Test Results
Univariate Analysis
Visual evidence of a correlation between democracy and interconnectivity is provocative. Figure 4.1 shows Freedom House democracy ratings for all the countries of the world. Darker shading indicate higher levels of democracy.
Figure 4.1-Democracy Rating
Figure 4.2-Interconnectivity Scores
The scatterplot with an accompanying regression line in Figure 4.3 displays
this relationship graphically and the following correlation matrix in Table
4.1, numerically.
Figure 4.3-Democracy and Interconnectivity
Matrix Showing First Order Correlations
Regression Models[11]
The correlation of GDP with democracy in this model is also statistically
significant although at a lower level. It is important to
note, however, that the sign is negative. This result could support arguments
of some scholars, as well as apologists for the Pinochet and Lee Kuan Yew
economic development theories, that democracy is not costless.[12] All else being equal, such as education,
interconnectivity and population, greater economic development might be
available only at the expense of democratization. Alternatively, a sufficient
standard of living may serve to deflect popular demands for more political
power. Another possibility may be that the relationship between economic
development is not linear as assumed in the regression model. Perhaps there is
a simple minimum threshold; or perhaps there is an "N" shaped progression in
democratic and economic development (Lipset et al., 1993); or perhaps there is
some other more complicated relationship.
Years of schooling and life expectancy also show statistical significance. In
the case of the latter, the negative sign is more difficult to explain although
this is the weakest of the significant predictors. The coefficient on
population is also significant, but the size of a country's population, largely
inaccessible to foreign intervention, offers scarcely few policy
recommendations (except perhaps to shine a glimmer of hope on the fractious
states of Yugoslavia and the former Soviet Union that potentially may have a
more democratic future than their larger predecessors.)
Model II contains a more parsimonious model retaining only GDP, the log of
population and the interconnectivity variable. These fewer variables continue
to explain more than 50 percent of the variation in democracy for 141
countries. After excluding three predictors, the small drop in
adjusted-R2 (0.047) underlines the relative importance of
interconnectivity. Alternatively, when retaining those three variables and
excluding interconnectivity, the goodness of fit measure decreases by more than
twice as much. In other words, interconnectivity alone may be more important
for predicting democracy level than these three independent variables
combined.
The possible effects of multicollinearity also deserve attention. The
correlation matrix in Table 4.1 indicates high correlations between many of the
independent variables, particularly those of specific interest to this
investigation: GDP, interconnectivity, and schooling. Collinearities between
independent variables will tend to reduce the efficiency of predictors, but
without bias. Reduced efficiency means that the reported statistical
significance may be less than the actual because the standard errors will be
excessively large. The absence of bias means that the estimated coefficients
will be neither systematically higher nor lower than their "true" values.
Correcting for the multicollinearity could result in an increase in the number
of statistically significant predictors and further strengthen statistical
inferences relative to those variables such as interconnectivity that are
already significant. On the other hand, the effect of GDP might be unduly
understated since some statistical significance is sacrificed to
interconnectivity and to the other included variables with which GDP is
collinear. The magnitude of the coefficient on GDP is, nevertheless, quite
small and is presumably reported without bias in the inclusive Model I.
Furthermore, comparing Models I and II, the coefficients do not vary much with
the consecutive inclusion or exclusion of the other independent variables.
In Africa, the coefficient on the interaction term is the highest, and
the t-statistics correspond to a 1 percent level of significance or better. In
Eurasia, the results are similar with a t-statistic also indicating 1 percent
as the lowest significance level. The coefficient is also substantial for
Latin America with a 10 percent significance level on model IV. The regression
lines that accompany the six scatterplots in Figure 4.4 approximate these
multivariate regression results for visual comparison. Western Europe shows
the most paltry correlation. In this region, the high interconnectivity levels
do not vary much and the high democracy ratings move even less.
Figure 4.4-Regional Regressions and Scatterplots
International interconnectivity data was not recorded until electronic networks
were already a global phenomenon. By the time Larry Landweber began his
landmark surveys in 1991, almost 100 countries and territories were already
connected. It was also not until 1991 that the Internet started sampling and
recording country data. Nevertheless, there was an earlier date on which
interconnectivity levels are known almost exactly: in the early 1980s when
electronic networking was still an American experiment almost exclusively
within the research community of the United States. At the end of 1983, for
instance, the level of interconnectivity for almost every country, other than
the U.S. was identically zero.[13] In that
year, FidoNet was invented and BITNET ("Because It's Time NETwork") was still
in its early stages of domestic development. The European version of BITNET,
EARN (European Academic and Research Network) was also established in 1983. In
the following year, the British and Japanese implemented their own national
versions of the Internet, JANET (Joint Academic Research Network) and JUNET
(Japanese Unix Network) respectively. In 1983, the total number of American
Internet hosts was still less than one thousand, but prepared to leap three
orders of magnitude to more than one million ten years latter (Hobbes, 1996).
This decade, from 1983 to 1993, although somewhat arbitrarily defined also
brackets the major surge in the number of democratic states at the end of the
1980s.
The dependent variable in the following tests is defined as the change in
democracy over the period from 1983 to 1993. To compute this difference across
the decade, democracy data are extracted, as before, from the Freedom House
surveys and then the democracy values for 1983/84 are subtracted from the
1993/94 values. Countries which appeared after 1983 assume the value of the of
the predecessor state. Namely these are the republics of the former Soviet
Union and the splinter states of Yugoslavia.[14] States that merged take the value of the dominant
partner, that is West Germany and North Yemen.
Possible values for the change in democracy over the decade are structurally
truncated at both ends of the scale because the democracy scale is finite. In
other words, if a country with the lowest democracy rating were to become less
free, the value for the change would nevertheless be zero. The reasoning and
value are the same if a country with the highest democracy rating were to
become more free. It is, thus, apparent that comparison between values of the
dependent variable can be valid only when the initial conditions are similar.
Therefore, the following analysis is segmented into three groups of countries
according to the Freedom House classification of countries in 1983/84: Not Free
(defined by a democracy rating below 33), Partially Free (between 33 and 67)
and Free (above 67).
Values of the independent interconnectivity variable can likewise be expressed
in terms of change since the initial condition is universally zero in all
countries except the United States. The change in interconnectivity is
everywhere non-negative yet spans several orders of magnitude. Therefore, to
show most clearly and represent most faithfully the relative relationships
between changes in democracy and gains in interconnectivity, the latter is
transformed as the square root of the number of networks per million
population. The previous metric, a partially limited rank order in discrete
levels from 0 to 16, mutes the magnitude effects that are important here. A
natural log transformation, which will be used later, necessitates an
undesirable separate treatment of the important data where there is identically
zero network connectivity. A disadvantage of the square root transformation is
that there is no ready physical interpretation or explanation. However, since
the purpose of this study is neither to seek the "right" functional form nor to
advocate any particular functional form, consistent results using various
transformations can suggest that a strong correlation is evidence of an
important relationship that exists between the dependent and independent
variables rather than simply a statistical accident contingent on a certain
functional form.
A finding of no correlation between the changes in democracy and
interconnectivity in these least democratic countries would cast substantial
doubt on the underlying basis of the correlations noted earlier. However, a
strong correlation, consistent with all previous results, is apparent from the
following scatterplot and corresponding regression line in Figure 4.5.
Interconnectivity proves to be a statistically significant predictor of the
change in democracy for the "Not Free" group at a level of confidence of more
than 99.9%.
Figure 4.5-Interconnectivity and Democratic Change
in "Not Free" Countries
Sufficiency, on the other hand, is not so easily discounted in the absence of
data to the contrary. The two countries closest to the edge of the "Empty
Corner" are extreme cases; Croatia was at war in 1993 and South Africa suffers
a bifurcated society consisting of a connected minority and a large unconnected
majority. Even under these trying circumstances, both of these countries
reported respectable democratic gains. Perhaps as these countries become more
"normal" the "Empty Corner" will expand further.
The R2 for the regression suggests that interconnectivity can
account for only about 20% of the change in democracy. The two outliers
mentioned above degrade line fitting, but the points scattered at the top
left-hand corner of the plot are the main cause of the relatively poor fit.
The measure of goodness of fit for this regression is not as high as others
have been but an increase in connectivity correlates with a steep increase in
democracy level as indicated by the slope of the regression line and the
magnitude of the regression coefficient. Introducing a hundred nodes per
million population corresponds with a leap in democracy equal to a third of the
scale.
The "Empty Corner" corroborates further the notion of the "Dictator's Dilemma,"
that greater connectivity can come only at the expense of political control.
President Gorbachev provided the most dramatic example of an authoritarian
leader who would seek economic benefits of information and communication while
hoping to maintain political control. The record shown in Figure 4.5 indicates
that neither he nor anyone else has yet been successful in exploiting networked
communication technologies while simultaneously avoiding political
liberalization.
The shatter of Gorbachev's Soviet Union resembles, in some aspects, a
controlled experiment. The initial state was arguably monolithic. The
constituent elements, subject to differing levels of intervention
(interconnectivity and otherwise), diverged toward disparate outcomes. The
results of this "quasi-experiment" are depicted in Figure 4.6. The point lying
farthest to the left on this plot, representing Estonia, is one of the five
points that hover around the top of the regression line in Figure 4.5.
Analysis of these five points can offer additional valuable insight. Since
they describe the regression line best, if all five were from the same cohort
of countries, the generalizability of the regression would be doubtful.
However, the five points represent countries from four different cohorts on two
continents: the former Soviet Union, former Yugoslavia, Latin America and
Eastern Europe.
It is subsequently instructive to compare these five points with the others
within their respective cohort of countries, such as Figure 4.6 depicts for the
countries of the former Soviet Union. If the elements within a cohort were to
line up vertically, no inference could be drawn because there would not have
been enough variance in the independent variable to reasonably test or
determine correlation. If the elements were to line up horizontally, the
inference would be one of two depending on whether the horizontal line was high
or low. A low horizontal regression line would be sufficient to reject the
correlation hypothesis within that particular cohort. It would indicate that
the independent variable had no effect on the dependent variable because even
high values of the former would correspond with no change in the latter. A
high horizontal line is more ambiguous. An elevated but flat distribution of
points could result if an outside factor instigated change uniformly throughout
the countries of the cohort, thus implying that interconnectivity was
irrelevant, or it could result if interconnectivity in fact contributed to
democratic change in leading countries and then the other less connected states
in the cohort were influenced by the example or proximity of the leaders.
Alternatively, a correlation within a cohort characterized by a line that rose
to the right would be consistent with the notion that among similar countries,
those which were the most interconnected were also those that experienced the
greatest democratic gains. Of the four cohorts of countries, a correlation
that rises to the right is clear in three of them although the samples consist
of as few as three elements.
Figure 4.6-Interconnectivity and Democratic Change
in the Former USSR
The break-up of Yugoslavia, like that of the Soviet Union, provides an
opportunity to compare clearly divergent outcomes from a nominally uniform
initial condition. The similarity extends to the results of this
"quasi-experiment" also. The Yugoslav results shown in Figure 4.7 repeat the
Soviet experience. The correlation between interconnectivity and change in
democracy level is striking although the sample size is too small for strong
statistical inference. Historic, economic,
demographic and cultural factors likely account for some of the
correlation but perhaps not all.
Figure
4.7-Interconnectivity and Democratic Change
in Former Yugoslavia
Figure 4.8-Interconnectivity and Democratic Change in
"Not Free" Latin America
Figure 4.9-Interconnectivity and Democratic Change in
"Not Free" Eastern Europe
Asia is the only region within this group that has sufficient variance of
interconnectivity for meaningful hypothesis testing on a regional level.
However, the dominant outlier, Singapore, is also from
Asia. Cohort analysis is, therefore, not helpful. The
leverage of this one influential outlier is enormous. Excluding Singapore
greatly alters the regression depicted in Figure 4.10. In its absence, the
statistical significance on interconnectivity rebounds dramatically from about
60% to more than 95% and the R2 goodness of fit measure quadruples.
Figure 4.10-Interconnectivity and Democratic Change
in "Partially Free" Countries
The segment of "Free" countries is probably the least interesting, at least
from a policy point of view. Nevertheless, the trends, while subdued because
these countries tend to have the most stable governments, are essentially the
same as those in the previous two groups. Figure 4.11 exhibits a strong
statistical significance in the correlation between interconnectivity and the
change in democracy. The slope is lower because countries in this segment are
less prone to dynamic political reversals. As before, some lesser connected
countries improved on the democracy scale but none of the most connected
countries registered substantial declines.
Figure 4.11-Interconnectivity and Democratic Change
in "Free" Countries
Figure 4.12-Interconnectivity and Democratic Change
Worldwide
The resulting regression coefficients are also listed on Table 4.2.
Interconnectivity remains a powerful predictor of democracy as before. The
magnitude of the coefficient for interconnectivity on democracy is even greater
than in the comparable OLS model. The level of significance remains
exceptionally high. Democracy, however, does not prove to have any significant
effect on interconnectivity. Thus, the suggestion that democracy leads to
interconnectivity is not supported while the hypothesis that there is no
positive effect cannot be rejected. The coefficient on population is still
negative and significant. The coefficient on GDP is also still negative and
nearly significant at the 10% level. The other outputs also closely parallel
those of Model I.
The alternative explanation for the strong correlation between
interconnectivity and democracy is that a third variable may influence both
simultaneously. The obvious candidate is economic development, which many
contend is an important prerequisite for democracy.[15] The correlation between interconnectivity and GDP, at
0.84, is also quite high, further encouraging tests on a hypothesis that
economic development is the third variable underlying the correlation between
democracy and interconnectivity. In practical terms, equipment necessary to
communicate electronically is expensive, especially for citizens of the Third
World regions that Western democratization policy would be most eager to
influence. The same economic resources that can finance participation in the
communications revolution could conceivably fuel demands for personal rights
and freedoms. In fact, in his seminal study, Lipset included measures of
communication, radios, telephones and newspapers per thousand persons in his
main index for wealth (1959: 75).
Again, a system of simultaneous equations can help unravel complex reciprocal
effects. Model VI includes all three dependent variables: GDP, democracy and
interconnectivity. The set of assumptions that this three equation model
embraces are: economic development and interconnectivity predict democracy;
democracy and economic development predict interconnectivity; and
interconnectivity and democracy predict economic development. The relative
effects of the predictors can be assessed and compared as before in Model V.
The interconnectivity equation utilizes the same two instrumental variables.
The independent variables in the democracy equation are the same as before
except that schooling is used to serve as an instrument for economic growth in
accordance with prevailing theory. Scholars surmise that education can
influence democracy by increasing personal and national wealth, as discussed
earlier. The 2SLS estimation results, shown in Table 4.2, are consistent with
all those that preceded. Statistical test results do not support the
hypothesis that economic development is a confounding third variable. Strongly
to the contrary, the regression coefficients for interconnectivity on democracy
and GDP are both substantial and statistically significant, again above the 0.1
percent level. Neither democracy nor GDP proves to influence interconnectivity
strongly. GDP again shows a negative correlation with democracy at a 10
percent significance level.
In each model, without exception, interconnectivity positively correlates with
democracy at high levels of significance. In each model, at lower but still
high significance, the correlation with population on democracy is negative.
Stories to explain both the country size and the interconnectivity phenomena
may share a common plot. Smaller size and greater interconnectivity may
similarly be conducive to democracy by facilitating coordinated civic action.[16] Although perhaps a cliché, the often
repeated analogy that information revolution technologies are shrinking the
world offers appropriate insight. It is interesting to note that the most
populous country that Freedom House labeled as completely "Free" became a
democracy in 1776 when its population was only a fraction of its current size.
At that time and at that size, available communication technologies, like
pamphleteering, were sufficient to gel public support into popular action.
As before, this analytic approach employs regression modeling to predict
democracy as a function of economic development, education, and population.
Metrics for the prevalence of traditional communication technologies are
additionally included for comparison. Per capita numbers of television sets
proxy for uni-directional communication, telephone instruments for
bi-directional and e-mail nodes for multi-directional.
Several of the variables have largely skewed distributions with a few wealthy
or large countries forming a long tail far to the right. These variables are
economic development, population and the three telecommunication technologies:
telephones, televisions and networks. All of these variables were similarly
modified for linear estimation using a natural logarithm transformation. In
1993, thirty-eight countries had zero connectivity, so a binary variable on the
presence of networks was also incorporated into the analysis. Nearly complete
data was available for 140 countries. For sixteen of these, including nine
former Soviet republics, there were no data on the number of television sets.
The analysis involves two forms of linear regression equations. Inclusive
equations contain all three communication variables. Each of the three
corresponding exclusive equations includes only one of the three communication
variables. The rationale for this approach is to best interpret unavoidable
collinearities among the supposedly independent predictors. As discussed
earlier, within the inclusive equation, collinearities between the
communication variables may cause reduced predictor efficiency but no bias. In
the exclusive models, on the other hand, efficiency is improved but effects of
wrongly excluded variables will be assigned incorrectly to the included
variables resulting in systematic bias. Together the inclusive and exclusive
models provide alternative vantage points to examine the data for a
comprehensive understanding of the complex interactions.
In accordance with the theory of the "Dictator's Dilemma," one could anticipate
the results of the investigation as follows: multi-directional communication
technologies as simultaneous enhancers of both autonomy and influence will
presumably be most highly correlated with democracy; uni-directional
communication technologies as the dictator's preferred communication instrument
of control are likely to be the least correlated or perhaps even negatively
correlated with democracy; and bi-directional communication as essential for
market development will probably be somewhere in between, inconclusive or not
statistically significant. These predictable results are precisely the test
outputs shown in Table 4.3.
The regression coefficient on economic development, on the other hand,
oscillates dramatically between the models as a result of its collinearities
with the communication variables. The only model in which GDP is statistically
significant is the network exclusive model. This implies that the effects of
collinearities between networks and economic development have less of a
degrading effect on the slope estimate. This may indicate that the emergence
of networks is less dependent on economic development. While the Maitland
commission report found that, regarding telephones, "Telecommunications have
often been seen as a luxury to be provided only after other investments"
(Maitland 1984: 8), electronic networks might be viewed as more of a necessity
which organizations, particularly those seeking to effect political change,
purchase before other amenities. Notably, the sign on the regression
coefficient for economic development is again negative. This inverse
relationship recalls previous analysis hinting that there might be some mutual
exclusivity in the policy choices to between supporting democracy and economic
growth. All else being equal, more of one might be obtainable only at the
expense of the other. Of course, the coefficient on economic development in
this exclusive model is biased by the excluded variables, but telephones and
televisions seem to operate in different directions. They may somewhat balance
each other out reducing the net bias. In the unbiased inclusive model, the
sign is also negative although the standard error on the GDP variable is large.
Not only do networked communications appear as the most statistically
significant predictor, they also predict the greatest percentage of the
variance in democracy relative to the other communication technologies, almost
sixty percent. Observing the adjusted R2 values, the exclusive
model with television predicts just over half of the variance in democracy and
with telephones, even less. The inclusive model produces the highest adjusted
R2, nearly two-thirds, again suggesting that the collinear effects
among the communication technologies are not severe.
The four other models examine the opposing cutting edge of the "Dictator's
Dilemma." Economic development replaces democracy as the dependent variable.
Again, the empirical results are consistent with theory. Networks correlate
with economic development at high levels of statistical significance, as do the
other communication variables. The magnitudes of the coefficients for
networks, however, are substantially less than those of either telephones or
televisions. In the inclusive model with networks, the adjusted R2
is the lowest of the four models. In effect, this corroborates the notion that
networking may be less dependent on wealth than the other communication
technologies. Also this lower correlation may be at least partially due to the
accepted use policy (AUP) of the NSFNet. The main backbone for main networks,
the NSFNet, having evolved in the research community forbade use of the
networks for commercial purposes. It is expected that as policies and
capabilities become more favorable to the conduct of business and trade, this
correlation may strengthen over time (Jacobson, 1994).
Results of the media specific analyses lead to two important conclusions.
First, national wealth cannot be a primary source of the significant
correlation between democracy and interconnectivity. If it were, presumably
the prevalence of telephones and televisions would produce similar statistical
profiles. Multi-directional networks are less tied with economic
development than either bi-directional telephones or uni-directional
televisions, yet of the three, only networks exhibit a powerful positive
correlation with democracy. Second, strengthening the relationship between
networks and economic activity could well tighten the vice grip of the
"Dictator's Dilemma." As such, this formulation of the choices that face
anti-democratic regimes offers stimulating policy suggestions for western
democracies who have national security interests in promoting global
democracy.
[2] For several examples see, the World Bank
(1991), Starr (1991), Helliwell (1992), Lipset, Seong and Torres (1993), Muller
and Seligson (1994), Boone (1994), and Rowen (1995).
[3] For similar precedents, see Rowen (1995) or
Muller and Seligson (1994)
[4] "Interconnectivity" is a term popularized by
Larry Landweber for his measures of the proliferation of global email networks.
[5] For examples see Lipset (1959), Helliwell
(1992), Lipset, Seong and Torres (1993), and Rowen (1995).
[6] For examples, see the World Bank (1991) and
Boone (1994).
[7] United Nations Development Programme (1993)
provided all the economic, education and health data which are used in these
analyses.
[8] These data are published in the Central
Intelligence Agency World Fact Book (1994). In a few cases, mostly in
Northern Europe and Africa, these data were not available. Where applicable,
the percentage of largest religious affiliation substituted for the missing
data.
[9] For characteristic arguments from both sides
of the debate, see Huntington (1993) and Schifter (1994).
[10] Data were either missing or relative to
inconsistent entities for many of the new countries resulting from the recent
breakups of Czechoslovakia and Yugoslavia. Therefore excluded from this study
are Slovakia, the Czech Republic, Bosnia-Herzogovina, and in some cases
Croatia, Slovenia and greater Serbia (called Yugoslavia.) Additionally,
critical missing data precluded the inclusion of Taiwan.
[11] "INT*" indicates an interaction term as
the product of the interconnectivity variable and the categorical region
variable.
[12] For more discussion on the potentially
negative economic consequences of democratization, see Shin (1994) or Rothstein
(1991).
[13] The first international connections to
the ARPANET were England and Norway in 1973 but even in these countries,
interconnectivity was very close to nil until after 1983 (Hobbes).
[14] As mentioned earlier, missing or
inconsistent data entries precluded the inclusion of the Czech and Slovak
Republics along with Bosnia-Herzogovina in these analyses. Additionally,
Namibia, which did not become an independent country until March 1990, is now
also among the exclusions.
[15] The seminal work on this topic is Lipset
(1959) but the literature is large. Also see, for example, Helliwell (1992),
Lipset, Seong and Torres (1993), and Rowen (1995).
[16] The enormous size of India, nominally a
democracy although Freedom House gives it only a "Partially Free" label, stands
out as an obvious exception. As such, it exemplifies the epistimological
difficulty of statistical inference to determine the causes of a complex effect
like democracy. Close British tutelage as an important colony and the
personality and presence of Matahma Gandhi are likely key factors. However,
the objective of this dissertation is rather more modest, to examine the
effects of a single cause.
Multivariate Regression Analysis
In large complex systems such as international politics, simple
relationships can rarely tell the whole story. Multiple linear regression can
be a powerful technique to provide insight into the complicated interactions.
As with other techniques, the answers are often influenced by the ways in which
the questions are asked. In other words, regression results can be model
specific. Therefore, versions of several models offer various perspectives
that can be integrated to form a comprehensive understanding of the
interactions. Ultimately, the multiple linear regressions in this research
provide further evidence that this correlation cannot be dismissed. Regression
results of six representative and most informative models are shown in Table
4.2. Models I and II show the resulting statistical output of ordinary least
squares (OLS) regressions. Model I is an inclusive model involving six
predictors. 
Regional Analysis
Models III and IV, with the addition of the regional interaction terms,
are analogous to I and II, respectively. These next two models show that the
positive correlation of interconnectivity with democracy is consistent across
and within regional boundaries. In all the regions the coefficient is
positive. In half of the regions, the coefficient is substantial and
statistically significant. 
Longitudinal Time Analysis
In the wake of the coincident revolutions, cross-sectional
investigations reveal a strong correlation between democracy and
interconnectivity. These one-moment-in-time pictures cannot show which
countries took part in the recent democratic revolution nor whether or not
these same countries were participants in the information revolution. The
salient question that the cross-sectional regressions fail to answer is how
much did democracy change in countries that increased interconnectivity?
Addressing this question requires temporal comparisons, the data for which,
unfortunately, is largely nonexistent. "Not Free" in '83
Among the three groups, the countries in the "Not Free" segment are of greatest
analytic interest, both from a theoretic and from a policy perspective. The
theory suggests that an increase in connectivity is likely to be accompanied by
an increase in democracy. These "Not Free" countries have the most room for
democratic improvement. The "Not Free" countries are also those for which
policy initiatives to support democracy can have the most dramatic impacts.
While the need for programs to encourage democratic development in these
countries is pressing, the success of such programs, historically, has been
most elusive, otherwise they would have moved out of this bottom category.
Additionally, this segment contains a majority of countries, more than double
the number of either "Partially Free" or "Free" countries.
"Partially Free" in 1993
The group of partially free states is the smallest and the most diverse
segment. It includes 31 countries with roughly equivalent representation each
from Africa, Asia, Latin America and the Middle East. Thus, characterizing
this group with a small number of variables is difficult. This is the only
segment of the three in which the statistical significance of the
interconnectivity variable as a predictor for democratic change is less than
99.5%. 
"Free" in 1993
Multiple Endogeneity Analysis
It may be tempting to infer causality from these strong correlations
between interconnectivity and both the change and level of democracy and then
conclude that interconnectivity influences democratization. However, to do so
would be erroneous. Causality could, in fact, flow in the opposite direction.
Democracies rely on an informed public and uninhibited communication and may
therefore seek interconnectivity. One way to explore this possibility
analytically is through a system of simultaneous equations with multiple
endogenous variables solvable by two-stage least squares (2SLS) estimation.
The two-equation model assumes that interconnectivity can influence democracy
and also that democracy can influence interconnectivity. Then one can compare
the relative statistical significance and sizes of the coefficients on these
variables in each of the two equations. To perform these tests, both democracy
and interconnectivity are both dependent variables in Model V. And to obtain a
unique solution, at least one additional variable, called an "instrumental
variable," must be included in the interconnectivity equation. Since
electronic mail is text-based and travels over telephones lines, appropriate
instruments are percent literacy and the number of telephone lines per capita.
Independent variables in the democracy equation are, as before, related to
economic growth, human development and ethnicity.Comparative Media Analysis
Results presented above using advanced regression techniques involving
two endogenous variables to ascertain more about the relative predictive power
of each were not consistent with the hypothesis that causality flowed uniquely
from democracy to interconnectivity. The various econometric models showed
interconnectivity to be a statistically significant predictor of democracy, but
not vice versa. Tangled collinearities pose an even greater problem to
tease out the relative role of economic development as a third endogenous
variable. Two stage least squares regressions consistently showed the
predominance of interconnectivity, but the available instrumental variables
such as literacy or number of telephones are all themselves highly collinear
with both endogenous variables. Thus, isolating any relationship is nearly
impossible since everything is connected to everything else. An
alternative exploratory tack is to determine the relative effects of the
various communication technologies on democratization. If uni-directional,
bi-directional, and multi-directional technologies differ in their relative
effects on democracy, the difference can be assigned to the attributes of the
technology itself since each communication technology is similarly, though not
necessarily identically, dependent on wealth.
MODEL
I
II
III
IV
V
VI
VII
VIII
LHS Variable
DEM
DEM
DEM
DEM
LGDP
LGDP
LGDP
LGDP
N
126
140
126
140
126
140
126
140
Adjusted R2
0.630
0.460
0.511
0.577
0.902
0.889
0.822
0.735
constant
71.03
(1.80)*50.03
(1.19)-19.39
(-0.61)82.64
(3.50)***7.32
(86.8)***7.26
(113)***6.76
(94.0)***6.81
(63.4)***
Log(GDP)
-5.43
(-1.03)-3.28
(-0.57)7.30
(1.61)-8.03
(-2.41)*
Ave. School Yrs.
3.82
(3.35)***6.58
(5.48)***6.78
(5.98)***3.69
(3.27)***0.011
(0.567)0.044
(2.61)**0.121
(6.17)***0.170
Log(Population)
-4.64
(-3.43)***-3.72
(-2.40)*-4.70
(-3.09)**-3.69
(-2.69)**
Log(Telephones)
4.23
(1.28)2.10
(1.19)0.346
(7.54)***0.450
(15.61)*** Log(Televisions)
-5.47
(-2.18)*
-4.92
(-1.99)*
0.108
(2.58)*
0.353
(9.25)***
Log(Networks)
4.10
(4.01)***
5.38
(5.40)***0.058
(3.48)***
0.106
(4.43)***
Binary Networks
51.30
(5.57)***
57.19
(6.22)***0.341
(2.21)*
1.050
(4.74)***
t-statistics are in parentheses
The first column reports the results of the inclusive model. The
network variables, both the binary variable showing presence and the continual
variable showing prevalence, predict democracy with the highest recorded level
of statistical significance, greater than 0.1 percent. The high statistical
significance is particularly notable because the collinearities with other
variables would tend to artificially decrease the apparent statistical
significance. In both cases the substantive values are also large. A one
percent increase in networks associates with a 4 point increase on the
democracy scale. The
mere existence of networks in a country predicts a democracy level above
half the entire democracy scale. The regression coefficient on the
television variable is also statistically significant but the sign is negative.
Increasing the number of televisions by 1 percent corresponds with a decrease
in democracy on the same order that increasing networks corresponds with an
increase in democracy. The coefficient on telephones is positive but not
statistically significant, therefore no meaningful conclusions can be made
regarding its magnitude or sign. The exclusive models repeat results nearly
identical to those of the inclusive model. The similarity suggests that the
collinearities between media as predictors for democracy, beyond the part that
is related to wealth, are not great. In other words, each media might
influence democracy in its own distinct way.
*** = Significance at the 0.1% level
** = Significance at the 1% level
* = Significance at the 10% level
[1] Many of the measurement and statistical
difficulties are addressed in considerable depth by Inkeles (1990) and Dahl
(1971).
Contents
Previous section
Next section