The Effects of Climate on Output per Worker: Evidence from the Manufacturing Industry in Colombia
| Autor | Mateo Salazar |
| Páginas | 55-89 |
55
DESARRO. SOC. 71, PRIMER SEMESTRE DE 2013, PP. X-XX, ISSN 0120-3584
Revista
Desarrollo y Sociedad
79
Segundo semestre 2017
PP. 55-89, ISSN 0120-3584
E-ISSN 1900-7760
The Effects of Climate on Output per Worker:
Evidence from the Manufacturing Industry
in Colombia
Los efectos del clima en la productividad de
los trabajadores: evidencia de la industria
manufacturera colombiana
Mateo Salazar1
DOI: 10.13043/DYS.79.2
Abstract
This paper quantifies the effect of an increase in temperature and precipitation
on the average output per worker in the Colombian manufacturing industry.
In order to address this issue with rigor, a methodology has been developed
using a theoretical model and an empirical estimation. The estimation of the
empirical model was made with economic data from the Annual Survey of
the Manufacturing Industry, the Monthly Manufacturing Sample and climate
data from IDEAM. The results show evidence that temperature (- 0.3 % / + 1 %)
has a negative effect and precipitation (+ 0.03 % / + 1 %) has a positive effect
1 Contact information: London School of Economics (email: m.salazar-rodriguez@lse.ac.uk). I am grateful
to Hernán Vallejo for his valuable comments and his advice on this project. I also want to thank Román
David Zárate and Román Andrés Zárate. Finally, would like to thank Adriana Camacho and Ana María
Ibáñez for providing the data used in the empirical exercises.
Este artículo fue recibido el 3 de agosto del 2016, revisado el 13 de septiembre del 2016 y finalmente
aceptado el 12 de junio del 2017.
The Effects of Climate on Output per Worker
56
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
on average output per worker. The results build on previous literature argu-
ing that worker's productivity is a channel through which climate and climate
change affect economic performance.
Key words: Climate change, ergonomics, productivity.
JEL classification: O44, Q54, J81.
Resumen
Este artículo cuantifica el efecto de un aumento de temperatura y precipita-
ción sobre la productividad de los trabajadores en la industria manufacturera
colombiana. La metodología se basa en un modelo teórico y una estimación
empírica. La estimación del modelo empírico se realiza con datos económicos
de la Encuesta Anual Manufacturera, la Muestra Mensual Manufacturera,
mientras que los datos climáticos provienen del Ideam. Los resultados mues-
tran un efecto negativo de la temperatura (- 0,3 % / + 1 %) y un efecto positivo
de la precipitación (+ 0,03 % / + 1 %) en la productividad de los trabajadores.
Los resultados se basan en la literatura que sostiene que la productividad
laboral es un canal a través del cual el clima y el cambio climático afectan el
desempeño económico.
Palabras clave: cambio climático, ergonomía, productividad.
Clasificación JEL: O44, Q54, J81.
Mateo Salazar 57
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
Introduction
It is a classical problem in economics to understand what drives and con-
straints economic development (Ramsey, 1928; Smith, 1776; Solow, 1956).
Many theories have been developed to approach this issue, but there is still
a debate between the two most popular lines of research: geographical and
institutional (Acemoglu, Johnson & Robinson, 2002; Rodrik, Subramanian &
Trebbi, 2002; Sachs, 2003). Closely related to both of these approaches are
Hall and Jones´ findings, which state that differences in capital accumulation,
productivity and worker productivity are closely related to differences in social
infrastructure2 (Hall & Jones, 1999). This paper uses quarterly municipal data
for Colombia to look for evidence to support the hypothesis that a healthy
environment makes up part of that social infrastructure, in this case through
sustained moderate climate. This paper will evaluate the impact of a healthy
environment on the manufacturing industry specifically because the losses
produced by changes in temperature are 29 times larger for sectors not related
to agriculture (Hsiang, 2010).
There is still debate about the exact impact the environment has on the econ-
omy and its relevance for public policy (Arrow, 2004; Daly, 1996; Dell, Jones &
Olken, 2012; Stern, 2006; Tol, 2009). In this context, it is important to develop
methodologies that provide precise and trustworthy results since “climate
change is the mother of all externalities: larger, more complex, and more uncer-
tain than any other environmental problem” (Tol, 2009). Assessing its economic
impacts is a very relevant matter in the literature on economic development.
Specifically, in terms of Colombian public policy, the climate change issue has
been becoming ever more relevant in Colombia. The Council of Economic and
Social Policy (CONPES) presented its official climate change document on July 14,
2011 (CONPES, 2011). The aim of CONPES is to establish an institutional arrange-
ment to articulate a strategy between sectors in order to facilitate and enhance
the formulation and implementation of policies, plans, programs, methodologies,
incentives and projects on climate change, including climate as the main vari-
able in the design and planning of development projects (Cadena et al., 2012).
2 Social infrastructure is defined as “the institutions and government policies that determine the economic
environment within which individuals accumulate skills, and firms accumulate capital and produce
output”’ (Hall & Jones, 1999).
The Effects of Climate on Output per Worker
58
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
The council intends to enhance mainly four strategies changing the way the
country understands climate change and sustainable development in general.
These four strategies are:
1. The National Adaptation Plan
2. The Low Carbon Development Strategy
3. The National Strategy for Emissions reduction due to Deforestation
and Forest Degradation
4. The Strategy for Financial Protection against Disasters
It is necessary for the country’s productive force in all municipalities to imple-
ment adaptation and mitigation actions without affecting the productive sectors
driving long-term growth in the Colombian economy. This study presents rigorous
evidence regarding another impact that climate change has on economic per-
formance, and, at the same time, it is an opportunity for the productive sector
to adapt to the imminent impending raise in temperature. This study quantifies
an additional impact that is also an incentive for industry to mitigate carbon
emissions in order to maximize their benefits in the long-run.
The relationship between climate and economic activity has traditionally been
approached using two kinds of models. The first group of models has studied
the impact of average temperature on aggregate economic variables using
cross-sections (Gallup, Sachs & Mellinger, 1999; Nordhaus, 2006; Sachs &
Warner, 1997). One good example is Dell et al. (2009) who found evidence of
national income falling 8.5% per degree Celsius in a world cross-section (Dell
et al., 2009). Nevertheless, other scholars argue that these results are driven by
associations of temperature and other national characteristics, which means
that the estimations are biased (Acemoglu et al., 2002; Rodrik et al., 2002).
The second group of models looks for micro climatic effects that, when com-
bined, have an effect on aggregate national income. These models are more
rigorous in terms of internal consistency, but the main critique of them is
the complexity of measuring all possible correlations. The set of candidate
Mateo Salazar 59
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
mechanisms through which temperature affects economic outcomes is very
large, and quantifying every single one is virtually impossible. A recent study
undertaken by Dell et al. (2009) describes a wide variety of potential channels
through which climate affects economic performance: agricultural produc-
tivity, health (Graff & Neidell, 2013), physical performance, cognitive per-
formance (Graff, J., Hsiang, S., & Neidell, M., 2015), crime and social unrest
(Burke, Hsiang, & Miguel, 2015); however, many of these are not measured
by quantitative models (Dell et al., 2012). In this study, the main result is that
production decreases by 1.1 % for every degree Celsius that the temperature
increases (-1.1 % / +1°C). For exports, the relationship varies from -2.0 % / +1°C
to -5.7 % / +1°C. The authors obtained these results for a large set of hetero-
geneous countries without quantifying the impact of productivity per worker
(Dell et al., 2013).
Such large variations in GDP cannot be only explained by agriculture (Hsiang,
2010). The main result of Hsiang’s paper shows that losses produced by changes
in temperature are 29 times larger for non-agro sectors than for the agro sector.
Even though that paper mentions ergonomics as a possible link between cli-
mate and GDP, the dependent variable is historic production. This means that
worker output is not measured directly as it is in the present paper.
The present article studies cognitive and physical performance are the two main
drivers of productivity shifts in the industry sector (Somanathan, Somanathan,
Sudardhan & Tewari, 2015). In the case of Colombia, the impact of climate on
output per worker has only been measured for small samples in very specific
areas. Studying the entire manufacturing industry was not considered because
isolating all the individual factors is very challenging. This research takes advan-
tage of the fact that the country has no discernible seasons, which provides an
opportunity to treat monthly climatic changes as natural experiments. This can
be done because they are largely unexpected, unlike in countries with seasons in
which temperature fluctuations are predictable. Additionally, the absence of a
winter season with extreme temperatures, has historically created few incentives
to develop infrastructure in a country where climate is not an overarching
issue. In the long-term, this lack of infrastructure will be a problem for climate
change adaptation policy.
The Effects of Climate on Output per Worker
60
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
In this study, we develop a theoretical and empirical methodology to evaluate
ergonomics3 as a channel through which climate impacts the average output
per worker4. The theoretical framework is based on the Y = AK -type5 of produc-
tion function that depends on temperature. The conclusion of the theoretical
model is that the impact temperature has on the average worker output needs
to be analysed in terms of both level and growth. Based on this, the estimation
strategy uses temperature and precipitation data for each municipality and for
each quarter in Colombia from 2000 to 2004 from the Institute of Hydrology,
Meteorology and Environmental Studies (IDEAM) and the Monthly Manufac-
turing Sample (MMS), which provides data on production and employment
for the same years. The study empirically determines that the average worker
output (levels and growth) are statistically dependent on intra-annual varia-
tions in local temperature. The estimation framework´s key characteristic is
that it uses industry-municipality and time fixed effects to only examine the
dynamic variations, which reduces potential sources of endogeneity.
There is evidence that suggests a correlation between temperature and income
in Colombia. This evidence is shown in Figure 1. In the map, one bar is worker
output and the other is temperature. Most departments have one big bar
together with a small one. Figure 2 illustrates a positive correlation between
the competitiveness ranking6 and temperature (Sánchez & Acosta, 2001). This
means that the departments that are ranked within the top five in Colombia
(including Bogotá D. C., Valle, Antioquia) tend to have lower temperatures
compared to Chocó, Córdoba and Sucre that are ranked last. Even if these are
only correlations that do not represent causality, they suggest a strong effect.
3 Ergonomics studies the design and arrangement of things people use in order to make their interaction
safer and more efficient.
4 The ergonomics of thermal stress on humans has been well studied, and there is a lot of literature on
this topic. Laboratory experiments show that when the temperature is higher than 26.62°C WGBT and
lower than 18.29°C productivity drops. The WGBT is a composite temperature used to estimate the
effect that temperature, humidity, wind speed and solar radiation has on humans. (U.S. Army Technical
Bulletin Medical 507/Air Force Pamphlet 48-152) (Pilcher, Nadler & Bush, 2002). This is evidence of
the non-linearity that is mentioned above.
5 The motivation for choosing this specific type of model is that the study focuses on the impact climate
has on workers output. The manufacturing industry was chosen because the climate will not affect the
output of capital, which is the other factor that is taken into account in traditional production functions.
6 The competitiveness ranking is based on an index that is calculated to measure all the departments in
Colombia´s level of economic performance.
Mateo Salazar 61
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
Figure 1. Correlation Between Temperature and Average Production per Worker
in Colombia.
Legend
2.5
Standarized average temperature
Standarized average production per worker
Source: DANE (2004) and IDEAM (2011). The shape to construct the map was provided by Agustin Codazzi.
The estimates report large, generally negative effects of higher temperatures on
a worker’s productivity. Changes in precipitation have relatively mild effects
on national growth, but is still important to include them because they affect
The Effects of Climate on Output per Worker
62
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
both temperature and income. This paper finds consistent results across a wide
range of specifications, including a robustness check.
Figure 2. Temperature- Competitiveness Ranking
1 to 5 6 to 10 11 to 15 16 to 20 21 to 23
27
25
23
21
19
17
15
Competitiveness ranking
Average temperature (C)
Note: Departments are the unit of analysis in this graph. The bar labelled “1 to 5”’, represents the average
annual temperature in the top-five ranked departments in 2001. The departments with the lowest com-
petitiveness ranking are Chóco, Córdoba and Sucre, the highest are Bogotá, Valle and Antioquia.
Source: Sánchez and Acosta (2001) and IDEAM (2011).
After this introductory section, the production function model will be pre-
sented, and the section will conclude with the level-growth theory. Section 2
describes the data sources, the methodology for constructing the indicators
used in the empirical exercises and the descriptive statistics of the merged data
set. Section 3 presents the empirical framework used to measure the effect of
temperature on the output per worker, the presentation of the results and also
an alternative model that is used as robustness check. Finally, conclusions are
presented together with some policy suggestions.
I. Temperature in the Production Function:
A Theoretical Model
This section analyses the channels through which temperature affects the
current output per worker and the growth of output per worker over time in
the manufacturing industry. In both cases, thermal stress having an impact
Mateo Salazar 63
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
on a workers’ performance seems reasonable.7 In the long run, the creation
of a working culture could be affected by the effect of thermal stress in the
short run.8
Contemplating the impact of temperature on the growth of average output per
worker in the long-run makes climate change relevant. If temperature shifts
have an impact on current economic performance and economic growth, climate
change will, in turn, affect these variables by affecting temperature patterns.
The future implications of climate change are very difficult to estimate and the
empirical scope of this study is historical and short-term. Notwithstanding,
this section explores a theory using which this long-term phenomenon could
affect economic performance; however, it still takes into consideration uncer-
tainties about the extent and nature of climate change. The analysis is based
on four main effects that need to be controlled in order to make consistent
long-term conclusions. First, it is possible that countries adapt to permanent
changes in climate. Second, as climate change becomes a global issue, it may
affect sea-levels, biodiversity and frequency of extreme climatic events that
could simultaneously impact economic variables. Third, there is a big chance
that mandatory mitigation actions will be implemented, and they may distort
economic performance. Fourth, convergence forces may offset the impact of
climate on the economy, especially within poor countries (Solow, 1956). The
compounded effect of these four factors can turn the small effects found in
this paper into long-term issues with major consequences.
The theoretical framework is a modification of the model presented in Dell et al.
(2009) that develops a long-term mathematical relationship between tem-
perature and average output per worker. The modification consists of using
a specific production function in order to theoretically justify the empirical
7 “Three of six non-agricultural industries suffer large and robust reductions in annual output that are
dominated by temperatures experienced during the hottest season and are non-linear in temperatures
during that season. The magnitude, structure and coherence of these responses support the hypothesis
that the underlying mechanism is a reduction in the productivity of human labor when workers are
exposed to thermal stress” (Hsiang, 2010).
8 The Pygmalion effect refers to the phenomenon that the higher the expectation placed upon a person
the better that person performs (Rosenthal & Jacobson, 1992); Temperature altering the Circadian
rhythm (in biology, circadian rhythms are oscillations of biological variables at regular intervals of
time); and implications on the allocation of time. Temperature affects the opportunity cost of working
over leisure. Following this logic, it will also affect labour supply (Graff & Neidell, 2014).
The Effects of Climate on Output per Worker
64
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
estimation of the missing parameter . It is important that in this case the
analysis is only undertaken for the manufacturing industry with the purpose
of isolating the effect that climate has on the output of capital.
The long-term effect of temperature on output per worker can be summarized
into two broad categories9: adaptation and convergence. These have opposite
effects given that convergence tends to increase the output per worker and
adaptation tends to produce the opposite result.
It is important to clarify that the concept of convergence comes from the
neoclassical theory and is based on the assumption that factors of production
grow at the same rate in all countries; this has its roots in the decreasing mar-
ginal returns to scale of the factors of production. Additionally, the Colombian
growth trend is upward sloping, and, hence, it can be said that the convergence
level is above current levels. Therefore, the convergence effect will increase
output per worker in the long-term, which is a natural inertia.
Moreover, the adaptation effect is directly related to temperature. The relation-
ship is based on the fact that, in the long-run, areas must adapt to changes
in geographic conditions such as climate. In the case of industries (the unit of
analysis in this paper), the adaptation effect impacts the capital and labour
sides of the production function. In terms of capital, there is an adaptation of
technology and physical capital. More specifically, industrial establishments
face costs caused by the variation of climate in the long-run. Hoverer, on the
labour side of the production function, the adaptation occurs through migra-
tions, fertility and mortality rates. People are forced to do things because of
geographic conditions; in general, there are alterations on the industry’s rela-
tive factor intensity.
The differential equation (1) is the starting point that summarizes the effect
of temperature on output per worker10. It can be interpreted as the evolution
9 Mitigation is another category that is neither related to adaptation or convergence. Universidad de los
Andes is now calculating the abatement cost curves (net present value of mitigating climate change)
under the Low Carbon Development Strategy framework. Once these values are available, it would be
interesting to redesign this model in order to calculate the impact such measures have on the long-term
growth of output per worker. The way the adaptation parameter is calculated ensures that impacts on
other environmental variables such as biodiversity are accounted for.
10 This equation is an application of the model derived in Dell et al. (2009).
Mateo Salazar 65
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
of output per worker over time, and it depends on the adaptation and con-
vergence effects.
dlogy
dt
gTtT Tlogy tlogy tfor t
i
ii
ii
=
()
()
() () ()
()
+γ −+γ+ρ+ϕ− ≥
∗ 00 (1)
where y is the income per capita, y* is the income per capita to which the
regions converge, Ti is the temperature and
Ti
is the average temperature of
the municipalities where the industry i is present. The subscript i represents
the industry and t the period. The coefficient captures the short-term effect
of temperature, and captures the degree of adaptation to the average tem-
perature in the long-term. Finally, represents the convergence rate.
Integrating the differential equation (1) and taking expectations, the follow-
ing is obtained,
Elogy tElogy tTT
ii
t
()
=
()
()
()
∗+
γ+ρ
ϕ−+
−ϕ
1exp (2)
Then, differentiating equation (2) with respect to temperature,
dE logy t
dT
i
i
()
=γ+ρ
ϕ (3)
Equation (3) shows that the changes in output per worker, with respect to a
change in average temperature, depends on the convergence parameter (),
the effect of temperature in the short-term () and the degree of adaptation to
average temperatures in the long-term ( ).
Dell et al. (2009) calculate dE logy t
dT
i
i
()
=0012. in a within-country context.
The convergence parameter, much analysed in the growth literature, is typically
estimated in the cross-country context in the range
00201..
< (Barro &
i Martin, 1995). The convergence rate is calculated between countries so, for
the purpose of this paper, the upper bound will be used as the convergence
within countries is higher (Caselli et al., 1996). The only two parameters left to
calculate are and . There is no estimation of the within country short-term
growth coefficient in the literature; therefore, the empirical section of this
paper is devoted to calculating it. The result for the Colombian manufacturing
The Effects of Climate on Output per Worker
66
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
industry presented in Section 4 is
γ=
−
0003.
11 (see Table 1). Due to the lack
of information, Dell et al. (2009) use country level estimate: γ=−
0001.
.
Finally, from equation (3) and, by undertaking a sensibility test, it is possible
to calculate a range for the adaptation parameter
0019 0 0029..
<.
Table 1. Aggregate Results for the Entire Manufacturing Industry
Number of firms Production
Level
(/+1%)
Growth
(%/+1°C)
Level
(/+1°%)
Growth
(%/+1°C)
Average Temperatureq-0.301*** -0.082 -0.290** 0.358
(0.101) (0.101) (0.136) (0.316)
Average Precipitationq0.032*** 0.018*** 0.038*** 0.001
(0.003) (0.004) (0.004) (0.027)
F-test 0.78 0.14 0.80 0.88
R20.45 0.01 0.47 0.00
Observations 6,969 8,760 5,376 6,798
Municipality fixed effect Yes Yes Yes Yes
Lagged dependent variable 4 No 4 No
Note: Beta coefficients presented for an increase in temperature and precipitation. Unit of observation are
industries in a municipality for a given quarter. Robust standard errors are presented in brackets. Durbin’s
h-test was performed to check for serial correlation. The fourth lag is no longer statistically significant,
hence no further lag is included. *** p < 0.01, ** p < 0.05, * p < 0.1.
Source: DANE (2011) and IDEAM (2011).
The rest of this section is used to explain how is calculated. This parameter
is defined as the within country short-term growth coefficient. In order to
estimate it, the methodology must disregard the long-term effects (adaptation
and convergence) and focus on the short term in order to develop a hypoth-
esis that is testable with the data available.
To begin, it is important to point out that recent empirical literature on eco-
nomic growth estimates specifications based on variants of the Solow model
in which the long-term growth rate of output per worker is determined by
technical progress, which is taken to be exogenous. The most popular model
used to evaluate this framework and to study the issue of convergence is
derived from transition dynamics to the steady state growth path; it was first
suggested by Mankiw, Romer and Weil (1992).
11 The result is the average country short-term coefficient for the Colombian manufacturing industry.
Mateo Salazar 67
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
The model presented below is based on the theoretical framework by Bond, Leb-
lebiciog and Schiantarelli (2009). For the specific scope of this study, a single-
sector economy was chosen for the simplicity of the Y = AK type production
function. The motivation for choosing this specific type of model is that the
study focuses on the impact climate has on workers output. The manufactur-
ing industry was chosen because climate will not affect the output of capital,
which is the other factor that is taken into account in traditional production
functions. In fact, the assumption here is that the other production factors
will not be affected by the variance of the climate in a specific municipality
between quarters. It is relevant to clarify this as the empirical methodology
compares a municipality to itself through time. For instance, consider the fol-
lowing production function that incorporates temperature:
Ye
AL
it
T
it it
it
= (4)
where Y is aggregate output, L measures workforce, A measures labour produc-
tivity and T measures weather. Equation (4) captures the relationship between
weather and production.
∆+γ
A
A
gT
it
it
ii
t
= (5)
Equation (5) represents the growth of labour productivity that is affected by
weather.
Now, dividing both sides of equation (4) by
L
it, taking logs, differencing with
respect to time and replacing equation (5) yields:
gg
TT
it
ii
tit
=
()
()
+β+γ −β −1 (6)
where g
it
is the growth rate of output per worker (dependent variable) in the
short-term. There is a direct link between equation (1) and (6). It is obvious that
equation (6) ignores the long-term effects such as convergence and adaptation,
but it includes the temperature lag in order to control for short-term lagged
impacts of temperature on output per worker. The level effects of weather
shocks on
git
, which come from equation (4), appear through . The growth
effects of weather shocks, which come from equation (5), appear through .
Thus, equation (6) implies that it is not only current levels of average output
per worker that are affected by temperature, but also the growth of the average
The Effects of Climate on Output per Worker
68
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
output per worker. It also implies that temperatures from previous periods
may have an impact. The aim of the empirical exercise is to estimate from
a variant of regression (6) for the case of Colombia.
II. Data
The task of measuring the impact climate has on the average output per worker
in the short-term() is done by merging and analysing several datasets. The
datasets that are described in this section complement each other in order to
use the most precise and disaggregated information available. Some indices
have had to be constructed in order to obtain the quarterly average output
per worker, which, ultimately, is the dependent variable. Since the richness
of the analysis lies in the comparison between hotter and colder periods, it is
vital for the data to be quarterly. The variance of temperature and precipita-
tion within years is larger than their variance between years, even in a country
without seasons such as Colombia.
The challenge is to construct a dataset with the information available for
Colombia that has quarterly climate variables for each municipality and pro-
duction variables for each industry (ISIC).
The National Bureau of Statistics (DANE) has constructed the Annual Man-
ufacturing Survey (AMS), which aims to obtain basic information from the
Colombian industrial sector in order to characterize its structure and evolution.
An unbalanced panel12 records data for all industrial establishments with
ten or more employed workers, or that had a production value of more than
$130.5 million pesos in 200513. Some of its variables are: number of employees,
expenditure in wages and salaries, total value of production, total expenses,
production costs, energy consumption, etc. The information is available for
the industry in each municipality, it covers the period from 1993 to 2004, and
has an annual frequency.
12 All regressions are calculated using the balanced panel. The panel is lightly unbalanced (the biggest
difference in observations between years is never more that 1%. Industrial establishments coming in
and out of business, as well as passing the sampling threshold, are the main reasons for the panel
being unbalanced). Therefore, the data is not censored.
13 This assumption had to be made due the National Bureau of Statistics’ (DANE) budget constraints.
Mateo Salazar 69
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
The Monthly Manufacturing Sample (MMS) constitutes another valuable
source of information. It contains monthly data on the value of production,
value of sales, size of the workforce, expenditure on wages and salaries, social
benefits given to employees and average hours worked. To preserve anonym-
ity, DANE publishes indices and variations. Using the Annual Manufacturing
Survey (AMS) as reference, the MMS was designed to include 1344 randomly
chosen establishments employing ten people or more. The resulting data is
a representative sample of the manufacturing industry. It has been divided
into 48 groups according to the third revision of the International Standard
Industrial Classification adapted for Colombia (ISIC Rev. 3 A.C). The base year
for this sample is 2001. From now on, all results will be presented in constant
2001 Colombian pesos.
For the climate variables, the information is obtained from the Institute of
Hydrology, Meteorology and Environmental Studies (IDEAM for its Spanish
acronym). This dataset is an unbalanced panel that contains monthly infor-
mation for 772 municipalities in Colombia for the years 1931 to 2005. Some
municipalities have more than one climatological station and therefore more
than one observation. The variables available are maximum, median and mini-
mum temperature, precipitation, humidity and solar radiation. The data before
1980 is incomplete but this will not matter as the merge will only take into
account the data between 2001 and 2004 because of information constraints.
The main purpose of the following sub-sections is to explain the merge between
the MMS and the AMS. This step is necessary because the empirical section
uses a dataset on the municipal level (to be able to match the weather data)
with quarterly coverage (to be able to exploit the intra-year variation of tem-
perature). The AMS, has municipal disaggregation but is annual. On the other
hand, the MMS is quarterly (at least the public version), but it is not disaggre-
gated by municipality. The methodology used in this section can be viewed as
a statistical matching technique, in which I observe subsamples of the same
industrial establishments in both surveys and make some minor assumption to
disaggregate the data to the level that I need (D’Orazio, Zio & Scanu, 2006).
In summary, the market share by municipality is calculated from the AMS and
is used to disaggregate the MMS.
The main assumption of this methodology is that the market share remains
constant within a year (the variation across years is low so the assumption is
The Effects of Climate on Output per Worker
70
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
reasonable). This is the main reason why the number of firms is used to calculate
the market share. It is possible that production varies considerably within a
year across municipalities; however, the number of companies in each industry
is relatively structural and remains largely constant throughout quarters.
A. Market Share and Labour Share of each
Municipality within an Industry
Since the quarterly data from the MMS is divided by ISIC but not by munici-
pality, I used the information from the AMS to obtain quarterly data for each
municipality. The indices used to do this are the market share and the labour
share of each municipality within an industry in a specific year. These indices
are calculated using the number of companies (total production is also used
as a robustness check) and the work force.
mktshare
produc
produc
imt
imt
imt
m
Ni
=
=
∑
1
(7)
laborsharelabor
labor
imt
imt
imt
m
Ni
=
=
∑
1
(8)
where produc and labor are the number of firms and work force respectively.
The subscripts i, m and t from now on correspond to industry, municipality
and year. The top limit of the sum Ni depends on the number of municipali-
ties in which the industry i is present. The market share and labour share will
be useful to separate the MMS indices in municipalities.
B. Quarterly Output per Worker
The MMS contains quarterly indices of production, employment and sales. The
index is the value of the variable in a quarter divided by the value of that same
variable in the first quarter of 2001 (2001q1: base period). As has been previ-
ously mentioned, these indices are not divided into municipalities. This means
that from the MMS we have
iprodu
ciq (where q is quarter) and from the AMS
mkts
hare
imt
. The problem is symmetric for labour. By merging the two datasets
and assuming that the market share is constant within years, it is possible to
calculate the
iprodu
c
imq
. Again, the main assumption is that that the market
Mateo Salazar 71
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
and labour shares do not change within years. The fact that the market and
labour shares do not change significantly between years suggests that the
assumption is reasonable (see Figure 3). To convert these indices into the
average output per worker, we used the following algebraic manipulation. By
dividing both indices, we can obtain the following:
iproduc
ilabor
produc
produc
labor
labor
imq
imq
imq
im
imq
im
=2001
20001
(9)
After some algebraic manipulation, the following was obtained:
prodl
produc
labor
iproduc
ilabor
produc
imq
imq
imq
imq
imq
im
== 2001
ll aborim2001
(10)
This is the quarterly average output per worker that is used as a dependent
variable in the empirical estimations. The growth rate of
prodl
imq, grodl
p
imq
is calculated with a standard methodology and will also be used as depen-
dent variable.
Figure 3. Inter-annual Variation in the Number of Industrial Establishments (Firms)
0 1000 2000 3000
3000
2000
1000
0
Number of firms in t-1
Numbers of firms in t
Source: DANE (2004).
The Effects of Climate on Output per Worker
72
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
C. Matching the Weather Data with Each Municipality
This paper calculates the municipal average temperature and precipitation
based on data from the weather station that is closest to the largest populated
area. The motivation to do so is in line with the issues identified in Auffen-
hammer, Hsiang, Schlenker and Sobel (2013). The size of each municipality
means that there is a risk of measurement error for the explanatory variable.
However, there are three reasons that explain why this is not a problem for
the identification strategy. First and foremost, since the paper only studies the
manufacturing industry, most of the establishments are located in the munic-
ipality’s largest town (or the cabecera municipal). Luckily, these places also
happen to have the majority of the weather stations. Using gridded or satellite
data would contribute more noise than actual information; hence, I use the
stations that are located in the largest populated areas of each municipality.
Second, the potential measurement error is random and potentially very small,
which generates a negligible attenuation bias of the estimator. Third, one of the
main problems when using averages is that coverage is usually spotty (i.e.
there are a lot of missing values or stations coming in and out of the sample).
Since the data used in the paper is gathered on a quarterly basis, a missing
value in a specific day or even week does not have a major impact on the
accuracy of the explanatory variable.
Finally, it is important to highlight that the resulting panel does not contain
establishment level data. Instead, it is aggregated on industry level. Therefore,
the unit of analysis is the industry in each municipality of the country for the
all quarters between 2001-2004.
D. Summary statistics
The summary statistics are presented in Table 2. The table summarizes the
data used in the regressions, which cover the sixteen quarters between 2001
and 2004, for nineteen industries in 125 municipalities. The panel is set as the
combination of the 633 different combinations of industries and municipalities
(panel variable) for the sixteen quarters mentioned above (time variable).
The table divides variation into the following categories: overall, between and
within. These categories arise because of the panel structure of the data. As
the name suggests, the overall category summarizes all the 8,760 observations
in the sample. The between variation is the variation between industries and
Mateo Salazar 73
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
municipalities. The within variation represents the variation that is exploited
in the empirical strategy, which is the variation within the same municipality
and industry and through quarters. The minimum and maximum numbers in
the between and within categories are demeaned, which is why some numbers
are negative.
Table 2. Summary Statistics
Variable Category Mean StdDev Min Max Obs
prodl (million pesos 2001) Overall 15.87 80.70 0 2076 8760
Between . 78.53 0 1714 633
Within . 18.78 -481.2 522.1 16
gprodl (%) Overall 0.0516 1.379 -1 124.9 8760
Between . 0.355 -0.250 8.404 633
Within . 1.332 -8.467 116.6 16
Average Temperature (°C) Overall 21.68 5.619 4.167 30.97 8760
Between . 5.636 4.733 29.22 633
Within . 0.658 19.05 24.26 16
Average Precipitation (mm) Overall 116.9 91.89 0 3007 8760
Between . 66.02 41.26 779.0 633
Within . 63.85 -278.7 2387 16
Note: Unit of observations are industries in a municipality for a given quarter. The Table contains the
summary statistics for the overall sample together with variation within and between panel units (industry-
municipality).
Source: DANE (2011) and IDEAM (2011).
The first variable shown in table 1 is the independent variable: prodl. As was
mentioned in the data section, this variable represents the average output per
worker in millions of pesos in the year 2001. The interpretation of the overall
mean is that the average worker produces 15,87 million pesos in a quarter.
The standard deviation is very large because it includes firms from all indus-
tries and all municipalities. There are very large and very small values for this
variable; for example, for the Manufacture of general purpose machinery in
Espinal, Tolima the value is very large because it refers to a capital intensive
industry, which registered enormous production with only two workers. This
The Effects of Climate on Output per Worker
74
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
variation is controlled by the industry-municipality fixed effect that makes
sure outcomes between industries and municipalities are never compared.
The within variation perfectly illustrates this point: the standard deviation is
much lower as it is only compared within an industry-municipality group for
the sixteen quarters. The empirical section also uses logarithms to control
for potential outliers.
The second variable is the growth rate gprodl, which shows that the manufac-
turing industry has been growing at a moderate pace (0.5 %) over the quar-
ters; this is roughly the same rate as the aggregate GDP.
The third and fourth variables summarize the weather in the 125 municipalities,
which are for the most part the largest urban areas where the manufacturing
industry is present. The main purpose of this table is to show that even though
Colombia has almost indiscernible seasons, there is significant variance within
years. The average temperature for the entire period of study is 21.68°C, with a
standard deviation of 5.62, a minimum value of 4.17 in Villamaría, Caldas and
a maximum value of 30.97 in Valledupar, Cesar. It is also important to note that
the data recorded is an average for the whole day, not only the working hours.
In terms of precipitation, the country average for all quarters is 116.9mm, with
El Carmen de Atrato, Chocó presenting the highest levels of rainfall and four
municipalities in La Guajira registering no rainfall during the period of study.
III. Estimation of the Effect of Temperature
on the Average Output per Worker
In this section, the estimation models are described and the results are discussed.
Furthermore, a robustness check is presented.
A. Empirical Framework
The empirical model of the study is based on Hsiang (2010) and borrows some
concepts from Bogliacino and Pianta (2009) and Dell et al. (2009) that use a
similar panel approach to this problem (Bogliacino & Pianta, 2009; Dell et al.,
2009; Hsiang, 2010). It is a regression with industry-municipality and time
fixed effects. The output per worker is explained as a function of its lags and
the climate variables.
Mateo Salazar 75
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
The objective of this study is to determine empirically if the mean output per
worker in individual industries has any statistical dependence on intra-annual
variations in the local temperature. Previous research used a cross-sectional
approach in which patterns of production are correlated with the average
state of the local atmosphere. A critique of this approach is that the aver-
age state of the atmosphere (a fixed parameter) may be correlated with other
fixed parameters (for example, altitude), which may then directly affect pat-
terns of production (Tol, 2009). This is the omitted-variables problem: with-
out describing all fixed variables affecting an outcome, statistical inference
on any single fixed variable may be biased (Greene, 2008).
Since it is almost impossible to control for all of these fixed variables, this study
inserts industry-municipality and time fixed effects in order to only examine
the effects related to dynamic variations. The average atmospheric states of
any two municipalities are never compared here. Instead, the influence the
atmosphere has on production is identified by looking at the response of pro-
duction to perturbations in the atmospheric state around its mean value. This
means that a municipality should only be compared to itself at different points
in time when it is experiencing different atmospheric states. To avoid the
omitted variable problem mentioned above, the precipitation is also included
in the regression; this, together with temperature constitute the atmospheric
state (Auffenhammer et al., 2013). Some argue that cyclones should also be
included as they are correlated to temperature and production (Hsiang, 2010).
In the case of Colombia, cyclones are not a major problem and are mostly
isolated atmospheric phenomena that are not relevant for the study of this
specific country.
Given the variables calculated in the previous section, the following two regres-
sions can be run with fixed effect of industry-municipality, lags, time fixed
effect and environmental variables.
prodl prodl TP
imq lim q
I
Ljmq jjmq jmq
=
()
()
()
=
() ()
∑α+
ρ+β+ησ +
−
−−
1
0
µµ
+θ +ε
im
qimq
j
J
=
∑
0
(11)
gprodl TP
iimq jmq jjmq jmq im qimq
j
J
=
()
() ()
=
∑ρ+β+ησ +µ +θ +ε
−−
0
(12)
The Effects of Climate on Output per Worker
76
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
The difference between regression (11) and (12) is the dependent variable. In
the first regression, it is the level of output per worker and in the second it
is the growth rate. For both regressions is the industry-municipality fixed
effect, q is the time fixed effect,
mq
is the temperature variance,
P
mq j
()
is the
precipitation and
T
mq j
()
is the temperature (all variables in equation (11) are
in logarithms). The parameter of interest is, therefore, j, which accompanies
the temperature indicator. Since the survey is a panel, the regression is run
where i is the industry, m the municipality and q is the quarter. The industry-
municipality fixed effect is very important because it captures all the stati-
cal differences in levels of production between industries and municipalities.
Finally, the time fixed effect captures all changes in average output per worker
that vary smoothly over time for all industries and municipalities equally. To
calculate J, the Durbin’s h-test is performed to check for serial correlation. The
lags are included as long as they are statistically significant. For L, the litera-
ture states that only one lag is necessary.
The empirical strategy estimates (11) and (12). The following null hypotheses
are tested in order to assess if temperature has no effect on growth:
HJ
00
00
>ρ
()
=: (13)
Hypothesis (13) is relevant because if it is not rejected, it would mean that there
is an absence of both level and growth effects within the entire manufacturing
industry. If we reject the null hypothesis, the parameter j will be our estima-
tor of the coefficient (), which is needed to complete the long-term model.
Following the conventions in the distributed-lag literature (Greene, 2008), the
growth null hypotheses is stated for the accumulated effect of temperature:
HJ j
j
L
0
0
00>ρ
()
=
=
∑
:º (14)
Null hypotheses (13) and (14) are tested for both regressions to explore the sign
and magnitude of the parameter of interest . It summarizes the evidence of
an effect of temperature on the growth of output per worker in the short-term.
Mateo Salazar 77
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
1. Non-linear Effects of Temperature
Ergonomic studies state that the effect of temperature on productivity is non-
linear because in high-temperature places the effect is greater: “If the economic
response to temperature is non-linear in agreement with ergonomic studies,
temperature changes during the hottest season should have a greater economic
impact than temperature changes in other seasons” (Pilcher et al., 2002). Figure 5
illustrates the mean percentage difference in performance between the neutral
temperature groups and five temperature subcategories defined by the author.
Cold CColdCHotCHot2108 110811828118 29 26 62 2:.;:..;:..;:
−° −°
226 63
32 17 33218
.
.; :.−° >°CHot C
Our interpretation is that performance is not altered when the mean tempera-
ture is in the
Hot1
subcategory. When the temperature is below that level, per-
formance is lowered because of the cool environment, and when it is beyond
that point, performance is lowered because of the hot environment. From the
total of 125 municipalities in the sample, 67 have higher temperatures than
Hot1
and 24 have lower, leaving 34 in the temperature comfort zone.
Some interesting conclusions emerge when the ergonomics findings are applied
to the Colombian case. The minimum average temperature of the areas where
the productive activities take place is very close to the bottom frontier of
Hot1
,
where performance is not altered. In fact, only four departments show tem-
peratures below that subcategory: Nariño, Boyacá, Cundinamarca and Caldas.
The purpose of this explanation is to show that Colombia will not be positively
affected by an increase in its average temperature driven by climate change;
on a department and municipality level the evidence is clear. This conclusion is
due to the fact that most of the productive activity in Colombia is undertaken
where the temperature is in subcategories
Hot1
and
Hot2
. The mean 21.68°C is
also evidence of this. The ergonomics findings in this area are relevant because
they suggest the existence of non-linearity on the impact of climate change.
The evidence in Figure 4 is in line with Pilcher et al. (2002) and shows possible
non-linearity in the Colombian data. It is a graphic representation of the cor-
relation between temperature and average output per worker for Manufacture
of apparel; preparation and dyeing of fur (this specific industry was chosen from
all industries for illustrative purposes only). The figure can be interpreted as a
rough graphical representation of equation (11) (without the fixed effects or
The Effects of Climate on Output per Worker
78
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
controls). Each dot represents the output per worker for a specific municipality
for a specific quarter. The presence of clusters is expected as temperature and
output per worker are highly correlated over time in specific municipalities; for
this reason, the industry-municipality fixed effect is so important and, standard
errors must be clustered and robust to control for heteroskedasticity and
autocorrelation. The coloured lines are regression lines for each municipality,
and the vertical dashed lines show the
Hot1
thresholds (where productivity is
optimal according to ergonomics findings). It is interesting to see how most
coloured lines to the left of the threshold have a positive slope. Those within
the
Hot1
threshold show no clear trend (perhaps negative), while those to the
right of the threshold show a clear negative trend. This is suggestive evidence
for the presence of non-linearity.
Figure 4. Non-linear Relationship Between Productivity and Average Temperature
4
2
0
-2
-4
-6
Temperature
Productivity (logs)
5 10 15 20 25 30
Source: DANE (2004) and IDEAM (2011).
In this paper, we develop one additional strategy to provide some evidence of
the non-linear effects. The approach is to break the sample into the temperature
categories identified in the lab experiments, expecting larger negative effects
in the groups that have the highest average temperatures.
Mateo Salazar 79
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
It is important to highlight that these methodologies only suggest the presence of
non-linearity. However, further research is necessary, and, in this respect, the data
used in this paper faces some challenges. It is difficult to capture non-linearity
with this data because there is not a large enough variation in temperature within
municipalities. Since the empirical strategy only compares the same industry-
municipality over time, the range of temperatures within these groups is never
wide enough to capture proper non-linearity. According to (Pilcher et al., 2002),
we would need to see municipalities that report quarterly variation ranging from
at least Cold1 (10.81-18.28) to Hot2 (26.63-32.17). The within standard deviation
in the dataset is 0.65 degrees, and the maximum is 5 degrees, from 17 to 22 in
Pereira, which is too small for this type of analysis.
B. Results
In this subsection, the results for both level regressions in equation (11)14 and
growth regressions in equation (12)15 are reported and discussed (see Table 1).
The first part of the section contains the results from the main regressions; the
null hypotheses mentioned above are then tested with this data, and the param-
eter
γ
is calculated. The results of the empirical strategy designed to assess
the non-linear effect hypothesis are presented at the end of this subsection
(see Figures 4 and 3). The tables in the Tables and Figures section contain the
complete results of the regressions. It is important to highlight that the null
hypothesis (13) is directly tested with the resulting coefficients, and the
null hypothesis (14) is tested with an F-test.
The first hypothesis that will be tested states that temperature does not affect
average output per worker, either through level effects or growth effects (equation
13) in the entire manufacturing industry. As was previously mentioned, this
hypothesis is tested for the aggregate regressions. Table 1 presents the results
that show that when the estimation is made on an aggregate level, there is a
negative statistically significant relationship between temperature fluctuations
and the level of output per worker (-0.3 % / +1 %); there is also a negative but
statistically insignificant relationship between temperature fluctuations and
the growth of output per worker (-8.2 % / +1°C). The results hold for an alter-
14 All variables are in logarithms and therefore results are in elasticities.
15 All variables are in original units. These variables were left in original units because the logarithm can
only be taken when all values are positive.
The Effects of Climate on Output per Worker
80
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
native definition of the dependent variables. In columns 1 and 2 of Table 1, the
dependent variable was calculated using the market share defined by the number
of firms, while columns 3 and 4 use production. The sign and magnitude of the
effects are similar in both cases. However, the F-test is very low, which does
not allow the null hypothesis (14) to be rejected. The fact that the level is sig-
nificant and the growth insignificant is expected because the time horizon of
the study is not long enough to capture significant changes.
An important result that has been presented in this paper is the fact that the
precipitation coefficients for all cases presented an opposite sign and smaller
magnitude compared to temperature. This finding supports the hypothesis that
climate alters the labour supply by shifting the opportunity cost of work over
leisure. However, this hypothesis is not directly tested in this paper.
Figure 5 illustrates the non-linear effects of productivity as measured in the
lab (Pilcher et al., 2002). However, measuring this type of non-linearity with
field data is more challenging. Table 3 and Figure 4 present the results of the
methodologies developed to assess the non-linearity of the effect of tempera-
ture on average output per worker. The industry regressions are run to check if
deviations from the mean temperature in normally hot and cold municipalities
have similar effects.
Figure 5. The Mean Percentage Difference in Performance Between the Neutral Tem-
perature Group and Five Temperature Subcategories
Cold2 Cold1 Hot1 Hot 2 Hot 3
Temperature subcategories
Percent difference
5
0
-5
-10
-15
-20
Cold CColdCHotCHot2:< 10.8° ;1: 10.81 18.28° ;1: 18.29 26.62° ;2:−−226.63 32.17° ); 3:> 32.18°−CHot C
Source: Pilcher et al. (2002).
Mateo Salazar 81
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
Table 3 shows that the negative coefficient is only observed in the
Hot2
category:
a temperature higher than 26.63°C. The coefficients in all remaining categories
are not statistically significant. This is in line with the original lab results, which
suggest that the impact is indeed larger in this context and provide some evi-
dence of the above mentioned non-linearity.
Table 3. Non-linear Effects- Heterogeneous Effects (%/+1°C)
(°
C
<10.8) (10.81
C
<18.28) (18.29
C
<26.62) (26.63
C
)
Average Temperature -0.086 -0.001 -0.000 -0.012**
(0.856) (0.001) (0.001) (0.005)
Average Precipitation 0.000 -0.000** -0.000*** -0.000***
(0.000) (0.000) (0.000) (0.000)
R2 0.98 0.99 0.99 0.99
Observations 77 2,179 3,430 1,838
Note: Beta coefficients are presented for the non-linear effect of temperature and precipitation on output
per worker. The linear temperature and precipitation are also included. Unit of observation are industries in
a municipality for a given quarter. Beta coefficients presented for an increase in the maximum temperature
and precipitation. The mean temperature is replaced by the maximum temperature. Robust and clustered
standard errors are presented in brackets. Durbin’s h-test was performed to check for a serial correlation.
The fourth lag is no longer statistically significant, hence no further lag is included. *** p < 0.01, ** p < 0.05,
* p < 0.1.
Source: DANE (2011) and IDEAM (2011).
C. Robustness Analysis
An additional empirical exercise confirms the robustness of the negative
effects of temperature on the main variables of interest. This section contains
an alternative specification as a robustness check.
1. Alternative Sample and Data Sources: Regression with
Geo-referenced Temperature Data and CEDE Yearly Municipal Panel
This section contains the same empirical framework applied to a data set with
different characteristics. This data set is a panel that was constructed by the
Center of Economic Development Studies (CEDE) between 1993 and 2010 on
a yearly basis. It contains data from the presidency, the National Planning
Department and DANE.
The Effects of Climate on Output per Worker
82
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
Using municipal level data for Colombia, this section shows the relationship
between Geo coded climate variables (obtained from Worldclim (Hijmans,
Cameron, Parra, Jones & Jarvis, 2005)) (mean temperature, mean precipita-
tion levels and other climatic variables) and income. Since there is no data for
income on a municipal level in Colombia, the independent variable is the proxy
tributary income of industry and commerce. This makes sense since the taxes
that firms have to pay are linearly related to the value added they produce.
In turn, these variables are divided by the total workforce of the municipality
to obtain the per-worker level, and the logarithm is then calculated in order to
make the regression in levels. The summary statistics can be found in Table 4.
The average temperature is similar to the one measured with the weather sta-
tion data. In the case of precipitation, the calculations are substantially differ-
ent. In both cases, the within variability is enough to allow for the application
of the fixed effects methodology. The mean tax income per worker is four mil-
lion pesos for all industrial establishments.
Table 4. Summary Statistics-Robustness Check 1
Variable Category Mean StdDev Min Max Obs
Tax income
(million pesos 2001)
Overall 4.306 2.744 -6.908 14.55 5169
Between . 2.502 -0.827 13.36 347
Within . 1.173 -2.303 9.062 14.90
Average Temperature (°C) Overall 22.13 5.489 4.208 30.32 5169
Between . 5.529 4.655 29.27 347
Within . 0.447 19.30 24.52 14.90
Average Precipitation(mm) Overall 176.4 118.0 0.500 3101 5169
Between . 116.3 31.92 810.6 347
Within . 46.55 -177.6 2551 14.90
Source: CEDE (2010) and Hijmans et al. (2005).
The inclusion of this robustness check responds to the pitfalls usually encoun-
tered in economic studies that use weather data (Auffenhammer et al., 2013).
Calculating the average of a set of weather stations (such as in the case of
this study), as well as the fact that some stations have intermittent coverage,
may introduce measurement error. As was explained in the data section, this
Mateo Salazar 83
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
is not the case because most manufacturing establishments are located in
the largest populated areas of the municipality, which is where most weather
stations are located. Additionally, the use of quarterly data diminishes the
risk of having biased results because a few days’ worth of data are missing.
However, using gridded weather data from Worldclim removes any suspicion
of measurement error.
The results of this subsection support the previously reported results. As shown
in Figure 6, the results that come from this data are in line with the general
empirical results. Both linear and non-parametric estimations show a negative
correlation between the tributary income of industry and commerce and the
mean temperature. Additionally, the non-parametric estimation illustrates a
non-linear effect that is in line with the lab results from Pilcher et al. (2002).
Figure 6. Linear and Non-parametric Estimations-tempprom
0.011
0.010
0.009
0.008
0.007
5 10 15 20 25 30
temp_prom
Fitted values lowess y_corr_tribut_IyC_pc temp_prom
Source: CEDE (2010) and Hijmans et al. (2005).
The empirical framework explained above is also applied to this data set, and
the results are available in Table 5. Similarly to the other results, there is a
negative statistically significant coefficient of mean temperature on output
The Effects of Climate on Output per Worker
84
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
per worker. Since the weather data in this case comes from a gridded approxi-
mation, the accuracy is inferior to that from the actual weather station data.
Table 5. Robustness Check 1 (%/+1°C)
Tributary expenses
Average temperature -0.039**
(0.020)
Average Precipitation 0.000
(0.000)
R2 0.68
Observations 4,618
Municipality fixed effect Yes
Lagged dependent variable Yes
Source: CEDE (2010) and Hijmans et al. (2005).
Conclusions
This study uncovered evidence relating to the effect of temperature and pre-
cipitation on the average output per worker. It builds on previous literature that
argues that in fact worker’s productivity is a channel through which climate and
climate change affect economic performance. The study is relevant in the eco-
nomic growth theory complementing Hall and Jones´ (1999) theory by saying
that social infrastructure should have a healthy environment as one of its com-
ponents, in this case through sustained moderate temperatures. Calculating
the short run effect of temperature on growth within the country is a key
component of the long-run model that was previously unknown in the case of
Colombia. In the long-run, climate change will increase average temperature,
and according to the results of this paper, will in turn affect output per worker.
In terms of public policy, the study presents empirical evidence for an alterna-
tive channel through which climate change will impact sectors of the manu-
facturing industry. It presents ergonomics as new area of interest for the Low
Carbon Development Strategy and the National Adaptation Plan. The quan-
tification of this impact constitutes an incentive for the industry to mitigate
carbon emissions in order to maximize the benefits in the long-term. Finally,
Mateo Salazar 85
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
these results could help environmental agencies and researchers to more accu-
rately calculate the costs of climate change in the long-run. They could also
provide valuable inputs for international and sectoral negotiations.
This paper does not constitute experimental evidence, and, therefore, causality
should be inferred carefully. Omitted variables such as unemployment or
violence are potential threats to the identification strategy. Experimental data
or the exploitation of a natural experiment should be carried out to generate
causal evidence.
The methodology could be used with more disaggregated data in order to
obtain more accurate and robust results. It is likely that since the unit of
analysis is a whole industry, some detailed effects have been overlooked. The
same methodology will surely be more useful with industrial establishments
as the unit of analysis. Further work could also be carried out, that is in fact
more robust and statistically significant, to analyse the effect of precipitation.
It would also be interesting to repeat this exercise with municipal quarterly
data directly taken from the MMS, instead of making assumptions in order
to merge the AMS with the MMS. Regarding the climatic variables, it would
be helpful to use temperature for working hours only. It would also be inter-
esting to see a dataset that can properly capture the non-linearities, and not
only though approximations.
References
1. Acemoglu, D., Johnson, S., & Robinson, J. (2002). Reversal of fortune:
Geography and institutions in making of modern world income distribu-
tion. The Quarterly Journal of Economics, 117 (4), 1231-1294.
2. Arrow, K. (2004). Are we consuming too much? Journal of Economic
Perspectives, 18(3), 147-172.
3. Auffenhammer, M., Hsiang, S., Schlenker, W., & Sobel, A. (2013). Using
weather data and climate model output in economic analyses of climate
change. Review of Environmental Economics and Policy, 7(2),181-198.
4. Barro, R., & i Martin, X. S. (1995.). Economic Growth. London: McGraw-Hill.
The Effects of Climate on Output per Worker
86
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
5. Bogliacino, F., & Pianta, M. (2009). The impact of innovation on labour
productivity growth in European industries: Does it depend on firms’
competitiveness strategies? (Working Paper). ITPS.
6. Bond, S., Leblebiciog, A., & Schiantarelli, F. (2009). Capital accumulation
and growth: A new look at the empirical evidence. Journal of Applied
Econometrics, 25(7), 1073-1099
7. Burke, M., Hsiang, S., & Miguel, E. (2015). Climate and conflict. Annual
Review of Economics, 7(1), 577-617.
8. Cadena, A., Rosales, R., Salazar, M., Rojas, A., Delgado, R., & Espinosa,
M. (2012). MAPS: Colombia case study (Working Paper). MAPS.
9. Caselli, F., Esquivel, G., & Lefort, F. (1996). Reopening the convergence
debate: A new look at cross-country growth empirics. Journal of
Economic Growth, 1(3), 363-389.
10. CEDE (1993-2010). Panel Municipal del CEDE- Universidad de los Andes.
(Working Paper). CEDE
11. CONPES (2011). 3700- Estrategia institutcional para la articulación
de políticas y acciones en materia de cambio climático en Colombia.
Technical report, DNP. http://oab2.ambientebogota.gov.co/es/
documentacion-e-investigaciones/resultado-busqueda/conpes-
3700-estrategia-institucional-para-la-articulacion-de-poli-ticas-y-
acciones-en-materia-de-cambio-climatico-en
12. D’Orazio, M., Di Zio, M., & Scanu, M. (2006). Statistical Matching.
London: Wiley.
13. Daly, H. (1996). Beyond Growth: The Economics of Sustainable Develop-
ment. Boston: Beacon.
14. DANE (1993-2004). Annual Manufacturing Industry. Technical Report.
Departamento Nacional de Estadística. https://www.dane.gov.co/files/
investigaciones/boletines/eam/boletin_eam_2015.pdf
Mateo Salazar 87
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
15. DANE (2001-2011). Monthly Manufacturing Sample. Technical Report.
Departamento Nacional de Estadistica. https://www.dane.gov.co/index.
php/comunicados-y-boletines/industria/mmm
16. Dell, M., Jones, B., & Olken, B. (2012). Climate change and economic
growth: Evidence from the last half century. American Economic Journal:
Macroeconomics, 4(3), 66-95.
17. Dell, M., Jones, B., & Olken, B. (2009). Temperature and income: Recon-
ciling new cross-sectional and panel estimates. American Economic
Review: Papers & Proceedings, 4(3), 66-95.
18. Dell, M., Jones, B., & Olken, B. (2013). What do we learn from the
weather? The new climate-economy literature. Journal of Economic
Literature, 52(3), 740-798.
19. D’Orazio, M., Zio, M. D., & Scanu, M. (2006). Statistical matching.
London: Wiley.
20. Gallup, J. L., Sachs, J. D., & Mellinger, A. (1999). Geography and economic
development (Working Paper). CID.
21. Graff, J., Hsiang, S., & Neidell, M. (2015). Temperature and human capital
in the short- and long-run. Cambridge: NBER.
22. Graff, J., & Neidell, M. (2013). Environment, health and human capital.
Journal of Economic Literature , 51(3), 689-730.
23. Graff, J., & Neidell, M. (2014). Temperature and the allocation of time:
Implications for climate change. Journal of Labor Economics, 32(1), 1-26.
24. Greene, W. (2008). Econometric analysis. Cambridge: Pearson.
25. Hall, R. E., & Jones, C. I. (1999). Why do some countries produce so
much more output per worker than others?. The Quarterly Journal of
Economics, 114 (1), 83-116.
The Effects of Climate on Output per Worker
88
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
26. Hijmans, R., Cameron, S., Parra, L., Jones, P., & Jarvis, A. (2005). Very
high resolution interpolated climate surfaces for global land areas.
International Journal of Climatology, 25(15), 1965-1978.
27. Hsiang, S. (2010). Temperatures and cyclones strongly associated with
economic production in the Caribbean and Central America. Proceedings
of the National Academy of sciences, 107(35), 15367-15372.
28. IDEAM. (2011). Temperature and precipitation data. IDEAM. https://
www.datos.gov.co/
29. Mankiw, G., Romer, D., & Weil, D. N. (1992). A contribution to the
empirics of economic growth. The Quarterly Journal of Economics,
107(2), 407-437.
30. Nordhaus, W. (2006). Geography and macroeconomics: New data and
new findings. Proceedings of the National Academy of Sciences of the
United States of America, 103(10), 3510-3517.
31. Pilcher, J., Nadler, E., & Bush, C. (2002). Effects of hot and cold tempera-
ture exposure on performance: A meta-analytic review. Ergonomics,
45(10), 682-698.
32. Ramsey, F. (1928). A mathematical theory of saving. The Economic
Journal, 38(152), 543-559.
33. Rodrik, D., Subramanian, A., & Trebbi, F. (2002). Institutions rule: The
primacy of institutions over geography and integration in economic
development. Journal of Economic Growth, 9(2), 131-165.
34. Rosenthal, R., & Jacobson, L. (1992). Pygmalion in the classroom: Teacher
expectation and pupils’ intellectual development. London: Crown House
Publishing Limited.
35. Sachs, J. (2003). Institution don’t rule: Direct effects of geography on
per capita income (Working Paper). NBER.
36. Sachs, J. D., & Warner, A. M. (1997). Sources of slow growth in African
economies. Journal of African Economies, 6(3), 335-376.
Mateo Salazar 89
DESARRO. SOC. NO. 79, BOGOTÁ, SEGUNDO SEMESTRE 2017, PP. 55-89, ISSN 0120-3584, E-ISSN 1900-7760, DOI: 10.13043/DYS.79.2
37. Sánchez, F., & Acosta, P. (2001). Proyecto indicadores de competitividad:
Colombia. Proyecto Andino de Competitividad. Documento de trabajo.
Reporte técnico. CID, Harvard University.
38. Smith, A. (1776). An inquiry into the nature and causes of the wealth of
nations. London: W. Strahan and T. Cadell.
39. Solow, R. (1956). A contribution to the theory of economic growth.
Quaterly Journal of Economics, 70(1), 65-94.
40. Somanathan, E., Somanathan, R., Sudardhan, A., & Tewari, M. (2015).
The impact of temperature on productivity and labor supply: Evidence
from Indian manufacturing (Working Paper). CDE.
41. Stern, N. (2006). Stern review: The economics of climate change. Tech-
nical report. Cambridge: Cambridge University Press.
42. Tol, R. (2009). The economic effects of climate change. The Journal of
Economic Perspectives, 23(2), 29-51.
