. It should be noted that they do report first stage results for rural women.
4
Placebo policy estimates and intention to treat (ITT) effects implied by the empirical model
match up precisely with the timing of the Pill’s introduction. Section 6 presents the main
empirical results on teenage childbearing, while Section 7 decomposes the effect of the Pill
with regard to marital status. Section 8 concludes.
2 The Introduction of the Pill in Sweden
The Pill was approved for contraceptive use by the Swedish Board of Health and Welfare
in May of 1964. Pill use grew rapidly. Three percent of women between the ages of 15
and 44 were on the Pill in 1965, ten percent a year later, and by 1969 a quarter of fecund
females were on the Pill.6 Only condoms and coitus interruptus surpassed the Pill as a means
of preventing pregnancy among Swedish couples.7 By 1976, 30 percent of women in their
reproductive years were using the Pill annually, and more than half of teenage girls were on
the Pill by 1979.8 The unique data on Pill use and the institutional setting surrounding the
Pill’s introduction and rapid adoption are described below.
2.1 Data on Pill Use
I quantify the specific channel of the Pill with data on oral contraceptive (OC) sales. Data
on OC sales by locality are constructed from 1970 onward using the quarterly Swedish Drug
Market publication from Läkemedelstatistik, AB. This publication presents complete infor-
mation on OC sales across 70 local markets that constitute the entire universe of OC sales in
Sweden. Although the Pill was introduced in 1964, local sales data is first available in 1970.
Previous studies have relied on retrospective surveys to determine whether and which types
of contraceptives women used at different points in time. Individual level surveys are useful
in eliciting information on which types of birth control methods women had experience with,
but this information on the extensive margin is an incomplete picture of women’s exposure
to the Pill. I take a different approach. I use actual sales data. Pill sales capture both the
6Swedish Board of Health and Welfare (1984).7See Lewin (2000). There are no recurring surveys on contraceptive use during this period so it is not
possible to see how pill use evolved relative to other contraceptives.8See Table 2 of Swedish Board of Health and Welfare (1984).
5
extensive margin, more and more women opt to try the Pill, as well as the intensive margin
that women continue to take the Pill. The direct quantification of Pill use proximate to the
Pill’s introduction is an important contribution of this study.
Pill use is quantified in terms of expenditures per woman or teen. Geographically disag-
gregated data does not distinguish between different types of OC and only reports sales in
Swedish Krona (SEK). Converting sales into doses requires a price index, and the additional
assumption that the composition of OC sales across communities is uniform. Converting sales
data into doses introduces measurement error, to avoid this the main results are presented
in terms of sales. Table 1 presents data on OC sales and price data on the most popular
brand of OC by year. Price data can be used to transform sales into doses. For example,
Follinyl was the most popular OC accounting for over 27 percent of sales in 1970. A 21 day
regimen of Follinyl was priced at 2.94 SEK, less than the price of a movie ticket. Annual OC
sales correspond to over 400,000 women using the leading brand annually. According to the
leading brand price index, roughly a quarter of fecund females were using the Pill in 1970,
consistent with the data reported by the Swedish Board of Health and Welfare (1984).
2.2 Swedish Institutional Setting
Use of the Pill was widespread, especially among young women. Swedish law provided women
as young as 15 access to contraceptive services without regard to parental knowledge or
consent. Maternal health clinics provided information about contraceptives and supplied di-
aphragms to women regardless of their marital status, and had since the 1950s. The Riksför-
bundet för Sexuell Upplysning (RFSU) kept a list of doctors known to provide contraceptives
but by the late 1960s they deemed this unnecessary as doctors were universally willing to
provide contraceptive services to women.9
The assumption of uniform institutions may be diffi cult to support in some settings, but in
the context of Sweden this assumption is accurate in both the de jure and de facto sense. The
laws regulating the sales of contraceptives do not differ by jurisdiction and are set by national
regulatory bodies. In addition, the medical and retail pharmaceutical sectors in Sweden are
highly regulated, and almost entirely operated by public entities that are subject to central
9See Linner (1967).
6
administration.10 Prescription drugs could only be distributed by a publicly administrated
network of pharmacies whose assortment of drugs, staffi ng, and hours of operation were
subject to public supervision. Drug prices were fixed and did not vary across markets. In
this unique setting where the supply curve for the Pill was flat, fixed and identical across
markets, differences in take-up are arguably demand driven. The importance of the demand
side of the market in determining Pill use will inform the IV approach.
3 A First Look at the Data
Using data on OC sales as a prism to shed light on the fertility decisions of teenagers helps
build the case that the Pill’s diffusion played a causal role in the steep decline in teenage
childbearing observed in Sweden. Pill use in Sweden was heavily skewed toward young women.
By the late 1970s more than half of teenaged girls were using the Pill. Given that such a
large share of teens came to use the Pill it is natural to ask whether there is any evidence in
the aggregate that the Pill altered fertility patterns among teenagers.
Time series and cross-section data illustrate how the introduction of the Pill coincided
with a steep drop in teenage fertility that was largest in communities where take-up of the Pill
was greatest. This analysis is combined with data on the fertility patterns of older women
who used the Pill at much lower rates to construct a differences-in-differences-in-differences
(DDD) estimator of the Pill’s impact on teen childbearing. The descriptive statistics point
to a strong negative relationship between the Pill and teen fertility.
3.1 Time Series Data on Teen Fertility
Figure 1 shows how teenage childbearing fell by nearly 2 births per 100 women, a 50 percent
decline, in the decade following the Pill’s introduction.11 Teenage marital fertility essentially
disappeared after the Pill, declining from 1.5 births per hundred teens in 1965 to less than
0.3 births in 1975. Non-marital fertility also declined.
10In addition to these features of the legal and medical environment, it should also be noted that sexeducation has been compulsory in Sweden since 1956, and that the guidelines for the sex education curriculumare set at the national level.11Oral contraceptives were first approved for use in May of 1964.
7
Figure 1: Teenage Fertility 1961-1975
The distribution of teen fertility rates across markets is plotted in Figure 2. The sharp
decline in teenage childbearing is reflected in the leftward shift in the distribution from 1965
to 1969. In 1965, 20 percent of teens lived in communities with three or fewer births per
100 teens. By 1969 80 percent of teens lived in communities with less than three births per
100 teens, and by 1973 almost all teens lived in areas with less than three births.
0.2
.4.6
.81
Pro
babi
lity
<= B
irth
Rat
e
0 2 4 6Births Per 100 Women Aged 1519
1961 1965 1969 1973
Note: Fertility per 100 women. Population weighted
Distribution of Teenage Fertility Rates
Figure 2: Distribution of Teen Fertility
8
1.5
0.5
1
0 5 10 15 20 25Pill Use Per Woman 1519
Log Differences Fitted values4
20
2
0 5 10 15 20 25Pill Use Per Woman 1519
Level Differences Fitted values
Note: Pill use measured in SEK per 1519 year old women. Fertility is per 100 women aged 1519.
10 Year Log (Left Panel) and Level (Right Panel) DifferencesPre/Post Pill Change in Fertility vs. Pill Use
Figure 3: Change in Teen Fertility vs. Pill Use
3.2 Time Variation Across Communities: Differences-in-Differences
Fertility declines were not uniform, but occurred to a larger extent in communities with high
take-up of the Pill. Figure 3 plots 10 year changes in teen fertility from before until after the
Pill against Pill use by market.12 Although teen births fell in most communities, declines
were largest where take-up of the Pill was greatest. Time differences net out fixed factors
across markets. The negative slope in Figure 3 indicates that permanent fertility differences
do not explain the negative relationship between teen fertility and Pill use.
3.3 Time Variation Across Communities and Age Groups: Differences-
in-Differences-in-Differences
Time variation across communities and age groups can be used to construct a differences-
in-differences-in-differences (DDD) comparison. Table 2 uses data on the age distribution of
12Log differences in the left panel, level differences in the right panel. The sizes of the circles in Figure 3indicate market population.
9
1.5
0.5
1
Log
Diff
eren
ces
0 5 10 15 20 25
.8.6
.4.2
0.2
0 2 4 6 8 10
1.5
1.5
0.5
0 2 4 6 8 10
42
02
Leve
l Diff
eren
ces
0 5 10 15 20 25Pill Per Woman Aged 1519
64
20
2
0 2 4 6 8 10Pill Per Woman Aged 3034
43
21
01
0 2 4 6 8 10Pill Per Woman Aged 3539
Note: Fertility expressed per 100 women of each age group.
10 Year ChangesPre/Post Pil l Differences in Ferti l i ty vs. Pi l l Use By Age
Figure 4: Differences in Fertility from Before and After the Pill vs. Pill Use by Age
Pill prescriptions to compare the fertility of low/high Pill use age groups in low/high Pill
use markets before and after the introduction of the Pill.13 I use data on the age distribution
of Pill use to define a high use age group, women aged 15-19, and two low use age groups,
women between 35-39 and 30-34. Figure 4 presents 10 year log differences (top panel) and
level differences (bottom panel) in births per 100 women by age before/after the Pill relative
to Pill expenditures. The negative correlation between fertility changes and Pill take-up seen
among teens is reversed among older women where relatively high Pill use communities have
smaller fertility declines.
Comparing fertility patterns of women who use the Pill at much lower rates than teens
allows us to account for the role of location specific trends in driving teen fertility changes.
Table 2 presents the data as a DDD comparison. Average changes in births before/after
the Pill in both low and high Pill use areas (defined as above or below median Pill use per
13Women between 30 and 34 accounted for 12 percent of scripts for the pill and women aged 35-39 togetheraccounted for 9 percent of pill sales. According to the Drug Information Committee of the Swedish Boardof Health and Welfare (1984) a quarter of OC prescriptions were written to teenagers. This is more thandouble the share of prescriptions written to women in their early thirties, and three times that of women intheir late thirties. (insert: relative to their population shares.)
10
woman 15-44) for high Pill use teens and low Pill use populations are presented. Differencing
between high and low Pill use areas approximates the slopes for each age group plotted in
Figure 4. Differencing again across age groups gives an estimate of the total effect of the
Pill, net of both age and community specific time effects. Dividing through by differences in
average Pill use across groups we have an estimate of the coeffi cient on Pill use in a linear
fertility model that includes both age and community specific time effects. The DDD analysis
suggests that community specific trends, such as a general decline in fertility in urban areas,
cannot explain the strong negative association between Pill take-up and teen fertility seen in
Figure 3.
3.4 Looking at the Data: A Summary
A negative relationship between the diffusion of the Pill and teenage childbearing is seen in the
data. The steep decline in teen fertility that coincides with the introduction of the Pill and the
strong negative relationship between fertility declines and Pill use are reinforced by the DDD
analysis which nets out the effect of market level trends. The strong negative relationship
holds regardless of whether fertility changes are measured in log or level differences. The
negative relationship is robust to the inclusion of age and community time effects, but there
remain a host of confounding factors that could explain the negative association seen in the
data. For example, trends in local marriage markets specific to teenage girls may drive fertility
changes, not the Pill.14 In the next section I write down a model that allows for unobserved
location specific time effects in teen fertility and outline an IV approach aimed at identifying
the causal effect of the Pill by identifying a part of Pill use that is predetermined by historical
factors from a century earlier and arguably unrelated to contemporaneous trends in labor
and marriage markets.
4 The Empirical Model
Consider a model where teen fertility in community i and time t, denoted Birthi,t, is a
function of Pill use in community i, a community specific fixed characteristic, γi, teen specific
14More generally, the concern is that there are omitted fertility trends specific to teens by each location.
11
local time effects, denoted by φi,t, and an idiosycratic fertility shock denoted by εi,t:
Birthi,t = βPilli,t−1 + γi + φi,t + εi,t.
Rewrite the model in differences that span the introduction of the Pill:
Birthi,t −Birthi,t−j = βPilli,,t−1 + φi,t − φi,t−j + εi,t − εi,t−j.
Initial Pill use is zero everywhere and hence omitted. The persistent community character-
istic, γi, is differenced out.
The challenge is to estimate β when the local trends in fertility related to other factors,
captured by the φi,t’s, are unobserved. This is addressed by using an instrumental variables
approach, where out-of-wedlock births (OWB) from a century earlier are used to instrument
for the diffusion of the Pill. The instrument for Pill use is defined the following way:
Zi,t =
0 t < 1964
OWBi t ≥ 1964
Time variation comes from the legalization of the Pill, common across communities. Cross
section variation in latent demand comes from historical illegitimacy patterns. The inter-
action of the Pill’s legalization and historical patterns that determine latent demand define
the instrument, Zi,t = OWBi × I(1 if t ≥ 1964, else 0).
Illegitimacy from a century earlier is highly correlated with take-up of the Pill, as shown in
the next section. I make the additional assumption that historical illegitimacy is uncorrelated
with contemporaneous fertility trends across communities, except through the influence on
Pill use. This means Zit − Zit−j is uncorrelated with both φi,t − φi,t−j and εi,t − εi,t−j, or
cov(φi,t − φi,t−j, Zit) = 0.
In words, this choice of instruments assumes that changes in teen motherhood during the
late 1960s are not affected by the level of illegitimacy from a century earlier except through
the take-up of the Pill.
By using historical variation in illegitimacy I isolate that part of Pill use which is pre-
determined. The strong relationship between illegitimacy in 1860 and Pill use provides a
12
channel through which exogenous differences in Pill demand can be identified. The empiri-
cal model explicitly takes into account the direct effect that persistent factors may have on
teenage fertility across communities through γi. This provides some plausibility to the ex-
clusion restriction. Although the identifying assumption cannot be verified, the next section
will show how illegitimacy in 1860 is uncorrelated with trends in teen fertility in the period
prior to the Pill’s legalization, consistent with the identifying assumption that underlies the
empirical model.
5 Historical Illegitimacy and the Pill
Since the 17th Century, large and persistent differences in out-of-wedlock births (OWB)
divided Sweden into distinct demographic regions.15 OWB patterns are interesting not only
as a demographic regularity invariant to mass migration and industrialization, but also as
a strong predictor of latent demand for the Pill.16 Illegitimacy rates from 1860 predict a
quarter of the variation in Pill use a century later. This section documents the correlation
between Pill take-up and OWB in 1860, paying particular attention to how OWB is related
to fertility trends before and after the Pill’s introduction.
5.1 The Geography of Historical Illegitimacy and Pill Use
Sundbärg (1910) compiled data through the 17th Century to illustrate how differences in
illegitimacy behavior persisted over hundreds of years. A single characteristic that defines
illegitimacy patterns is diffi cult to discern; high rates of illegitimacy are seen north and
south of the limes norrlandicus, along coasts and plains. Mining and industrial regions have
relatively high occurrence of unwed birth, but many areas with high levels of illegitimacy were
primarily agrarian. Heckscher (1949) argued that urbanization led to the differential pattern
of out-of-wedlock fertility, but Frykman (1975) shows how areas with high illegitimacy had
population densities no different from central Småland, an area of low illegitimacy.
15See Sundbärg (1910).16Frykman (1975) presents a detailed analysis of non-marital fertility trends and the ethnological back-
ground regarding their social roots. Sklar (1977) also discusses the importance legal and economic develop-ments in the 19th century with regard to illegitimacy in Sweden.
13
Figure 5: Maps of Out-of-Wedlock Births in 1860 and Pill Use per Woman in 1970
14
Figure 5 maps illegitimate births per 100 live births across markets in 1860 and demand
for the Pill a century later. These maps illustrate the positive correlation that underlies the
strong first stage result presented in the next section. The maps illustrate how the correla-
tion between unwed births and Pill use a century later is not solely driven by North/South
differences. Even within regions, illegitimacy and Pill use are closely related. Similarly, urban-
ization does not seem to be the defining factor driving illegitimacy and Pill take-up. Although
Stockholm has the highest rates of Pill use and illegitimacy, Gothenburg, the second largest
city, is not one of the top ten Pill demand markets nor is it an area of high illegitimacy in
1860.
5.2 Illegitimacy in 1860 and Teen Fertility: Before the Pill
If teen specific time effects are correlated with OWB in 1860 this would violate the identifying
assumption. I can’t test this directly, but I can test whether changes in teen fertility before
the Pill are correlated with OWB in 1860. Table 3 summarizes results from regressions of
changes in teen fertility on historical illegitimacy, year fixed effects, and regional time trends.
The top panel uses measures of teen fertility in logs and levels, the middle panel looks at
younger and older teens, and the bottom panel breaks out teen fertility by marital status.
There is no significant correlation between fertility changes before the Pill and OWB in 1860
regardless of the measure used. The lack of a significant correlation between teen fertility
before the Pill’s introduction and illegitimacy from a century earlier is reassuring, as it is
consistent with the identifying assumption behind the IV model.
5.3 Illegitimacy in 1860 and Teen Fertility: Placebo Treatments
Placebo experiments are another way to test whether OWB in 1860 is correlated with local
time effects before the Pill. I define a placebo instrument identical to the instrument defined
in Section 4 except for the timing of the Pill’s introduction:
Z62i,t =
0 t < 1962
OWBi t ≥ 1962
15
Specification 1 in Table 4 reports the intention to treat (ITT) coeffi cients on OWB in 1860
for a fixed effects regression on a panel of teen fertility from 1961-1963. The coeffi cient
on the placebo instrument is not significantly different from zero. The second specification
extends the data window by one year, conducting the same placebo treatment. Specification
3 estimates the effect of a placebo treatment corresponding to the Pill being introduced
in 1963. The coeffi cients are not significantly different from zero.17 Placebo tests provide
further reassurance that the instrument is not correlated with trends in fertility before the
Pill’s introduction.
5.4 Illegitimacy in 1860 and Teen Fertility: The Reduced Form
The Pill’s legalization in 1964 should induce fertility responses well before disaggregated
sales data become available in 1970.18 I test whether the instrument is working through the
hypothesized channel, legal access to the Pill, by estimating a reduced form model in the
period following legalization. The model estimated in Table 5 parallels the model in Section
4 where Pilli,t has been replaced with the instrument Zi,t.19 Table 5 reports the coeffi cients
on OWB in 1860 as the data window is extended from 1965 through 1974. The coeffi cient
estimates reported in Table 5 suggest that the Pill had a negative and significant effect
on fertility, corresponding to a drop of almost 17 percent in the decade after the Pill was
approved for contraceptive use.20 Reduced form estimates provide further support for the
identifying assumption, namely that legalization of the Pill in 1964 drives the importance
of OWB in 1860 as a predictor of teen fertility declines, not trends coinciding with data
availability in 1970.
A similar reduced form approach can be used to trace out how the Pill altered teen
fertility at specific points in time relative to the pre-legalization period. Figure 6 plots ITT
17The results are the same when placebo tests are run on specifications in levels.18See Section 2.1 for further details on the data.19The equation estimated in Table 5 is Birthi,t = βZi,t−1 + γi + δt + η
∗j t+ εi,t where Zi,t is defined as
Zi,t =
{0 t < 1964
OWBi t ≥ 1964
and ηj denotes region specific time trends. Robust standard errors are clustered at the market level. All ofthe ITT estimates reported in Table 5 are weighted by the teen population.20The total effect is computed by multiplying β̂ by the population weighted average value of OWB in 1860
16
effects and confidence bands from rolling one year response estimates. The first two points
correspond to the placebo estimates reported in specifications 1 and 3 of Table 4 scaled
by the teen population weighted illegitimacy rate in 1860. The vertical line indicates the
introduction of the Pill. There is an immediate reduction in teen fertility of almost 7 percent
in 1965, a year of partial treatment, relative to the pre-legalization period. The effect grows
rapidly through 1969 when the estimated reduction reaches 27 percent. The estimated effects
gradually decline during the 1970s, reaching 15 percent in 1974. The estimated magnitudes of
the ITT estimates mirror the dynamics of aggregate Pill use reported by the Swedish Board
of Health (1984). The share of women aged 15-44 using the Pill on an annual basis is plotted
by the dashed line, and illustrates a rapid take up of the Pill that peaks in the late 1960s and
then declines gradually through 1974. The reduced form estimates capture the increasing
intensity of treatment and subsequent decline observed in the aggregate data.
17
Reduced Form Fertility Effect Estimates and the Share of Women Using the Pill by Year
Teen fertility is strongly predicted by OWB in 1860 in the years following the Pill’s in-
troduction, but not in the years immediately preceding legalization. Reduced form estimates
from placebo treatments and the actual legalization of the Pill make a strong case for the
instrument shifting Pill take-up, and in turn fertility, while being uncorrelated with contem-
poraneous fertility trends. The timing corresponds precisely to the introduction of the Pill,
matching the relative magnitudes of the time pattern of initial partial treatment as well as
the data on the Pill’s diffusion in the aggregate.
18
6 The Pill’s Impact on Teenage Childbearing
The reduced form estimates discussed in the previous section not only make a strong case for
the instrument working through the hypothesized channel but also suggest that confidential
access to the Pill reduced teenage childbearing for the average teen by 17 percent in the
decade following the Pill’s approval for contraceptive use.21 With data on Pill use across
markets we can take the analysis a step further, characterizing the relationship between Pill
use, not just Pill access, and teen fertility. By estimating the full model presented in Section
4 we can quantify the causal channel between Pill use and teen fertility.
Table 6 present OLS and IV estimates of the Pill’s effect on teen fertility based on a log
linear version of the empirical model presented in Section 4, while Table 7 reports results for
a linear in levels version of the model.22 Coeffi cient estimates on Pill use are negative and
highly significant regardless of whether the results are weighted by the population of teen
women (Tables 6 and 7 columns 1-2) or whether region trends are included.23 Specification 2
in Table 6 includes both population weights and region trends, and is the baseline specification
used throughout the paper. OLS estimates of this baseline specification (Table 6, column 2,
top panel) suggest that every Krona increase in average Pill use reduced teen fertility by 2.4
percent.
IV estimates are roughly twice the magnitude of OLS estimates (second panel, Tables
6 and 7). The IV coeffi cient from the baseline log linear model, β̂IVlog = −0.053, implies a
predicted fertility decline of 69 percent if all teens behaved like the marginal teen identified
by the IV estimates, somewhat greater than the observed decline.24 The coeffi cient from
the comparable regression in levels, β̂IVlevel = −0.109, implies a predicted fertility decline of
1.42 births per 100 teens, less than the actual decline seen in Figure 1. Both OLS and IV
21See Table 5 for the ITT estimates and predicted fertility effects. Since treatment intensities vary wecan compute an interval of fertility responses ranging from a reduction of 6 percent in the community withthe lowest illegitimacy in 1860 to a 40 percent reduction in Stockholm, the highest historical illegitimacycommunity.22The equations estimated in Tables 6 and 7 are Birthi,t = βPilli,t−1+γi+δt+η
∗j t+εi,t where ηj denotes
region specific time trends. Robust standard errors are clustered at the market level.23Regions are defined as in Sundbärg (1910).24See Table 6, column 2, middle panel. The bottom panel of Table 6 reports the first stage of the IV
regression. Coeffi cient estimates on historical illegitimacy are highly significant, and the F-statistics reportedin the middle panel are well above standard thresholds for weak instruments.
19
estimates point to a significant negative relationship between Pill use and the teen birth rate.
First-stage results are reported in the bottom panel of Tables 6, and OWB in 1860 is highly
significant in all first stage regressions.
These results provide further evidence that the strong negative relationship between Pill
use and teen fertility is causal. The larger magnitude of IV estimates, relative to OLS, may
be driven by several factors. Attenuation bias, introduced by measurement error in teen
Pill use, may contribute to the difference in magnitudes between the estimates. Larger IV
estimates may also be related to the instrument shifting the behavior of a highly responsive
teen population. Predicted fertility responses based on the IV estimates are larger than
the reduced form estimates in Table 5 and Figure 6. This is consistent with the fertility
reduction for the marginal Pill user being greater than the fertility effect of Pill access for
the average teen. The IV coeffi cients provide a consistent estimate of the fertility reduction
among teenage Pill users and suggest that the Pill was a significant contraceptive innovation
for Swedish teens.
6.1 Alternative Instruments
Other historical instruments plausibly satisfy the exclusion restriction as well. Table 8
presents first and second stage results for alternative historical instruments.25 The first col-
umn of Table 8 presents the baseline results using OWB in 1860, as reported in column 2 of
Table 6. This specification includes regional trends and is weighted by the teen population.
The second specification presents results when nonmarital fertility is measured in 1910. The
first stage is strong and the coeffi cient on Pill use is negative and highly significant regardless
of when OWB is measured. Specifications 3 and 4 use butter prices from the 19th Century
to instrument for Pill take-up. The use of butter prices follows Schultz (1985) who showed
how terms of trade shocks affecting the dairy sector, the primary employer of women in 19th
Century Sweden, altered women’s wages and in turn their marital fertility decisions. Butter
prices provide a very different source of variation in Pill take-up that arguably satisfies the
identifying assumption. The first stage is strong when using butter prices alone, or in combi-
25Note that regional time trends are included in all of the specifications in Table 8. Results are similarwhen region trends are excluded.
20
nation with OWB in 1860. Specification 4 reports overidentification test results (J-statistic).
The test does not reject the null hypothesis that the instruments are uncorrelated with the
residuals from the estimation equation. This provides another piece of evidence consistent
with the exclusion restriction.
6.2 Alternative Age Groups
My focus on teen fertility parallels Guldi (2008). Yet, much of the previous literature has
focused on 18-21 year old women. Although age specific Pill use measures do not exist, I can
measure fertility for these subpopulations before and after the Pill. Table 9 presents results
from OLS and IV regressions of fertility among 15-17 year old women, 18-19 year olds, 18-21
year olds, and 15-21 year old women on teen Pill use, year and market fixed effects, region
trends, and population weights. Point estimates are largest for the youngest age group, but
generally similar to the baseline results in Table 6. IV estimates are twice the magnitude of
OLS estimates across all age groups. Reduced form estimates, reported in the lower panel of
Table 9, are highly significant. For the case of 15-21 year old women, the predicted fertility
reduction from Pill access is over 12 percent. This is larger than estimates reported in Guldi
(2008) who found that confidential access to the Pill reduced fertility for a similar population
by 8.5 percent.
In addition to quantifying the fertility effects of Pill access, I can also compute the pre-
dicted fertility decline associated with using the Pill. For 15-21 year old women the OLS
coeffi cient is -0.017; instrumenting yields a coeffi cient of -0.036.26 Given average Pill ex-
penditures, this implies a predicted fertility reduction of 22 to 47 percent. The estimated
fertility responses to Pill use are much greater than reduced form estimates. This suggests
that reduced form estimates of the fertility effects of confidential access for the average teen
significantly understate the behavioral effects of Pill use relative to OLS and IV estimates.
26Both coeffi cients are significant at the 0.1 percent level.
21
6.3 Robustness Checks
6.3.1 Alternative Clustering
One may be concerned that shocks are correlated across geographically proximate markets.
One way to take this into account is to cluster at a more aggregated level to allow for
correlation of residuals across markets. Standard errors are little changed when clustering at
the county level instead of the more disaggregated market level.
6.3.2 Alternative Pill Use Measures
To be written.
6.4 Estimate Comparisons
The previous literature on the Pill has relied on reduced form estimation based on natural
experiments which shifted access to the Pill for different populations at different points in
time. Guldi (2008) focuses on teenagers confidential access to the Pill. As seen in Table 5 and
Figure 6, reduced form estimates of the impact of Pill access on teen fertility are generally
larger for the Swedish case.
The main contribution of this paper, though, is to estimate the impact of Pill use on teen
fertility. Although reduced form estimates such as Guldi (2008) and Bailey (2010), for the
case of married teens, are suggestive of Pill use fertility effects the causal channel is not well
understood. Bailey, Hershbein and Miller (2012) report first stage results for rural women
using changes in family planning policy as an instrument to shift Pill use, but their full model
estimation is the exception in this literature. By making a case for historical illegitimacy as a
valid instrument for Pill take-up, I can quantify the causal channel between Pill use and teen
fertility in the wake of the Pill’s introduction, a relationship that relatively little is known
about. The predicted fertility reduction associated with Pill use is 69 percent in the baseline
specification (Table 6, column 2). This suggests that reduced form estimates understate the
fertility effects of the Pill for the marginal Pill user identified by IV.
Are the estimates reported in Tables 6-9 comparable to previous research on Pill use
among Swedish teens? The Pill’s influence on teenage childbearing in Sweden has been
22
studied by Grönqvist (2009). He uses the introduction of subsidies on teenagers purchases
of certain types of OC in a subset of Swedish municipalities beginning in 1989, 25 years after
the Pill’s introduction. Grönqvist (2009) estimates the effect of exposure to a subsidy on
the log teen fertility rate, the dependent variable in the baseline specification and extensions
presented here. Grönqvist estimates a coeffi cient of -0.076 with a standard error of 0.054,
and concludes that exposure to OC subsidies reduced teen fertility by 7.5 percent. The
subsidy exposure effect is not directly comparable to the ITT or full model estimates I
present. One way to compare the fertility effects induced by the Pill’s introduction to the
subsidies considered by Grönqvist is to compute the implied subsidy rate that would have
led to a fertility reduction similar to those predicted by the baseline estimates reported in
Table 6. Grönqvist (2009) reports an average subsidy of 75 percent. Assuming linearity, a
10 percentage point increase in the subsidy rate translated into a one percent reduction in
teen fertility. In comparison, the IV estimates in Table 6 (column 2) imply that confidential
access to the Pill, although unsubsidized, led to a 69 percent reduction in teen fertility. To
generate a similar behavioral response using subsidies would require a subsidy rate of nearly
700 percent. Even in comparison to the ITT estimates reported in Table 5 and Figure 6,
the subsidy exposure effects reported by Grönqvist (2009) are a half or a third the size of
estimates based on the Pill’s introduction. The behavioral responses brought about by the
introduction of the Pill are larger than those brought about by later subsidy policies.
7 Marital and Nonmarital Childbearing and the Pill
Theory has focused on how contraceptive innovations alter nonmarital childbearing. Figure 7
plots log differences before/after the Pill for both marital and nonmarital childbearing versus
Pill use. The strong negative relationship between total fertility and Pill use is mirrored in
the nonmarital fertility patterns plotted in the right panel. Yet, marital childbearing behaves
very differently; marital births per teen and Pill use appear to be positively correlated, as
seen in the left panel of Figure 7. Did the Pill both increased marital fertility and reduce
nonmarital fertility among teens? In the following sections I report estimates of the Pill’s
effect on childbearing among teenagers by marital status and relate these estimates to the
23
43
21
0
0 5 10 15 20 25Pill Use per Woman 1519
Change in Marital Births Fitted values
.5
0.5
11.
5
0 5 10 15 20 25Pill Use per Woman 1519
Change in Nonmarital Births Fitted values
Note: Fertility is measured by mother's marital status but expressed per 100 of the total teen female population.
10 Year Log Differences Before and After the PillChange in Fertility vs. Pill Use by Marital Status
Figure 6: Changes in Teen Fertility by Marital Status
theoretical and empirical literature. In order to understand the forces that drive the fertility
patterns seen in Figure 7 an analysis of teenage marriage behavior is also presented.
7.1 Nonmarital Childbearing
Akerlof et al (1996) emphasize how the diffusion of improved contraceptives and the use of
abortion may contribute to increased nonmarital childbearing as norms enforcing marriage
in the case of pregnancy are eroded. Column 1 report results for a regression of the log of
nonmarital births per single teens on Pill use, year and market fixed effects, and regional time
trends. Column 2 reports results for a similar model where nonmarital fertility is expressed
per the total teen population. Estimates by OLS and IV are large, negative and highly
significant regardless of how nonmarital fertility is measured. Given average Pill use in 1974,
the OLS estimates imply that nonmarital fertility dropped by 48 to 50 percent as a result of
the Pill’s diffusion. IV estimates are even larger.
The Pill may have led to a decline in nonmarital fertility relative to the teen population,
but the share of nonmarital births among teens may have increased. In 1964, 59 percent of
teen births occurred outside of marriage. By 1974 this rate had increased to 87 percent.
Swedish time series data is consistent with Akerlof et al (1996) in the sense that the OWB
24
share increased after the Pill was introduced. Yet, a positive correlation between the share of
births occurring outside of marriage and the Pill is not seen in the panel regressions presented
in Table 10 (column 3). Regressing the share of teen births that occur outside of marriage
on Pill use per teen, year and market fixed effects, and regional time trends yields a negative
and significant coeffi cient on Pill use. The diffusion of the Pill led to a decline in nonmarital
fertility both relative to the teen population and as a fraction of teen births. Although a
general increase in the non-marital fertility share coincided with the introduction of the Pill
there is little evidence to support the view that the diffusion of the Pill led to increased
nonmarital fertility.
7.2 Marital Childbearing
Table 11 estimates the impact of the Pill on births per the married teen population in both
logs (column 1) and levels (column 3). Column 1 reports OLS (top panel) and IV (bottom
panel) estimates of the Pill’s impact on log births per married woman. The IV estimates,
β̂IVlog = −0.022 suggest that a one SEK increase in average Pill use translated into an decline
in teen marital fertility of 2.2 percent. Given average expenditures of 13 SEK this translates
into a marital fertility reduction of 29 percent. Column 3 reports estimates of a similar
specification where marital fertility is measured in levels. The IV estimate in column 3,
β̂IVlevel = −1.681 implies that every Krona increase in average Pill use led to a decline of 1.68
births per 100 married women. This implies a reduction of 22 births per 100 married teens,
half of the 1964 level.
The significant and large negative relationship between marital fertility per the population
of married women is consistent with the findings of Bailey (2010), though larger in both
absolute and relative terms than the reduced form effects she estimates. Bailey (2010) focuses
on marital fertility relative to the population of married women, using sales bans as a way
to identify the marital fertility effects of Pill access. Bailey estimates the impact of sales
bans for every year from 1951 through 1980. The largest fertility effect among teenagers
occurs in 1964, with sales ban states having almost one more birth per 100 married teens
than states without sales bans. Marital births per married teen were very similar in the U.S.
and Sweden during this period. In 1963 there were 49 live births per 100 married teens in
25
the U.S. Hence, percentage increases in teen marital fertility induced by sales bans were on
the order of 2 percent in 1964, an order of magnitude less than the estimated effect of Pill
use reported in column 1 of Table 11.
Having established that the Pill led to fewer births per married teen, and that the mag-
nitude of this effect was large, we turn the effect of the Pill on marital fertility per the total
teen population. If the share of teens that marry is inelastic to contraceptive innovations,
or if the Pill led to a decline in the share of teens entering marriage, we would expect the
share of marital births per the teen population to decline. A negative relationship is not
born out in the data. Instead, we find a positive correlation between Pill use and marital
births, as seen in Figure 7.27This positive correlation stands in sharp contrast to the negative
relationship between Pill use and both total and nonmarital teen fertility. Estimating the
impact of the Pill on marital births per the population of married teens misses the larger
trend toward increased marital fertility brought about by increased entry into marriage, an
effect undocumented in the previous literature and discussed in the next section.
The Pill increased fertility within marriage among Swedish teens. This effect is seen in
columns 2 and 4 of Table 11 where the impact of the Pill on marital births per the total teen
population is estimated. IV estimates of the log specification, reported in the bottom panel
of column 2, suggest that for every Krona increase in average Pill use marital births per
the teen population increased by 5.5 percent. Column 3 reports estimates for a specification
in levels, and the IV estimates in the bottom panel point to a similar relationship where a
one Krona increase in Pill use increased marital fertility by over 0.03 births per 100 teens,
or an increase in marital fertility per teen of 29 percent relative to 1964 levels. The steady
decline in marital childbearing per teen seen in the time series data does not appear to be
driven by the diffusion of the Pill. The next section presents evidence on how the Pill altered
teen marriage behavior and contrasts these result to studies such as Goldin and Katz (2002)
which have emphasized how the Pill led to marital delay among college educated women in
the U.S.27The positive relationship between Pill use and teen marital fertility is not mechanically driven by a
decline teen marriage rates, as both marital and nonmarital birth rates are measured relative to the totalteen population in Figure 7.
26
.06
.04
.02
0.0
2
0 5 10 15 20 25Pill per Woman 1519
Women 1519 Fitted values
.3.2
.10
.1
0 5 10 15 20Pill per Woman 2024
Women 2024 Fitted values
.01
.005
0.0
05
0 5 10 15 20 25Pill per Woman 1519
Women 1517 Fitted values.1
5.1
.05
0.0
5
0 5 10 15 20 25Pill per Woman 1519
Women 1819 Fitted values
Note: Proportion married is the ratio of married women to al l women in each age group.
Change in Proportion Married vs. Pill Use
Figure 7: Female Married Population by Age
7.3 Teen Marriage and the Pill
Marriage among teens was on the rise in Sweden from 1960 when 2.7 percent of women aged
15-19 were married through 1968 when 5.8 percent married. Marriage rates declined sharply
thereafter, only 2.5 percent of teens were married in 1969 and 1.3 percent by 1975. Despite
this aggregate decline, Figure 7 depicts a weak positive relationship between Pill take-up and
teenage marriage. Figure 7 plots log differences in the share of married females relative to Pill
use for several age groups. The top panel contrasts teen marital behavior to that of women
aged 20-24, a population that more closely parallels the college graduates studied by Goldin
and Katz (2002). Teen marriage appears to be positively related to Pill use, while marriage
among women in their 20s shows a negative relationship. The bottom panel decomposes teen
marriage for younger and older teens; a positive correlation is seen among both populations.
Although a thorough treatment of the Pill’s effect on marriage is beyond the scope of the
current study, the empirical model in Section 4 can be used to estimate the impact of the
Pill on teen marriage. Table 12 reports results from OLS and IV regressions of the share of
27
teens that are married on current and lagged Pill use, as well as year and market fixed effects
and region time trends.28 OLS and IV estimates of the Pill’s effect on marriage are reported
for both women and men. For women, both OLS and IV results point to a positive and
significant relationship between adoption of the Pill and the decision to marry during their
teens. The results are similar for men, though not significant when estimated by OLS. The
estimates in Table 12 suggest that a one SEK increase in Pill use among teens led to a 10
percent increase in the teen marriage rate. Goldin and Katz (2002) emphasize how the Pill
enabled young college educated women in the U.S. to delay marriage. Although the Pill may
have been important for fertility delay among women in their 20s, or women who completed
a college degree, I do not find evidence that the Pill contributed to marital delay among
Swedish teenagers in the decade after the Pill. To the contrary, the Pill appears to have
increased teenagers likelihood to marry, and in turn increased marital birth among teens.
8 Conclusion
This paper investigates teenage childbearing in Sweden and estimates how Pill use altered
fertility in the decade after its introduction. A novel IV strategy is presented which uses
variation in nonmarital childbearing from a century earlier to identify exogenous variation
in take-up of the Pill. IV estimates provide evidence that the diffusion of the Pill led to a
statistically and economically significant reduction in teenage childbearing of 50 percent or
greater. The fertility effect of Pill access for the average teen substantially understates the
impact of the Pill for the marginal Pill user identified by the IV estimates. By identifying
the effect of Pill use on teen fertility this paper quantifies an empirical relationship that the
reduced form literature has not been able to characterize.
Full model estimates are diffi cult to compare directly to the previous literature as most
studies have reported reduced form estimates of the impact of Pill access. For example, Guldi
(2008) estimates that Pill access reduced fertility for 15-21 year old women by 8.5 percent.
Table 9 presents reduced form estimates of the effect of Pill access on births for a similar
population which implies that Pill access reduced fertility by over 12 percent. Although
28Fertility results use lagged pill use, but it could be argued that current pill use is the relevant variablefor marriage decisions.
28
the reduced form estimates I present in Section 5.4 are larger than many in the previous
literature, their striking feature is how small the implied behavioral effects are relative to the
full model estimates discussed in Section 6.
Theory is ambiguous as to how women’s use of highly effective contraceptive methods
such as the Pill changed the composition of fertility across marital groups. The data is
clear; the diffusion of the Pill led to a decline in nonmarital fertility both relative to the teen
population and as a fraction of teen births. The data do not support Akerlof et al (1996),
but instead the female empowerment model of Chiappori and Orrefice (2008).
I find evidence that the Pill reduced marital fertility among married teens. Point estimates
are an order of magnitude larger than the estimates reported in Bailey (2010). Yet, the
negative fertility effects of the Pill among married teens does not translate into a decline in
marital fertility in the aggregate. To the contrary, I estimate that marital fertility per the
total teenage population increased in Pill use. Increased selection into marriage among young
women is the driving force behind this rise in marital fertility. This facet of teen fertility
has not been discussed in the previous literature and provides a point of comparison to the
marital delay mechanisms emphasized by Goldin and Katz (2002) in their study of college
educated women in the U.S.
Although the Pill may have influenced a variety of decisions over women’s life course,
its primary effect must be seen with regard to fertility. In the absence of a measureable
fertility response, ancillary effects on marriage, education and income become tenuous. Using
the introduction of the Pill in Sweden to estimate the effect of Pill use on teen fertility
and documenting a large and significant effect with regard to delay of fertility during the
teenage years is an important contribution to the literature. Moreover, this paper presents
an empirical design which opens the door to future studies regarding the long run impact of
the Pill on women’s schooling, career choices, and child outcomes.
References
[1] Akerlof, G., J. Yellen, andM. Katz (1996), "An Analysis of Out-of-Wedlock Childbearingin the United States." The Quarterly Journal of Economics, 111(2): 277-317.
29
[2] Ananat, E. and D. Hungerman (2012), "The Power of the Pill for the Next Generation:Oral Contraception’s Effects on Fertility, Abortion, and Maternal and Child Character-istics." The Review of Economics and Statistics. 94(1):37-51.
[3] Bailey, Martha J. (2006) “More Power to the Pill: The Impact of Contraceptive FreedomonWomen’s Life Cycle Labor Supply.”The Quarterly Journal of Economics, 121(1):289—320.
[4] Bailey, Martha J. (2010), ""Momma’s Got the Pill": How Anthony Comstock andGriswold v. Connecticut Shaped US Childbearing." American Economic Review, 100(1):98—129.
[5] Bailey, Martha J. (2013), "Fifty Years of Family Planning: New Evidence on the Long-Run Effects of Increasing Access to Contraception." Brookings Papers on EconomicActivity, ():341-395.
[6] Bailey, M., Hershbein, B. and A. Miller (2012), "The Opt-In Revolution? Contraceptionand the Gender Gap in Wages." American Economic Journal: Applied Economics, 4(3):225-254.
[7] Becker, G. (1991), A Treatise on the Family, Harvard University Press, Cambridge.
[8] Chiappori, P. A., and S. Oreffi ce (2008), "Birth Control and Female Empowerment: AnEquilibrium Analysis." Journal of Political Economy, 116(1).
[9] DiCenso A., G. Guyatt, A. Willan, and L. Griffi th (2002). "Interventions to Reduce Un-intended Pregnancies among Adolescents: Systematic Review of Randomized ControlledTrials." British Medical Journal 324:1426-34.
[10] Frykman, J. (1975) "Sexual Intercourse and Social Norms: A Study of Illegitimate Birthsin Sweden 1831-1933." Ethnologia Scandinavica, 1975: 110-150.
[11] Frykman, J. (1977) Horan i Bonde Samhället. LiborLäromedel, Lund.
[12] Goldin, C. and L. Katz (2002), "The Power of the Pill: Oral Contraceptives andWomen’sCareer and Marriage Decisions."Journal of Political Economy,110(4):730-770.
[13] Grönqvist, Hans (2009), "Putting teenagers on the Pill: the consequences of subsidizedcontraception." Institute for Labor Market Policy Evaluation Working Paper No. 2009:8.
[14] Guldi, M. (2008), "Fertility Effects of Abortion and Birth Control Pill Access for Mi-nors." Demography, 45(4):817-827.
[15] Heckscher, Eli (1949), Sveriges Ekonomiska Historia Från Gustav Vasa: Andra DelenDet Moderna Sveriges Grundlägning Fösta Halvbandet. Albert Bonniers, Stockholm.
[16] Joyce, T. (2013), "If Only Policy Analysis Were So Easy." Journal of Policy Analysisand Management, 32(4):897-99.
30
[17] Lewin, Bo (2000), Sex in Sweden : on the Swedish sexual life 1996. Stockholm : NationalInstitute of Public Health.
[18] Liljeström, Rita (1974), A Study of Abortion in Sweden: A Contribution to the UnitedNations World Population Conference. Royal Ministry of Foreign Affairs. Stockholm:Norstedt.
[19] Linner, Birgitta (1967), Sex and Society in Sweden. New York: Pantheon Books.
[20] Läkemedelstatistik, A.B. "Swedish Drug Market: Statistical Survey of Registered Phar-maceutical Specialties in Sweden." Stockholm. 1970(I)-1974(IV).
[21] Myers, C. K. (2012), "Power of the pill or power of abortion? Re-examining the effects ofyoung women’s access to reproductive control." IZA Discussion Paper No. 6661. Bonn,Germany: Institute for the Study of Labor.
[22] Schultz, T. Paul (1985), "Changing World Prices, Women’s Wages, and the FertilityTransition: Sweden, 1860-1910." Journal of Political Economy, 93(6): 1126-1154.
[23] Statistics Central Byrån (1864), "Befolkning Statistisk Årsbok 1860." SOS, Stockholm.
[24] Statistics Central Byrån (1904), "Befolkning Statistisk Årsbok 1900." SOS, Stockholm.
[25] Statistics Central Byrån (1914), "Befolkning Statistisk Årsbok 1910." SOS, Stockholm.
[26] Sundbärg, Gustav (1907), Bevölkerungsstatistik Schwedens 1750-1900. Reprinted in Ur-val 3, Statistika Central Byrån.
[27] Sundbärg, Gustav (1910), Emigrationsutredningen Bilaga V: Ekonomisk-statistiskbeskrifning öfver Sveriges olika landsdelar. Royal Publisher, P. A. Norstedt and Sons,Stockholm.
[28] Swedish Board of Health and Welfare (1984), "Oral Contraceptives." National Board ofHealth and Welfare Workshop, Drug Information Committee, 1984:3, Stockholm.
31
Table 1: Pill Sales in Detail
Annual Sale and DosageYear Total OC Sales Annual Doses Female Population OC Use
1000 SEK Woman Years Aged 15-44 Percent of Women1970 14,948 423,452 1,585,915 26.71971 13,785 391,270 1,590,503 24.61972 14,697 374,973 1,591,942 23.61973 12,986 329,548 1,592,887 20.71974 14,079 334,983 1,600,290 20.9
Leading Brand Price IndexYear Leading Introduced Share of Price per
Brand OC Sales Month1970 Follinyl 1968 Q4 27.3 2.941971 Follinyl 1968 Q4 33.0 2.941972 Follinyl 1968 Q4 36.9 3.271973 Follinett 1971 Q3 25.6 3.281974 Follinett 1971 Q3 31.9 3.50Note: Price based on Q4 prices and sales data. Follinyl and Follinett produced by Recip.
32
Table 2: A Differences-in-Differences-in-Differences Analysis
Log Differences in Births Before-After the PillLow Pill Use Area High Pill Use Area Difference
Low Pill Users (30-34) -0.26 -0.21 0.04(0.14) (0.13)
Low Pill Users (35-39) -0.49 -0.45 0.04(0.22) (0.18)
High Pill Users (15-19) -0.21 -0.26 -0.05(0.27) (0.26)
Difference (Teens-30-34) 0.05 -0.05 -0.09Slope of Log Fertility in Pill Use per Woman (Age 30-34 Control Group): -0.04Difference (Teens-35-39) 0.28 0.19 -0.10Slope of Log Fertility in Pill Use per Woman (Age 35-39 Control Group): -0.04
Level Differences in Births Before-After the PillLow Pill Use Areas High Pill Use Area Difference
Low Pill Users (30-34) -2.10 -1.75 0.36(1.16) (1.04)
Low Pill Users (35-39) -1.76 -1.49 0.27(0.76) (0.57)
High Pill Users (15-19) -0.69 -0.80 -0.10(0.87) (0.81)
Difference (Teens-30-34) 1.41 0.95 -0.46Slope of Log Fertility in Pill Use per Woman (Age 30-34 Control Group): -0.20Difference (Teens-35-39) 1.06 0.69 -0.37Slope of Log Fertility in Pill Use per Woman (Age 35-39 Control Group): -0.16Note: Births are measured per 100 women aged 15-19.
33
Table 3: Changes in Teen Fertility Before the Pill
Teen Fertility in Logs and LevelsLog Level Log Level
OWB 1860 -0.0008 -0.0029 -0.0034 -0.0097(0.0009) (0.0026) (0.0030) (0.0083)
Teen Fertility by AgeAge 15-17 Age 18-19 Age 15-17 Age 18-19
OWB 1860 -0.0010 -0.0010 -0.0002 -0.0045(0.0013) (0.0011) (0.0033) (0.0036)
Teen Fertility by Marital StatusNonmarital Marital Nonmarital Marital
OWB 1860 -0.0001 -0.0009 -0.0014 -0.0056(0.0010) (0.0011) (0.0026) (0.0043)
Additional Controls:Year FE Yes Yes Yes YesRegion Trend Yes Yes Yes YesWeighted Yes Yes No NoClusters 70 70 70 70N 210 210 210 210Note: All teens refers to women aged 15-19. Births are expressedper 100 teens. Unless otherwise specified fertility is measured in logs.* p<.05, ** p<.01, *** p<.001
34
Table 4: Teen Fertility Placebo Tests
(1) (2) (3)
OWB 1860 X 0.0004 -0.0011I(1 if t=1962, else 0) (0.0015) (0.0016)
OWB 1860 X -0.0027I(1 if t=1963, else 0) (0.0016)
Additional Controls:Market FE Yes Yes YesYear FE Yes Yes YesRegion Trend Yes Yes YesClusters 70 70 70N 210 280 280Weighted Yes Yes YesNote: Log births per 100 teens is the dependent variable.* p<.05, ** p<.01, *** p<.001
35
Table 5: Historical Illegitimacy and Teen Fertility: Reduced Form Estimates Over ExtendedData HorizonsTerminal Period 1965 1966 1967 1968 1969
OWB 1860 X -0.006*** -0.009*** -0.010*** -0.012*** -0.013***I(1 if t<=1964, else 0) (0.001) (0.001) (0.001) (0.001) (0.002)
N 350 420 490 560 630R-squared 0.891 0.886 0.875 0.860 0.863
Predicted Fertility Effect -0.068 -0.101 -0.108 -0.129 -0.144
Terminal Period 1970 1971 1972 1973 1974
OWB 1860 X -0.014*** -0.014*** -0.015*** -0.015*** -0.015***I(1 if t<=1964, else 0) (0.002) (0.002) (0.001) (0.001) (0.001)
N 700 770 840 910 980R-squared 0.863 0.856 0.854 0.859 0.858
Predicted Fertility Effect -0.150 -0.155 -0.162 -0.167 -0.168
Additional Information:Clusters 70 70 70 70 70Note: Log births per 100 teen women is the dependent variable. All specifications includeyear and market fixed effects, as well as regional time trends. Robust standard errors,clustered at the market level, are reported in brackets. All estimates are weighted bythe female teen population. See Section 4 for more details.* p<.05, ** p<.01, *** p<.001
36
Table 6: Log Births per 100 Women Aged 15-19 and Pill Use the Previous YearOLS
Pill Use in Previous Year -0.029** -0.024*** -0.019** -0.014*(0.009) (0.006) (0.007) (0.007)
R-squared 0.855 0.861 0.798 0.803
IV:1860 OWB X Pill Legal
Pill Use in Previous Year -0.058*** -0.053*** -0.045*** -0.034**(0.006) (0.009) (0.011) (0.011)
F-statistic 51.44 55.30 18.46 16.44
First Stage: Pill Use on OWB 1860 X Pill Legal
OWB 1860 X I(1 if year>=64, else 0) 0.280*** 0.301*** 0.460*** 0.437***(0.04) (0.04) (0.11) (0.11)
R-squared 0.317 0.384 0.237 0.272
Region Trend No Yes No YesWeighted Yes Yes No NoClusters 70 70 70 70N 560 560 560 560Note: The dependent variable in the top panels is log births per 100 women aged 15-19. Pilluse is the average expenditure (SEK) on oral contraceptives per teen. Year and market fixedeffects are included in all specifications. Robust standard errors are clustered at the marketlevel and reported in brackets. OWB 1860 is the ratio of unwed births to total live births in1860. The instrument is defined as the interaction of OWB 1860 and a dummy equal to 1if the pill is legal (1964 and later) and zero otherwise. The F-statistic reports the value of theF-test of excluded instruments.* p<0.05, ** p<0.01, *** p<0.001
37
Table 7: Births per 100 Women Aged 15-19 and Pill Use the Previous YearOLS
Pill Use Previous Year -0.066** -0.045** -0.052* -0.027(0.024) (0.016) (0.021) (0.021)
R-squared 0.838 0.852 0.778 0.793
IV:1860 OWB X Pill Legal
Pill Use Previous Year -0.153*** -0.109*** -0.165*** -0.097**(0.015) (0.030) (0.036) (0.033)
F-statistic 51.44 55.30 18.46 16.44
Region Trend No Yes No YesWeighted Yes Yes No NoClusters 70 70 70 70N 560 560 560 560Note: The dependent variable in the top panel is births per 100 women aged 15-19. Pill useis the average expenditure (SEK) on oral contraceptives per teen. Year and market fixedeffects are included in all specifications. Robust standard errors are clustered at the marketlevel and reported in brackets. OWB 1860 is the ratio of unwed births to total live births in1860. The instrument is defined as the interaction of OWB 1860 and a dummy equal to 1if the pill is legal (1964 and later) and zero otherwise. The F-statistic reports the value of theF-test of excluded instruments. See the bottom panel of Table 6 for first stage results.* p<0.05, ** p<0.01, *** p<0.001
38
Table 8: Log Teen Fertility and Pill Use: Alternative InstrumentsOWB OWB Butter OWB in 1860
Instruments: in 1860 in 1910 Prices and Butterin 1862 Prices in 1862
Second Stage: Log Births per 100 Women 15-19 on Pill Use
Pill Use in Previous Year -0.053*** -0.051*** -0.047*** -0.050***(0.009) (0.011) (0.011) (0.009)
F-statistic 55.30 22.35 10.17 34.03
J-Statistic 0.233
First Stage: Pill Use on Historical Instruments
OWB in 1860 X 0.301*** 0.229***I(1 if year>=64, else 0) (0.04) (0.04)
OWB in 1910 X 0.427***I(1 if year>=64, else 0) (0.09)
Butter Prices in 1862 X 0 .111** .081*I(1 if year>=64, else 0) (0.03) (0.03)
R-squared 0.384 0.488 0.367 0.481
Additional Information:N 560 560 560 560Clusters 70 70 70 70Note: All specifications include year and market fixed effects as well as regiontime trends. Robust standard errors, clustered at the market level, are reportedin brackets. Results are weighted by the population of women aged 15-19. OWBrefers to out-of-wedlock births measured per 100 live births. TheJ-statistic has a p-value of 0.629. Hence we cannot reject the null hypothesis thatthe instruments are uncorrelated with the residuals of the estimation equation.* p<0.05, ** p<0.01, *** p<0.001
39
Table 9: Log Fertility and the Pill: Alternative Age Groups
Dependent Age 15-17 Age 18-19 Age 18-21 Age 15-21Variable: Fertility Fertility Fertility Fertility
OLS
Pill Use in Previous Year -0.026* -0.021*** -0.015*** -0.017***(0.010) (0.006) (0.003) (0.004)
R-squared 0.648 0.849 0.874 0.867
IV: OWB 1860 X Pill Legal
Pill Use in Previous Year -0.063*** -0.045*** -0.029*** -0.036***(0.014) (0.008) (0.004) (0.005)
F-statistic 54.71 48.27 53.50 53.50
Reduced Form: OWB 1860 X Pill Legal
OWB 1860 -0.019*** -0.013*** -0.009*** -0.011***(0.003) (0.002) (0.001) (0.001)
R-squared 0.654 0.852 0.872 0.868
Additional Information:N 560 560 560 560Clusters 70 70 70 70Note: All specifications include year and market fixed effects, region trends,and teen population weights. Robust standard errors are clustered at themarket level and reported in brackets. (N=560,Cluster=70)* p<0.05, ** p<0.01, *** p<0.001
40
Table 10: Pill Use and Teen Nonmarital Fertility
Log Nonmarital Log Nonmarital NonmaritalDependent Births per Births per Births ShareVariables: 100 Single 100 Teens of All
Teens Teen Births
OLS
Pill Use in Previous Year -0.036*** -0.037*** -0.008***(0.007) (0.007) (0.002)
R-squared 0.762 0.762 0.921
IV:1860 OWB X Pill Legal
Pill Use in Previous Year -0.064*** -0.066*** -0.010*(0.012) (0.012) (0.004)
F-statistic 55.38 55.38 55.30
Additional Information:Weights Single Teens Single Teens All TeensClusters 70 70 70N 560 560 560Note: All specifications include year and market fixed effects and regiontime trends. Robust standard errors are clustered at the market leveland reported in brackets.* p<0.05, ** p<0.01, *** p<0.001
41
Table 11: Pill Use and Teen Marital Fertility
Log Marital Log Marital Marital MaritalDependent Births per Births per Births per Births perVariables: 100 Married 100 Teens 100 Married 100 Teens
Teens Teens
OLS
Pill Use in Previous Year -0.013** 0.019 -0.445 0.025***(0.005) (0.010) (0.287) (0.007)
R-squared 0.731 0.907 0.761 0.914
IV:1860 OWB X Pill Legal
Pill Use in Previous Year -0.022* 0.055*** -1.681*** 0.033**(0.009) (0.016) (0.433) (0.012)
F-statistic 44.01 44.01 43.99 43.99
Additional Information:Weights Married Teens Married Teens Married Teens Married TeensClusters 70 70 70 70N 560 560 560 560Note: All specifications include year and market fixed effects and regiontime trends.* p<0.05, ** p<0.01, *** p<0.001
42
Table 12: Pill Use and Teen Marriage
Dependent Log Married Log Married Log Married Log MarriedVariables: per 100 Teen per 100 Teen per 100 Teen per 100 Teen
Women Women Men Men
OLS
Pill Use in Current Year 0.034** 0.017(0.011) (0.017)
Pill Use in Previous Year 0.044** 0.032(0.014) (0.020)
R-squared 0.887 0.892 0.714 0.711
IV:1860 OWB X Pill Legal
Pill Use in Current Year 0.075*** 0.102***(0.016) (0.026)
Pill Use in Previous Year 0.102*** 0.126***(0.017) (0.031)
F-statistic 56.03 55.38 49.47 46.66
Additional Information:N 560 560 560 560Clusters 70 70 70 70Note: All specifications include year and market fixed effects as well asregion time trends. Robust standard errors, clustered at the market level,are reported in brackets. All estimates are weighted by the teen population.* p<0.05, ** p<0.01, *** p<0.001
43
Table 13: Teen Fertility Placebo Tests Using Alternative Butter Price Instrument
(1) (2) (3)
Butter Prices in 1862 X 0.0006 -0.0003I(1 if t=1962, else 0) (0.0010) (0.0009)
Butter Prices in 1862 X -0.0015I(1 if t=1963, else 0) (0.0010)
Additional Information:Clusters 70 70 70N 210 280 280Note: Log births per 100 teens is the dependent variable. Allspecifications include year and market fixed effects as well asregion trends. Robust standard errors, clustered at the marketlevel, are reported in brackets. All specifications are weightedby the teen population.* p<.05, ** p<.01, *** p<.001
Table 14: ITT Effects for the Alternative Butter Price InstrumentTerminal Period 1965 1966 1967 1968 1969Butter Prices in 1862 X -0.001 -0.002* -0.002* -0.003** -0.003**I(1 if t<=1964, else 0) (0.0009) (0.0009) (0.0009) (0.0010) (0.0011)N 350 420 490 560 630R-squared 0.888 0.877 0.864 0.845 0.847Terminal Period 1970 1971 1972 1973 1974Butter Prices in 1862 X -0.003** -0.003** -0.004** -0.004** -0.004**I(1 if t<=1964, else 0) (0.0011) (0.0011) (0.0012) (0.0012) (0.0012)N 700 770 840 910 980R-squared 0.849 0.841 0.839 0.845 0.844Additional Information:Clusters 70 70 70 70 70Note: Log births per 100 teen women is the dependent variable. All specifications includeyear and market fixed effects, as well as regional time trends. Robust standard errors,clustered at the market level, are reported in brackets. All estimates are weighted bythe female teen population.* p<.05, ** p<.01, *** p<.001
44