10
measures intensity of gears operating in the coastal waters. The unit of measurement is metric tons
of catch from trawling and dredging gears in a country for a given time divided by the area of its
Exclusive Economic Zone (EEZ) in square km. The data is available for 1950-2006. Due to its
skewed distribution, the variable is logarithmically transformed.
Following our theoretical argument of the impact of democracy on environmental perfor-
mance at different stages of economic development, we want to control for national income levels
at a given time. The measure we use is real GDP per capita in constant 2005 prices, chain series
(Heston, Summers and Aten, 2009). Chain series remove effects from price changes and include
only the values of production volumes, which is very useful for the time-series analysis (Teorell et
al., 2011). The indicator is available from 1950 to 2007 and is log-transformed due to its skewed
distribution.
In order to model different stages of economic development for countries, we divide na-
tions at different points in time into groups according to their gross national income (GNI) per
capita, following the World Bank methodology (World Bank, 2011). Low-income countries have a
GNI below $1,005 per capita, lower middle-income countries have a GNI between $1,006 and
$3,975 per capita, upper middle-income countries have a GNI between $3,976 and $12,275 per
capita, and high-income countries have a GNI above $12,276 per capita (World Bank, 2011). GNI
per capita is calculated with the World Bank Atlas Method, which allows for smoothing exchange
rate fluctuations when comparing countries. This measure does not account for “welfare and suc-
cess in development,” but is recognized as “the best single indicator of economic capacity and pro-
gress" (World Bank, 2011).
Specification and methodology
In order to model the impact of our independent variables on changes in MTI across countries and
years, we use time-series cross-sectional (TSCS) analysis. Since we are interested in changes of
trophic levels and not the absolute levels as such, the dependent variable is here measured as the
first difference of MTI instead of annual values.
We make sure to deal with problems inherent to TSCS data. The Hausman test confirms
the existence of unobserved unit heterogeneity, indicating that country-specific effects are correlat-
ed with our independent variables. This implies that a random effects model will be inconsistent
when applied to our data and confirms the necessity to use a fixed effects model for correct estima-
11
tion (Greene, 1997). A Dickey-Fuller test for a unit root in a time series sample shows that our data
is stationary. Potential autocorrelation of the data is initially dealt with by using the first difference
of MTI. The Wooldridge-Drukker test confirms that autocorrelation disappears after performing
differencing of the dependent variable.
In order to make sure that independent variables are measured before the change in the
dependent variable takes place, we use a one-year lag of all the independent variables in our models.
We use one-year lags in combination with the first differencing of the dependent variable, as used
by Bohrnstedt (1969, cited in Liker, 1985, p.87).
As mentioned, the raw data of openness to trade, population, GDP per capita and trawling
intensity required logarithmic transformation before inclusion into the model due to skewed distri-
bution. Based on the discussion above and after the necessary adjustments to our model, our final
specification can be presented in the following equation:
;
where
i corresponds to each country in the sample and
t refers to the year.
and corresponds to the change in the marine trophic index for a
given country in a given year,
is an intercept term for i,
(j=1,2,3,4,5) denotes the coefficients
to be estimated,
is a Freedom House/Polity index for democracy for a given country in a giv-
en year, O
it
is openness to trade (country, year), P
it
stands for population (country, year), G
it
refers to
real GDP per capita for a certain country in a given year,
is trawling intensity in the EEZ of
each country per year, and
is an error term for each unit of analysis.
The equation will be estimated using generalized least squares (GLS) with a fixed effect and
robust standard errors per country and per year (Wooldridge, 2002). An alternative way to estimate
the equation would be to use OLS regressions with panel-corrected standard errors as suggested by
Beck and Katz (1995). However, taking into account the necessity to include fixed-effects estima-
tion into our model and control for significant but unobservable unit-specific effects, we have to
give preference to the GLS regression, since introducing fixed-effects specification into Beck and
Katz’s model in our case is problematic.
The MTI assigns values for each major marine coast or island colony of a nation. For this
reason some problems arose in our analysis, since our independent variables are measured at the
national level and are not available specifically for coastal regions or island colonies of a nation.
1
,
t
i
it
it
MTI
MTI
MTI
i
j
it
D
it
T
it