The power of the little blue pill: innovations and implications of lifestyle drugs in an aging population.
Lariviere, Jacob ; Wolff, Hendrik
I. INTRODUCTION
How do innovations in medical technology targeted to the elderly
affect the decision making of an aging population? This is an
increasingly important question as more resources and R&D are
dedicated to the health care of a growing elderly age cohort in the
United States. Although the direct health benefits of medical
innovations are well studied, the effects of a new class of drug on
behavioral decisions of the elderly remains largely unexplored in the
literature (Cutler 2007; Cutler and McClellan 2001). (1)
The 1998 approval of Viagra, informally known as the "little
blue pill," offers an ideal natural experiment to identify how one
particular medical innovation targeted at the elderly affects their
behavior. Before 1998, men suffering from erectile dysfunction (ED) had
very limited treatment possibilities. Upon Food and Drug Administration
(FDA) approval in April 1998 Viagra was rapidly adopted. Sales soared
and made the drug a tremendous commercial success for its
producer-Pfizer. Since Viagra is targeted to older men, its introduction
enables an increased supply of sexually active older men. As a result,
the equilibrium decision making of the older population may have been
affected leading to several changes in a large set of behavioral
outcomes.
We examine Viagra's effect on five different dimensions:
adoption of the new class of oral ED drugs and collapse of the
traditional ED market, fatherhood decisions by age, the age composition
of the marriage/divorce market, sexually transmitted diseases (STDs),
and sexual assault and rape rates. Each variable of interest is chosen
to highlight different channels through which this lifestyle drug could
affect the target population, as well as produce spillover effects on
nontarget populations. First, we test for Viagra's impact on the
number of children fathered by older men. This is a variable of interest
because it highlights the drug's capacity to correct a physical
disorder and the resulting immediate impact on child bearing decisions.
(2) Second, we test for Viagra's effect on sorting in the marriage
market by analyzing changes to the distribution in the age spread among
couples as well as divorce rates across age cohorts. These outcomes
highlight the importance of the drug on the household production
function. Third, we test for Viagra's effect on one particular
STD--gonorrhea. STDs are of interest because it studies the drug's
capacity to impact risky behavior associated with instant gratification.
Finally, we test for Viagra's impact on rape committed by the
target population (e.g., older men). This is an important variable due
to the magnitude of consequences for victims. It also highlights the
drug's capacity to impact unlawful behavior in individuals who
previously were not physically able to commit such behavior. In summary,
then, we test whether this particular lifestyle drug has an observable
effect on four different behavioral channels: newly enabled family
planning decisions, equilibrium in the marriage market, newly enabled
risky behavior associated with instant gratification, and newly enabled
criminal behavior.
The United States has the highest Viagra usage per capita of any
country in the world. We therefore collected a rich set of both U.S.
state- and county-level data for our outcome variables of interest. For
each variable we use a difference-in-difference estimator to test how
Viagra differentially affects outcomes in our treatment and control
groups. Upon Viagra's approval in the United States in April 1998,
nearly 90% of Viagra users were over 45 years of age. Therefore, our
treatment group is individuals of age 45 and above and our control group
is individuals of age 25-40. (3) We show convincing evidence that
short-run fluctuations like macroeconomic conditions affect both the
treatment and control groups in similar ways. To handle variables with a
small number of observations, we employ both a trimming rule that drops
certain counties in addition to a Poisson estimator which can more
readily handle data with a low number of observations in the data.
To interpret our estimates as causal, we have to show the following
chain of events: first, the FDA approval leads to the adoption of the
drug in the target population (both married and unmarried older men). We
show convincing evidence that this occurs. Second, the target population
changes their sexual behavior upon adoption of the drug. Unfortunately,
there is no nationally representative dataset which documents the sexual
behavior of the elderly which we could examine around the introduction
of Viagra in April 1998. In lieu of direct data on sexual behavior, we
directly collected state- and county-level data on outcome variables of
interest. Given that we collect data on outcome variables as opposed to
sexual behavior directly, there must be no other simultaneous changes
which affect Viagra's target population around the date of
Viagra's introduction. This motivates our differences-indifferences
identification strategy in which the control group (individuals aged
25-40) controls for changes in sexual activity in the nation as a whole.
Third, the observed change in behavior by the target population must
lead to measurable changes in our outcomes of interest in the target and
nontarget population (e.g., older women).
We find that Viagra had a positive and significant effect on
gonorrhea rates among men in the older population: rates increased by
15%-28% in this group relative to the control. This result is robust
across a variety of specifications and robustness checks. We find weak
evidence that there was a smaller increase in women, but those estimates
are less consistent and precise. We find no evidence that Viagra's
introduction had any significant effect on any other outcome. As a
result, we conclude that this lifestyle drug had a significant effect on
short-term lifestyle decisions (e.g., risky sexual encounters). We find
no convincing evidence, though, that it affected deviant (e.g., sexual
assault) or long-run (e.g., natality, divorce) lifestyle decisions.
Overall, in a back of the envelope cost-benefit analysis, we find that
the welfare impacts of Viagra with respect to our outcomes of interest
are positive and large.
These results are important for several reasons. First, medical
studies have attempted to determine if Viagra is associated with
increased risk of STDs. Jean et al. (2010) compares the STD rate of the
group of older men taking ED medications with older men not taking ED
medications. However, there are vast differences between these two
groups and due to self-selection their estimates cannot be considered as
causal. Our study is designed to pinpoint the causal percentage change
of Viagra's availability on our variables of interest, something
that the medical literature falls short on.
Second, the well-being of an aging population is an increasingly
important topic given the vast demographic shifts occurring in the
United States, Europe, and other industrializing countries. (4) Medical
innovations for the elderly increase the feasible choice set for their
actions. Knowing precisely how this increased choice set affects
socioeconomic decision making for older populations is therefore an
important economic problem. The 15 years since Viagra's
introduction otter a glimpse as to what economists might observe going
forward as medical innovation increases. (5)
Third, there could be spillover and indirect effects of medical
innovations on nontarget groups. It is vital to understand the results
for both males and females. Finding effects of Viagra on males
demonstrates the existence of a direct effect. However, it important to
understand how lifestyle drugs affect the agents who do not take the
drugs but rather indirectly enjoy their benefits or suffer from their
costs.
Fourth, the "Power of the Pill" articles have shown that
innovation in medical technology has substantial impacts on many
important economic choices made earlier in life by the targeted group
(Bailey 2006; Goldin and Katz 2006). Our article is different because it
is the first to look at the effect of medical innovations for decisions
made later in life, when individuals have more wealth and experience
with decision making. We find that in the context of Viagra, if an
individual does not have the ability to engage in sexual behavior and
suddenly is given that ability, outcomes related to sexual behavior tend
to be no different except for relatively low cost risky sexual decisions
associated with instant gratification. We find some evidence in this
context, then, that self-control can be a challenge even for experienced
decision makers.
The remainder of this article is organized as follows; Section II
gives historical background on ED medication. Section III presents the
econometric model. Section IV presents results. Section V offers
discussion and concluding remarks.
II. BACKGROUND AND DATA
This section provides background about the introduction of Viagra
and the age composition of its users which are both necessary for
discussion of the econometric model in Section III. We also discuss the
four variables of interest in this section, their relevance, and the
data used for analysis. For our variables of interest we collected
separate datasets from different sources. As a result, the years for
which data are available vary by outcome of interest.
Prior to the launch of Viagra (medical name Sildenafil) in April
1998, ED could only be treated with invasive methods involving
injections, penile prosthesis, penis pumps, or vascular reconstructive
surgeries (Montague et al. 2005). The most common ED drug at the time
was Alprostadil. Alprostadil is a penile suppository applied either into
the urethra or injected into the penis about 10 minutes before the
erection is needed. Figure 1 below shows that while in March 1998 more
than 900,000 units of Alprostadil were prescribed. 1 month later, in
April 1998 this figure dramatically declined to 462,000 units to then
continually decline to insignificant numbers today. (6) The aphrodisiac
Yohimbine, a sexual stimulant, experienced even more drastic declines.
This dramatic drop in medication sales can be clearly explained
with the approval of Sildenafil by the FDA, becoming the first oral
medication to treat ED, sold under the brand name Viagra starting in
April of 1998. The new drug was celebrated in the press and public media
as revolutionizing the sexual behavior of the elderly population. The
New York Times, for example, in the year 1998 featured over 193 articles
on Viagra. (7) Viagra was advertised directly to consumers in the United
States on TV, famously being endorsed by sport stars like soccer player
Pele and former U.S. Senator Bob Dole. Viagra was voted the "word
of the year" in both Germany and in the United States. Hence, many
of the former customers of the drugs Alprostadil or Yohimbine
immediately started taking Viagra. Moreover, many new patients must have
adopted Viagra, as evidenced by Figure 2, which shows shipments to
pharmacies for Viagra, Alprostadil, and Yohimbine. Figure 2 demonstrates
that simultaneously with the drop of the former ED drugs, in April 1998
ED drugs experienced an unprecedented consumption level due to the
launch of Viagra selling over 14 million units. The figure shows that
Alprostadil and Yohimbine played an insignificant role whereas Viagra
dominated the market after its introduction.
[FIGURE 1 OMITTED]
A. Data on Viagra Users
An important factor to consider when determining the effects of the
release of Viagra are the demographic characteristics of the drug's
user. To identify demographic information of Viagra users, we collected
data from the Prescribed Medicine File from the Household Component from
a U.S. Department of Health and Human Services database called the
Medical Expenditure Panel Survey (MEPS). (8) It is a collection of
results from nationally representative surveys that are distributed to
families, individuals, medical providers, and employers throughout the
United States. From 1998 to 2001, (9) MEPS collected data on various
demographic characteristics of a random subsample of 158 Viagra users.
We present summary statistics of this survey in this subsection.
[FIGURE 2 OMITTED]
Figure 3 shows the age distribution of Viagra users. The average
Viagra user was at 57 years. The youngest participant in the survey was
at 18 years, while the oldest was at 87 years of age. Overall, the
distribution of ages appears fairly normal. Specifically, almost 90% of
Viagra users in the survey are 45 years of age and older. Owing to this
age distribution, we establish the treatment group in our analysis as
individuals aged 45 and above and the control individuals aged 25-40.
(10) We drop individuals between the ages 41-44. We view this age
category as neither in the treatment nor control as some of these
individuals may be prescribed Viagra, but not in the concentrations of
males aged 45 or over. For a different reason, we drop the 18-24 age
group: sexual activity of the 18-24 age group is in all likelihood a
poor control for people aged 45 and above relative to individuals aged
25-40. For example, the 25-40 age demographic has marriage rates much
closer to the above 45 age group than 18-24-year-olds.
[FIGURE 3 OMITTED]
B. Data on Variables of Interest
We examine how Viagra's introduction affects four different
variables of interest. First we examine how Viagra affected gonorrhea
rates in the drug's target population (e.g., males aged 45 and
above). (11) STDs are of interest because it studies the drug's
capacity to impact risky behavior associated with instant gratification.
(12) If Viagra did affect STDs, it is reasonable to expect that
gonorrhea rates would be the clearest example: it is one of the most
common STDs as it is a bacterial infection that can easily be
transmitted by fluids. We examine gonorrhea rates for men of age 45 and
above, but also for women. (13) It is important to understand how
lifestyle drugs affect the agents who do not take the drug directly but
rather indirectly enjoy their benefits or suffer from their costs. We
obtained state-level gonorrhea infection rate data by age and gender via
e-mail exchange from the Centers for Disease Control (CDC). These data
are available from the authors upon request.
Second, we test for how Viagra's introduction affects natality
rates in the target population. This is a variable of interest because
it highlights the drug's capacity to correct a physical disorder
and the resulting impact on immediate family planning decisions. Viagra
makes conceiving a child much less costly for the target population
(males with ED). Therefore, natality rates could increase upon
Viagra's introduction because couples who previously had physical
impediments to conception would now be better enabled to conceive. (14)
State-level natality data by father's age are drawn from the CDC
Vital Statistics. We define natality rates as the number of newborns by
fathers in age category j in state k divided by 100,000 men in age
category j in state k.
Third, we examine how Viagra's introduction affects criminal
sexual activity. This is an important variable due to the magnitude of
consequences for victims. It is possible that the publicity in the
public media about Viagra and sex of elderly changed the sexual crime
behavior of the elderly population. Hence, whether there has been an
increase in such arrests in the United States since the release of
Viagra, and if so amongst what age groups, is a relevant question. We
collected data for (i) sex offenses arrests (that excludes forcible
rapes) and (ii) forcible rape arrests in the United States by
suspect's sex and age group. Sexual offenses include the following
crime subcategories: adultery/fornication, incest, buggery, indecent
exposure, seduction and indecent liberties, sodomy or crimes against
nature, statutory rape (not forced), and any attempts of the stated
categories. (15) County-level arrests for sexual offense and rape
statistics are collected from the Uniform Crime Reports of the National
Archive of Criminal Justice Data for almost 900 U.S. counties.
Finally, we test for Viagra's effect on divorce rates. This
outcome highlights the importance of this particular medical innovation
on the household production function and is also a long-run planned
decision. Viagra is a lifestyle drug that enables increased coital
capacity for its users. If the relative benefit of getting married,
staying in a marriage, or getting divorced is asymmetrically affected by
Viagra then we would expect a significant effect of Viagra's
introduction on marriage market in the treatment group. To test this
hypothesis, divorce and marriage data were compiled using the Current
Population Survey (CPS) of the Bureau of Census for the Bureau of Labor
Statistics. This dataset is a time series of repeated cross-sectional
draws of individuals in the U.S. population.
III. ECONOMETRIC MODEL
We estimate several econometric models to identify the causal
effect of the introduction of Viagra on several outcomes of interest for
men and women middle aged and older. Specifically, we estimate the
drug's effect on rates of a STD (gonorrhea), natality rates and
sexual offenses rates including rape, and divorce rate. We use
state-level data for gonorrhea and natality, county-level data for
crime, and individual-level data for divorce. (16) We are forced to use
state-level data in some cases since, to our knowledge, no more
granular-level data exists.
For every variable of interest, our preferred econometric model
uses a difference-indifference estimator. Specifically, we estimate
least squares and Poisson models with the following specification:
(1) [y.sub.sgt] = h ([alpha] + [[mu].sub.s] + trend + 1 {g [greater
than or equal to] 45) 5
+ 1 {t [greater than or equal to] 1999} [gamma] + 1 {g [greater
than or equal to] 45}
x1 {t [greater than or equal to] 1999}[beta + [[epsilon].sub.sgt]).
In Equation (1), s indexes a location (state or county), g indexes
an age group, and t the year of observation. We allow for state or
county fixed effects, [[mu].sub.s]. When the data dictate, we include a
linear time trend takes the value of one for observations in the first
year of our dataset and n for observations in the nth year of data. The
function h() indicates that we sometimes estimate Equation (1) as a
semi-logarithmic model and sometimes as a Poisson model. We estimate the
difference in outcomes for the control group for the post-1998 period
relative to the pre-1999 period, [gamma], and the fixed effect of being
in the target age group for Viagra, [delta]. The main coefficient of
interest is [beta] which indicates the change in outcome variable of
interest for the treatment group attributable to the approval and sale
of Viagra. The differences-in-differences identification strategy
assumes that in the absence of treatment, the outcome variable of both
the treatment and control groups would have evolved the same over time.
We cluster standard errors at the state level to account for serial
correlation within a state. Further all county data regressions are
clustered by state which also accounts for the potential of spatial
correlation within each state.
We estimate Equation (1) using two methods. First, estimated by
"least squares," we put the outcome variables of interest in
log rates. While log rates are useful for ease of interpretation, it is
problematic for us in some cases: in our data we observe some years with
zero rates in some age groups for certain variables of interest,
especially in lower population states and counties. (17) Taking the log
of these rates leads to dropped observations and severe nonlinearity
around zero rates. We address this by trimming our datasets in the log
rate specification so that no state or county with fewer than 10
observed outcomes is ever observed. For example, if a particular state
ever has fewer than 10 observed cases of gonorrhea in either the
treatment or control group we trim that state from the dataset. We find
qualitatively similar results when using other trimming rules.
Second, in addition to estimating Equation (1) using ordinary least
squares (OLS) and log rates, we also use Poisson regressions with
standard errors clustered by state. This is a natural setting for a
Poisson regression since our dependent variables are count data (e.g.,
number of gonorrhea cases, number of births). This specification affords
us the opportunity to use all of our data, as opposed to dropping some
observations due to our trimming rule.
The key identifying assumption in our article is the timing of
treatment. Viagra was first sold commercially in mid-April 1998. We take
1999 to be the starting date of treatment for three reasons. First it is
the first complete year that Viagra was legal. Second, our aim is to
identify long-run effects. Three, any learning about the drug was likely
to have occurred in the second half of 1998. In any case, the
qualitative results are robust to allowing the treatment date to start
in 1998.
In interpreting our results, we want to stress caution in
identifying causal long-run impacts of Viagra due to some drawbacks of
our data. First, we do not observe a panel dataset on both an
individual's sexual activity and Viagra usage over time. Therefore,
our results are market-level outcomes. From a policy perspective,
though, market-level outcomes are the effects of interest since any
public policy targeted toward lifestyle drugs would operate at the
market level. Second, we did not conduct a randomized controlled trial,
which would have been ideal for assessing the direct link between ED
drug use and our variables of interest. Third, the timing of
Viagra's availability was uniform in all 50 states. This eliminates
cross-sectional variation in the timing of treatment as a source of
identification in our analysis. As with any differences-in-differences
approach, then, if the timing of treatment is correlated with any other
event which affected the outcome of interest, it will bias our
estimates. (18) As a result, we urge further study with individual-level
panel data. Fourth, although our collected CDC, CPS, and crime data
report the universe of all reported individuals in the United States,
the prevalence of some outcome variables is still low and precludes a
highly powered analysis at the individual level. More generally, from a
policy perspective further work may better characterize those users of
ED drugs who are at highest risk for an STD. For example, screening,
whether in the form of brief conversations or formal STD testing, would
be most effective if targeted toward those at highest risk.
IV. RESULTS
A. Sexually Transmitted Diseases
We estimate the model with state-level data for both males and
females by age cohort using the CDC dataset. Figure 4 displays the
aggregated data for gonorrhea rates by age group and sex. Due to
differing levels across the two age cohorts the plot has different
y-axes with the treatment group (e.g., older males) on the left y-axis
and the control group on the right y-axis. Panel (a) in Figure 4 shows
that gonorrhea rates increased substantially after 1999 for elderly
males peaking in 2007 at 40% above of their 1999 levels. In comparison,
the control group (age: 25 to 40) saw only a slight increase. It is
possible that some of the increase in the control group's rates may
be transmission of gonorrhea from older to younger cohorts over time. As
a result, then, our DD estimates (comparing the 40% increase relative to
the marginal increase) are a lower bound of the effect of Viagra in the
elderly population. Female gonorrhea rates are shown in panel (b). Here
patterns are less clear. After Viagra's approval and sale gonorrhea
rates in older women remain elevated while rates in young women
decrease.
In Figure 4, two additional features are apparent. First, the early
1990s experienced a substantial decline in the STD rates. This decline
coincides with a time period of increased caution and educational
campaigns to curb HIV infections in the United States (Health and Vital
Statistics 2010). Starting in 1998 and 1999, with the introduction of
Viagra, the age group trends reverse, with the elderly group showing
striking increases of over 40% for males by 2007.19 Moreover, Figure 4
provides evidence that the increase in gonorrhea is not due to a higher
detection (because males may run additional STD tests at the time when
Viagra is prescribed at the doctor's office). Instead, note that
the increase in gonorrhea rate occurs gradually, and not immediately in
the older cohort. This is in sharp contrast to the very immediate
adoption of the drug in April of 1998 as shown in Figure 1. If the
higher detection rate were responsible for the higher CDC gonorrhea
statistic, then one would expect a jump in the STD figure as well.
[FIGURE 4 OMITTED]
Finally, Figure 4 shows a significant effect of the great recession
after 2007. As the great recession affected accumulated wealth through
the stock market and housing stock savings, it appears that the elderly
group is more responsive to this wealth effect compared to the younger
age cohort. (20) We will return to this phenomenon when analyzing the
natality data below.
Estimation results are shown in Tables 1 and 2 for males and
females, respectively. In these and all subsequent tables we present
several versions of the econometric model and all our tables take the
following form unless otherwise stated: all least squares specifications
use log of rates as the dependent variable and progressively add more
fixed effects and a linear time trend to control for the time varying
effects. Because of using log rates, each of our OLS specifications trim
the sample as described above (drops state/counties ever reporting fewer
than 10 occurrences of the outcome variable). Our most flexible
specification in log rates is column 3 which includes a linear time
trend and state fixed effects. Columns 4 and 5 show results from
estimating the Poisson model on the subset of the trimmed data (column
4) that is used in the log rates specification and the full (untrimmed)
dataset in column 5. The last two columns limit the entire sample so
that only the years closest to the introduction of Viagra (1995-2005)
are used in the analysis. We view the results from columns 6 and 7 as a
robustness check to verify that the estimated effect of Viagra on our
variables of interest is not being driven by data many years before or
after Viagra's availability. For each variable of interest we also
include additional robustness checks in the Appendix that cut the sample
differently. Our main treatment effect estimate of interest is the top
row corresponding to [beta] of Equation (1) labeled as
"Viagra" in each table.
Table 1 displays the regression results for male gonorrhea rates as
the left-hand-side variable. Gonorrhea rates in the older population of
men unambiguously increases by between 15% and 28%. (21) Estimates are
somewhat smaller when the entire dataset is used in the Poisson
specifications in columns 5 and 7 compared to the corresponding least
squares specifications in columns 4 and 6. For both the Poisson and OLS
regression specifications, the estimated treatment effect of Viagra is
overall consistent conditional on the number of years being used in the
analysis being constant, that is, excluding the years of the highest STD
difference in 2006 and 2007 drops the estimated treatment effect from
23% in column 5 to 15% in column 7. The Appendix further provides
estimates on different time windows and shows that across all
specifications the Poisson treatment effects range from 13% to 24%.
Table 2 shows the same estimation results for females. It is
important to note that in these specifications for females, our
estimates are lower bounds since older infected men could have
transmitted gonorrhea to younger women. Overall, we find only very weak
evidence of an effect for females. In the Poisson regressions in column
4 (which include all years from 1990 to 2011, but trims the data to at
least 10), we find a significant effect of Viagra's introduction on
older female gonorrhea rate. Once we, however, estimate this via least
squares (column 3, or use all data (column 5) this significant result
disappears. In other specifications with the entire time series, the
coefficient of interest has the expected sign but is imprecisely
estimated. In the robustness check in which we include only a subset of
the years, the effects tighten to be nearly zero.
[FIGURE 5 OMITTED]
B. Natality Rates
State-level natality data by father's age cohort are drawn
from the CDC Vital Statistics and are shown in Figure 5. As stated
above, we define natality rates as the number of newborns by fathers in
age category j in state k divided by 100,000 men in age category j in
state k. (22) Figure 5 displays log natality rates by age grosup. Three
notes on these data are in order. First, there is an obvious drop in
natality rates from 2004 to 2006 for the treatment group. This is due to
unfortunately missing data in the CDC dataset for the 55+ age group in
some states. Second, for the control group, we see a pre-trend decline
in the natality rate over the first years of the sample period followed
by an increase which mirrors the older age group's increase. Third,
both groups' natality rates fall with the onset of the great
recession in 2008. To address these three data issues, in the following
regressions we systematically vary the time window of included years.
Table 3 shows regression results from the data displayed in Figure
5. Overall, we find no convincing evidence that Viagra had a significant
effect on natality rates for men in Viagra's target population in
our models. Using the entire dataset, in only two specifications is
there a marginally significant (10% level) decrease in natality rates
for the treatment group. Panel (b) and (c) then further display the
regression results for the years 1994 to 2003 as well as 1990 to 2007
and 1995 to 2005. In particular the data from 1994 to 2003 is used (i)
to balance the length of the pre-Viagra and post-Viagra periods, (ii)
eliminate the different pre-1994 trends, and (iii) ensure that we have
data for all age groups (as the dataset starts to become irregular in
2004). Across all these specifications, taken together, we find no
convincing evidence that Viagra had a significant effect on natality
rates for men in Viagra's target population.
Although we fail to reject the null hypothesis that Viagra had a
significant increase on birth rates for the treated group, the 95% of
the estimated confidence interval for the Poisson regression with all
data and a time trend is (-.077, .015). This means that the largest
possible increase Viagra could have caused is 1.5%. We take this as
evidence that even if Viagra were inducing additional births by older
males, the rate is small. (23)
C. Sexual Offenses and Rape
We collected county-level rape and sexual assault arrest data for
male perpetrators by age group from the Uniform Crime Reports of the
National Archive of Criminal Justice Data. Figure 6 shows the trends in
rape and sexual assault in log rates by age group. Both panels show a
significant downward trend in criminal activity. It is not immediately
apparent from the figure that Viagra's approval and sale caused
significantly higher sexually related reported criminal activity by the
target population.
[FIGURE 6 OMITTED]
Table 4 shows the regression results for county-level rape arrests
over the entire dataset. The three simplest regression specifications
find a significantly positive effect of Viagra and rape arrests. These
effects, though, are eliminated when more flexible specifications are
performed. In the two most flexible specifications, columns 5 and 7, we
find insignificant effects of Viagra on rape arrests. We show in the
Appendix the same qualitative results when we balance the dataset around
Viagra's introduction date.
Table 5 shows the regression results for county-level sexual
offense arrests over the entire dataset. We find similar results to
those from rape: in the simplest specifications we find a positive and
significant effect of Viagra on sexual offenses that is eliminated in
more flexible specifications. As shown in the Appendix, when using a
balanced dataset around Viagra's
introduction these results stay insignificant. In summary, we find
no strong evidence that Viagra affected rates of rape or sexual offenses
for the target population of the drug.
Although we fail to reject the null hypothesis that Viagra had a
significant increase on crime rates for the treated group, the 95% of
the estimated confidence interval for the Poisson regression with all
data is (-.112, .026). This means that the largest possible increase
Viagra could have caused in our econometric model is 2.6%, or 88 rapes
per year. Rapes by males over 45 were 1,300 in 1998, which represents
18% of all total rapes between both age groups.
D. Divorce Rates
To test if Viagra had an effect on divorce rates we collected data
from the CPS of the Bureau of Census for the Bureau of Labor Statistics.
These data are somewhat different than our other data in that the CPS is
collected as a random subsample of the U.S. population each year. The
dataset has age and marital status for the years 1990 to 2008. The
average sample size per year for males is 50,879 and for females 35,242.
Data showing divorce rates by age are displayed in Figure 7. The
graph shows divorces per 1,000 married households. For both age groups,
there is a pre-trend before Viagra's approval in 1998. The average
number of divorced households is roughly 135 and 123 per 1,000 married
households for the treatment and control, respectively over the sample
period. It is not clear from a visual inspection that there is any
significant break in this pre-trend due to Viagra's introduction.
There is a discrete drop in the divorce rate between 2000 and 2001 in
the control group, but that drop is both more than two full years after
Viagra's introduction and coincides with a U.S. recession in that
year. We performed regressions for both male and female divorce rates by
age cohort as above. Although we do not report the output in tables, we
estimate the effect of Viagra on male divorce to be 2.01 but that
estimate is insignificant (clustered standard error is 2.3). Noting that
the divorce rate for men in the treatment group in 1999 is 140 per
1,000, this represents a statistically insignificant change of roughly
1.4%. Similarly, we perform the same specification for women and
estimate a coefficient of .059 and that estimate is also insignificant
(standard error is .053). Taken together, these estimates indicate that
Viagra had no statistically significant effect in the marriage market.
[FIGURE 7 OMITTED]
Although we find no effect on the aggregate divorce rate, it could
still be that Viagra affected the age composition of divorces and
marriages. Tables 6 and 7 address these possible effects. We collected 3
years of data (1989, 1997, and 2007) from the CPS in order to check for
Viagra effects in the age composition of marriage in the treatment
group. The reason to start with 1989 is that we did not want to
attribute long-run trends occurring before 1997 to Viagra. We also stop
the sample in 2007 to avoid any systematic interactions of
marriage/divorce patterns by age due to the great recession. Table 6
shows a cross tab of average and standard deviation of age for all
cohabitating married couples by male's age. Consistent with
intuition, younger people are marrying later. The population of older
married males appears to be getting younger. This trend started before
Viagra's introduction. The average age of females married to men 45
or older has stayed roughly constant.
Breaking the data down further, we calculate percentiles of
wife's age conditional on having a husband aged 45 or more in Table
2. The table shows that there is no significant effect in the bottom
half of the distribution after the introduction of Viagra that breaks
from pretrends. We take this as further evidence that any effects occur
the marriage market due to Viagra's introduction are likely to be
second order. Taking the results of Tables 1 and 2 together shows that
the age spread between older males and their spouses is unaffected by
the introduction of Viagra.
In summary, although we fail to reject the null hypothesis that
Viagra had a significant increase on divorce rates for the treated
group, the 95% of the estimated confidence interval for the percentage
change in divorce rates is (-.018, .047). This means that the largest
possible increase Viagra could have caused in our econometric model is
4.7% or a decrease of 1.8%. There are reasonable arguments for Viagra
leading to an increase in divorce (Viagra makes being single relatively
more attractive) or a decrease in divorce (Viagra improves marriages).
Either percentage change, though, is dominated by long-run trends for
the older age cohort.
E. Welfare
Our results can inform a rudimentary welfare calculation of the
effect of Viagra on the target population. For four variables of
interest--natality, divorce, sex crimes, and age spread--we find no
evidence that Viagra had a statistically significant effect on the
target population. We find that Viagra's introduction increased
gonorrhea rates 15%-28% for both men and women older than 45. According
to the CDC, the average gonorrhea rate in 2011 in the 45-55 age group
ranged from 20.8 to 30.3 per 100,000 and 6.0 to 9.7 for 55-65 age group
with negligible rates for ages 65+.24 Taking the upper bounds for both
of the gonorrhea rate increases due to Viagra it implies that Viagra is
responsible for at most an additional 4,381 cases of gonorrhea per year
in the target population.25 Gonorrhea is commonly treated with Cefixime,
which has inexpensive generic versions for no more than $40. Therefore,
this yearly negative cost due to Viagra's introduction is on the
order of $140,000. This number is clearly a lower bound on cost, though,
given the physical and psychic costs of gonorrhea. However, the number
is orders of magnitude lower than annual sales of Viagra, which is a
lower bound for consumer value associated with its use. As a result, we
conclude that a back of the envelope welfare calculation finds that
Viagra's introduction led to an overall welfare increase.
V. CONCLUSIONS
We use the introduction of a well-publicized and heavily prescribed
lifestyle drug, Viagra, to test how that drug affects outcomes for the
target age cohort. We are studying four types of lifestyle choices of
economic importance: enabled risky behavior associated with instant
gratification (STD rates), enabled child bearing and family planning
decisions (natality and divorce), and enabled criminal behavior (sexual
assault and rape). We find evidence that the drug increased STDs in the
target population but that no other variable of interest was affected.
As a result, we find that this particular medical innovation targeted
toward older age cohorts only led to changes in short-term decision
making rather than long-run decision making. Overall, in a back of the
envelope cost benefit analysis, we find that the welfare impacts of
Viagra with respect to our outcomes of interest are positive and large.
These results have important implications for the economics
literature. First, as the population ages, more resources will be
invested in improving the elderly people's quality of life by the
development of medical innovations. As other new groundbreaking medical
innovations are introduced, our results imply that there might not be
necessarily substantial changes in economic decisions with long-term
impacts by the elderly. Second, these results contribute to a large
economic literature on how endowments and wealth influence decision
making since older individuals have, on average, more wealth than
younger cohorts (Kahneman, Knetsch, and Thaler 1991; Rabin 1998). We
find that Viagra affected relatively low cost risky decisions associated
with instant gratification only. This points to self-control problems,
even for experienced decision makers. These questions could be important
avenues for future research to better understand decisions in an aging
population.
ABBREVIATIONS
CDC: Centers for Disease Control
CPS: Current Population Survey
ED: Erectile Dysfunction
FDA: Food and Drug Administration
MEPS: Medical Expenditure Panel Survey
OLS: Ordinary Least Squares
STD: Sexually Transmitted Disease
doi: 10.1111/ecin.12147
APPENDIX
TABLE A1
Impact of Viagra on Log Rate of Gonorrhea Cases 1990-2011: U.S.
Males
1996-2011
Least Squares Poisson
(8) (9)
Viagra 0.110 0.132
(0.054) * (0.062) **
45+ -3.716 -3.627
(0.077) *** (0.077) ***
t > 1998 0.069 0.040
(0.037) * (0.035)
Linear trend 0.003 0.010
(0.006) (0.005) *
State FE Yes Yes
Adjusted [R.sup.2] 0.98
[R.sup.2] 0.98 0.92
N 608 1,632
1990-2007
Least Squares Poisson
(10) (11)
Viagra 0.310 0.244
(0.046) *** (0.065) ***
45+ -3.887 -3.723
(0.066) *** (0.076) ***
t > 1998 -0.017 0.166
(0.067) (0.037) ***
Linear trend -0.017 -0.042
(0.007) ** (0.009) ***
State FE Yes Yes
Adjusted [R.sup.2] 0.97
[R.sup.2] 0.97 0.90
N 864 1,834
1995-2007
Least Squares Poisson
(12) (13)
Viagra 0.184 0.181
(0.050) *** (0.051) ***
45+ -3.761 -3.660
(0.067) *** (0.078) ***
t > 1998 -0.150 -0.093
(0.038) *** (0.035) ***
Linear trend 0.032 0.023
(0.007) *** (0.007) ***
State FE Yes Yes
Adjusted [R.sup.2] 0.98
[R.sup.2] 0.98 0.92
N 494 1,326
1995-2011
Least Squares Poisson
(14) (15)
Viagra 0.155 0.165
(0.051) *** (0.048) ***
45+ -3.761 -3.660
(0.067) *** (0.078) ***
t > 1998 0.028 -0.008
(0.039) (0.036)
Linear trend 0.002 0.008
(0.005) (0.006)
State FE Yes Yes
Adjusted [R.sup.2] 0.98
[R.sup.2] 0.98 0.91
N 646 1,734
Notes: All columns have standard errors clustered at the state
level. Least squares specifications drop 32 states that ever have
less than 10 cases in a given age category in a given year. Note
that in 1999 there were 30,687 reported cases of gonorrhea in
males aged 25 -40 and 1,326 in males 45+.
* p < .1; ** p < .05; *** p < .01.
TABLE A2
Impact of Viagra on Log Rate of Gonorrhea Cases 1990-2011: U.S.
Females
1996-2011
Least Squares Poisson
(8) (9)
Viagra -0.046 -0.012
(0.036) (0.034)
45+ -2.183 -2.152
(0.034) *** (0.049) ***
t> 1998 0.264 0.104
(0.037) *** (0.050) **
Linear trend -0.041 -0.028
(0.008) *** (0.006) ***
State FE Yes Yes
Adjusted [R.sup.2] 0.96
[R.sup.2] 0.96 0.91
N 608 1,632
1990-2007
Least Squares Poisson
(10) (11)
Viagra 0.142 0.121
(0.051) ** (0.050) **
45+ -2.306 -2.253
(0.043) *** (0.070) ***
t> 1998 0.130 0.247
(0.069) *** (0.037) ***
Linear trend -0.048 -0.076
(0.006) *** (0.011) ***
State FE Yes Yes
Adjusted [R.sup.2] 0.96
[R.sup.2] 0.96 0.91
N 684 1,834
1995-2007
Least Squares Poisson
(12) (13)
Viagra 0.033 0.023
(0.037) (0.040)
45+ -2.197 -2.155
(0.031) *** (0.059) ***
t> 1998 0.042 0.001
(0.045) (0.045)
Linear trend -0.012 -0.017
(0.009) (0.008) **
State FE Yes Yes
Adjusted [R.sup.2] 0.97
[R.sup.2] 0.97 0.92
N 494 1,326
1995-2011
Least Squares Poisson
(14) (15)
Viagra -0.033 -0.010
(0.037) (0.039)
45+ -2.197 -2.155
(0.031) *** (0.059) ***
t> 1998 0.237 0.070
(0.041) *** (0.056)
Linear trend -0.042 -0.029
(0.008) *** (0.006) ***
State FE Yes Yes
Adjusted [R.sup.2] 0.96
[R.sup.2] 0.960 0.91
N 646 1,734
Notes: All columns have standard errors clustered at the state
level. Least squares specifications drop 32 states that ever have
less than 10 cases in a given age category in a given year. Note
that in 1999 there were 55,071 reported cases of gonorrhea in
females aged 25-40 and 8,169 in females aged 45+.
* p < .1; ** p < .05; *** p <.01.
TABLE A3
Impact of Viagra on Log Rate of Rape Arrests of U.S. Males:
1994-2010
1996-2011
Least Squares Poisson
(8) (9)
Viagra -0.030 -0.010
(0.045) (0.039)
45+ -1.805 -1.879
(0.060) * (0.049) ***
t> 1998 -0.020 0.028
(0.028) (0.040)
Linear trend -0.043 -0.027
State FE (0.004) *** (0.004) ***
Adjusted [R.sup.2] 0.84
[R.sup.2] 0.84 0.52
N 2,250 16,110
1994-2007
Least Squares Poisson
(10) (11)
Viagra -0.014 -0.008
(0.051) (0.040)
45+ -1.803 -1.874
(0.064) *** (0.063) ***
t> 1998 -0.059 -0.027
(0.038) (0.049)
Linear trend -0.036 -0.019
State FE (0.005) *** (0.005) ***
Adjusted [R.sup.2] 0.84
[R.sup.2] 0.85 0.54
N 2,100 15.036
1995-2007
Least Squares Poisson
(12) (13)
Viagra -0.014 -0.007
(0.051) (0.040)
45+ -1.814 -1.871
(0.069) *** (0.060) ***
t> 1998 -0.059 -0.029
(0.038) (0.050)
Linear trend -0.037 -0.016
State FE (0.006) *** (0.006) **
Adjusted [R.sup.2] 0.85
[R.sup.2] 0.85 0.54
N 1,950 13,962
1995-2011
Least Squares Poisson
(14) (15)
Viagra -0.022 -0.012
(0.042) (0.047)
45+ -1.819 -1.875
(0.061) *** (0.059) ***
t> 1998 -0.025 0.029
(0.029) (0.041)
Linear trend -0.043 -0.027
State FE (0.004) *** (0.004) ***
Adjusted [R.sup.2] 0.84
[R.sup.2] 0.84 0.52
N 2,400 17,184
Notes: All columns have standard errors clustered at the state
level. Least squares specifications drop 32 states that ever have
less than 10 cases in a given age category in a given year.
* p < .1; ** p < .05; *** p < .01.
TABLE A4
Impact of Viagra on Log Rate of Sexual Offense Arrests of U.S.
Males: 1994-2010
1996-2011
Least Squares Poisson
(8) (9)
Viagra -0.049 -0.026
(0.029) * (0.034)
45+ -1.139 -1.219
(0.041) (0.035) ***
t > 1998 0.025 0.038
(0.039) (0.031)
Linear trend -0.031 -0.020
State FE (0.004) *** (0.004) ***
Adjusted [R.sup.2] 0.76
[R.sup.2] 0.77 0.55
N 8,550 26,130
1994-2007
Least Squares Poisson
(10) (11)
Viagra -0.042 -0.004
(0.032) (0.035)
45+ -1.140 -1.160
(0.028) *** (0.027) ***
t > 1998 -0.031 -0.017
(0.038) (0.033)
Linear trend -0.020 -0.009
State FE (0.005) *** (0.005) ***
Adjusted [R.sup.2] 0.76
[R.sup.2] 0.77 0.56
N 7,980 24,388
1995-2007
Least Squares Poisson
(12) (13)
Viagra -0.042 -0.004
(0.032) (0.035)
45+ -1.136 -1.179
(0.033) *** (0.030) ***
t > 1998 -0.031 -0.018
(0.038) (0.033)
Linear trend -0.019 -0.007
State FE (0.005) *** (0.005)
Adjusted [R.sup.2] 0.76
[R.sup.2] 0.77 0.56
N 7,410 22,646
1995-2011
Least Squares Poisson
(14) (15)
Viagra -0.048 -0.036
(0.025) * (0.033)
45+ -1.141 -1.200
(0.034) *** (0.031) ***
t > 1998 0.031 0.048
(0.039) (0.031)
Linear trend -0.030 -0.020
State FE (0.004) *** (0.004) ***
Adjusted [R.sup.2] 0.75
[R.sup.2] 0.76 0.54
N 9,120 27,872
Note. Least squares specifications drop 32 states that ever have
less than 10 cases in a given age category in a given year.
*p < .1; ** p < .05; *** p < .01.
REFERENCES
Bailey, M. "More Power to the Pill: The Impact of
Contraceptive Freedom on Women's Life Cycle Labor Supply."
Quarterly Journal of Economics, 121(1), 2006, 289-320.
Cutler, D. M. "The Lifetime Costs and Benefits of Medical
Technology." Journal of Health Economics, 26(6), 2007. 1081-100.
Cutler, D. M., and M. McClellan. "Is Technological Change in
Medicine Worth It?" Health Affairs, 20,2001. 11-29.
Gaydos, C. A., M. Theodore, N. Dalesio, B. J. Wood, and T. C.
Quinn. "Comparison of Three Nucleic Acid Amplification Tests for
Detection of Chlamydia trachomatis in Urine Specimens." Journal of
Clinical Microbiology, 42(7), 2004, 3041-45.
Goldin, C., and L. Katz. "The Power of the Pill: Oral
Contraceptives and Women's Career and Marriage Decisions."
Journal of Political Economy, 110(4), 2002, 730-70.
Harris, A. R., W. N. Evans, and R. M. Schwab. "Education
Spending in an Aging America." Journal of Public Economics, 81,
2001,449-72.
Health and Vital Statistics. "Use of Contraception in the
United States: 1982-2008." U.S. Department of Health and Human
Services, Series 23, Number 29, 2010.
Jena, A. B., D. P. Goldman, A. Kamdar, D. N. Lakdawalla, and Y. Lu.
"Sexually Transmitted Diseases Among Users of Erectile Dysfunction
Drugs: Analysis of Claims Data." Annals of Internal Medicine,
153(1), 2010, 1-7.
Kahneman, D., J. L. Knetsch, and R. H. Thaler. "Anomalies: The
Endowment Effect, Loss Aversion and Status Quo Bias." Journal of
Economic Perspectives, 5(1), 1991, 193-206.
Levy, G. "The Politics of Public Provision of Education."
Journal of Political Economy, 120(4), 2005, 1507-34.
Lindau, S. T., S. A. Leitsch, K. L. Lundberg, and J. Jerome.
"Older Women's Attitudes, Behavior, and Communication About
Sex and HIV: A Community-Based Study." Journal of Women's
Health, 15. 2006, 747-53.
Meyers, D., T. Wolff, K. Gregory, L. Marion, V. Moyer, H. Nelson,
D. Petitti, and G. F. Sawaya. "USPSTF Recommendations for STI
Screening." American Family Physician, 77(6), 2008, 819-24.
Montague, D. K., J. P. Jarow, G. A. Broderick, R. R. Dmochowski, J.
P. Heaton, T. F. Lue, A. J. Milbank, A. Nehra, and I. D. Sharlip.
"The Management of Erectile Dysfunction: An AUA Update."
Journal of Urology, 174(1), 2005, 230-39.
Patel, D., B. Gillespie, and B. Foxman. "Sexual Behavior of
Older Women: Results of a Random-Digit-Dialing Survey of 2.000 Women in
the United States." Sexually Transmitted Diseases, 30, 2003,
216-20.
Paul, J. P., L. Pollack, D. Osmond, and J. A. Catania. "Viagra
(Sildenafil) Use in a Population-Based Sample of U.S. Men Who Have Sex
with Men." Sexually Transmitted Diseases, 32, 2005, 531-33.
Rabin, M. "Psychology and Economics." Journal of Economic
Literature, 36(1), 1998, 11-46.
Skiest, D. J., and P. Reiser. "Human Immunodeficiency Virus
Infection in Patients Older Than 50 Years. A Survey of Primary Care
Physicians' Beliefs, Practices, and Knowledge." Archives of
Family Medicine, 6, 1997, 289-94.
Stall, R., and J. Catania. "AIDS Risk Behaviors Among Late
Middle-Aged and Elderly Americans. The National AIDS Behavioral
Surveys." Archives of Internal Medicine, 154, 1994, 57-63.
Swearingen, S. G., and J. D. Klausner. "Sildenafil Use, Sexual
Risk Behavior, and Risk for Sexually Transmitted Diseases, Including HIV
Infection." American Journal of Medicine, 118, 2005, 571-77.
(1.) There are at least two channels through which medical
innovations could affect decision making of the elderly. First, medical
innovations can enable specific behavior that was previously made
impossible due to age or disease For example, a hip replacement or
medication can enable exercise. Second, if medical innovations extend
the expected life span of a population, it can indirectly affect
decisions related to financial planning and years worked. We are
concerned with the first channel in this article.
(2.) We want to highlight here we mean immediate impacts. In the
longer run there could be additional family planning effects as younger
men choose to delay having children longer due to the existence of ED
drugs. Further, fatherhood can clearly be unplanned as well for both
older and younger men. As a result, we are testing for an effect of
Viagra's introduction generally on all births, both planned and
unplanned, fathered by the older age cohort.
(3.) We show that the age composition of Viagra users stays roughly
constant from the first years of its introduction in 1998 through 2007.
(4.) In a different context, recent research finds larger
concentrations of older populations can cause modest decreases in
education spending and/or changes in the political economy equilibrium
of different voting blocs (Harris, Evans, and Schwab 2001; Levy 2005).
The goal of this article, then, is to examine how one particular
lifestyle drug targeted at the elderly population changes the aging
population's socioeconomic decisions.
(5.) Artificial organs, for example, are another major medical
innovation currently being tested which could lead to major changes in
the lifestyle decisions of the elderly and even young people.
(6.) ED medication data were purchased from IMS Health.
(7.) Titles for these articles included The Nation: Thanks a Bunch,
Viagra and The Pill That Revived Sex, Or at Least Talking About It.
(8.) These data are stored in the yearly Prescribed Medicine Files
at http://www.meps.ahrq.gov/mepsweb/.
(9.) We also collected the same information from MEPS for 2005-2007
and found that the age composition of Viagra users stayed roughly
constant over time. That data are available from the authors upon
request.
(10.) There are some media reports that recreational Viagra usage
increased in younger men in the 2000s
(http://usatoday30.usatoday.com/news/health/2001-03-21viagra-abuse.htm).
This implies that our estimates are lower bounds for the effect of
Viagra on the treatment group if Viagra also had effects on the control
group.
(11.) To transfer the raw data of levels into age cohort-by-state
and cohort-by-county rates, we use the Intercensal Estimates of the
Resident Population by the U.S. Census Bureau, Population Division.
(12.) Jena et al. (2010) is one relevant study from the medical
literature which relates STD rates of Viagra users to nonusers. Jena et
al. (2010) does not use the approval of the drug as part of the
identification technique. As a result, their treated population does not
reflect the treatment group isolated with our identification strategy.
The contribution of our study, then, is that it presents the first
empirical paper that does document the causal relationship between ED
medications and STD rates among the elderly, using the most
comprehensive dataset collected for the United States. In fact Jena et
al. (2010) suggest that a study such as ours, where STDs are compared
prior and post the introduction of Viagra, should be performed. The
medical literature further descriptively outlines some differences
between ED users and nonusers and we list these findings here: due to
some combination of reduced fertility rates and preferences, the elderly
are up to one sixth less likely to use condoms than people in their
twenties and one fifth as likely to be tested for STDs (Lindau et ai.
2006; Patel, Gillespie, and Foxman 2003; Stall and Catania 1994). To
that end, primary care physicians are less likely to discuss STD
prevention with older patients (see Skiest and Reiser 1997). However, ED
drugs users report having both more recent sexual partners and higher
STD rates than nonusers (see Paul et al. 2005; Swearingen and Klausner
2005). Although direct data do not exist, these findings imply that if
older individuals use Viagra, they both have more unprotected sex with
more sexual partners than older individuals who do not use Viagra. As a
result, there is reason to believe that the introduction of Viagra led
to a change in sexual behavior of people who took it. That is not to say
there was not self-selection in which individuals choose to take Viagra
but it is plausible that even with self-selection changes in sexual
behavior are feasible. Unfortunately, due to a lack of data, more
precise empirical statements are not possible.
(13.) We collected data for chlamydia but that data series begins
in 1996. Also, there was a significant improvement of chlamydia
screening which occurred during the study period (Gaydos et al. 2004).
This partially explains a dramatic increase in chlamydia diagnoses over
the sample period. Further, there was a change in public health policy
which suggested that younger women were tested significantly more over
this time period (Meyers et al. 2004). As a result, there was a dramatic
rise in chlamydia rates observed in the control group during this time
period. Disentangling these last two effects from the Viagra effect is
challenging. As a result, we do not include chlamydia in the analysis.
(14.) As noted above, we consider family planning generally as
natality rates are a function of both planned and unplanned pregnancies.
(15.) Sex solicitation rates are also not included in this category
but in a "catch all" category that includes many other arrests
not prevalent to this research. Solicitation rates may be of interest
but are only recorded at the local level and time restraints do not
allow us to acquire these rates.
(16.) In some cases we aggregate to the state level in order to
include rural counties with few observations.
(17.) For example, in Wyoming there are no recorded sexual assaults
for males older than 45.
(18.) For example, we cannot separately identify if the role of
media coverage of Viagra or advertising for Viagra directly affected the
variables of interest.
(19.) It is important to note, though, that STD rates did not
immediately increase in the older cohort in 1998, but rather the rate of
growth of STD infections increased. We see two potential reasons for
this. First, bacteria need time to diffuse in the network of sexual
partners through unsafe sexual encounters. The more people carry an STD
the higher the probability of infection. Second, the composition of
Viagra users could have changed over time, even if the age of
distribution of Viagra users stays constant. For example, there could
have initially been a social stigma associated with Viagra so that it
was not accepted in casual situations. Conversely, it could have been
prescribed more initially to couples in a long-term relationship.
Unfortunately, there is no dataset we are aware of that is a repeated
cross section of Viagra users over time that could help us answer this
more precisely.
(20.) This could also be the result of decreased medical care due
to the recession and therefore missed diagnoses. Generally, the goal of
this article is not to attribute STD rates to business cycles so we
leave this question to future research.
(21.) Given that we find gonorrhea rates increase for older men
relative to younger men, there is the question of what percentage was
due to increased infidelity of married men versus increased sexually
activity of single men. Our data do not let us address this interesting
question directly.
(22.) We lagged our births in the following way: Our births in 1999
are matched to Viagra availability in 1998 (which is ideal, given that
Viagra became available on April 1).
(23.) Given that older males' fathering rates are only 201
births per 100,000 (while the rate is above 7,400 per 100.000 in the
25-40 age cohort), the overall economic effect is not large, as it would
account for an additional three births only per 100,000. If it were
true, though, that Viagra had a significant and negative effect on
natality rates, the maximum estimated effect size, -7.7% (or 15 newborns
per 100.000), is nontrivial.
(24.) http://www.cdc.gov/std/statsll/tables/21.htm. Last accessed
October 17, 2013.
(25.) Average gonorrhea rates by age from the CDC and population by
age data from the census from 2011,
http://www.census.gov/population/age/data/201 lcomp.html. Last accessed
October 17. 2013.
JACOB LARIVIERE and HENDRIK WOLFF, We are indebted to Max
Auffhammer, Ron Lee, and David Zilberman for valuable discussions and
suggestions. We are also grateful for helpful student research
assistance by Reid Johnsen, Mingyuan Hua, Ling Ma, Jennifer Meredith,
Dorian Sidhu, and Stephanie Thomas. Any mistakes are the authors'.
LaRiviere: Assistant Professor, Economics and Baker Center for
Public Policy, University of Tennessee, Knoxville, TN 37996. Phone (865)
974-8114, E-mailjlarivil@utk.edu Wolff: Assistant Professor, Department
of Economics, University of Washington, Seattle, WA 98195. Phone (510)
220-7961, E-mail hgwolff@uw.edu
TABLE 1
Impact of Viagra on Log Rate of Gonorrhea Cases 1990-2011: U.S.
Males
1990-2011
Least Squares Least Squares Least Squares
(1) (2) (3)
Viagra 0.281 0.281 0.281
(0.044) *** (0.044) *** (0.044) ***
45+ -3.887 -3.887 -3.887
(0.065) *** (0.066) *** (0.066) ***
t > 1998 -0.182 -0.182 -0.013
(0.056) *** (0.057) *** (0.049)
Linear trend -0.015
(0.004) ***
State FE No Yes Yes
Adjusted [R.sup.2] 0.93 0.97 0.97
[R.sup.2] 0.93 0.97 0.97
N 836 836 836
1990-2011
Poisson Poisson
(4) (5)
Viagra 0.274 0.229
(0.053) *** (0.064) ***
45+ -3.896 -3.723
(0.051) *** (0.076) ***
t > 1998 0.024 0.027
(0.035) (0.036)
Linear trend -0.028 -0.022
(0.004) *** (0.007) ***
State FE Yes Yes
Adjusted [R.sup.2]
[R.sup.2] 0.92 0.89
N 836 2242
1995-2005
Least Squares Poisson
(6) (7)
Viagra 0.169 0.155
(0.046) *** (0.061) **
45+ -3.761 -3.660
(0.068) *** (0.078) ***
t > 1998 -0.098 -0.031
(0.042) ** (0.039)
Linear trend 0.022 0.010
(0.010) ** (0.010)
State FE Yes Yes
Adjusted [R.sup.2] 0.98
[R.sup.2] 0.98 0.92
N 418 1,122
Notes: All columns have standard errors clustered at the state
level. Least squares specifications drop 32 states that ever have
less than 10 cases in a given age category in a given year. Note
that in 1999 there were 30,687 reported cases of gonorrhea in
males aged 25-40 and 1,326 in males aged 45+.
* p < .1; ** p < .05; *** p <.01.
TABLE 2
Impact of Viagra on Log Rate of Gonorrhea Cases 1990-2011: U.S.
Females
1990-2011
Least Squares Least Squares Least Squares
(1) (2) (3)
Viagra 0.077 0.077 0.077
(0.046) (0.046) (0.047)
45+ -2.306 -2.306 -2.306
(0.042) *** (0.043) *** (0.043) ***
t> 1998 -0.388 -0.388 0.196
(0.063) *** (0.064) *** (0.047) ***
Linear trend -0.053
(0.005) ***
State FE No Yes Yes
Adjusted [R.sup.2] 0.84 0.93 0.95
[R.sup.2] 0.84 0.93 0.95
N 836 836 836
1990-2011
Poisson Poisson
(4) (5)
Viagra 0.146 0.088
(0.059) ** (0.047) *
45+ -2.349 -2.253
(0.045) *** (0.070) ***
t> 1998 0.183 0.141
(0.038) *** (0.039) ***
Linear trend -0.062 -0.060
(0.004) *** (0.008) ***
State FE Yes Yes
Adjusted [R.sup.2]
[R.sup.2] 0.92 0.91
N 836 2,242
1995-2005
Least Squares Poisson
(6) (7)
Viagra 0.016 0.008
(0.037) (0.039)
45+ -2.197 -2.155
(0.031) *** (0.059) ***
t> 1998 0.097 0.049
(0.044) ** (0.047)
Linear trend -0.023 -0.027
(0.010) ** (0.010) ***
State FE Yes Yes
Adjusted [R.sup.2] 0.97
[R.sup.2] 0.97 0.92
N 418 1,122
Notes: All columns have standard errors clustered at the state
level. Least squares specifications drop 32 states that ever have
less than 10 cases in a given age category in a given year. Note
that in 1999 there were 55,071 reported cases of gonorrhea in
females aged 25-40 and 8,169 in females aged 45+.
* p < .1; ** p < .05; *** p < .01.
TABLE 3
Impact of Viagra on Rate of Births by Age of
U.S. Male, 1990-2010
1990-2010
Least Least
Squares Squares
(1) (2)
Viagra -0.028 -0.028
(0.022) (0.017) *
45+ -3.745 -3.745
(0.016) *** (0.012) ***
t > 1998 0.098 0.098
(0.009) *** (0.011) ***
Linear trend
State FE No Yes
Adjusted [R.sup.2] 0.98 0.99
N 1,836 1,836
1990-2010
Least
Squares Poisson
(3) (4)
Viagra -0.028 -0.031
(0.016) * (0.023)
45+ -3.745 -3.704
(0.012) *** (0.017) ***
t > 1998 0.052 0.058
(0.017) *** (0.005) ***
Linear trend 0.004 0.004
(0.001) *** (0.001) ***
State FE Yes Yes
Adjusted [R.sup.2] 0.99 0.99
N 1,836 1,836
1994-2003
Least Least
Squares Squares
(5) (6)
Viagra -0.049 -0.049
(0.029) * (0.022) **
45+ -3.707 -3.707
(0.021) *** (0.015) ***
t> 1998 0.086 0.086
(0.013) *** (0.015) ***
Linear trend
State FE No Yes
Adjusted [R.sup.2] 0.98 0.99
N 1,020 1,020
1994-2003
Least
Squares Poisson
(7) (8)
Viagra -0.049 -0.049
(0.022) ** (0.030)
45+ -3.707 -3.672
(0.015) *** (0.021) ***
t> 1998 0.027 0.011
(0.025) (0.004) ***
Linear trend 0.012 0.015
(0.004) *** (0.001) ***
State FE Yes Yes
Adjusted [R.sup.2] 0.99 0.99
N 1,020 1020
1990-2007
Squares Poisson
(1) (2)
Viagra -0.018 -0.023
(0.018) (0.026)
45+ -3.745 -3.704
(0.012) *** (0.017) ***
t> 1998 0.036 0.059
(0.019) * (0.006) ***
Linear trend 0.006 0.003
(0.002) *** (0.001) ***
State FE Yes Yes
Adjusted [R.sup.2] 0.99 0.99
[R.sup.2] 0.99 0.99
N 1,530 1,530
1995-2005
Squares Poisson
(3) (4)
Viagra -0.051 -0.049
(0.031) * (0.032)
45+ -3.705 -3.671
(0.017) *** (0.024) ***
t> 1998 0.024 0.003
(0.026) (0.004)
Linear trend 0.013 0.018
(0.004) *** (0.001) ***
State FE Yes Yes
Adjusted [R.sup.2] 0.99 0.99
[R.sup.2] 0.99 0.99
N 918 918
Notes: All columns have standard errors clustered at the
state level. Least squares regressions use log rates as left-
hand-side variable. Poisson specifications use count data.
2003 is last year before missing data for 55+ age cohort's
fatherhood data.
* p < .1; ** p < .05; *** p < .01.
TABLE 4
Impact of Viagra on Log Rate of Rape Arrests by U.S. Male
Population by County: 1994-2010
1994-2010
Least Squares Least Squares Least Squares
(1) (2) (3)
Viagra 0.186 0.186 -0.026
(0.032) *** (0.032) *** (0.042)
45+ -1.734 -1.734 -1.808
(0.053) *** (0.053) *** (0.059) ***
t > 1998 -0.382 -0.382 -0.021
(0.033) *** (0.034) *** (0.030)
Linear trend -0.043
(0.004) ***
State FE
Adjusted [R.sup.2] 0.54 0.83 0.83
[R.sup.2] 0.54 0.83 0.84
N 2,550 2,550 2,550
1994-2010
Poisson Poisson
(4) (5)
Viagra -0.034 -0.013
(0.051) (0.044)
45+ -1.739 -1.877
(0.072) *** (0.062) ***
t > 1998 0.003 0.024
(0.042) (0.042)
Linear trend -0.043 -0.028
(0.005) *** (0.004) ***
State FE
Adjusted [R.sup.2]
[R.sup.2] 0.73 0.53
N 2,550 18,258
1995-2005
Least Squares Poisson
(6) (7)
Viagra -0.002 0.008
(0.064) (0.048)
45+ -1.804 -1.861
(0.077) *** (0.057) ***
t > 1998 -0.077 -0.070
(0.044) * (0.054)
Linear trend -0.033 -0.007
(0.007) *** (0.008)
State FE
Adjusted [R.sup.2] 0.85
[R.sup.2] 0.86 0.55
N 1,650 11,814
Notes: All columns have standard errors clustered at the state
level. Least squares specifications drop 32 states that ever have
less than 10 cases in a given age category in a given year.
* p < .1; ** p < .05; *** p < .01.
TABLE 5
Impact of Viagra on Log Rate of Sexual Offense Arrests of U.S.
Males: 1994-2010
1994-2010
Least Squares Least Squares Least Squares
(1) (2) (3)
Viagra 0.104 0.104 -0.048
(0.023) *** (0.023) *** (0.025) *
45+ -1.089 -1.089 -1.143
(0.027) *** (0.028) *** (0.030) ***
t > 1998 -0.230 -0.230 0.029
(0.046) *** (0.046) *** (0.040)
Linear trend -0.031
(0.004) ***
State FE
Adjusted [R.sup.2] 0.26 0.74 0.75
[R.sup.2] 0.26 0.75 0.76
N 9,690 9,690 9,690
1994-2010
Poisson Poisson
(4) (5)
Viagra -0.012 -0.043
(0.029) (0.035)
45+ -1.085 -1.181
(0.025) *** (0.027) ***
t > 1998 0.053 0.047
(0.033) (0.031)
Linear trend -0.030 -0.020
(0.004) *** (0.004) ***
State FE
Adjusted [R.sup.2]
[R.sup.2] 0.70 0.54
N 9,690 29,614
1995-2005
Least Squares Poisson
(6) (7)
Viagra -0.037 0.010
(0.038) (0.034)
45+ -1.134 -1.169
(0.033) *** (0.033) ***
t > 1998 -0.059 -0.028
(0.038) (0.032)
Linear trend -0.013 -0.005
(0.006) ** (0.006)
State FE
Adjusted [R.sup.2] 0.77
[R.sup.2] 0.78 0.57
N 6,270 19,162
Notes: All columns have standard errors clustered at the state
level. Least squares specifications drop 32 states that ever have
less than 10 cases in a given age category in a given year.
* p < .1; ** p < .05; *** p < .01.
TABLE 6
Mean Age of Married Female by Husband's Age
Male 25-40 Male 45+
Year Male Female Male Female
1989 33.15 31.62 60.38 56.48
4.36 5.24 10.56 11.57
1997 33.72 32.66 59.92 56.11
4.34 5.71 11.02 11.87
2007 33.71 32.85 59.64 56.33
4.34 5.93 10.55 11.32
Notes: 1989 data: n= 137,108 (age: 25-40) and 191,092
(age: 45+); 1997 data: n = 98,765 (age: 25-40) and 168,596
(age: 45+); 2007 data: n = 90,611 (age: 25-40) and 215,407
(age: 45+). All data are from CPS IPUMS. Numbers in italics
display the standard deviation of the age distributions.
TABLE 7
Age Distribution of Married Females
Conditional on Husband 45+
Percentile
Year 1% 5% 10% 25% 50%
1989 31 39 43 47 56
1997 32 39 43 47 54
2007 33 40 43 48 55
Notes: Data are from CPS IPUMS. Percentiles are
wife's age condition on male spouse aged 45+. 1989 data
n = 191,092 (age: 45+). 1997 data n = 168,596. 2007 data
n = 215,407 (age: 45+).
