Eye Disease, the Fertility Decline, and the Emergence of
Global Income Differences*
Thomas Barnebeck Andersen†, Carl-Johan Dalgaard†† and Pablo Selaya‡
First draft: April 18, 2010.
This version: February 5, 2014
Abstract: This research advances and empirically establishes the hypothesis that regional variation in
the historical incidence of eye disease has influenced the current global distribution of per capita
income. By reducing work life expectancy, high historical eye disease incidence has served to
diminish the incentive to accumulate skills, thereby delaying the fertility transition and the take-off to
sustained economic growth. As a consequence of a differential timing of the take-off to growth,
prompted by differences in the inherent return to skill formation, global income disparities have
emerged.
Keywords: Comparative development, eye disease, climate, fertility decline
JEL Codes: O11; I00; Q54
* We would like to thank Oded Galor, Moshe Hazan, Peter Sandholt Jensen, Nicolai Kaarsen, David Mayer,
Stelios Michalopoulos, Fidel Perez-Sebastian, James Robinson, Jon Temple, David Weil and seminar
participants at a number of conferences and workshops. Lise Hansen provided excellent research assistance.
This research was supported by the European Commission within the project “Long-Run Economic Perspectives
of an Aging Society” (LEPAS) in the Seventh Framework Programme under the Socio-economic Sciences and
Humanities theme (Grant Agreement: SSH7-CT-2009-217275). A previous version of the paper was circulated
under the title “Eye Disease and Development.”
† Department of Business and Economics, University of Southern Denmark. Email: [email protected]
††
Department of Economics, University of Copenhagen, and CEPR. Email: [email protected]
and [email protected]
‡ Department of Economics, University of Copenhagen. Email: [email protected]
1
1
Introduction
The last few decades have witnessed the emergence of a new strand of growth research that attempts
to elucidate the mechanics of development over the very long run: “from Malthusian stagnation to
sustained growth” (e.g., Galor and Weil, 2000; Galor and Moav, 2002; Lucas, 2002; Hansen and
Prescott, 2002; for a survey, see Galor, 2011). The key proposition of this literature in terms of
comparative development is that the differential timing of the fertility transition importantly shaped
the global distribution of income. Theoretically, the fertility transition is paramount in that it both
reduces capital dilution, enabling per capita growth to take hold, and unleashes a reinforcing virtuous
circle involving rising human capital levels and technological innovation. The present research
examines the empirical relevance of this fundamental proposition.
We assess the predictive power of the above-mentioned theoretical work by studying the long-run
consequences of global differences in the inherent return to skill formation, which can be attributed to
variation in the historical incidence of eye disease. We hypothesize and extensively document that in
places with high incidence of debilitating eye disease the time horizon over which skill investments
can be recuperated shortens substantially. In theory, low returns to skill investments translate into a
delayed onset of the fertility transition and in turn into a delayed take-off to sustained growth (see e.g.
Galor, 2010). Consequently, the historical incidence of eye disease should be a (fundamental)
determinant of the timing of the fertility transition, subsequent human capital investments, and
contemporary prosperity.
Our empirical analysis reveals that historical eye disease incidence is a robust determinant of
contemporary prosperity and that this reduced-form link is mediated through the timing of the fertility
transition and long-run investments in human capital. We thereby find support for the central
propositions of the literature on growth over the very long run, namely that historical differences in
the tendency to invest in human capital led to a differential timing of the fertility transition, which
helped generate the vast cross-country difference in per capita income that can be observed today.
In order to measure the historical incidence of eye disease we employ an environmental determinant
of a cluster of eye diseases: solar ultraviolet radiation (UV-R). Epidemiologically, UV-R has been
shown to be a determinant of several forms of eye disease of which the most important is cataract. The
proposition that greater exposure to UV-R leads to cataract has been established theoretically, through
experimental work, and through a substantial number of epidemiological studies that relate UV-R
exposure to cataract incidence in human populations (e.g., Javitt et al., 1996; Brian and Taylor, 2001;
West, 2007). The UV-R/cataract link is particularly important as cataract is the single most important
determinant of blindness; in 2002, 48% of global blindness was attributable to cataract alone
(Lansingh et al., 2007). But UV-R is also suspected of influencing the incidence of two other eye
2
diseases: pterygium and macular degeneration (e.g., Gallagher and Lee, 2006). Like cataract, both of
these diseases negatively influence visual acuity; therefore, they might also have had a harmful effect
on the return to skill formation. In the analysis below we document that our measure of UV-R is a
robust cross-county correlate of cataract prevalence. This provides assurance that our indicator of the
historical incidence of eye disease indeed is relevant, in keeping with a voluminous literature in
epidemiology.1
In order to ascertain whether eye disease is capable of substantially affecting the return to skill
formation, we invoke existing ophthalmological surveys of cataract prevalence, which we use to
calculate work life expectancy in different geographic regions. We find that the observed prevalence
rates can create a gap in expected work life by up to 14 years when comparing high and low UV-R
regions. Whereas cataract only appears as an old-age condition in Western Europe, it emerges
considerably earlier in life, and proceeds to increase at greater speed with age, in regions closer to the
equator. It is therefore plausible that differences in historical eye disease incidence have served to
create substantial variation in the perceived return to skill accumulation.
On this background we proceed to examine the influence of UV-R on comparative development, by
way of cross-section regression analysis. We find that that countries more exposed to UV-R are
significantly poorer today, as compared to countries less exposed. This result is robust to the inclusion
of a demanding set of geography controls, including (absolute) latitude, elevation, precipitation, and
average temperature. Taken at face value, the estimated effect of UV-R on contemporary income per
capita is economically significant. In the cross-country setting, one standard deviation increase in UVR lowers early 21st century GDP per capita by roughly 130%. This is a large effect; and it is too large
to plausibly reflect the direct impact of eye disease on individual-level earnings, as argued below. But
if differences in UV-R, by generating cross-country variation in the perceived return to skill
accumulation, has influenced the relative timing of the take-off to sustained growth, a much larger
impact on current income per capita can be rationalized.
In order to examine whether this is a meaningful account of the UV-R/income gradient we perform a
series of consistency checks. First, we document that the gradient emerges during the 20th century; it
did not exist in previous centuries. Accordingly, UV appears only to become a relevant determinant of
productivity after the first emergence of the fertility transition. Second, we proceed to investigate the
determinants of the timing of the fertility transition. Our analysis reveals that UV-R is a robust
1
Cataract is a clouding of the lens, which leads to blurred vision and ultimately to blindness. Pterygium is a
(benign) growth of the conjunctiva, which influences an affected individual’s vision if it reaches the cornea.
When the macula degenerates, the individual’s vision becomes blurred, ultimately rendering it impossible to see
fine details.
3
determinant of the onset of this transition. Further, the fertility transition is itself a strong determinant
of current cross-country income differences, consistent with the predictions of the above-mentioned
theoretical literature. Importantly, according to our estimates the link between UV-R and the timing of
the fertility transition is quantitatively large enough to reasonably account for our reduced-form
estimate of the influence of UV-R on current income per capita. Third, consistent with a pivotal role
of the fertility transition in unleashing a process of human capital accumulation, we find that UV-R is
a significant determinant of human capital investments since 1870. At the same time, UV-R looses
significance once we control for the timing of the fertility transition, which itself is strongly linked to
human capital investments in keeping with theoretical priors.
In sum, our empirical analysis provides evidence that initial differences in work life expectancy in
skilled occupations, attributable to differences in historical incidence of eye disease, have contributed
to a differential timing of the fertility transition. As a consequence of comparative differences in the
onset of the fertility transition, global differences in prosperity have emerged, mediated by differences
in human capital accumulation.
We trust that historical eye disease is neither the full nor the most important explanation of the
differential timing of the fertility transition and thus ultimately of contemporary income differences.
However, disease ecology with respect to eye disease—i.e., UV radiation—provides us with a
plausible source of exogenous variation in the inherent return to skill formation. As a consequence, it
enables us to study the relevance of a key mechanism from which the “Great divergence” is believed
to have originated. This key mechanism, linking contemporary income differences to the differential
timing of the fertility transition and accompanying variation in human capital investments, appears to
be strongly borne out in the data.
An obvious concern is that there might be alternative interpretations of the empirical link between
UV-R and contemporary economic development, which cannot be ruled out a priori. First, one may
worry that UV-R captures another affliction: skin cancer. If the incidence of skin cancer is higher in
regions more exposed to UV-R, our reduced form estimate might be convoluting an impact from
mortality. More broadly, it seems possible that UV-R may pick up the impact of other climate-related
diseases. That is, perhaps our UV-R estimate is capturing the influence from a larger set of diseases
that just happens to be pervasive in regions highly exposed to UV-R. Second, one may worry that
UV-R is spuriously correlated with relatively time invariant determinants of productivity of a nonclimatic nature, such as institutions and/or cultural values and norms.
In addressing the first concern, we document that the UV-R/income gradient is robust to controlling
for skin cancer as well as a series of tropically clustered diseases such as malaria and hookworm.
4
Consequently, it seems unlikely that the correlation between UV-R and economic development is
attributable to a confounding influence from skin cancer or tropically clustered diseases. In order to
address the second concern, we demonstrate that the cross-country UV-R/income gradient also
survives the inclusion of direct indicators of institutions as well as theoretically plausible indirect
determinants of institutions and culture. In a more decisive set of tests, we move beyond the use of the
country as the unit of analysis. Instead we employ a global data set on economic activity for all
terrestrial grid cells from the Yale G-Econ project (see Nordhaus et al., 2006). This dataset enables us
to examine the association between UV-R and economic activity while pruning the data for country
fixed effects. We expect country fixed effects to pick up the influence from political institutions and
country-specific cultural traits. Still, a case can be made that cultural traits may vary within countries.
In an effort to deal with this concern, we invoke language group fixed effects in the sub-national
analysis. Insofar as language is a sensible indicator for a wider set of cultural attributes, this should
allow us to control for within-country variation in the said attributes. In this setting, where we solely
rely on within country variation (or within language group variation) we continue to find that UV-R is
detrimental to economic development. This result carries over when we, following Henderson et al.
(2011), employ satellite data on earthlights at night as an alternative proxy for regional per capita
income.
Finally, in an effort to gauge the relevance of the hypothesized mechanism behind the UV-R/income
gradient at the sub-regional level, we examine the link between UV-R and income per capita within
two countries: the US and China. As demonstrated in Hansen et al. (2014), the differential timing of
the fertility transition within the US was highly influenced by schooling. Hence, in theory the
mechanism under scrutiny should be operative, for which reason we would expect to see empirically
that UV-R is a determinant of economic activity within the US. In contrast, the fertility transition in
China was highly influenced by government policy: the “later, longer, fewer” policy of the early
1970s and the one-child policy that was enacted in the late 1970s (Bongaarts and Greenhalgh, 1985).
Hence, geographical determinants of the return to human capital should be of less consequence to
comparative development in China. Consistent with these priors, our analysis demonstrates an impact
from UV-R within the US, but not within China.
These additional checks show that the link between UV-R and income can neither be attributed to
skin cancer nor to other diseases such as malaria and hookworm, which previous studies have shown
to exert an influence on growth.2 Moreover, the UV-R/income nexus does not appear to be driven by a
confounding influence from other key geographical determinants of prosperity, institutions and
culture. As a result, we are led to the conclusion that the most plausible explanation for the influence
2
See e.g. Gallup and Sachs (2001) on malaria; Bleakley (2007) on hookworm.
5
of UV-R on prosperity is that historical incidence of eye disease has had an important effect on the
contemporary global distribution of income per capita. Our cross-country analysis makes plausible
that this influence is largely attributable to the differential onset of the fertility transition and
subsequent human capital investments. The sub-regional analysis of the US and China provides
further corroborating evidence of this mechanism linking UV-R to economic activity.
The paper is related to several strands of literature. First, it is directly related to the literature on
growth in the very long run, as laid out in the opening paragraphs of the paper. On the empirical front
a few papers has examined the fundamental mechanism highlighted above, which links the return to
education, the fertility transition, and contemporary development. Using regional data from Prussian
countries as of the 19th century, Becker et al. (2010) demonstrate that a perceived high return to
education worked to lower fertility levels. They obtain identification by employing the geographic
distance to Wittenberg as an instrument for education, based on the theory that the Protestant
Reformation—which had the city of Wittenberg as epicenter—led to a greater perceived return on
education, for religious reasons. The study also demonstrates that lower fertility rates were conducive
to educational investments. Identification of the impact of human capital on fertility relies on the
study by Becker and Woessman (2009), which demonstrates that differences in the extent of human
capital investment between Protestant and Catholic Prussian countries importantly explain regional
income disparities in the late 19th century in the region.3 Taken together, Becker et al. (2010) and
Becker and Woessman (2009) point to a conclusion that is similar to the one we reach. Namely that a
higher perceived return to human capital helped lower fertility rates and advance educational
investments, thereby leading to income divergence between “high return” and “low return” areas. At
the cross-country level Murtin (2013), employing GMM estimation for the purpose of identification,
finds that education helped instigate the demographic transition. Hansen et al. (2014) uncover similar
results using cross-state data for the US, as mentioned above. The present research complements the
findings of Murtin (2013) in terms of the influence of human capital on the timing of the fertility
transition. In contrast to the present study, however, Murtin (2013) does not examine the role of the
fertility transition in enabling human capital accumulation and in generating global income
disparities.4
Second, the present paper is also related to a recent body of empirical literature that argues that deep
historical factors have helped shape the contemporary income distribution (e.g., Olsson and Hibbs,
3
The study by Becker et al. (2010) seeks to identify the impact of fertility on educational investments by
employing the male-female sex ratio, capturing marriage market tightness, as an instrument for average fertility
rates.
4
See also Glaeser et al. (2004) on the role of human capital investments for long-run development in a crosscountry setting, and Gennaioli et al. (2013) for regional evidence. See Herzer et al. (2012) and Angeles (2010)
on the determinants of the fertility transition using cross-country data.
6
2005; Nunn, 2008; Feyrer and Sacerdote, 2009; Ashraf and Galor, 2012; Michalopulos and
Papaioannou, 2013; for a survey, see Nunn, 2013). Whereas previous research has examined the
impact of for instance the exodus from Africa (Ashraf and Galor), the extent of slave export in Africa
(Nunn), the timing of the Neolithic (Olsson and Hibbs), colonialization (Feyrer and Sacerdote), and
indigenous ethnic institutions (Michalopulos and Papaioannou), the present paper examines the
impact from historical incidence of eye disease.
Finally, our paper contributes to the macro literature, which examines the impact of mortality and
morbidity on development (e.g., Gallup and Sachs, 2001; Young, 2005; Acemoglu and Johnson,
2007; Weil, 2007; Ashraf, Lester and Weil, 2008; Lorentzen, McMillan and Wacziarg, 2008; Aghion,
Howitt and Murtin, 2010; Cervellati and Sunde, 2011; Kalemli-Ozcan and Turan, 2011). While
previous contributions have measured health by variables such as life expectancy, height, and HIV
infection rates, we focus on eye disease. Overall, our empirical work suggests that morbidity holds
strong explanatory power vis-à-vis contemporary income differences.5
We proceed as follows. In the next section we discuss why eye disease may influence long run
productivity; Section 3 discusses our empirical strategy; Section 4 contains our main empirical
analysis, whereas Section 5 examines alternative interpretations of the link between UV-R and
income such as auxiliary diseases, institutions, and culture. Finally, Section 6 concludes.
2
Why eye disease should matter to labor productivity
2.1 The qualitative significance of eye disease for the process of development
As observed in the Introduction, the present study focuses on forms of eye disease that are influenced
by UV-R. Of these eye diseases, cataract deserves special attention because it is the single most
important cause of blindness worldwide. The global significance of cataract is illustrated by the fact
that the World Health Organization specifically targeted it in the context of its “Vision 2020 – the
Right to Sight” campaign, which was launched in 1999 and aims to eliminate preventable blindness
by the year 2020.
Cataract is opacity of the lens of the eye, which leads to impaired vision and ultimately to blindness.
The condition is progressive and may (after its time of onset) proceed slowly, over a time horizon of
5
At the same time, our results also imply that contemporaneous improvements in (this kind of) morbidity may
not have large effects on growth going forward, since the impact we observe today is likely the accumulated
outcome of past events. In this sense our results strikes something of a middle ground between previous
contributions that suggest the impact from health on productivity is modest or negative, at least in the short to
medium run (see Young, 2005; Acemoglu and Johnson, 2007; Ashraf, Lester and Weil, 2008), and contributions
that uncover a strong positive impact on growth (e.g., Gallup and Sachs, 2001; Lorentzen, McMillan and
Wacziarg, 2008).
7
years, or rapidly, in a matter of months. In terms of risks of contracting cataract, age is the strongest
factor because environmentally induced damage accumulates over time. In the end, most people
experience cataract if they live long enough. Yet the timing of its onset varies considerably across
individuals and countries. We return to this fact in a detailed manner in Section 2.2.
The only treatment of cataract is eye surgery, which historically was a rather precarious proposition.6
During the 20th century the surgical techniques improved enormously, but the procedure is still the
work of a specialist. Unfortunately, such specialists are scarce in many developing countries. In
Africa, for instance, the relative number of ophthalmologists is minuscule: fractions as low as
1:1,000,000 inhabitants have been reported (Foster, 1991). Inevitably, this extreme supply constraint
limits the possibility of cataract treatment in many poor places even today.7 Much like cataract,
surgery is needed for the treatment of pterygium; macular degeneration, by contrast, can only be
prevented.
Accordingly, corrective eye surgery is unlikely to have played an important role historically, and even
during the 20th century access to adequate treatment is likely to have been severely limited in many
places around the world. It is therefore conceivable that eye disease in general and cataract incidence
in particular influenced comparative development. More concretely, one may envision at least two
separate channels through which eye disease may have influenced living standards: a static- and a
dynamic channel.
The static channel derives from reduced labor market effort by working-age individuals afflicted by
eye disease. The static channel is unlikely to be quantitatively important however. A sense of
magnitudes can be obtained by assuming that the fraction of the population suffering from cataract
contributes nothing to prosperity; this is obviously an exaggeration designed only to provide an upper
bound for the impact of cataract via this participation channel. Hence if cataract was eliminated GDP
per capita would rise with the share of afflicted working-age individuals in the population. Using data
deriving from a study from India discussed below this would amount to an overall increase in income
per capita by 4.3%. If we were (able) to include information about the incidence of additional (UV-R
related) eye diseases in the calculation (e.g., pterygium and macular degeneration) this number could
undoubtedly be increased, yet we would expect it to remain small in magnitude compared to current
global income per capita differences.
6
A preferred method for dealing with cataract historically involved the displacement of the lens using a needle;
a method called couching. It is noteworthy that this procedure has been practiced at least since 1000 B.C. (e.g.,
Corser, 2000), testifying to the fact that in spite of shorter life spans cataract was a well-known condition
requiring treatment even in antiquity.
7
Another problem is that the quality of the treatment (if available) is often low in poor countries. For example,
evaluating cataract surgery in urban India, 50% of the outcomes were classified by international experts as
“poor” or “very poor”, reflecting only limited post-operation vision (Dandona et al., 1999).
8
A potentially viable dynamic channel derives from the literature that models the transition to the
modern economic growth regime (e.g., Galor and Weil, 2000; Galor and Moav, 2002; Lucas, 2002;
Hansen and Prescott, 2002). The aim of this literature is to expound the forces that triggered the
abrupt change in income per capita growth, which first occurred in Western Europe sometime late in
the 18th century. The assertion of this literature is that currently observed differences in income per
capita is (largely) a consequence of the differential timing of the take-off, rather than being the result
of post take-off sources of influence as elucidated by the neoclassical growth model. Moreover,
virtually all contributions to this literature view the fertility transition as a key maker for the onset of
sustained growth. The theoretical reasoning motivating a critical impact from the fertility transition on
the growth acceleration is easy to grasp. Prior to the fertility transition increases in income stimulated
fertility and thus translated into higher population levels, which, as a result of diminishing returns,
kept income per capita levels from rising persistently. In other words, Malthusian forces lead to
stagnating living standards (e.g., Ashraf and Galor, 2011; Clark, 2007). After the fertility transition,
however, rising income is associated with declining fertility. The reversal of the income/fertility
nexus, which is the outcome of the fertility transition, has several critically important effects on the
growth process (e.g., Galor, 2011). The fertility transition serves to reduce capital dilution, and thus to
increase resources per capita, which stimulates labor productivity. Moreover, it enables intensified
child investments in the form of human capital accumulation. By stimulating productivity, higher
human-capital investments pave the way for a virtuous circle involving rising per capita income,
further reductions in fertility, and greater child investments. In addition, the fertility transition
temporarily increases the relative size of the working age population, thereby stimulating growth in
income per capita.
The leading theory for the onset of the fertility transition is that a gradually rising return on human
capital accumulation eventually triggered a substitution of child quantity (family size) for child
quality (capital investments per child) at the household level (Galor, 2011, Ch. 4). According to this
theory, the inherent return to skill accumulation is key to an understanding of comparative differences
in the timing of the onset of the fertility decline, and thus to an understanding of cross-country income
inequality (Galor, 2010). This is where eye disease may have played a role. By lowering the expected
work life over which skill investments can be recuperated, an early onset of, say, debilitating cataract
will work to lower the return on human capital accumulation. As a consequence of a lower return to
skills, high incidence of eye disease may serve to delay the onset of the fertility transition. If the
fertility transition is an important driving force behind the take-off, an income gap emerges between
countries with respectively high and low incidence of eye disease. A century later, such a divergence
(deriving from a differential timing of the fertility transition and thus the take-off to sustained growth)
ought to be detectable in the data. A formal model, which predicts that variations in health status may
9
have led to a differential timing of the take-off, along the lines of the argument sketched above, is
developed in Hazan and Zoabi (2006).
To see the argument more clearly, and with an eye to the empirical analysis to come, consider the
following crude representation of the long-run growth process. For a county i at time t > si, the level
of (log) GDP per worker, yit, can be written as yit = yi0 + (t-si)·g, where si is the country specific timing
(year) of a take-off in growth in labor productivity, or the timing of the fertility transition as argued
above.8 The implicit assumption is that between time zero and si the economy stagnates; yi0 can be
viewed as the subsistence level of income, or, alternatively, as the equilibrium level of income per
capita prior to the take-off. For all t > si the economy grows at the rate g > 0; i.e., we assume that all
countries that have taken off share the same value of g. Suppose that the timing of the take-off is
explained by some underlying characteristic, xi, and by other factors, si , assumed to be uncorrelated
with xi. That is, si = si + τ ⋅ xi , where τ is a parameter capturing the impact of x on s. In the argument
above, xi would be historical incidence of eye disease (cataract) or key environmental determinants
thereof, UV radiation in particular, whereas si would capture other determinants of the timing of the
fertility transition that are unrelated to eye disease.
Now consider running a cross-country regression of yit on xi, where yit is governed by the two
equations above. Specifically, we estimate yit = a + b⋅ xi + ε it . Assuming that yi0 is uncorrelated with

xi, the OLS estimate, b , for the impact of x on y is given by:
(
)
 σ 2
 E yit xi
N
t
x,t
bt =
= −τ g
2
N σ x2
σx
 t , a subset of N, is the number of countries that have managed the take-off as of time t, σ 2
where N
x,t
 t countries, and σ 2 is the variance of x across all N
is the variance of the characteristic x across the N
x
countries. The intuition for this result is straightforward. Since we assume that xi is uncorrelated with
yi0 , the OLS coefficient must be zero if no countries have managed the take-off; as seen above,
 t = 0 produces b = 0 . However, as countries start taking off in a systematic way related to x , a link
N
i
between yit and xi emerges. In the long run, assuming all countries have experienced their take-off,
8
This mechanical way of capturing the impact of a differential timing of the take-off on 21st century income
outcomes is inspired by Lucas (2000).
10

b = −τ ⋅ g ; a unit change in x instigates τ years of delayed take-off, which has g percent as a yearly
penalty in terms of labor productivity.9
The main point is that even if characteristic xi has a very limited (static) impact on the level of the
growth path, measured by yi0 (in the example above this effect is nil), we may nevertheless find a
substantial impact on yit due to the influence of xi on the timing of the take-off. That is, even if the
static (participation) effect from eye disease is limited, a substantial impact on income per capita can
emerge if the incidence of eye disease influenced the timing of the take-off.
2.2 The quantitative significance of cataract for expected work life
The idea laid out above is that UV radiation, by affecting expected work life as a skilled worker via
eye disease incidence, influenced the timing of the fertility transition and thereby comparative
development. Since cataract is the most important eye disease, which is epidemiologically related to
UV radiation, the key question becomes whether this particular eye disease is in fact quantitatively
important.10
In order to answer this question we need to examine the extent to which cataract may have worked to
reduce expected work life in high UV regions compared to low UV regions. In order to perform such
a calculation, we need age-specific incidence rates for different geographical areas. Alongside a few
assumptions about how the age-specific incidence rates influences the “exit rate” from skilled
occupations, we can work out expected work life, with and without cataract, across different
geographical areas.
In terms of the required input data we rely on two ophthalmological surveys of visual impairment
carried out in two very different geographical locations: the Indian state of Punjab and Rotterdam in
the Netherlands. Punjab is located in a high UV region, whereas Rotterdam is located in a low UV
region.11 As a result, we would expect to see significant differences in cataract incidence across these
two locations. This expectation is supported by the data depicted in Figure 1. The figure, which plots
measures of age-specific prevalence rates of cataract in the two areas, reveals a marked difference: In
Punjab, essentially the same prevalence rate is found in the age group 40-49 as what can be detected
in Rotterdam among individuals in the age group 70-79. We will use these data below to calculate the
9
For simplicity, we are ignoring convergence, which may nonetheless be important post take-off. However, as
long as income convergence is not complete, the timing of the take-off will matter to observed cross-country
income differences.
10
In the Supplementary Appendix (Figure B1) we have included vision simulations of severe cataract that can
be compared to normal vision. As should be clear, cataract is a highly debilitating condition that seriously
impairs the vision.
11
In the context of our satellite data on UV-R described below, we find that Punjab and Rotterdam are located
respectively at the 60th and 32nd percentile in the global (grid based) distribution of UV-R.
11
difference in expected work life (as a skilled worker) between Punjab and Rotterdam. But before we
turn our attention to these calculations, we need to address an important question: How well do the
said prevalence rates speak to historical eye disease at these two locations?
Figure 1: Age-specific cataract in Punjab (solid line) and age-specific visual impairment in
Rotterdam (dotted line). Notes: Punjab data are from Chatterjee et al. (1982) and Rotterdam data are
from Klaver et al. (1998). The y-axis gives cataract prevalence in percentages, whereas the x-axis
provides the age categories.
The ophthalmological fieldwork in Punjab, which is reported in Chatterjee et al. (1982), was carried
out in 1976/77. In the 1970s India was a very poor country with the majority of the population earning
their livelihood in agriculture, which in effect exposed them to non-trivial amounts of UV-R.12
Consequently, the recorded prevalence rates are probably a sensible approximation to a generic preindustrialized context in the area.
Rotterdam is likely a different story. This survey, which is reported in Klaver et al. (1998), was
carried out in 1990/93. Ideally, we want to have a survey for Rotterdam (or a similar geographic
location) prior to industrialization for a better comparison with Punjab. The reason is that many
contemporary inhabitants of Rotterdam are working indoors, implying less exposure to UV-R and
therefore potentially lower age-specific cataract than what was the case historically. At the same time,
cataract prevalence is also affected by life style factors such as smoking, which almost surely work to
increase cataract prevalence in 1990/93 compared to the pre-industrial level of prevalence. In any
event, a pre-industrial survey is not available, so it is frankly unclear if the prevalence rates reported
in the Rotterdam survey exceed, or fall short of, the pre-industrial counterparts.
In order to deal with the problem of missing pre-industrial prevalence data in a conservative way, we
deliberately make assumptions that bias our calculations (reported below) against finding a major
difference in expected work life between Punjab and Rotterdam. In particular, we exploit two
12
In the 1970s the employment rate in agriculture in India hovered around 80% (e.g., Bhalla, 1989).
12
circumstances in connection with the survey designs: (i) the Rotterdam survey actually considers a
cluster of eye diseases, and not just cataract as in the Punjab survey; (ii) the Rotterdam survey focuses
on persons of 55 years or older, and not individuals aged 30 or older as in the Punjab survey. In the
calculations below, however, we assume that the Rotterdam prevalence rates only refer to cataract. In
so doing we are artificially inflating cataract prevalence in Rotterdam, thereby lowering the likely gap
in expected work life between Rotterdam and Punjab due to cataract. Moreover, we assume that
individuals in Rotterdam belonging to the age group 30-54 experience the same cataract prevalence as
the age group 55-59.13 In practice, the prevalence rates will be much smaller. The potential positive
bias in our calculations of the expected work life differential between Punjab and Rotterdam, due to
the late timing of the Rotterdam survey, and the negative bias that we impose, may well balance out.
To make further progress we need to make a few additional assumptions. First, we assume that
individuals use cross-sectional prevalence rates as an indication of the risk of cataract. This is
reasonable if individuals look to the experience of older family members when forming their
expectations. A fully rational individual, however, might employ cohort-specific prevalence in the
same context, but such data is not available to us. In any case, it is quite common in the literature to
employ cross-age-group information to gauge life cycle developments; see e.g. Hall and Jones (2007).
Consequently, we follow this practice. Second, we assume that individuals leave the skilled labor
force upon contracting cataract according to the prevalence rates recorded in the surveys. In order to
understand why this assumption is justifiable a few additional remarks on the surveys are required. A
critically important aspect of both surveys is that they speak to the prevalence of severe cataract. Both
studies involved careful eye examinations of the subjects, allowing the ophthalmologists to record
visual acuity. As a result, we know that the prevalence rates in Punjab and Rotterdam speak to
individuals with corrected visual acuity of 20/60 or worse. In practice, 20/60 visual acuity means that
the individual is only able to see the first few lines on the familiar Snellen chart; it therefore implies a
substantially reduced vision, which inevitably makes it very hard, or even impossible, for an
individual to perform tasks in skilled occupations, as this usually requires an ability to read.14 From
this perspective it seems reasonable to treat the prevalence rates as “exit rates” from skilled
occupations.15 Third, we assume that individuals expect to work until they decease. As a result, we do
not distinguish between the adult population and the labor force. With the historical period we are
13
In all calculations the prevalence rate in age group 20-29 is nil.
Ophthalmologists distinguish between “corrected” and “uncorrected” visual acuity. In the former case subjects
are allowed to wear glasses (if available). Formally, a visual acuity of 20/60 means that at a 20 feet distance to
the familiar test chart for eyesight, the individual can read letters that a person with 20/20 vision (the reference
standard) can read at a 60 feet distance.
15
According to the WHO “low vision” is defined as visual acuity between 20/60 and 20/200. The surveys
discussed in the text therefore fall in this category. According to the National Eye Institute, “low vision” means
that “[a]ctivities like reading, shopping, cooking, writing, and watching TV may be hard to do” (cf.
http://www.nlm.nih.gov/medlineplus/visionimpairmentandblindness.html).
14
13
focusing on in mind, we view this as a reasonable assumption. Consequently, individuals may leave
the skilled labor force for one of two reasons: either they contract severe cataract or they simply die.
With these three assumptions in place, we first calculate expected work life in Punjab and Rotterdam
with and without cataract, and then we proceed to calculate the difference in expected work life
between these two locations that is a result of cataract. Since cataract prevalence is negligible before
the age of 20, we focus on calculating expected work life at the age of 20. We proceed by invoking
the following recursive equation, in which e(x) is remaining expected work life at age x:
e ( x ) = λ ⋅ q ( x ) + ⎡⎣1− q ( x ) ⎤⎦ ⋅ ⎡⎣1+ e ( x + 1) ⎤⎦
In the equation, λ is the share of the year a person, who exits the labor force during this particular
year, will actually end up having spent in the labor force during this year; we set λ = 0.5. q(x) is the
age specific exit probability, and 1-q(x) is therefore the probability of remaining in the labor force
from age x to age x+1. The exit probability will reflect age-specific mortality and eye disease
incidence, as explained above.
In our calculations we limit life to the age of 70. This ensures that we obtain a level of life expectancy
at age 20 that roughly coincides with the one observed in Punjab around the time of the survey. From
this assumption follows that q(70) = 1, in turn implying that e(70) = 0.5 since λ = 0.5. With e(70) thus
established, we iterate backwards with e(69) = 0.5q(69) + (1-q(69))(1+e(70)) using the age-specific
exit probabilities (i.e., the q(x)’s) until we arrive at e(20). As already noted, we stop iterating at e(20)
since cataract prevalence before the age of 20 is negligible. In order to calculate the q(x)’s in the
baseline scenario without cataract risk, we use age-specific mortality rates for Punjab as reported in
Chatterjee et al. (1982). We assume that the same age-specific mortality rates are about appropriate
for pre-industrial Rotterdam. Using these we obtain e(20) = 47, which means that a person aged 20
can expect to work (live) for 47 additional years, i.e., until the age of 67.16
To gauge the impact of severe cataract on expected work life as a skilled laborer we simply add the
probability of contracting cataract (measured as the cataract prevalence rate) to mortality risks in
order to get a modified exit probability. More specifically, q(69) for a person living in Punjab will be
2.45% (age-specific mortality rate) plus 42% (the age-specific cataract rate in the age interval 60-69),
and so on and so forth. Doing the backwards iteration gives us e(20) = 32 years with Punjab based
age-specific cataract rates and e(20) = 46 with Rotterdam based visual impairment prevalence rates. In
16
According to Saikia et al.(2012) actual life expectancy at age 20 was 69.8 in Punjab in 1979. For comparison,
life expectancy at birth was 65 years at the time of the survey (Chatterjee et al., 1982).
14
the case of Punjab, cataract risk therefore reduces an individual’s expected work life as a skilled
laborer by 15 years, whereas in the case of Rotterdam the reduction is only a single year. Hence the
difference in expected work life as a skilled laborer as of the age of 20 between these two locations is
15-1 = 14 years.17
Is 14 years a large number? One way to think about this question is to observe that the most dramatic
increases in life expectancy, and therefore in work life expectancy, associated with the international
epidemiological transition (caused, among other things, by the discovery of penicillin) were about 20
years (Ashraf et al., 2008). Hence the cross-country difference in expected work life resulting from
cataract is roughly comparable to the expansion in expected work life resulting from the discovery of
penicillin, DDT, etc. Another way to appreciate the result is via the study by Cervellati and Sunde
(2013). The authors calibrate a prototypical unified growth model, upon which they study the
consequence of introducing cross-country differences in initial work life, which is tantamount to
differences in initial life expectancy in their set up. Cervellati and Sunde find that a difference in
initial work life of only five years can generate a difference in the timing of the take-off to modern
growth (and in the fertility transition) of nearly 150 years. In this light 14 years is a huge number.
Hence on both metrics —i.e., judged either by historical mortality shocks, which are conventionally
believed to be of first order, or by outcomes from calibrated unified growth models—the influence
from cataract on comparative expected work life across pre-industrial societies is substantial. This
supports the assertion that historical differences in the incidence of debilitating eye disease may have
created important differences in the perceived return to skill formation. Consequently, it supports the
viability of the dynamic channel, discussed above, linking UV radiation, historical eye disease, and
contemporary comparative development. Whether this mechanism is borne out in the data is the
question to which we now turn.
3
Empirical Strategy
The basic specification that we take to the data is of the following form:
log(yi) = β0 + β1 log(Ei) + Zi’γ + εi
(1)
where y is labor productivity (GDP per worker) or GDP per capita, E is the historical incidence of eye
disease, and Z is a vector of additional controls.
17
Observe that since life expectancy at birth in Punjab, at the time of the survey, was 65 (Chatterjee et al.,
1982), and since (severe) cataract prevalence below the age of 20 is virtually zero, the reduction in work life in
Punjab from cataracts, evaluated at birth, would be roughly 65-52 = 13 years. In Rotterdam, by contrast,
cataracts would not affect expected work life, evaluated at birth; that is, assuming the same life expectancy at
birth as in Punjab.
15
As is well known, the level of income per capita is explained at the proximate level by the availability
of capital (physical and human) and productivity (technology and macroeconomic efficiency).
Following the literature on “fundamental determinants of productivity” we do not control for these
proximate sources of growth. Rather, we attempt to understand comparative development by
introducing variables that ultimately explain why some countries have more capital, higher
productivity, and therefore have attained higher levels of income per capita (e.g., Acemoglu, 2009,
Ch. 4). The key hypothesis of the present study is that the climate- induced historical incidence of eye
disease is one such fundamental determinant.
In measuring E we face the challenge that survey data on historical eye disease incidence is
unavailable. As a result, we employ an indirect approach in capturing eye disease incidence using data
on UV-R.18 The use of UV-R is motivated by its epidemiological impact on various eye diseases. First
and foremost, UV-R is known to influence the incidence of cataract. Theoretical mechanisms
connecting cataract with UV-R have been established; see, e.g., Dong et al. (2003) and references
cited therein. Second, randomized controlled trials with animals have confirmed the impact of UV-R
on the formation of cataract (e.g., Ayala et al. 2000). Third, epidemiological studies have
demonstrated that greater exposure to UV-R produces an earlier onset of cataract in human
populations (e.g., Hollows and Moran, 1981; Taylor et al., 1989; West et al., 1998). It seems fair to
say that a consensus has been reached on the issue.19 UV-R is also suspected of influencing the
incidence of two other eye diseases, namely pterygium and macular degeneration (e.g., Gallagher and
Lee, 2006). It should be noted, however, that there is an ongoing debate as to which extent UV-R
influences pterygium and macular degeneration. Consequently, at this point in time we cannot rule out
that UV-R may be capturing a cluster of eye diseases: cataract, pterygium and macular degeneration.
For this reason we proxy the historical incidence of eye disease, E, by employing data on UV-R
exposure.
With regards to Z we follow the literature on the fundamental determinants of productivity, which
emphasize three major underlying causes of diverging long-run comparative development. These are
institutions, culture, and geography/climate (Acemoglu, 2009, Ch. 4).
All estimations in this paper are performed by OLS, where measures of UV-R are used to control for
historical eye disease incidence. It should be clear that reverse causality is unlikely to be a concern.
18
Ultraviolet (UV) radiation is a form of electromagnetic radiation, which is found in sunlight. There are three
types of UV radiation: A, B and C. These three varieties of UV radiation are distinguishable by their
wavelength: UVA radiation has the longest wavelength (yet shorter than visible light), UVC the shortest, with
UV wavelength being in between. Of the three forms of UV radiation, UVC is considered the most harmful to
humans. This form of electromagnetic radiation is fortunately filtered out by the atmosphere, which leaves only
UVA and UV with the potential to affect life forms on Earth.
19
Surveys of the literature are found in Javitt et al. (1996) and West (2007).
16
Rather, the key issue is whether it can reasonably be argued that our UV-R variable is capturing eye
disease and not other covariates with (fundamental determinants of) living standards.20
It will become apparent when we present our data on UV-R that it features a very strong latitude
gradient; the simple correlation between our measure of UV-R and absolute latitude is -0.95. Since
latitude may capture a host of income determinants, we include it in Z. In our full specification,
identification is therefore obtained from the residual variation in UV-R that is orthogonal to absolute
latitude. Two other climate/geography traits create variation in UV-R beyond absolute latitude,
namely cloud cover and elevation. In places with more cloud cover, UV-R is lower; and at higher
altitudes, UV-R is higher. Since cloud cover and nation-specific topography do not track latitude
fully, these features provide variation in UV-R that is orthogonal to latitude.
It is worth reflecting on whether these residual sources of variation are problematic from the point of
view of isolating an effect from eye disease. Clearly, the elevation of a country above sea level may
have independent effects on long-run productivity. Diamond (1997) for instance discusses the
challenges involved in developing complex societies in mountainous regions. If high altitude regions
had a historical growth disadvantage, the ramifications may be felt to this very day. This would
naturally render the interpretation of a correlation between (residual) UV-R and current economic
development unclear. We confront this issue is several ways. First, we control for the timing of the
Neolithic revolution. If Diamond (1997) is correct this should capture the indirect economic
ramifications of elevation. Second, moving beyond the Diamond thesis, elevation may have a
contemporary direct effect on productivity via trade costs. We attempt to capture trade costs by
including distance to coast and navigable river and by adding a direct measure of average elevation.
Moreover, climatic conditions change with altitude, which we attempt to capture by controlling for
both average temperature and precipitation. Furthermore, in an effort to ensure that our estimates are
not confounding a spurious link between UV-R and climatic conditions with particular relevance to
agriculture, we also control for soil quality, the share of the country situated in the tropical climate
20
Some may object to the notion that reverse causality can be ruled out, observing that the ozone layer
influences UV-R. Thus, if economic activity impacts on the former, the latter should be regarded as endogenous
to income. But there is clearly no simple link between human activity at a particular location and the ozone layer
at the selfsame location. A good illustration of this is the ozone holes over the North Pole and the Antarctica,
which surely are not the result of local human activity. Accordingly, while UV-R may be endogenous at the
global level, local UV-R is not endogenous to local economic activity. As a result, reverse causality is not a
concern in our analysis.
17
zone, and the average number of frost days per year.21 In the baseline control set we also include the
size of the country, and we add continental fixed effects so as to capture unobserved heterogeneity.22
When we control for this set of variables, the variation we exploit should essentially be that related to
variation in cloud cover. Clouds obviously have other roles to play aside from shielding humans from
UV-R; they may, for instance, influence agricultural productivity via precipitation and perhaps even
temperature. However, as we do control for precipitation, temperature, and even agricultural
productivity, this particular basis for concern should be eliminated.
Despite the extensive list of controls, there are at two lingering concerns. First, the UV-R variable
may be spuriously correlated with other (non UV-R related) diseases that just happen to be more
pervasive in high UV-R areas. Second, UV-R may be correlated with institutions or even cultural
values, which in complex ways derive from climatic conditions. In order to deal with the first concern
we examine (in Section 5) the robustness of our results to the inclusion of a range of tropically
clustered diseases that are epidemiologically independent of UV-R. We also examine an affliction,
which is epidemiologically related to UV radiation, namely skin cancer. In order to gauge the
relevance of the second concern in a cross-country setting, we check the resilience of the detected
UV/income gradient to additional indirect and direct controls for culture and institutions. Finally, we
move beyond the use of cross-country data. More specifically, we re-examine the link between UV-R
and income (in Section 5) by employing data on economic activity for all terrestrial grid cells. In this
setting we can control for country fixed effects, which should partial out the potentially confounding
influence from institutions and culture. In addition, we are also able to control for comparative
differences in cultural values at the local (sub-national) level by invoking language fixed effects.
4
Empirical Analysis
The cross-country analysis falls in three parts: Section 4.1 presents our data, Section 4.2 contains our
main results, while Section 4.3 examines the viability of the take-off hypothesis as an interpretation of
our results from Section 4.2.
4.1 Data
Our dependent variables in this section are GDP per worker and per capita (PPP$) in 2004; current
(2004) cataract incidence; and the timing (year) of the fertility decline. Most of these data are
21
Ashraf and Galor (2011) use the same soil quality variable to control for agricultural productivity; Gallup and
Sachs (2000) demonstrate a detrimental impact from tropical climate on agricultural yields; whereas Masters
and McMillan (2001) employ “frost” in a similar vein.
22
There are many reasons why “scale” could matter to economic development, motivating the inclusion of
country area in the set of geographic controls: Olsson and Hansson (2011) develop a theory linking institutional
development to country size; country area is also known to influence the intensity of trade and travel (e.g.,
Frankel and Romer, 1999; Andersen and Dalgaard, 2011).
18
commonly used in the literature and therefore require little further presentation; sources and brief
descriptions are found in Appendix 1. Still, a few remarks on cataract incidence are warranted.
Figure 2. Global distribution of the UV-R variable. Notes: See Appendix 1 for details on the index.
Our incidence-of-cataract measure for each country is the number of years lost to disability (YLD) in
2004, expressed as a ratio of per 100,000 people in the population (WHO, 2008). Formally,
YLD = I ⋅ w ⋅ L , where I is new incidences per year, w is a weight measuring the severity of the
condition, and L is the average duration of the condition. The weight w is the same everywhere, and
so is L. Consequently, the cross-country variation in the variable stems from I.
Our key independent variable is UV radiation. NASA produces daily satellite-based data for
ultraviolet exposure. The UV index captures the strength of radiation at a particular location, and it is
available in the form of geographic grids and daily rasters with pixel size of 1-degree latitude x 1degree longitude. We rely on data for daily local-noon irradiances for 1990 and 2000 to produce
average yearly UV-R levels for each country. That is, in our analysis below we employ an average for
the 1990 and 2000 observation.23 Figure 2 provides a visual illustration of the UV data; the correlation
with latitude mentioned in Section 3 is visually obvious.
Further details on the data, including the controls discussed in the last section, summary statistics, and
correlations between the controls and UV-R exposure are, as noted, found in Appendices 1 and 2.
23
Though we invoke an average, the correlation between UV-R in 1990 and 2000 is above 0.99. In general, it
seems that the intensity of surface UV-R has been relatively stable on earth during the last 2 billion years
(Cockell and Horneck, 2001). Hence in a cross-section context current comparative UV-R levels are likely to
represent a good indicator for UV-R conditions a few centuries ago.
19
4.2 Main results
The results from estimating equation (1) by OLS are reported in Tables 1 and 2, where the dependent
variable is GDP per worker and GDP per capita, respectively. The first column of Table 1 (Table 2)
shows the bivariate association between (log) UV-R and (log) GDP per worker (GDP per capita): An
increase in UV-R by one percent is associated with a reduction in labor productivity by approximately
1.3 percent (Table 1) and 1.5 percent in the context of GDP per capita (Table 2).
[Tables 1 and 2 about here]
In columns 2 to 7 of the two tables we add controls sequentially and in columns 8-9 we include all of
them at once. Note that we are measuring the influence from latitude and elevation in two different
ways, which explains why we have two “full specifications” (columns 8-9 of Tables 1-2). First we
allow them to enter linearly; and, second, we control for latitude and elevation fixed effects. The
latitude fixed effects are constructed such that each 10-degree latitude bin is allowed to hold a
separate impact on GDP per worker and GDP per capita. Similarly, every 500 meters of additional
altitude is allowed an individual effect. Since UV-R is not increasing linearly in latitude and elevation
this additional check serves to prune UV-R in a demanding way from its correlation with latitude and
elevation. Regardless of which full specification we examine, however, we find that UV-R is
significant at five percent or better in all columns.24
The controls have only a modest effect on the partial correlation between UV-R and living standards,
despite the fact they themselves are relevant (see bottom of tables for F-tests), and despite the fact that
the controls are strongly correlated with UV-R. For example, when all controls are added
simultaneously in Table 1, columns 8 and 9, the UV-R elasticity is -1.50 and -1.46, depending on how
latitude and elevation are controlled for, which is close to the simple bivariate estimate of roughly 1.3.25
Despite our best efforts to control for all relevant observable factors, we obviously cannot rule out that
some omitted factor is correlated with both UV-R and income levels. To gauge just how concerned
we should be about omitted variables, we invoke the insights of Altonji et al. (2005) that selection on
observables can be used to assess the likely importance of bias arising from unobservable factors. We
24
As a further robustness check we have examined whether the results appear to be driven by any particular
continent. As shown in Table B1 in the Supplementary Appendix, the results are qualitatively unaffected by
dropping continents one at a time.
25
As demonstrated in Appendix 2, Table A.2, when all controls are included simultaneously they account for
95% (and 97% with latitude and elevation fixed effects) of the variation in UV-R. Much of the reduction in the
size of the UV-R estimate is thus plausibly attributable to the fact that UV-R is strongly correlated with e.g.
latitude, which influences economic prosperity in various independent ways. On physical grounds, the
remaining UV variation plausibly reflects variation in cloud cover, as discussed in Section 3.
20
essentially ask the following question: How much stronger must selection on unobservables be
relative to selection on observables in order to explain away the entire estimated effect from UV-R?26
Specifically, consider two regressions: one with a full control set (F), which should be representative
of all possible controls, and one with no control set (N). Denote the UV-R point estimates from the


first and second regression as β F and β N , respectively, then what we are interested in is the ratio



β F / ( β N − β F ) . The larger (numerically) this ratio, the less plausible it is that omitted variables can
explain away the entire effect of UV-R on prosperity. The intuition is simple: the smaller the
denominator, the less the UV-R estimate is affected by selection on observables, and the larger the
metric. Also, the larger the numerator, the larger is the total effect that needs to be explained away.
Consequently, using columns 1 and 9 of Table 1 we get that the ratio is =-1.46/(-1.32-(-1.46)) = 10.43. Accordingly, in order to attribute the entire OLS estimate of UV-R’s impact on GDP per
worker in column 9 of Table 1 to omitted variables, selection on unobservables (i.e., the covariance
between UV-R and unobservables) should be more than ten times greater than selection on
observables (i.e., the covariance between UV-R and observables). This seems very unlikely given the
extensive control set in column 9 of Table 1.27
In the last two columns of both Table 1 and 2 we replace UV-R with cataract, which is arguably the
most important eye disease in the cluster that should be epidemiologically related to UV-R. Consistent
with the hypothesis under examination, we also find a strong correlation between cataract incidence
and prosperity. It is worth noting that the R2 in column 8 (respectively 9) of either table is very similar
to that of column 10 (respectively 11). This suggests that UV-R and cataract are contributing roughly
in equal proportion to the overall fit of the model, consistent with UV-R chiefly affecting living
standards mainly via cataract.
Following up on the link between UV-R and eye disease, Table 3 provides the results from regressing
cataract incidence on UV-R. We observe that UV-R is indeed significantly correlated with cataract
incidence in all specifications; typically at the 1% level of confidence, though when we add all of our
auxiliary controls (collectively spanning 93% of the variation in UV-R) the significance level widens
to 5%. The results provide some assurance that the findings from Tables 1 and 2 reflect the stifling
effect on development from the historical incidence of eye disease.
[Table 3 about here]
26
We employ a specialization of the Altonji et al. (2005) approach to the linear model not assuming joint
normality, which is developed in Bellows and Miguel (2008). See also Nunn and Wantchekon (2011).
27
Using instead Table 2 we obtain a ratio of -5.88.
21
Suppose then that the point estimate for UV-R is in fact capturing the causal impact of eye disease on
economic development: Is the impact economically significant? Judging from column 9 of Table 2 we
find an elasticity of UV-R with respect to GDP per capita of -1.7. To get a sense of the economic
significance, observe that one standard deviation reduction in (log) UV-R damage (about 0.5) implies
about 0.85 log points increase in GDP per capita, which translates into an increase in the level of GDP
per capita by roughly a factor of 2.3 (= exp(0.5×1.7)), or about 130 percent. Is this a large effect? The
study by Ashraf et al. (2008) may serve as a useful benchmark for comparison. Using an augmented
Solow model the authors calibrate the long-run impact on aggregate labor productivity from a large
health improvement, corresponding to an increase in life expectancy from 40 to 60 years. The
imposed individual-level productivity impact from health improvements is anchored in micro
estimates. According to the Ashraf et al. simulations, aggregate long-run labor productivity may rise
by around 15%. In this light the estimate obtained above seems very large indeed. However, the
calibration approach of Ashraf et al. involves an economy, which has already experienced the take-off
into sustained growth. If morbidity has served to delay the onset of sustained growth, the accumulated
impact on labor productivity could well be much larger than what a calibrated Solow model suggests.
In fact, this is precisely what the simulation study by Cervellati and Sunde (2013) suggests. But how
viable is the take-off interpretation of the reduced form link between UV-R and contemporary
prosperity?
4.3 Exploring the take-off interpretation
Is there a differential impact of UV-R on GDP per capita and per worker? As a first step note that
the results from Tables 1 and 2 admit a simple check of the take-off interpretation. As explained in
Section 2, the fertility transition has three substantive effects on growth: (i) it increases resources per
capita; (ii) it stimulates human capital accumulation, and thus indirectly productivity growth via
technological change; and (iii) it leads to a temporary demographic dividend, whereby the size of the
labor force relative to population increases. Importantly, the third effect only influences GDP per
capita; it has no impact on GDP per worker. Consequently, the impact from UV-R on GDP per
worker, if the estimates reflect the take-off mechanism, must be strictly smaller than the impact from
UV-R on GDP per capita. Comparing columns 1-9 in the two tables shows that this pattern is borne
out in the data; the point estimates for UV-R are consistently larger (in absolute value) in Table 2
compared to Table 1.
Is there a time-varying correlation between UV-R and GDP per capita? As a second check we
examine the historical evolution of the UV-R/income gradient. If the take-off hypothesis is viable, and
if the direct impact of eye disease on productivity is negligible, then we should not expect to see a link
22
between UV-R and income prior to the fertility transition; only once the transition actually occurs,
1
1
and countries start to take-off, should we expect to see a clear link.28
.5
USA
ARG
CZE
DNK
NLD
NZL
HUNGHA
AUS
PER
URYGRC
CAN
IND BRA
JPN
ESP
BEL
COL
IDN
ITAFIN
ROM
CHL
PRT
NOR
FRA
0
NOR
VEN
CHE
SWE
DNK
BGR
GBR DEU
CZEPHLNLD
AUT
LKA
NZL
IND
GHA
CHN
AUS
JPN MEX
HUN
ARG
GRC
SWE
URY
CAN
PER BRA
IDN
ITA
ALB
POL
BEL
CHL
BGR
COL
ROM
-­‐1
-­‐1
POL
CHE
ESP
FIN
ALB
-­‐.5
0
FRA
PRT
SVK
e( lmad1950 | X )
SVK
GBR
USAAUT
MEX
LKA
PHL
DEU
CHN
-­‐.5
e( lmad1900 | X )
.5
VEN
-­‐.15
-­‐.1
-­‐.05
0
e( loguvpopw | X )
.05
.1
-­‐.15
coef = -­‐.70532609, (robust) se = .90533282, t = -­‐.78
-­‐.1
-­‐.05
0
e( loguvpopw | X )
.05
coef = -­‐3.268849, (robust) se = 1.4607612, t = -­‐2.24
.1
Figure 3. The partial correlation between UV-R and log GDP per capita in 1900 and 1950, respectively. Notes:
The figure depicts the (partial) correlation between UV-R and GDP per capita in 1900 (left panel) and 1950
(right panel), respectively, while controlling for the influence of a full set of controls, including latitude an
elevation fixed effects. The sample of countries is restricted to be the same in 1900 and 1950. Hence the slopes
of the two depicted regression lines correspond to the estimates reported in, respectively, Table 4, column 5 (for
1900) and Table 4, column 7 (for 1950).
Accordingly, using data on GDP per capita from Maddison (2003) we re-estimate the full
specification (corresponding to Table 2, column 9) for the years 1820, 1900, and 1950. The results are
found in Table 4, columns 4-7. A consistent pattern emerges in the sense that starting from 1820, the
size of the partial correlation rises (in absolute value) until it turns significant by 1950, where the
estimate is of the same order of magnitude as those reported in Table 2. From column 7 in Table 4 we
see that the significance of the estimate remains when we restrict the 1950-sample to countries for
which GDP per capita data were also available in 1900. Put differently, the significance of UV-R in
1950 is not simply a matter of more data being available. As seen from Figure 3, the change in the
partial correlation between UV-R and income from 1900 to 1950 is visually quite striking.29
[Table 4 about here]

See Section 2; if N ≈ 0 (i.e., no countries have taken-off), then !b ≈ 0 .
29
Some may speculate whether this table is not showing “too much”. According to Galor and Weil (2000), the
take-off was in full operation by 1900. From this perspective, it may seem puzzling that we do not detect a
significant influence from UV-R as off 1900 (perhaps already as off 1820) if UV-R influences the timing of the
take-off. This is not really a puzzle, however, for two reasons. First, the industrial revolution was initially
confined to Europe. As a result, the continental fixed effects will pick up most of the information as long as the
take-off is highly geographically concentrated. Secondly, the size of the estimate for UV-R is affected by the
number of countries taking off and by the variation in UV-R across the countries that have taken off (see Section
2). Since the forerunners in the industrial revolution were a relatively small group of countries, and because
Europe is a very small place climatically speaking, the variation in UV-R is relatively modest. Consequently, a
modest estimate is expected prior to the 1900s. But as the industrial revolution diffused, selectively, to other
continents and more countries, one would expect to see that (a) the point estimate for UV-R rises and (b) that
statistical significance eventually emerges.
28
23
Studying the impact of UV-R on early levels of prosperity is not without drawbacks however. To
begin, there is very limited data available for GDP per capita in the 19th century. Moreover, even if we
ignore the issue of data availability, the results reported in columns 4 and 5 of Table 4 do not
necessarily refute an influence from UV-R on historical levels of productivity. The reason is that
higher levels of productivity during the period prior to the demographic transition would be converted
into greater population density rather than into income gains; the Malthusian model appears
appropriate for the period prior to the fertility decline (e.g., Ashraf and Galor, 2011; Clark, 2007).
While the Malthusian forces that governed the growth process for most of human history had started
to vanish during the 19th century, one may nevertheless be concerned that income levels in 1820 is a
poor indicator of productivity. A more decisive check of the historical influence of UV-R would
therefore involve tests of its impact on population density prior to the demographic transition. If UVR influenced historical productivity levels it should be significantly correlated with population
density. However, conditional on our controls, UV-R is not correlated with population density in the
years 1 C.E., 1000 C.E., and 1500 C.E, respectively, as documented in Table 4, columns 1-3.
Overall, the results reported in Table 4 show that UV-R’s impact on current prosperity appears to be
of a relatively recent origin; the negative impact emerges during the 20th century. This supports the
hypothesis that the impact of UV-R on prosperity is largely mediated through the differential timing
of the take-off across the globe.
UV-R and the timing of the fertility transition In our third set of checks we begin by asking if UV-R
is a predictor of the timing of the fertility transition, as suggested by the theory. Contingent on an
affirmative answer we can proceed to a more demanding consistency check by asking if the estimated
delay in the fertility transition, and thus in the take-off to sustained growth, can account for the above
reduced form estimates. To see how this check works, note that if we assume that countries, post
transition, grow at a rate between 2 and 3 percent per year on average, and that they stagnate prior to
the transition, then the “required” delay from one standard deviation increase in UV-R (0.5 log points)
is τ = log(0.5⋅1.4) / g , or between about 23 (g = 0.03) and 35 years (g = 0.02). Hence, to account for
the results in Table 1, one standard deviation increase in UV-R should delay the fertility transition by
2 to 4 decades. A final check consists in inquiring if the influence from historical eye disease, as
captured by UV-R, ceases to be a significant determinant of current income once the timing of the
fertility transition is taken into account.
24
To limit the risk that omitted variable bias influences our estimates, Table 5 (columns 1-9) mimics the
control strategy from Tables 1 and 2, and it reports the results of estimating the link between UV-R
and the date of the fertility decline.
[Table 5 about here]
The general message from the table is that countries exposed to more UV-R have experienced the
fertility decline at a later date. In column 1 we note that UV-R can account for around 60% of the
variation in the date of fertility decline; when all our controls are added simultaneously we can
account for about 80% of the global variation in the timing of the fertility decline.
UV-R is always significant and carries the expect sign, with one exception: in column 9 the estimate
appears to loose significance. While UV-R is significant in the full specification where latitude and
elevation enters (log) linearly as controls (column 8), the same is not true when we introduce latitude
and elevation fixed effects (column 9). The most obvious reason why UV-R turns insignificant is
variance inflation due to multicollinearity. The variance inflation factor for UV-R in column 9 is
32.44 (not reported in the table); i.e., multicollinearity in the specification is increasing the variance
on the UV-R estimate by a factor of 32 compared with the counterfactual of zero multicollinearity.
The simplest way to check this explanation is to re-run the regression without the inclusion of controls
for latitude and elevation. By omitting latitude and elevation, we reduce the extent of variance
inflation due to multicollinearity. At the same time, nothing is lost in terms of explanatory power by
dropping latitude and elevation, as they are jointly insignificant in both columns 8 and 9; the relevant
F-tests are reported at the bottom of Table 5. Column 10, which reports the regression result, reveals a
strongly significant estimate for UV-R. Undoubtedly, this is explained by a dramatic fall in the
variance inflation factor on the UV-R estimate; it drops from 32 to 13 (not reported in the table).
Since a lowering of variance inflation—while keeping the overall explanatory power of the model
constant (due to the insignificance of latitude and elevation)— markedly increases the significance of
UV-R, it seems safe to conclude that multicollinearity is responsible for the one-off insignificance of
UV-R in column 9.
Turning to economic significance, UV-R does seem to have a substantial impact on the fertility
decline. Consider column 10 of Table 5: Taken at face value, the estimate implies that an increase in
UV-R of one percent delays the fertility decline by roughly 40 years. Alternatively, one standard
deviation increase in (log) UV-R (approximately 0.5 log points) delays the transition by roughly 20
years, which is in the ballpark of the delay needed to account for our reduced form results in Table 1
(i.e., 2-4 decades). Moreover, as revealed by columns 11-14 of Table 5, the fertility decline is itself
strongly and negatively correlated with current GDP per worker and GDP per capita; the point
25
estimates suggest that each additional year of delayed fertility transition has a 2% cost in terms of
forgone income per capita, with a standard error of about 0.5.30
The final issue is whether historical eye disease adds explanatory power (with respect to
contemporary income) if we control for the timing of the fertility decline. As shown in the
Supplementary Appendix (Table B2), the influence from historical eye disease, as explained by UVR, is greatly diminished once the timing of the fertility transition is controlled for; indeed, it generally
turns statistically insignificant. At the same time, the fertility transition remains a strong predictor of
GDP per capita as well as GDP per worker. Overall, these results support the hypothesis that UV-R
exerts an impact on contemporary income via the timing of the fertility transition.
UV-R and post take-off skill accumulation and fertility The manner in which the timing of the
fertility transition is thought to matter for labor productivity is by propelling human capital
accumulation and also by increasing per worker resources via diminished capital dilution. A final set
of consistency checks of the take-off account therefore consists in examining (i) whether UV-R
explains post take-off human-capital accumulation and fertility, and (ii) whether the influence of UVR on human capital and fertility disappears once the timing of the fertility transition is controlled for.
On the human capital front we obtained data on years of schooling in the population from Murtin and
Morrison (2009); data covers the period from 1870-2005. With this data in hand we ask whether UVR holds predictive power vis-à-vis growth in years of schooling 1870-2005, which is a meaningful
indicator of (average) human capital investments over this period. Since the fertility transition took
place after 1870 in almost all countries around the world (Reher, 2004), we should expect to see UVR being negatively correlated with average human capital investments, due to its influence on the
timing of the fertility transition.
Our testing strategy mimics the approach taken above. That is, we examine the robustness of the UVR/human-capital investment link by introducing our extensive set of geographical confounders
sequentially as well as simultaneously. In addition, we also allow initial levels of human capital to
enter the set of controls since the process of human capital accumulation is subject to mean reversion
over the period.
In the interest of brevity we have relegated the regression results to the Supplementary Appendix. The
main result is that UV-R indeed predicts human capital investments from 1870-2005, cf. Table B.3.
30
Dalgaard and Strulik (2013) obtain a roughly similar estimate. However, their controls follow the structure of
the Solow model; i.e., they are not motivated by the literature on fundamental determinants, as is the case in the
present analysis. But the fact that this result is robust to different empirical strategies is worth noting.
26
Furthermore, the influence from UV-R diminishes once the timing of the fertility transition is
introduced as an additional control, cf. Table B5. As expected, the timing of the fertility transition is
strongly and negatively correlated with human capital investments; countries that underwent the
transition later have invested less in human capital between 1870 and today.
On the fertility side we explore the link between UV-R and average fertility rates, 1960-2010; data on
fertility rates are taken from World Development Indicators. By delaying the onset of the fertility
transition, UV-R should in theory be positively correlated with fertility rates during the second part of
the 20th century, conditional on our controls. And this is indeed what we find, as seen from Table B4.
Moreover, once we add the timing of the fertility transition to the set of controls UV-R looses
significance, as expected cf. Table B6.
It is worth observing that UV-R occasionally stays significant in explaining human capital and
fertility, even though the timing of the fertility transition is included in the set of controls. This can be
seen as an indication that UV-R, by lowering the return to skill accumulation, also works to lower the
post-transition steady state level of human capital towards which individual countries are converging.
One consequence of such an impact would be slower growth in human capital for the standard
convergence reasons; another consequence would be higher fertility rates. This would also seem to fit
with two other observations: First, the timing of the fertility transition can largely, but not fully,
account for the reduced form estimate of UV-R. Second, cataract is occasionally significant in
explaining GDP per capita, once we control for the timing of the fertility transition, cf. Table B2.
Consequently, the occasional significance of UV-R in explaining human capital investments and
fertility levels—conditional on the timing of the fertility transition—may indicate a modest impact of
UV-R on post fertility-transition economic growth.
Caveats notwithstanding, the results discussed in this section provide further support in favor of the
take-off account. High levels of UV-R have served to delay the onset of the fertility transition, thereby
stifling human capital investments since 1870 as well as supporting high fertility rates during the 20th
century.
5 Threats to Identification
This section falls in two parts. In Section 5.1 we discuss the potential problem that UV-R
epidemiologically affects skin cancer and that UV-R, by exhibiting a strong climate gradient (cf.
Figure 2), may be spuriously correlated with tropically clustered diseases. Subsequently, in Section
5.2, we address the problem that UV-R could be spuriously correlated with institutions and/or culture,
which are two fundamental determinants of productivity.
27
5.1 Skin Cancer and Tropically Clustered Diseases
Skin cancer is caused by sun exposure: overexposure to UV-R, to be more precise. At the same time
UV-R plays a benign role by being the human body’s main source of vitamin D; a key vitamin which
influences the immune system, and thus ultimately longevity. Accordingly, UV-R could (through
either mechanism) influence mortality and thereby potentially labor productivity. However, for
evolutionary reasons UV-R is unlikely to be a cross-country determinant of longevity through these
mechanisms. Over millennia evolutionary pressures have changed the human skin pigmentation so
that a balance has been struck between the beneficial and harmful effects of UV-R on longevity.
Consequently, in “high UV regions” skin color turned darker, while in “low UV regions” it became
lighter.31 Obviously, this does not mean that sun exposure is inconsequential for skin cancer:
excessive UV-R exposure is indisputably a major cause of malignant melanoma.32 But what it does
mean is that UV-R is unlikely to causally affect longevity in a cross-country setting via its effects on
vitamin D supply and skin cancer, since evolution has traded these two factors off against each other
during the selection process determining local skin color.33 Nevertheless, it remains an empirical
question whether skin cancer in practice diminishes the influence of UV-R on contemporary levels of
GDP per worker or capita. As demonstrated in Table 6 (with latitude and elevation fixed effects) and
Supplementary Appendix Table B7 (with log linear latitude and elevation), adding skin cancer
prevalence to the full set of controls has no implications for the point estimate associated with UVR.34 The identification of UV-R with eye disease is therefore unlikely to be jeopardized by skin cancer
and vitamin D absorption.
[Table 6 about here]
A somewhat greater concern is a potential mapping between our UV-R variable and other diseases
with higher incidence in tropical climate zones where UV radiation is most intense. It could be the
case that UV-R is spuriously correlated with other diseases that in turn exert an impact on
productivity. In particular, previous research has highlighted a set of tropically clustered diseases,
which might influence growth. These are malaria, hookworm, and HIV/AIDS.35 To this list we further
31
See Diamond (2005) for a clear exposition of these points and references to the relevant literature.
Malignant melanoma is by far the most dangerous type of skin cancer, but it is also least common. There are
two other types of skin cancer: basal cell cancer and squamous cell cancer. Basal cell cancer, the most common
type of skin cancer, almost never spreads; squamous cell cancer is more dangerous, but not nearly as dangerous
as a melanoma.
33
In support of this assessment, a previous draft of this manuscript (Andersen et al., 2011) examined whether
UV-R is predictive of skin cancer. Conditional on our full set of controls UV-R is in fact not significantly
correlated with skin cancer prevalence, whereas UV-R in the same setting predicts cataract.
34
The data for the alternative diseases such as skin cancer that we refer to in this section also derive from the
WHO and represents YLD, just as our cataract data. See the Appendix 1 for a description of the data.
35
On Malaria, see Gallup and Sachs (2001)—for a skeptical assessment of malaria’s influence, see DepetrisChauvin and Weil (2013); on Hookworm, see Bleakley (2007); on HIV, see e.g. Papageorgiou and Stoytcheva
(2005)—and for a skeptical assessment of HIV’s influence, see Young (2005). HIV/AIDS is obviously not a
32
28
add intestinal nematode infection as well as trachoma, the latter being an infectious eye disease that is
more prevalent in the geographic tropics yet epidemiologically has nothing to do with UV-R. As seen
from Table 6 (and Table B7 in the Supplementary Appendix), the inclusion of any of these diseases
alongside our full set of geographic controls does not render the influence of UV-R insignificant.
Naturally, it is impossible to completely rule out that UV-R is picking up the influence from some
other disease. Still, we view these checks as a good indication that our regressions in Section 4 are not
simply convoluting the influence from tropically clustered diseases.
5.2 Institutions and Culture
Cross-country evidence So far the analysis has not dealt explicitly with two sets of fundamental
determinants, which may influence the link between UV and prosperity, and these are institutions and
cultural values. Of course, institutions and cultural values are not exogenous but represent the
outcome of historical processes. As a result, the analysis above may actually have accounted for their
influence inadvertently. More specifically, if institutions and culture are determined by underlying
climatic and/or geographical characteristics, the latter controls may be capturing at least some of the
influence from the former on prosperity in Tables 1 and 2.36 Nevertheless, in this section and the next,
we attempt to provide more rigorous checks.
In Supplementary Appendix Table B8 we report the results from including a much richer set of
fundamental determinants of culture and institutions in our full specification as well as from
additionally including direct measures of institutions alongside UV-R. In the latter case we must
recognize that institutions variables are endogenous, for which reason the reported (OLS) results are
(likely) uninformative about the impact of institutions on prosperity. However, these checks do allow
us to gauge the sensitivity of the UV-R/prosperity nexus to the inclusion of institutions measures into
the regression specification. Consequently, Tables B8 and B9 in the Supplementary Appendix report
the results from adding the risk of expropriation, the rule-of-law index, and political rights to the set
of controls. In all cases UV-R remains significant.
Prompted by the literature on the natural resource curse, we also examine the influence of UV-R on
income when measures of natural resource abundance (oil and gem stone) are added to the regression.
Again we find that the influence of UV-R on income persists. In yet another set of checks we add
tropical disease, but it tends to be more prevalent in populations located in the geographical tropics, SubSaharan Africa in particular.
36
See example Durante (2010) and Michalopoulos (2012) for evidence of climate’s impact on culture, and
Olsson and Hansson (2011) for the impact of geography on institutions.
29
underlying determinants of trust: slave exports and migratory distance from Ethiopia.37 As in the
checks involving institutions, including these variables has little influence on the link between UV-R
and GDP per capita (respectively GDP per worker). Finally, Tables B8 and B9 also report the results
from employing (several different sets of) regional fixed effects, rather than the continental fixed
effects hitherto adopted. Once again, the impact of UV-R remains significant.
While all these results testify to the strength of the UV-R/income correlation, it is obviously difficult
to show decisively, in a cross-country setting, that UV-R is not picking up some variation that should
be rightly attributed to culture and/or institutions. However, with data capturing within country
variation in economic activity and UV radiation, one is in a better position to prune the data from the
impact of (countrywide) institutions and culture. In the next section we therefore move beyond the use
of countries as unit of analysis.
Pixel-level analysis In this part of the empirical analysis we examine the determinants of economic
activity at the level of a pixel. Specifically, we divide the world into squares (i.e., pixel’s) with
dimensions 1-degree latitude by 1-degree longitude; this is approximately 100 square kilometers at the
equator. Figure 3 depicts the geographic distribution of GDP per capita as of 2005 at the 1x1 degree
resolution. For robustness we examine pixels with greater dimensions: 2x2 degrees latitude/longitude
and 4x4 degrees latitude/longitude. In addition, following Henderson et al. (2012) we employ an
alternative indicator of economic activity: satellite data on earthlights at night. In sum, we measure
economic activity in two different ways as well as at three different levels of aggregation.
The empirical strategy follows that of the country-level regressions; in the present setting we therefore
also control for latitude, elevation, distance to coast or river, precipitation, temperature and (pixel)
area.38 The main advantage of the analysis conducted in this section, however, is that we can control
much more rigorously for institutions and culture. Specifically, as a first step we check the strength of
the correlation between UV-R and local economic activity by including a full set of country fixed
effects. Controlling for country specific unobserved variation amounts to controlling for country
specific institutions and cultural values. Most likely this should be enough to gauge whether the UV-R
gradient is convoluting the influence from these other fundamental determinants of productivity.
Nevertheless, in light of recent research on the influence of culture on long-run prosperity (e.g.,
Tabellini, 2010; Michalopoulos and Papaioannou, 2013), one may worry that culture varies within
countries. As a result, we also include a full set of language fixed effects. The logic is that if the
37
Nunn and Wantchekon (2012) argue that slave exports, especially in Africa, led to societies characterized by
lower trust levels. Ashraf and Galor (2013) argue that migratory distance from Africa, by affecting genetic
diversity in society, has an effect on trust today.
38
Since the geographical size of a pixel varies across the global (as one moves away from the equator),
controlling for area is relevant in the present context as well.
30
spoken language varies within a country then this likely signals cultural variation as well. This
approach implies that when we examine economic activity at the 1x1 resolution, we include in excess
of 1000 (language) fixed effects in an effort to control for within country variation in cultural traits.
The details on the construction of the language fixed effects are found in Appendix 1.
Figure 3. Real Gross Product Per Capita (2005 PPP-USD). Source: Yale G-ECON Project. Notes: See
Appendix 1 for details.
It is worth observing that the controls collectively capture most of the variation in UV-R. When we
control for country fixed effects, in addition to the geographic controls, we span 95% of the UV-R
variation in the 1x1 setting; and slightly more than that when we instead invoke language fixed effects
(cf. Appendix 2, Table A.3). As in the cross-country analysis we are inclined to interpret the residual
variation in UV-R as largely reflecting the variation in UV-R that can be ascribed to prevailing cloud
conditions.
Table 7 reports the regression results when the dependent variable is (log) GDP per capita for 2005.
The first three columns focuses on results from the 1x1 resolution; the next three columns report
results from using pixel’s with dimensions 2x2 degree latitude/longitude; and in the final three
columns the table reports the 4x4 degree latitude/longitude units. At each level of aggregation we first
examine the partial link between UV-R and GDP per capita without any fixed effects; then with
country fixed effects, and then, finally, with a full set of language fixed effects.
[Table 7 about here]
31
As is evident, the controls and UV-R explain the bulk of the global variation in living standards.
Importantly, UV-R is significant in all cases, i.e. both with and without country and language fixed
effects. The strength of the partial correlation does not appear to be sensitive to the particular level of
aggregation chosen; in fact, it remains largely unchanged as we move from the 1x1 level to the 4x4
level.
Table 8 reports the results from when we employ earthlights as a proxy for economic activity. As can
be seen upon inspecting the table, results are qualitatively very similar. As in the case of GDP per
capita, the UV-R gradient appears at all three levels of aggregation and the partial correlation is
relatively stable. Overall, the regional analysis corroborates the results from the pure cross-country
analysis in suggesting a detrimental impact from UV-R on prosperity.39
[Table 8 about here]
But the results do differ in one important respect: the economic size of the impact from UV-R on
GDP per capita. As apparent from columns 2 or 3 in Table 7, when UV-R is increased by one percent,
GDP per capita drops by 0.16%. This is a considerably smaller effect than the estimate of 1.6%
obtained in the cross-country analysis (cf. Table 2, column 8). Another way to see the difference is by
observing that one standard deviation reduction in UV-R (roughly 0.85 log points) implies an increase
in GDP per capita of about 15% (= exp(0.85*0.16)-1); down from 130% in the pure cross-country
analysis.
What should we make of this? An obvious interpretation is that the cross-country analysis may be
tainted by some omitted variable bias, and apparently these omitted variables work to increase the
economic significance of UV-R. If this interpretation is correct, the results from Table 7 are more
likely to convey accurate information about the causal influence from eye disease on long-term
development than the results from Tables 1 and 2. Another interpretation, however, would be that the
results from Table 7 are underestimating the impact from eye disease. Migration may be a bigger
issue in the context of the pixel analysis compared to the cross-country ditto. If individuals tend to
migrate to regions with higher productivity, which could be caused by less UV-R in the first place,
this will reduce interregional income variation and thus temper the impact from UV-R. Of course, it
may be the case that both omitted variables and migration contributes to the reduction in the estimate
for UV-R.
39
As human capital accumulation is thought to be a key mechanism linking UV-R and economic development
(see Section 2), it is worth observing that Gennaioli et al. (2013) provide theory and direct evidence in favor of a
first-order impact of human capital on regional development.
32
The conservative conclusion from the analysis would be to assume the former interpretation is more
important, which implies that an elasticity of around 0.2 (rather than around 1.5) is a more plausible
estimate for the impact of UV-R on prosperity. This remains a very substantial impact however. As
noted above, the simulation study by Ashraf et al. (2008) finds that an increase in life expectancy by
about 20 years eventually leads to an increase in GDP per capita which is similar to what a reduction
in one standard deviation in UV-R produces, judged from the results in Table 7. In this respect, the
within country estimates reinforces the overall conclusion that historical eye disease incidence has had
a powerful impact on contemporary cross-country income differences.
US versus China A drawback of the subnational analysis above is that it does not allow us to assess
directly whether the mechanism behind the UV-R correlation is the same as the one uncovered in the
cross-country setting, namely that a greater inherent return to skill investments, by affecting the
timing of the fertility transition, served to generate contemporary income differences. But indirect
evidence can be brought to bear.
The United States is one of the so-called forerunners in terms of the fertility transition. A recent study
by Hansen et al. (2014) analyzes the causes of the transition using cross-state cohort data. Their main
result is that schooling in particular seems to have played a key role in instigating a differential timing
of the fertility decline across US states. If UV-R, as hypothesized above, were indeed a determinant of
the perceived return to schooling, one would therefore expect to see a UV-R/income gradient within
the US.
A country worth contrasting with the US is China. According to demographers China went through
the fertility decline in the 1970s (Reher, 2004). Unlike the US, however, this seems to have been
driven by governmental policies. In particular, the “later, longer, fewer” policy of 1971—which
encouraged people to have children later in life, increase birth spacing, and thus simply have fewer
children—and the one-child policy—which was announced in 1979—are often identified by
demographers to have been key drivers (e.g., Coale, 1984; Bongaarts and Greenhalgh, 1985; Xizhe,
1989). The work by Li and Zhang (2007) supports this assessment. The authors observe that the onechild policy was subject to an important qualification: it only applied to Han Chinese; ethnic
minorities did not have to abide by the policy. Accordingly, if the policy was in fact effective, the
minority share of the regional populations should be a good predictor of cross-regional fertility.
Specifically, in areas with more ethnic minorities fertility should be higher insofar as the policy was
binding. This does in fact seem to be the case. Li and Zhang further document that the fertility decline
importantly stimulated economic growth in China. For present purposes, however, the main point is
that the measures of population control enacted by the Chinese government were pervasive,
33
apparently effective, and unlikely to be correlated with UV-R within China. Therefore, we would
expect to see that UV-R is considerably less strongly correlated with regional levels of development
within China than what would be the case for the US.
In Table 9 we report the results from estimating the impact of UV-R on regional economic activity
within China and the US. As the indicator for economic activity we employ earthlights at night; we
estimate the model at all three levels of aggregation; and we estimate it with the same controls as
those invoked in the context of Tables 7 and 8, except that we obviously have no need to invoke
country fixed effects. Hence in the first two columns we compare China and the US at the 1x1 level,
without any fixed effects; the subsequent two columns invoke language fixed effects. This procedure
is repeated at the 2x2 and 4x4 level.
The results are quite striking. Whereas the negative link between UV-R and regional development is
strongly borne out in the US case, UV-R and economic activity is virtually uncorrelated (conditional
on controls) within China. Consistent with priors, it seems that the mechanism detected at the country
level has not been operative within China. Most likely because the fertility decline was importantly
driven by government policy. In contrast, and consistent with the regional analysis in Hansen et al.
(2014) for the US, UV-R is strongly and negatively correlated with economic activity in the US. This
is of course fully consistent with an impact of UV-R on the perceived return to skill investments,
which in turn has been important in generating a differential timing of the fertility decline, ultimately
leaving its mark on contemporary comparative development within the US.
[Table 9 about here]
6 Conclusion
This paper has examined the long-run consequences of global differences in the inherent return to
skill formation, which can be attributed to variation in the historical incidence of eye disease. Drawing
on research from the field of epidemiology, we proposed to use UV radiation to capture the historical
incidence of eye disease. Our key result is that UV radiation holds strong explanatory power vis-à-vis
contemporary income differences, both across and within countries. In the cross-country setting, one
standard deviation increase in UV-R lowers early 21st century GDP per capita by roughly 130%. This
is a large effect; and it is much too large to reflect the direct labor force participation effect of eye
disease. However, if differences in UV-R, by generating cross-country variation in the perceived
return to skill accumulation, has influenced the relative timing of the take-off to sustained growth, a
much larger impact on current income per capita can be upheld.
34
To substantiate that eye disease can in fact importantly affect the return to skill formation, we invoked
ophthalmological surveys of eye disease prevalence. We showed that actual prevalence rates are
capable of inducing gaps in expected work life of up to 14 years when comparing high and low UV-R
regions. This difference in expected work life is roughly comparable to the expansion in expected
work life resulting from the international epidemiological transition.
To further substantiate our interpretation of the UV-R/income gradient we performed a series of
demanding consistency checks. We first showed that the gradient only emerges during the 20th
century; it did not exist in previous centuries. Consequently, UV only became a relevant cause of
productivity after the first emergence of the fertility transition. We then documented that UV-R is a
robust determinant of the onset of the transition, and that the fertility transition itself is strong
predictor of current cross-country income differences. Quantitatively, the link between UV-R and the
timing of the fertility transition is large enough to reasonably account for our reduced-form estimate
of the influence of UV-R on current income per capita. Finally, we documented that UV-R is a
significant determinant of human capital investments since 1870. At the same time, UV-R looses
significance once we control for the timing of the fertility transition, which itself is strongly linked to
human capital investments in keeping with theoretical priors. This is of course fully consistent with a
pivotal role of the fertility transition in unleashing a process of human capital accumulation.
We do not claim that eye disease is the full, or the most important, explanation of the differential
timing of the fertility transition and thus ultimately of contemporary income differences. We what do
claim is that UV-R induced variation in disease ecology with respect to eye disease gives us a credible
source of exogenous variation in the inherent return to skill formation, which enables us to study the
relevance of a fundamental mechanism from which the “Great Divergence” is supposed to have
originated. This fundamental mechanism, linking contemporary income differences to the differential
timing of the fertility transition and accompanying variation in human capital investments, appears to
be strongly borne out in the data. In this sense, our paper strongly supports the central propositions of
the literature on growth over the very long run (e.g., Galor and Weil, 2000; Galor and Moav, 2002;
Lucas, 2002; Hansen and Prescott, 2002).
35
REFERENCES
Acemoglu, Daron. 2008. Introduction to Modern Economic Growth. Princeton: Princeton University
Press.
Acemoglu, Daron, and Simon Johnson. 2007. “Disease and development: the effect of life
expectancy on economic growth.” Journal of Political Economy, 115(6): 925-985.
Acemoglu, Daron and Melissa Dell. 2010. “Productivity Differences Between and Within
Countries.” American Economic Journal: Macroeconomics, 2(1), pp. 169–188.
Aghion, Philippe, Peter Howitt, and Fabrice Murtin. 2010. “The relationship between health and
growth: when Lucas meets Nelson-Phelps.” NBER working Paper No. 15813.
Altonji, Joseph G., Todd E. Elder, and Christopher R. Taber. 2005. “Selection on Observed and
Unobserved Variables: Assessing the Effectiveness of Catholic Schools.” Journal of Political
Economy, 113(1): 151-184
Andersen, Thomas B., and Carl-Johan Dalgaard. 2011. “Flows of People, Flows of Ideas and the
Inequality of Nations.” Journal of Economic Growth, 16(1): 1-32.
Andersen, Thomas B., Carl-Johan Dalgaard, and Pablo Selaya. 2011. “Eye Disease and
Development.” Discussion Paper No. 11-22, Department of Economics, University of
Copenhagen.
Angeles 2010 Angeles, Luis. 2010. “Demographic transitions: Analyzing the effects of mortality on
fertility.” Journal of Population Economics, 23: 99-120.
Ashraf, Quamrul, and Oded Galor. 2011. “Dynamics and Stagnation in the Malthusian Epoch.”
American Economic Review, 101(5): 2003-2041.
Ashraf, Quamrul, Ashley Lester, and David Weil. 2008. “When Does Improving Health Raise
GDP?” NBER Macroeconomics Annual, 23(1): 157-204.
Ayala, Marcelo N., Ralph Michael, and Per G. Söderberg. 2000. “Influence of exposure time for
UV radiation-induced cataract.” Investigative ophthalmology & visual science, 41(11): 353943.
Azam, David Corderi, Kyle Hood, Nadejda Makarova Victor, Mukhtar Mohammed, Alexandra
Miltner, and Jyldyz Weiss. 2006. “The G-Econ Database on Gridded Output:
Methods and Data.”
http://gecon.yale.edu/sites/default/files/gecon_data_20051206.pdf
Becker, Sascha O., and Ludger Woesmann. 2009. “Was Weber Wrong? A Human Capital Theory
of the Protestant Reformation.” Quarterly Journal of Economics, 124(2): 531-596.
Becker, Sascha O., Francesco Cinnirella, and Ludger Woesmann. 2010. “The trade-off between
fertility and education: evidence from before the demographic transition.” Journal of Economic
Growth, 15 (3): 177-204.
Bellows, John, and Edward Miguel. 2008. “War and Collective Action in Sierra Leone.”
Unpublieshed Working Paper, http://emlab.berkeley.edu/~emiguel/pdfs/miguel_sl.pdf
Bhalla, Shiela, 1989. Employment in Indian agriculture: retrospect and prospect. Social Scientist, 321.
Bleakley, Hoyt. 2007. “Disease and Development: Evidence from Hookworm Eradication in the
American South.” Quarterly Journal of Economics, 122 (1): 73-117.
Bongaarts, John, and Susan Greenhalgh. 1985. “An Alternative to the One-Child Policy in
China.” Population and Development Review, 11(4): 585-617
Brian, Gary, and Hugh Taylor. 2001. “Cataract blindness - challenges for the 21st century.”
Bulletin of the World Health Organization, 79 (3): 249-56.
Cervellati, Matteo, and Uwe Sunde. 2010. “Longevity and Lifetime Labor Supply: Evidence and
Implications Revisited”. Working Paper (St. Gallen).
36
Cervellati, Matteo, and Uwe Sunde. 2011. “Life Expectancy and Economic Growth: The Role of
the Demographic Transition.” Journal of Economic Growth, 16 (2): 99-133.
Chatterjee, Arin, Roy C. Milton, and Sydney Thyle. 1982. “Prevalence and aetiology of cataract in
Punjab.” British Journal of Ophthalmology, 66: 35-42.
Clark, Gregory. 2007. A Farewell to Alms: A Brief Economic History of the World. Princeton
University Press.
Coale, Ansley. 1984. “Rapid Population Change in China, 1952-1982.” Washington, D.C.,
Committee on Population and Demography, Report No. 27, National Research Council,
National Academy Press.
Cockell, Charles S., and Gerda Horneck. 2001. “The History of the UV Radiation Climate of the
Earth - Theoretical and Space-based Observations.” Photochemistry and Photobiology, 73(4):
447-51.
Corser, Noel. 2000. “Couching for Cataract: Its Rise and Fall.” The Proceedings of the 9th Annual
History of Medicine Days, edited by W. A. Whitelaw, pp. 35-41.
Dandona, Lalit, Rakhi Dandona, Thomas J. Naduvilath, Catherine A. McCarty, Partha
Mandal, M. Srinivas, Ashok Nanda, and Gullapalli N. Rao. 1999. “Population-based
assessment of the outcome of cataract surgery in an urban population in Southern India.”
American Journal of Ophthalmology 127 (6): 650–658.
Dalgaard, Carl-Johan, and Holger Strulik. 2013. “The History Augmented Solow Model.”
European Economic Review 63 134-49.
Depetris-Chauvin, Emilio and David N.Weil. 2013. Malaria and Early African Development:
Evidence from the Sickle Cell Trait. NBER working paper No. 19603
Diamond, Jared. 1997. Guns, Germs and Steel. New York: W.W. Norton.
Diamond, Jared. 2005. “Evolutionary biology: Geography and skin color.” Nature, 435: 283-284.
Dong, Xiuqin , Marcelo Ayala, Stefan Löfgren, and Per G. Söderberg. 2003. “Ultraviolet
Radiation–Induced Cataract: Age and Maximum Acceptable Dose.” Investigative
Ophthalmological and Visual Science, 44(3): 1150-54.
Durante, Ruben. 2010. “Risk, Cooperation and the Economic Origins of Social Trust: An Empirical
Investigation.” http://www.rubendurante.com/durante_jmp.pdf
Feyrer, james, and Bruce Sacerdote. 2009. “Colonialism and Modern Income—Islands as Natural
Experiments.” Review of Economics and Statistics, 91(2): 245-262.
Foster Allen. 1991. “Who will operate on Africa’s 3 million curably blind people?” Lancet,
337(8752): 1267–1269.
Foster, A., and S. Resnikoff. 2005. “The impact of Vision 2020 on global blindness.” Eye, 19: 11331135
Frankel, Jeffrey A., and David Romer. 1999. “Does trade cause growth?” American Economic
Review, 89(June): 379-399.
Gallagher, Richard P., and Tim K. Lee. 2006. “Adverse effects of ultraviolet radiation: A brief
review.” Progress in Biophysics and Molecular Biology, 92(1): 119-131
Gallup, John L., and Jeffrey D. Sachs. 2000. Agriculture, climate, and technology: why are the
tropics falling behind? American Journal of Agricultural Economics, 82: 731-737.
Gallup, John L., and Jeffrey D. Sachs. 2001. “The economic burden of malaria.” American Journal
of Tropical Medicine and Hygiene, 64(1, 2)S: 85-96
Gallup, John L., Jeffrey D. Sachs, and Andrew Mellinger. 1999. “Geography and Economic
Development”. CID Harvard University Working Paper No. 1.
Galor, Oded. 2005. “The Transition from Stagnation to Growth: Unified Growth Theory.” In
Handbook of Economic Growth, Vol IA, ed. Philippe Aghion and Steven N. Durlauf, 171-293.
Amsterdam, The Netherlands: Elsevier North-Holland
37
Galor, Oded. 2010. “2008 Lawrence R. Klein Lecture - Comparative Economic Development:
Insights from Unified Growth Theory.” International Economic Review, 51(1): 1-44.
Galor, Oded. 2011. Unified Growth Theory. Princeton: Princeton University Press.
Galor, Oded, and David N. Weil. 2000. “Population, technology and growth: From Malthusian
stagnation to the demographic transition and beyond.” American Economic Review 90(4): 806828.
Galor, Oded, and Omar Moav. 2002. “Natural selection and the origin of economic growth.”
Quarterly Journal of Economics, 117(4): 1133-1191.
Gennaioli Nicola, Rafael La Porta, Florencio Lopez-de-Silanes, and Andrei Shleifer, 2013.
“Human Capital and Regional Development.” Quarterly Journal of Economics, 128: 105-164.
Glaeser, Edward L., Rafael La Porta, Florencio Lopez-de-Silanes, and Andrei Shleifer. 2004.
“Do Institutions Cause Growth?” Journal of Economic Growth, 9: 271-303
Hall, Robert, and Charles I. Jones. 2007. The Value of Life and the Rise in Health Spending.
Quarterly Journal of Economics, 122: 39-72.
Hansen, Gary D., and Edward C. Prescott. 2002. “Malthus to Solow.” American Economic Review,
92(4): 1205-1217.
Hansen, Casper W., Peter Sandholt Jensen, and Lars Lønstrup. 2014. “The Fertility Transition in
the US: Schooling or Income?” Economics Working Papers 2014-02, School of Economics ad
Management, University of Aarhus
Hazan, Moshe. 2009. “Longevity and Lifetime Labor Supply: Evidence and Implications”
Econometrica, 77: 1829-1863.
Hazan, Moshe, and Hosny Zoabi. 2006. “Does Longevity Cause Growth? A Theoretical Critique.”
Journal of Economic Growth 11(4): 363-76.
Henderson, Vernon, Adam Storeygard and David Weil. 2012. “Measuring Economic Growth
From Outer Space”. Forthcoming: American Economic Review, 102(2): 994-1028.
Herzer, Dierk, Holger Strulik, and Sebastian Vollmer. 2012. “The long-run determinants of
fertility: one century of demographic change 1900-1999.” Journal of Economic Growth, 17(4):
357-385
Hollows, Fred, and David Moran. 1981. “Cataract-the ultraviolet risk factor.” The Lancet,
318(8258): 1249-1250.
Kalemli-Ozcan, Sebnem, and Belgi Turan. 2011.” HIV and Fertility Revisited.” Journal of
Development Economics, 96(1): 61-65
Klaver, Caroline C., Roger C. Wolfs, Johannes R. Vingerling, Albert Hofman, and Paulus de
Jong. 1998. “Age-Specific Prevalence and Causes of Blindness and Visual Impairment in an
Older Population.” Archives of Ophthalmology, 116(May): 653-658
Li, Hongbin, and Junsen Zhang, 2007. “Do high birth rates hamper economic growth?” Review of
Economics and Statistics, 89: 110-117.
Javitt, Jonathan C., Fang Wang, and Sheila K. West. 1996. “Blindness due to Cataract:
Epidemiology and Prevention.” Annual Reviews of Public Health, 17: 159-77
Lansingh, Van C., Marissa J. Carter, and Marion Martens. 2007. “Global Cost-effectiveness of
Cataract Surgery.” Ophthalmology, 114(9): 1670-1678.
Lorentzen, Peter, John McMillan, and Romain Wacziarg. 2008. “Death and development.”
Journal of Economic Growth, 13(2): 81-124.
Lucas, Robert E. Jr. 2000. “Some Macroeconomics for the 21st Century.” Journal of Economic
Perspectives, 14(1): 159-68.
Lucas, Robert E. Jr. 2002. Lectures on Economic Growth. Cambridge Massachusetts: Harvard
University Press.
Maddison, Angus. 2003. The World Economy: Historical Statistics. Paris, France: OECD.
38
Masters, William, and Margaret McMillan. 2001. “Climate and Scale in Economic Growth.”
Journal of Economic Growth (6): 167-86.
McCord, Gordon, Dalton Conley and J. D. Sachs. 2010. “Improving Empirical Estimation of
Demographic Drivers: Fertility, Child Mortality & Malaria Ecology,” Science Research
Network Working Paper.
Michalopoulos, Stelios, 2012. “The Origins of Ethnolinguistic Diversity.” American Economic
Review. 102 , 1508-39.
Michalopoulos, Stelios, and Elias Papaioannou. 2013. “Pre-colonial ethnic institutions and
contemporary African Development.” Econometrica 81: 113-52.
Morrison, Christian, and Fabrice Murtin. 2009. “A century of education.” Journal of
Human Capital, 3(1): 1-42.
Murtin, Fabrice. 2013. “Long-term Determinants of the Demographic Transition, 1870-2000.”
Review of Economics and Statistics, 95(2): 617-631.
Nordhaus, William, Qazi Azam, David Corderi, Kyle Hood, Nadejda Victor, Mukthar
Mohammed, Alexandra Miltner, and Jyldyz Weiss. 2006. “Detailed description of
derivation of G-ECON data,” http://gecon.yale.edu/data-and-documentation-g-econ-project
Nunn, Nathan. 2008. “The Long-Term Effects of Africa’s Slave Trades.” Quarterly Journal of
Economics, 123(1): 139-176.
Nunn, Nathan, and Leonard Wantchekon. 2011. “The Slave Trade and the Origins of Mistrust in
Africa.” American Economic Review 101 (December): 3221-3252.
Olken, Benjamin, Melissa Dell and Ben Jones. 2009. ”Temperature and Income: Reconciling New
Cross Sectional and Panel Estimates.” American Economic Review Papers and Proceedings,
99(2): 198-204.
Olsson, Ola, and Douglas A. Hibbs. 2005. “Biogeography and Long-Run Economic Development.”
European Economic Review, 49(4): 909-938.
Olsson, Ola, and Gustav Hansson. 2011. “Country Size and the Rule of Law: Resuscitating
Montesquieu.” European Economic Review, 55(5): 613-629
Papageorgiou, Chris and Petia Stoytcheva. 2005. What Do We Know About the Impact of AIDS
on Cross-Country Income So Far? Louisiana State University, Discussion Paper No. 2005-1.
Putterman,
Louis.
2006.
“Agricultural
Transition
Year
Country
Data
Set.”
http://www.econ.brown.edu/fac/louis_putterman/agricultural%20transition%20data%20set.pdf
Reher, David S. 2004. “The demographic transition revisited as a global process.” Population Space
and Place, 10, 19-42.
Saikia, Nandita, Abhishek Singh and Faujdar Ram. 2012. “Adult male mortality in India: An
application of the widowhood method.” Working Paper, University of Delhi. Tabellini, Guido. 2010. “Culture and Institutions: Economic Development in the Regions of
Europe.” Journal of the European Economic Association, 8: 677-716.
Taylor, Hugh R., Sheila K. West, Frank S. Rosenthal, Beatriz Muñoz, Henry S. Newland, Helen
Abbey, and Edward A. Emmett. 1988. “Effect of ultraviolet radiation on cataract formation.”
New England Journal of Medicine, 319: 1429-1433.
Weil, David N. 2007. “Accounting for The Effect of Health on Economic Growth.” Quarterly
Journal of Economics, 122(3): 1265-1306.
West, Sheila K. 2007. “Epidemiology of Cataract: Accomplishments over 25 years and Future
Directions.” Ophthalmic Epidemiology, 14(4): 173–178.
West Sheila K., Donald D. Duncan, Beatrice Muñoz, Gary S. Rubin, Linda P. Fried, Karen
Bandeen-Roche, and Oliver D. Schein. 1998. “Sunlight exposure and risk of lens opacities in
a population-based study: The Salisbury Eye Evaluation Project.” Journal of the American
Medical Association, 280(8): 714-718.
39
World Health Organization. 2008. “The global burden of disease: 2004 update”.
http://www.who.int/entity/healthinfo/global_burden_disease/GBD_report_2004update_full.pdf.
Xizhe, Peng. 1989. “Major Determinants of China's Fertility Transition.” China Quarterly, 117: 1-37. Young, Alwyn. 2005. “The Gift of the Dying: The Tragedy of AIDS and the Welfare of Future
African Generations.” Quarterly Journal of Economics, 120(2): 423-466.
40
APPENDIX 1
The appendix includes definitions and data sources for all variables in our dataset.
A. THE UV RADIATION VARIABLE
The measure labeled “UV-R” in the text has been developed by NASA, and is the end result of a
somewhat complex mathematical calculation.40 It basically involves “inputs” such as earth-sun
distance, time of sunrise and sunset, cloud optical thickness, total ozone column and more, which all
can be resolved at a high level of resolution. NASA then expresses the resulting UV-R measure in
units such that it speaks directly to how exposed people on the planet are to sunburn, as a consequence
of UV-R. Or, put differently, the variable becomes an “index of the potential for biological damage
due to solar irradiation, given the column ozone amount and cloud conditions on each day.” In
practice this is done by attaching greater weight to UV-R with shorter compared to longer
wavelengths. The weighting function is chosen such that it speaks to the susceptibility of a Caucasian
of getting sunburned, as a consequence of UV-R, regardless of whether or not there are any
Caucasians at any particular location. This particular choice of units is presumably motivated by a
demand by e.g. meteorologists, who wish to give the public a sense of the daily risk of sunburn, with
an eye to potential risks of eventually contracting skin cancer. As a result, NASA formally labels our
measure of UV-R the “Erythemal Exposure Data”; “erythema” means sunburn. Nevertheless, the
variation we extract from the data in the context of our regressions derives from UV-R, and not from
this particular choice of units (i.e., weights assigned to UV-R wavelengths), as they are the same
everywhere on the planet. Moreover, the range of UV-R wavelengths that goes into the measure are
also the relevant wavelengths as far as cataracts are concerned, i.e., below 400 nm (see e.g. Roberts,
2011, “Ultraviolet radiation as a risk factor for cataract and macular degeneration”, Eye and Contact
Lens, 37(4): 246-9)
Three other general remarks on our measure are important. First, while it does involve an “input”
measuring the thickness of the ozone column, this is unlikely to make the resulting measure
endogenous. As noted in the paper, local economic activity does not map into local ozone thickness in
any simple way. Second, the measure does not factor in the effects from smoke plumes, originating
from for example biomass burning. This would otherwise have been another way in which the UV-R
measure could be endogenous to local conditions. Third, since e.g. distance to the sun depends on
latitude and elevation, the UV-R index is strongly correlated with these factors. As a result, we control
for them in the regressions. The remaining variation across countries and regions is largely the result
of the influence from cloud conditions.
For the 1x1 latitude-longitude pixel level analysis (Tables 7-9 in the paper, and A3-A4 in Appendix
2), we rely on the simple average of the raw UV-R data. Following Dell, Olken and Jones (2009) and
Acemoglu and Dell (2010), we compute population-weighted averages for the 2x2 and 4x4 pixel level
analysis as well as for the cross-country analysis. We use gridded population levels from the GECON
Yale 3.4 database to compute the population-weights in the 2x2 and 4x4 pixel level analysis.41 For the
cross country analysis (Tables 1-6 in the paper, A2 in the appendix, and B1-B9 in the supplementary
appendix), we compute population weights based on NASA’s Socioeconomic Data and Applications
Center “Gridded Population of the World dataset”,42 and on country grid definitions from the U.S.
Board on Geographic Names’ database of foreign geographic names and features.43
B. CROSS COUNTRY ANALYSIS DATA
40
See http://ozoneaq.gsfc.nasa.gov/doc/erynotes.pdf.
GECON Yale data are available at http://gecon.yale.edu.
42
NASA’s Socioeconomic Data and Applications Center is hosted by Earth Institute at Columbia University.
The gridded population data are available at
http://sedac.ciesin.columbia.edu/gpw/global.jsp?file=gpwv3&data=pdens&type=ascii&resolut=one&year=15.
43
Data from the U.S. Board on Geographic Names data are available at
http://geonames.usgs.gov/domestic/download_data.htm.
41
41
Labor productivity, income per capita, and historical measures of prosperity
•
•
•
Real GDP per worker and per capita in 2004, from the Penn World Tables.
Real GDP per capita 1700-1950, from Maddison (2003).
Population density in year 1, 1000, 1500, from Ashraf and Galor (2011).
Cataract (and other diseases) incidence
The World Health Organization (WHO) quantifies the burden of any specific disease as the equivalent
number of years of “healthy” life lost due to the incidence (mortality and morbidity) of that disease.
This measure of Disability-Adjusted Life Years (DALY), can be interpreted as an estimate of the gap
between current health status and an ideal health situation where the entire population lives to an
advanced age, free of disease and disability.44
Our measure for the incidence of cataract in each country corresponds to the number of DALYs due
to the incidence of this disease in 2004, expressed as a ratio of 100,000 people in the population. Data
for the incidence of other diseases corresponds to DALYs per 100,000 people for trachoma, skin
cancer (melanoma and other skin carcinomas), HIV/AIDS, malaria, and hookworm disease, and
intestinal nematode infections.
All data from WHO (2008) are available at
http://www.who.int/healthinfo/global_burden_disease/2004_report_update/en/index.html.
Timing of the fertility decline
Year of the fertility transition for countries around the world are from Rehrer (2004). The criteria for
pinpointing the date of the transition: “[…] has been set at the beginning of the first quinquennium
after a peak, where fertility declines by at least 8% over two quinquennia and never increases again to
levels approximating the original take-off point” (Reher, 2004, p. 21).
Control variables (cross country analysis):
Geography and climate:
•
•
•
•
•
•
•
Continent dummies (Africa, Asia, North America, South America, Europe, Oceania) and latitude,
from Nunn and Puga (2010).
Elevation mean (average of elevation extremes). Source: CIA Factbook. Data available at
http://www.nationmaster.com.
Mean distance to coast or rivers, from Gallup, Sachs and Mellinger (1999).
Agricultural suitability index, from Ashraf and Galor (2011).
Percentage of land in tropical and subtropical zones, from Ashraf and Galor (2011).
Area weighted average number of frost days per year, 1901-2012, constructed from the Climatic
Research Unit’s (CRU) gridded dataset available at
http://badc.nerc.ac.uk/view/badc.nerc.ac.uk__ATOM__dataent_1256223773328276.
Area-weighted average air temperature (C degrees) and total precipitation (‘000 mm/year), 19802008. Constructed from the GECON 3.4 dataset. Data available at http://gecon.yale.edu.
Pre-industrial history:
•
Time passed since the Neolithic revolution, from Putterman (2008).
Human capital and fertility:
44
See http://www.who.int/healthinfo/global_burden_disease/metrics_daly/en/index.html.
42
•
•
Average schooling in 1870, 2000, 2010, from Morrison and Murtin (2010)
Fertility rate 1960’s, 1970s, 1980s, 1990s, 2000s (total births per woman), from World Bank’s
WDI.
Indicators of institutions, natural resources, trust, culture:
•
•
•
•
•
•
•
•
Malaria Ecology index, from McCord, Conley, and Sachs (2010)
Freedom House’s rating of political rights, 2002.
Rule of law 1996-2000, from Nunn and Puga (2010).
Percentage of mineral fuels in manufacturing exports, 2000, from World Bank’s WDI.
Gem diamond extraction 1958-2000 (1000 carats), from Nunn and Puga (2010).
Migratory distance from Ethiopia, from Olsson and Ahlerup (2009).
Slave exports 1400-1900, from Nunn and Puga (2010).
Fraction of population of Euro descent, from Putterman and Weil (2010).
Regional indicators:
•
World Bank’s regional definitions:
§ East Asia and Pacific
§ Europe and Central Asia
§ Latin America and Caribbean
§ Middle East and North Africa
§ North America
§ South Asia
§ Sub-Saharan Africa
•
WHO regional definitions:
§ Africa
§ Americas
§ Eastern Mediterranean
§ Europe
§ South East Asia
§ Western Pacific
§ Africa
C. GRIDDED DATA (1X1 degree latitude-longitude resolution)
Lights at night
Earth imagery data showing the intensity of lights at night.
The data are produced by satellites and sensors operated under the US Department of Defense’s
Version 4 Defense Meteorological Satellite Program Operational Linescan System (DMSP-OLS),
available at
http://www.ngdc.noaa.gov/eog/dmsp/downloadV4composites.html.
We resampled the raster corresponding to nighttime lights imagery in 2004, from a raw resolution of
30 arc seconds to a 1 x 1 degrees grid, using bilinear interpolation in ArcGIS.
43
Figure A1. Nighttime lights imagery. Raw data, 30 arc seconds resolution. Satellite F15, 2004,
average visible band digital number. From DMSP-OLS Nighttime Lights Global Composites
(Version 4).
http://maps.ngdc.noaa.gov/viewers/dmsp_gcv4/
Real grid cell product per capita
Real (PPP 1995 USD) gross product per capita, by cell of 1-degree latitude x 1-degree longitude,
constructed with data from GEcon Yale data version 3.4, available at http://gecon.yale.edu.
Control variables (pixel level analysis):
Language Fixed Effects
For each pixel we assign a unique dummy variable, which picks up a particular ethnic language, as
recorded in the World Language Mapping System Database version 3.01 (WLMS). WLMS contains
polygons for the linguistic homelands of 7,219 ethnic languages spoken in the world.45 In all the cases
where a pixel only involves one language the mapping is straightforward. In some instances, however,
multiple ethnic languages “share” a pixel. In these instances we assign the pixel to the language,
which is geographically the most widespread within the pixel. Moreover, in some pixels there are no
languages recorded in WLMS. What this means is that in these pixels there are no particular ethnic
languages being spoken. Therefore, we assign a separate dummy, which then constitutes the excluded
category within each country.
As an illustration, consider Bolivia. In Bolivia we find 8 ethnic languages spoken across 40 pixels (or
40% of the territory).46 The remaining pixels in the country are coded as places where there are no
predominant ethnic language spoken in Bolivia. In a similar way we construct pixel level language
dummies for the rest of the pixels in the world.
In our grid of 1x1 resolution, we end up with 1,222 language fixed effects (see for example column 3
in Table 8), of which 1038 correspond to a specific ethnic language.
45
The World Language Mapping System dataset is available at http://www.worldgeodatasets.com/language/.
These languages are Aymara, Chiquitano, Eastern and Western Bolivian Guarani, Guarayu, and North and
South Bolivian Quechua.
46
44
Data from WMLS v 3.01.
Geography and climate controls (GEcon Yale dataset)
•
•
•
•
•
•
•
Latitude (degrees)
Elevation (m above sea level)
Temperature (average annual level 1980-­‐2008, C degrees)
Precipitation (average annual level 1980-­‐2008, ‘000 mm)
Area (sq km)
Distance to ice-free ocean (km)
Distance to major navigable river (km).
APPENDIX 2
[Insert Tables A1-A4]
45
Table 1
Real GDP per worker, cataract incidence, and exposure to UV radiation
1
2
3
4
6
7
8
9
-1.50***
[0.42]
-1.46**
[0.61]
-1.32***
[0.12]
-1.07***
[0.27]
-1.09***
[0.20]
-1.36**
[0.56]
-1.50***
[0.26]
-1.35***
[0.12]
-1.63***
[0.19]
(log) Cataract prevalence, 2004
Observations
R-squared
Partial R-squared
Additional controls
10
11
-0.30***
[0.065]
-0.30***
[0.067]
143
0.67
0.09
143
0.74
0.10
(log) Real GDP per worker, 2004
Dependent variable:
(log) UV damage
5
143
0.37
143
0.51
0.11
143
0.39
0.15
143
0.54
0.04
143
0.37
0.15
143
0.50
0.39
143
0.43
0.20
-
Continent FEs
Latitude &
elevation (levels)
Latitude &
elevation (FEs)
Precipitation &
temperature
Distance to
ocean & rivers
Agricultural
suitability,
Tropical area,
Frost
4
2
14
2
4
3
Number of additional controls
143
0.67
0.07
143
0.72
0.05
All controls (with All controls (with All controls (with All controls (with
lat & elev levels) lat & elev FEs) lat & elev levels) lat & elev FEs)
15
27
15
27
Joint significance of the additional control variables (p-values for the H0: all regressors [except UV damage] are jointly insignificant):
Continent FEs
Latitude & elevation (levels)
Latitude & elevation (FEs)
Precipitation & Temperature
Distance to ocean & rivers
Agr. suitability, Tropical area, Frost
All controls (with latitude & elevation levels)
All controls (with latitude & elevation FEs)
0.00
0.10
0.00
0.70
0.00
0.01
0.00
0.00
0.00
0.00
Notes: OLS regressions. UV damage is an index of Erythemal exposure, constructed from pixel level daily averages of integrated ultraviolet irradiance A and B in 1990 and 2000, weighted by the susceptibility of caucasian skin
to sunburn (erythema), and the proportion of total population in each pixel. It can be interpreted as an index of the potential for biological damage due to solar irradiation, given the column ozone amount and cloud
conditions on each day. Raw UV exposure daily data for 1990 and 2000 produced by NASA. Cataract incidence is measured as the number of years lost due to disability, for incident cases of this disease (expressed as a rate
per 100,000 people between 15 and 59), estimated by WHO (2004). The R-squared from an OLS regression where all listed additional controls where partialled is reported as the Partial R-squared in each column. All
regressions include a constant term. Robust standard errors in brackets. ***, ** and * denote statistical significance at 1, 5 and 10% levels, respectively.
Table 2
Real GDP per capita, cataract incidence, and exposure to UV radiation
1
2
3
4
Dependent variable:
(log) UV damage
5
6
7
8
9
-1.63***
[0.46]
-1.72***
[0.62]
Additional controls
11
-0.36***
[0.069]
-0.36***
[0.066]
145
0.68
0.11
145
0.75
0.13
(log) Real GDP per capita, 2004
-1.47***
[0.13]
-1.18***
[0.28]
-1.34***
[0.20]
-1.82***
[0.61]
-1.71***
[0.28]
-1.55***
[0.13]
-1.95***
[0.23]
(log) Cataract prevalence, 2004
Observations
R-squared
Partial R-squared
10
145
0.40
145
0.53
0.11
-
Continent FEs
Number of additional controls
4
145
0.40
0.15
145
0.55
0.04
Latitude &
Latitude &
elevation (levels) elevation (FEs)
2
14
145
0.41
0.15
145
0.50
0.39
145
0.44
0.20
Precipitation &
temperature
Distance to
ocean & rivers
Agricultural
suitability,
Tropical area,
Frost
2
4
3
145
0.67
0.07
145
0.73
0.05
All controls (with All controls (with All controls (with All controls (with
lat & elev levels) lat & elev FEs) lat & elev levels) lat & elev FEs)
15
27
15
27
Joint significance of the additional control variables (p-values for the H0: all regressors [except UV damage] are jointly insignificant):
Continent FEs
Latitude & elevation (levels)
Latitude & elevation (FEs)
Precipitation & Temperature
Distance to ocean & rivers
Agr. suitability, Tropical area, Frost
All controls (with latitude & elevation levels)
All controls (with latitude & elevation FEs)
0.00
0.37
0.00
0.34
0.00
0.02
0.00
0.00
0.00
0.00
Notes: OLS regressions. UV damage is an index of Erythemal exposure, constructed from pixel level daily averages of integrated ultraviolet irradiance A and B in 1990 and 2000, weighted by the susceptibility of
caucasian skin to sunburn (erythema), and the proportion of total population in each pixel. It can be interpreted as an index of the potential for biological damage due to solar irradiation, given the column ozone
amount and cloud conditions on each day. Raw UV exposure daily data for 1990 and 2000 produced by NASA. Cataract incidence is measured as the number of years lost due to disability, for incident cases of this
disease (expressed as a rate per 100,000 people between 15 and 59), estimated by WHO (2004). The R-squared from an OLS regression where all listed additional controls where partialled is reported as the Partial
R-squared in each column. All regressions include a constant term. Robust standard errors in brackets. ***, ** and * denote statistical significance at 1, 5 and 10% levels, respectively.
Table 3
Cataract incidence and UV exposure
1
2
3
4
5
6
7
8
9
(log) Cataract prevalence, 2004
Dependent variable:
(log) UV damage
2.37***
[0.14]
1.55***
[0.30]
2.39***
[0.19]
2.51***
[0.82]
1.94***
[0.28]
2.45***
[0.17]
2.25***
[0.26]
1.43***
[0.46]
2.10***
[0.58]
Observations
R-squared
Partial R-squared
145
0.63
145
0.80
0.24
145
0.65
0.45
145
0.68
0.11
145
0.67
0.23
145
0.66
0.62
145
0.66
0.31
145
0.82
0.07
145
0.84
0.09
Continent FEs
Latitude &
elevation
(levels)
Latitude &
elevation (FEs)
Precipitation &
temperature
Distance to
ocean & rivers
Agricultural
suitability,
Tropical area,
Frost
4
2
14
2
4
3
Additional controls
Number of add controls
-
All controls
All controls (with
(with lat & elev
lat & elev FEs)
levels)
15
27
Joint significance of the additional control variables (p-values for the H0: all regressors [except UV damage] are jointly insignificant):
Continent FEs
0.00
Latitude & elevation (levels)
Latitude & elevation (FEs)
Precipitation & Temperature
Distance to ocean & rivers
Agr. suitability, Tropical area, Frost
All controls (with latitude & elevation levels)
All controls (with latitude & elevation FEs)
0.00
0.00
0.00
0.00
0.00
0.00
0.00
Notes: OLS regressions. Cataract incidence is measured as the number of years lost due to disability, for incident cases of this disease (expressed as a rate per
100,000 people between 15 and 59), estimated by WHO (2004). UV damage is an index of Erythemal exposure, constructed from pixel level daily averages of
integrated ultraviolet irradiance A and B in 1990 and 2000, weighted by the susceptibility of caucasian skin to sunburn (erythema), and the proportion of total
population in each pixel. It can be interpreted as an index of the potential for biological damage due to solar irradiation, given the column ozone amount and cloud
conditions on each day. Raw UV exposure daily data for 1990 and 2000 produced by NASA. The R-squared from an OLS regression where all listed additional
controls where partialled is reported as the Partial R-squared in each column. All regressions include a constant term. Robust standard errors in brackets. ***, **
and * denote statistical significance at 1, 5 and 10% levels, respectively.
Table 4
Measures of historical prosperity and exposure to UV radiation
1
2
3
4
Population density in:
Dependent variable:
5
6
7
Real GDP per capita in:
1 CE
1000 CE
1500 CE
1820
1900
1950
1950
(log) UV damage
-0.71
[0.69]
-0.84
[0.61]
-0.92
[0.65]
0.12
[0.32]
-0.71
[0.91]
-1.87***
[0.61]
-3.27**
[1.46]
Observations
R-squared
Partial R-squared
Additional controls
Number of add controls
127
0.79
0.01
All controls (with
27
127
127
0.76
0.78
0.01
0.02
latitude & elevation FEs)
27
27
42
0.93
0.01
26
41
111
0.88
0.74
0.03
0.10
All controls (with latitude & elevation FEs)
26
27
41
0.91
0.36
26
Joint significance of the additional control variables (p-values for the H0: all regressors [except UV damage] are jointly insignificant):
All controls
(with lat & elev FEs)
0.00
0.00
0.00
0.00
0.00
0.00
0.00
Notes: OLS regressions. Historical population density and levels of GDP per capita data are from Maddison (2000). UV damage is an index of Erythemal
exposure, constructed from pixel level daily averages of integrated ultraviolet irradiance A and B in 1990 and 2000, weighted by the susceptibility of caucasian
skin to sunburn (erythema), and the proportion of total population in each pixel. It can be interpreted as an index of the potential for biological damage due
to solar irradiation, given the column ozone amount and cloud conditions on each day. Raw UV exposure daily data for 1990 and 2000 produced by NASA. The
R-squared from an OLS regression where all listed additional controls where partialled is reported as the Partial R-squared in each column. All regressions
include a constant term. Robust standard errors in brackets. ***, ** and * denote statistical significance at 1, 5 and 10% levels, respectively.
Table 5
Timing of the fertility decline and exposure to UV radiation
1
2
3
4
Dependent variable:
(log) UV damage
5
6
7
8
9
10
Year of the fertility decline
52.6***
[3.76]
34.4***
[6.29]
53.0***
[5.01]
46.6***
[14.7]
55.9***
[7.17]
53.1***
[4.37]
Additional controls
Number of add controls
119
0.64
-
119
0.77
0.26
119
0.64
0.48
119
0.70
0.09
119
0.66
0.36
119
0.69
0.61
Continent
FEs
Latitude &
elevation
(levels)
Latitude &
elevation
(FEs)
Precipitation &
temperature
Distance to
ocean &
rivers
3
2
14
2
4
12
(log) Real GDP per
capita, 2005
61.7***
[5.64]
34.2***
[10.4]
22.9
[16.1]
119
0.69
0.47
119
0.81
0.08
119
0.83
0.02
119
0.80
0.02
Agricultural
All controls
All controls
All controls
suitability,
(with lat & elev (with lat & elev
(no lat & elev)
Tropical area,
levels)
FEs)
Frost
3
14
26
13
14
(log) Real GDP per
worker, 2005
37.5***
[9.35]
Year of the fertility decline
Observations
R-squared
Partial R-squared
11
12
-0.021***
[0.0056]
-0.020***
[0.0055]
-0.019***
[0.0054]
-0.018***
[0.0054]
119
0.72
0.17
119
0.76
0.16
118
0.70
0.14
118
0.75
0.13
All controls
(with lat & elev
levels)
All controls
(with lat & elev
FEs)
14
26
All controls (with All controls (with
lat & elev levels) lat & elev FEs)
14
26
Joint significance of the additional control variables (p-values for the H0: all regressors [except UV damage] are jointly insignificant):
Continent FEs
0.00
Latitude & elevation (levels)
Latitude & elevation (FEs)
Precipitation & Temperature
Distance to ocean & rivers
Agr. suitability, Tropical area, Frost
All controls (with latitude & elevation levels)
All controls (with latitude & elevation FEs)
All controls (no latitude & elevation controls)
0.78
0.49
0.00
0.35
0.01
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
Notes: OLS regressions. Data for the timing of the fertility decline are from Reher (2004). UV damage is an index of Erythemal exposure, constructed from pixel level daily averages of integrated ultraviolet irradiance A and B in 1990 and 2000,
weighted by the susceptibility of caucasian skin to sunburn (erythema), and the proportion of total population in each pixel. It can be interpreted as an index of the potential for biological damage due to solar irradiation, given the column ozone
amount and cloud conditions on each day. Raw UV exposure daily data for 1990 and 2000 produced by NASA. The R-squared from an OLS regression where all listed additional controls where partialled is reported as the Partial R-squared in each
column. All regressions include a constant term. Robust standard errors in brackets. ***, ** and * denote statistical significance at 1, 5 and 10% levels, respectively.
Table 6
Real GDP per capita/worker, skin cancer, tropically clustered diseases, and exposure to UV radiation
1
2
Dependent variable:
(log) UV damage
(log) Melanoma and other skin cancers
3
4
5
7
8
9
(log) Real GDP per capita, 2004
-1.71***
[0.63]
0.015
[0.13]
(log) Trachoma
-1.65***
[0.62]
-2.02***
[0.65]
-1.33**
[0.63]
-1.39**
[0.61]
-1.23*
[0.63]
-1.46**
[0.62]
0.026
[0.13]
-0.28***
[0.087]
-1.33**
[0.67]
0.013
[0.11]
0.006
[0.050]
0.10**
[0.045]
0.41
[0.31]
-0.21***
[0.077]
-0.61*
[0.31]
147
0.75
0.12
147
0.78
0.23
145
0.72
0.05
145
0.72
0.05
27
27
27
27
0.00
0.00
0.00
0.096**
[0.044]
(log) Hookworm disease
-0.25***
[0.088]
(log) Malaria
-0.20***
[0.070]
(log) Intestinal nematode infections
147
0.73
0.06
147
0.73
0.06
27
27
147
147
147
0.74
0.74
0.75
0.09
0.11
0.13
All controls (with latitude & elevation FEs)
27
27
27
10
11
12
13
14
-1.08*
[0.59]
-0.23**
[0.095]
-1.14*
[0.64]
0.013
[0.11]
0.014
[0.050]
0.089**
[0.043]
0.30
[0.35]
-0.20**
[0.078]
-0.47
[0.35]
145
0.74
0.09
145
0.77
0.20
27
27
0.00
0.00
(log) Real GDP per worker, 2004
-0.037
[0.051]
(log) HIV
Observations
R-squared
Partial R-squared
Additional controls
Number of additional controls
6
-1.44**
[0.61]
-1.70***
[0.63]
-1.17*
[0.60]
-1.17*
[0.60]
-0.012
[0.051]
0.073*
[0.042]
-0.20**
[0.091]
-0.19***
[0.072]
145
145
145
0.73
0.73
0.75
0.07
0.08
0.13
All controls (with latitude & elevation FEs)
27
27
27
Joint significance of the additional control variables (p-values for the H0: all regressors [except UV damage] are jointly insignificant):
All controls (with lat & elev FEs)
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
Notes: OLS regressions. UV damage is an index of Erythemal exposure, constructed from pixel level daily averages of integrated ultraviolet irradiance A and B in 1990 and 2000, weighted by the susceptibility of caucasian skin to sunburn (erythema), and
the proportion of total population in each pixel. It can be interpreted as an index of the potential for biological damage due to solar irradiation, given the column ozone amount and cloud conditions on each day. Raw UV exposure daily data for 1990 and
2000 produced by NASA. The incidence of skin cancer, trachoma, HIV/AIDS, hookworm disease, malaria, and intestinal infections are measured as the number of years lost due to disability, for incident cases of each disease (expressed as a rate per
100,000 people between 15 and 59), estimated by WHO (2004). The R-squared from an OLS regression where all listed additional controls where partialled is reported as the Partial R-squared in each column. All regressions include a constant term.
Robust standard errors in brackets. ***, ** and * denote statistical significance at 1, 5 and 10% levels, respectively.
Table 7
Real GDP per capita and exposure to UV radiation: Pixel level analysis
(log) Real GDP per capita, 2005
Dependent variable:
1x1
Granularity
2x2
4x4
1
2
3
4
5
6
7
8
9
(log) UV damage
-0.55**
[0.26]
-0.16**
[0.063]
-0.15**
[0.059]
-0.57**
[0.24]
-0.17***
[0.064]
-0.17**
[0.068]
-0.58**
[0.23]
-0.19***
[0.072]
-0.24***
[0.078]
Observations
R-squared
Partial R-squared
Additional controls
Number of add controls
Fixed effects
Number of fixed effects
Std errors clustered by:
16,972
16,972
16,972
0.50
0.91
0.92
0.03
0.01
0.005
All controls (with lat & elev levels)
7
7
7
Country
Language
159
1,167
Country
Country
Language
5,381
5,381
5,381
0.51
0.93
0.93
0.03
0.01
0.01
All controls (with lat & elev levels)
7
7
7
Country
Language
159
814
Country
Country
Language
1,930
1,930
1,930
0.52
0.95
0.95
0.03
0.01
0.02
All controls (with lat & elev levels)
7
7
7
Country
Language
159
414
Country
Country
Language
Joint significance of additional controls (p-values for the H0: all regressors [except UV damage] are jointly insignificant):
All controls
(with lat & elev levels)
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
Notes: OLS regressions. Each observation is for a geographic pixel of 1x1, 2x2, or 4x4 degrees of latitude and longitude, respectively. GDP per capita constructed with data from Yale
GECON 3.4 database. UV damage is an index of Erythemal exposure, constructed from pixel level daily averages of integrated ultraviolet irradiance A and B in 1990 and 2000, weighted
by the susceptibility of caucasian skin to sunburn (erythema), and for the case of 2x2 and 4x4 aggreagted pixel analysis, weighted by the proportion of total population in each unit of
aggregation. It can be interpreted as an index of the potential for biological damage due to solar irradiation, given the column ozone amount and cloud conditions on each day. Raw UV
exposure daily data for 1990 and 2000 produced by NASA. The R-squared from an OLS regression where all listed additional controls where partialled is reported as the Partial R-squared
in each column. All regressions include a constant term. Standard errors clustered by country or predominant language area reported in brackets. ***, ** and * denote statistical
significance at 1, 5 and 10% levels, respectively.
Table 8
Lights by night and exposure to UV radiation: Pixel level analysis
(log) Intensity of lights by night, 2004
Dependent variable:
1x1
Granularity
2x2
4x4
1
2
3
4
5
6
7
8
9
(log) UV damage
-0.40***
[0.082]
-0.24**
[0.11]
-0.21**
[0.10]
-0.47***
[0.053]
-0.30***
[0.100]
-0.28***
[0.093]
-0.49***
[0.055]
-0.29***
[0.10]
-0.29***
[0.11]
Observations
R-squared
Partial R-squared
Additional controls
Number of add controls
Fixed effects
Number of fixed effects
Std errors clustered by:
18,288
18,288
18,288
0.29
0.38
0.43
0.04
0.01
0.01
All controls (with lat & elev levels)
7
7
7
Country
Language
184
1,222
Country
Country
Language
5,670
5,670
5,670
0.39
0.53
0.57
0.08
0.02
0.02
All controls (with lat & elev levels)
7
7
7
Country
Language
184
861
Country
Country
Language
2,050
2,050
2,050
0.40
0.63
0.66
0.08
0.03
0.02
All controls (with lat & elev levels)
7
7
7
Country
Language
184
448
Country
Country
Language
Joint significance of additional controls (p-values for the H0: all regressors [except UV damage] are jointly insignificant):
All controls
(with lat & elev levels)
0.00
0.00
0.00
0.00
0.45
0.13
0.00
0.28
0.31
Notes: OLS regressions. Each observation is for a geographic pixel of 1x1, 2x2, or 4x4 degrees of latitude and longitude, respectively. The dependent variable is the raw average of the
visible band (digital number) of the radiance of lights at night in 2004, produced by the Defense Meteorological Satellite Program (DMSP) at NASA. UV damage is an index of Erythemal
exposure, constructed from pixel level daily averages of integrated ultraviolet irradiance A and B in 1990 and 2000, weighted by the susceptibility of caucasian skin to sunburn
(erythema), and for the case of 2x2 and 4x4 aggreagted pixel analysis, weighted by the proportion of total population in each unit of aggregation. It can be interpreted as an index of the
potential for biological damage due to solar irradiation, given the column ozone amount and cloud conditions on each day. Raw UV exposure daily data for 1990 and 2000 produced by
NASA. The R-squared from an OLS regression where all listed additional controls where partialled is reported as the Partial R-squared in each column. All regressions include a constant
term. Standard errors clustered by country or predominant language area reported in brackets. ***, ** and * denote statistical significance at 1, 5 and 10% levels, respectively.
Table 9
Lights by night and exposure to UV radiation: Pixel level analysis in China and USA
1
2
3
4
5
1x1
Granularity
(log) UV damage
Observations
R-squared
Partial R-squared
Additional controls
Number of add controls
Fixed effects
Number of fixed effects
Standard errors:
Robust/clustered by:
7
8
9
10
11
12
(log) Intensity of lights by night, 2004
Dependent variable:
Country
6
2x2
4x4
USA
China
USA
China
USA
China
USA
China
USA
China
USA
China
-1.46***
[0.22]
0.042
[0.082]
-1.42***
[0.013]
0.039
[0.097]
-1.51***
[0.28]
0.078
[0.11]
-1.57***
[0.014]
0.068
[0.071]
-1.45***
[0.34]
0.18
[0.17]
-1.46***
[0.0014]
0.15
[0.14]
1,079
0.20
0.00
362
0.30
0.07
303
0.43
0.001
116
0.45
0.12
7
Language
46
7
-
7
Language
36
7
-
Language
Robust
Language
Language
Robust
Robust
Language
Language
0.00
0.00
0.00
0.00
0.00
0.00
1,243
0.16
0.03
7
-
Robust
1,079
1,243
0.15
0.17
0.00
0.03
All controls (with lat & elev levels)
7
7
Language
21
Robust
Language
303
362
0.31
0.31
0.002
0.07
All controls (with lat & elev levels)
7
7
Language
20
Robust
89
116
0.55
0.46
0.01
0.11
All controls (with lat & elev levels)
7
7
Language
5
89
0.64
0.01
7
Language
23
Joint significance of the additional control variables (p-values for the H0: all regressors [except UV damage] are jointly insignificant):
All controls
(with lat & elev levels)
0.00
0.00
0.00
0.00
0.00
0.00
Notes: OLS regressions. Each observation is for a geographic pixel of 1x1, 2x2, or 4x4 degrees of latitude and longitude, respectively. The dependent variable is the raw average of the visible band (digital number) of the radiance of lights at night in
2004, produced by the Defense Meteorological Satellite Program (DMSP) at NASA. UV damage is an index of Erythemal exposure, constructed from pixel level daily averages of integrated ultraviolet irradiance A and B in 1990 and 2000, weighted by
the susceptibility of caucasian skin to sunburn (erythema), and for the case of 2x2 and 4x4 aggreagted pixel analysis, weighted by the proportion of total population in each unit of aggregation. It can be interpreted as an index of the potential for
biological damage due to solar irradiation, given the column ozone amount and cloud conditions on each day. Raw UV exposure daily data for 1990 and 2000 produced by NASA. The R-squared from an OLS regression where all listed additional
controls where partialled is reported as the Partial R-squared in each column. All regressions include a constant term. Robust/clustered by country or predominant language area standard errors in brackets. ***, ** and * denote statistical
significance at 1, 5 and 10% levels, respectively.
Table A1: Panel A
Summary statistics: Cross country data
N obs.
Mean
St dev
Median
Min
Max
145
147
147
147
147
147
147
147
147
145
147
147
147
147
147
147
147
147
147
23,527.5
10,848.1
192.2
0.31
0.27
0.24
0.02
0.17
20.7
1.55
1.00
17.7
890,000
0.43
1.00
4.86
0.39
0.36
9.4
23,851.8
12,366.1
77.7
15,409.0
5,856.0
204.9
25.0
0.99
0.71
8.3
2,150,000
0.46
1.11
2.40
0.24
0.43
10.4
20.3
1.38
0.83
21.4
251,000
0.28
0.68
4.10
0.38
0.01
3.4
934.1
353.7
42.7
0
0
0
0
0
-41.8
0.03
0.03
-4.6
2,635
0.02
0.02
0.40
0.00
0.00
0.0
107,000.0
68,390.4
328.5
1
1
1
1
1
64.5
4.31
3.31
28.9
17,200,000
2.37
9.41
10.50
0.95
1.00
29.8
121
72
72
72
72
147
147
147
147
147
147
1966
1.4
7.5
8.1
7.8
5.4
5.0
4.4
3.7
3.1
4.3
30
1.8
3.5
3.4
3.4
1.9
2.0
2.0
1.9
1.6
1.8
1975
0.8
7.3
8.1
7.7
6.1
5.7
4.5
3.3
2.6
4.6
1865
0.0
1.0
1.3
1.1
1.8
1.6
1.4
1.2
1.2
1.6
2000
6.2
13.1
13.6
13.3
8.1
8.2
9.1
8.0
7.3
7.7
147
147
147
147
147
147
147
347
19
28
3,282
17
78
17
307
17
75
8,315
20
130
20
253
13
0
174
5
2
5
8
0
0
1
0
0
0
949
70
440
60,288
67
512
68
42
41
111
145
140
127
813.2
2,002.7
2,582.5
6.3
3.6
2.6
356.0
1,167.7
3,595.2
8.9
4.4
3.7
696.0
1,729.0
1,259.0
2.4
1.2
0.8
397.0
533.0
289.0
0
0
0
1,838.0
4,492.0
28,878.0
46.6
19.9
23.8
147
147
147
147
147
147
147
147
147
147
147
147
147
147
147
147
120
147
147
147
145
3.73
3.59
-0.15
0.10
0.30
0.16
0.11
0.01
0.05
0.27
0.27
0.17
0.12
0.31
0.05
0.09
17.65
7,600.2
8,515
106,000
0.35
6.77
2.14
1.00
0.12
3.00
-0.33
27.89
35,313.7
6,902
401,000
0.42
4.63
0.0
5,645
0
0.03
0.00
1.00
-2.20
0
0
0
0
0
0
0
0
0
0
0
0
0
0.00
0.0
0
0
0
31.55
7.00
2.02
1
1
1
1
1
1
1
1
1
1
1
1
1
99.64
264,000.0
26,836
3,610,000
1
Main variables
Real GDP per worker, 2004
Real GDP per capita, 2004
Exposure to UV, av per sqm, 1990 and 2000, pop weighted. (NASA)
Continent indicator: Africa
Continent indicator: Asia
Continent indicator: Europe
Continent indicator: Oceania
Continent indicator: America
Latitude - Nunn and Puga (2010)
Average (max min) elev m above sea level (Source: Nationmaster)
Average area-weighted precipitation ['000 mm] (Source: GECON 3.4)
Average area-weighted temperature [C degrees] (Source: GECON 3.4)
Total area [sq km] (Source: GECON 3.4)
Mean distance to coast
Mean distance to river
Time passed since Neolithic revolution ('000 years)
Agricultural Suitability Index - Ashraf and Galor (2011)
% of Land in Tropical and Subtropical Zones - Ashraf and Galor (2011)
Average number of frost days (area-weighted frost-days)
Fertilty and education
Year of fertility decline - Rehrer (2004)
Average scholing in 1870 - Morrison and Murtin (2009)
Average scholing in 2000 - Morrison and Murtin (2009)
Average scholing in 2010 - Morrison and Murtin (2009)
Average scholing in 2000 and 2010
Fertility rate 1960's, total (births per woman) (WDI)
Fertility rate 1970's, total (births per woman) (WDI)
Fertility rate 1980's, total (births per woman) (WDI)
Fertility rate 1990's, total (births per woman) (WDI)
Fertility rate 2000's, total (births per woman) (WDI)
Average fertility 1960-2000
Years (per 100000 people between 15 and 59, 2004) lost due to disability from the incidence of:
Cataracts
Melanoma and other skin cancers
Trachoma
HIV/AIDS
Hookworm disease
Malaria
Intestinal nematode infections
Measures of historical prosperity
Est GDP/cap 1820, Maddison (2000)
Est GDP/cap 1900, Maddison (2000)
Est GDP/cap 1950, Maddison (2000)
Population density year 1500
Population density year 1000
Population density year 1
Indicators of institutions, natural resources, trust, culture:
Malaria Ecology, pop-weighted, Sept 2003 version
Freedom House rating of political rights, 2002
Rule of law 1996-2000 - Nunn and Puga (2010)
WB region indicator: East Asia and Pacific
WB region indicator: Europe and Central Asia
WB region indicator: Latin America and Caribbean
WB region indicator: Middle East and North Africa
WB region indicator: North America
WB region indicator: South Asia
WB region indicator: Sub-Saharan Africa
WHO region indicator: Africa
WHO region indicator: Americas
WHO region indicator: Eastern Mediterranean
WHO region indicator: Europe
WHO region indicator: South East Asia
WHO region indicator: Western Pacific
% of mineral fuels in manufacturing exports, 2000, WDI
Gem diamond extraction 1958-2000 (1000 carats) - Nunn and Puga (2010)
Migratory distance from Ethiopia - Olsson and Ahlerup (2009)
Slave exports 1400-1900 - Nunn and Puga (2010)
Fraction of population of Euro desent (Weil/Putterman)
Notes. Data sources and definitions in Appendix 1.
Table A1: Panel B
Summary statistics: Pixel data
N obs
Mean
St dev
Median
Min
Max
16,972
18,288
18,960
18,960
18,960
18,960
18,960
18,960
13.9
3.5
149.0
31.6
692.4
6,995.7
0.78
1.74
26.3
4.9
95.0
31.8
805.2
3,696.0
0.69
1.29
8.1
2.0
140.4
38.0
414.8
7,088.8
0.61
1.56
0
1
8.5
-56.0
4.0
2.2
0
0
2,344.8
63.0
428.6
83.0
6,350.0
12,393.2
2.98
9.99
5,381
5,752
5,969
5,969
5,969
5,969
5,969
5,969
13.6
3.5
152.1
30.3
663.0
22,380.1
0.73
1.74
23.8
3.6
96.0
31.7
770.9
15,521.1
0.68
1.33
7.6
2.3
150.4
36.5
401.9
21,330.3
0.53
1.54
0
1
0.0
-55.5
21.0
5.0
0
0
1,317.9
62.0
410.8
82.5
5,393.4
49,430.3
2.95
9.99
1,930
2,064
2,128
2,128
2,128
2,128
2,128
2,128
12.9
3.6
159.1
28.4
638.9
62,794.6
0.65
1.74
15.3
3.6
94.4
31.5
721.7
57,677.5
0.67
1.40
6.9
2.3
165.2
32.3
406.1
45,282.9
0.42
1.52
0
286.1
1
57.0
0.0
400.0
-55.5
82.5
22.4
5,250.0
5.0 198,000.0
0
2.90
0
9.99
1x1 degree
Gross cell product per capita, 2005 ('000 USD)
Lights digital number, 2004 (F152004. Source: DMSP NASA)
Exposure to UV radiation - av 1990, 2000 (Source: NASA)
Latitude (degrees)
Elevation (m above sea level)
Area of grid cell (sq km)
Distance to ice-free ocean (km)
Distance to major navigable river (km)
2x2 degrees
Gross cell product per capita, 2005 ('000 USD)
Lights digital number, 2004 (F152004. Source: DMSP NASA)
Exposure to UV radiation - pop weighted av 1990, 2000 (Source: NASA)
Latitude (degrees)
Elevation (m above sea level)
Area of grid cell (sq km)
Distance to ice-free ocean (km)
Distance to major navigable river (km)
4x4 degrees
Gross cell product per capita, 2005 ('000 USD)
Lights digital number, 2004 (F152004. Source: DMSP NASA)
Exposure to UV radiation - pop weighted av 1990, 2000 (Source: NASA)
Latitude (degrees)
Elevation (m above sea level)
Area of grid cell (sq km)
Distance to ice-free ocean (km)
Distance to major navigable river (km)
Notes. Data sources and definitions in Appendix 1.
Table A2
Correlates of exposure to UV radiation - Cross country data
1
2
3
4
Dependent variable:
1[Continent = Africa]
1[Continent = Asia]
1[Continent = Europe]
1[Continent = Oceania]
1[Continent = America]
0.33*
[0.17]
0.049
[0.18]
-0.81***
[0.18]
dropped
..
0.25
[0.18]
-0.37***
[0.044]
0.100**
[0.043]
(log) Elevation ('000 m)
a
426.01
(0.00)
Elevation FEs
a
0.47***
[0.066]
0.41***
[0.047]
dropped
..
0.42***
[0.14]
0.49***
[0.061]
-0.057***
[0.015]
0.13***
[0.021]
0.18
[0.15]
0.16
[0.13]
-0.10
[0.15]
dropped
..
0.20
[0.14]
7.09
a
15.45
a
(0.00)
7.09
a
a
a
a
0.000012
[0.000020]
0.053***
[0.0036]
(average 1990-2008) Temperature (C degrees)
Distance to coast (km)
-0.14
[0.10]
0.069
[0.069]
0.04
[0.026]
-0.066***
[0.013]
Distance to rivers (km)
(log) Country area (sq km)
Year of Neolithic Transition ('000 years)
(log) Agricultural suitability index
-0.045*
[0.024]
0.20**
[0.081]
-0.036***
[0.0041]
KG Tropical/subtropical region
(average annual 1901-2012) Frost days
Number of add controls
8
a
(average 1990-2008) Precipitation ('000 mm)
Controls
7
a
(0.00)
Observations
R-squared
6
Exposure to UV radiation
(log) Latitude
Latitude FEs
5
(0.00)
-0.000050* -0.00004
[0.000027] [0.000027]
0.034***
0.0089
[0.0050]
[0.0056]
0.06
-0.0024
[0.039]
[0.034]
-0.0056
0.021
[0.017]
[0.015]
-0.029***
-0.0071
[0.0083]
[0.0077]
0.0042
0.0083
[0.0089]
[0.0091]
0.021*
0.016
[0.012]
[0.013]
0.00014
-0.063*
[0.053]
[0.037]
-0.0034
-0.0078**
[0.0039]
[0.0034]
147
0.75
145
0.56
147
0.95
147
0.75
147
0.16
147
0.73
145
0.95
147
0.97
Continent FEs
Latitude &
elevation
(levels)
Latitude &
elevation
(FEs)
Precipitation
&
temperature
Distance to
ocean &
rivers
Agricultural
suitability,
Tropical
area, Frost
All controls
(with lat &
elev levels)
All controls
(with lat &
elev FEs)
4
2
14
2
4
3
15
27
Joint significance of the additional control variables (p-values for the H0: all regressors are jointly insignificant):
Continent FEs
0.00
Latitude & elevation (levels)
Latitude & elevation (FEs)
Precipitation & Temperature
Distance to ocean & rivers
Agr. suitability, Tropical area, Frost
All controls (with latitude & elevation levels)
All controls (with latitude & elevation FEs)
0.00
0.00
0.00
0.00
0.00
0.00
0.00
Notes: OLS regressions. UV damage is an index of Erythemal exposure, constructed from pixel level daily averages of integrated ultraviolet
irradiance A and B in 1990 and 2000, weighted by the susceptibility of caucasian skin to sunburn (erythema), and the proportion of total
population in each pixel. It can be interpreted as an index of the potential for biological damage due to solar irradiation, given the column
ozone amount and cloud conditions on each day. Raw UV exposure daily data for 1990 and 2000 produced by NASA. All regressions
include a constant term. Robust standard errors in brackets. ***, ** and * denote statistical significance at 1, 5 and 10% levels,
respectively.
a
: F-statistic (and p-value in parentheses) for the joint significance of latitude and elevation FEs reported in columns 3 and 8.
Table A3
Correlates of exposure to UV radiation - Pixel data
Exposure to UV radiation
Dependent variable:
1x1
Granularity
1
(log) Latitude
(log) Elevation ('000 m)
2
(log) Pixel area (sq km)
Distance to ocean (km)
Distance to major rivers (km)
Number of controls
4
5
-0.66***
[0.012]
0.22***
[0.0081]
0.20***
[0.0076]
-0.50***
[0.0072]
-0.043***
[0.0034]
-0.11***
[0.0037]
0.29***
[0.0030]
0.050***
[0.00026]
-0.036***
[0.0032]
0.020***
[0.0018]
0.034***
[0.0033]
-0.011***
[0.0014]
0.052***
[0.00022]
-0.028***
[0.0029]
(average 1990-2008) Precipitation
Controls
3
-0.72***
[0.0072]
0.19***
[0.0044]
(average 1990-2008) Temperature
Observations
R-squared
2x2
6
4x4
7
8
9
-0.64***
[0.018]
0.20***
[0.014]
0.16***
[0.010]
-0.51***
[0.013]
-0.035***
[0.0058]
-0.071***
[0.0053]
0.27***
[0.0058]
0.054***
[0.00039]
-0.026***
[0.0050]
0.020***
[0.0028]
0.054***
[0.0055]
-0.014***
[0.0022]
0.056***
[0.00030]
-0.042***
[0.0045]
10
11
12
0.14***
[0.017]
-0.52***
[0.024]
-0.032***
[0.0092]
-0.060***
[0.0082]
0.27***
[0.0094]
0.055***
[0.00063]
-0.018**
[0.0078]
0.024***
[0.0043]
0.059***
[0.0096]
-0.011***
[0.0031]
0.056***
[0.00051]
-0.041***
[0.0068]
18,960
0.61
18,960
0.82
18,960
0.25
18,960
0.92
5,849
0.59
5,849
0.84
5,849
0.23
5,849
0.93
2,113
0.58
2,113
0.85
2,113
0.22
2,113
0.93
Latitude &
elevation
levels
Precipitation &
temperature
Area, distance
to ocean &
major rivers
All controls
Latitude &
elevation
levels
Precipitation &
temperature
Area, distance
to ocean &
major rivers
All controls
Latitude &
elevation
levels
Precipitation &
temperature
Area, distance
to ocean &
major rivers
All controls
2
2
3
7
2
2
3
7
2
2
3
7
Joint significance of the additional control variables (p-values for the H0: all regressors are jointly insignificant):
Latitude & elevation levels
0.00
Precipitation & temperature
Area, distance to ocean & major rivers
All controls
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
Notes: OLS regressions. Each observation is for a geographic pixel of 1x1, 2x2, or 4x4 degrees of latitude and longitude, respectively. UV damage is an index of Erythemal exposure, constructed from pixel level daily averages of
integrated ultraviolet irradiance A and B in 1990 and 2000, weighted by the susceptibility of caucasian skin to sunburn (erythema), and for the case of 2x2 and 4x4 aggreagted pixel analysis, weighted by the proportion of total
population in each unit of aggregation. It can be interpreted as an index of the potential for biological damage due to solar irradiation, given the column ozone amount and cloud conditions on each day. Raw UV exposure daily
data for 1990 and 2000 produced by NASA. All regressions include a constant term. Robust standard errors in brackets. ***, ** and * denote statistical significance at 1, 5 and 10% levels, respectively.
Table A4
Correlates of exposure to UV radiation - Pixel data (including country and language FEs)
Dependent variable:
Exposure to UV radiation
Granularity
(log) Latitude
(log) Elevation ('000 m)
(average 1990-2008) Temperature
(average 1990-2008) Precipitation
(log) Pixel area (sq km)
Distance to ocean (km)
Distance to major rivers (km)
Observations
R-squared
Controls
Number of controls
Fixed effects
Number of fixed effects
Std errors clustered by:
1x1
2x2
4x4
1
2
3
4
5
6
-0.15**
[0.056]
0.22***
[0.027]
0.034***
[0.0050]
-0.068***
[0.017]
0.0083**
[0.0041]
0.058
[0.035]
-0.14***
[0.036]
-0.15**
[0.061]
0.20***
[0.021]
0.033***
[0.0063]
-0.071***
[0.019]
0.015**
[0.0061]
0.060*
[0.034]
-0.16***
[0.033]
-0.11**
[0.044]
0.22***
[0.027]
0.040***
[0.0034]
-0.076***
[0.016]
0.0059**
[0.0028]
0.060*
[0.032]
-0.13***
[0.038]
-0.11**
[0.043]
0.20***
[0.023]
0.041***
[0.0035]
-0.078***
[0.016]
0.013***
[0.0043]
0.064**
[0.029]
-0.13***
[0.034]
-0.12***
[0.043]
0.22***
[0.027]
0.043***
[0.0037]
-0.082***
[0.018]
0.0093**
[0.0037]
0.043
[0.036]
-0.11***
[0.040]
-0.12***
[0.045]
0.21***
[0.024]
0.043***
[0.0041]
-0.091***
[0.019]
0.014***
[0.0052]
0.046
[0.034]
-0.12***
[0.036]
18,960
0.95
18,960
0.96
5,849
0.96
5,849
0.97
2,113
0.96
2,113
0.97
All controls
All controls
All controls
All controls
All controls
All controls
7
Country
185
Country
7
Language
1,223
Language
7
Country
185
Country
7
Language
862
Language
7
Country
185
Country
7
Language
449
Language
0.00
0.00
Joint significance of the additional control variables (p-values for the H0: all regressors are jointly insignificant):
All controls
0.00
0.00
0.00
0.00
Notes: OLS regressions. Each observation is for a geographic pixel of 1x1, 2x2, or 4x4 degrees of latitude and longitude, respectively. UV
damage is an index of Erythemal exposure, constructed from pixel level daily averages of integrated ultraviolet irradiance A and B in 1990 and
2000, weighted by the susceptibility of caucasian skin to sunburn (erythema), and for the case of 2x2 and 4x4 aggreagted pixel analysis,
weighted by the proportion of total population in each unit of aggregation. It can be interpreted as an index of the potential for biological
damage due to solar irradiation, given the column ozone amount and cloud conditions on each day. Raw UV exposure daily data for 1990 and
2000 produced by NASA. All regressions include a constant term. Standard errors clustered by country or predominant language area are
reported in brackets, respectively in each column. ***, ** and * denote statistical significance at 1, 5 and 10% levels, respectively.