After successfully completing the preliminary tests, we proceed to create a composite ICT index by normalizing the indicators. This normalization results in a mean of zero (0) and a standard deviation of one (1), accommodating the different scales of these indicators [51]. Table 4 illustrates the differences, proportion, cumulative values, and eigenvalues of these components. Following the Kaiser rule of eigenvalues, as employed in [2] and [52], we retain three components with eigenvalues greater than one (1) to form the composite ICT index. These three components collectively capture a significant portion (61.6%) of the information in our data. Both Table 4 and Fig. 2 confirm these findings.
4.2 Estimated Results of the impact of ICT and renewable energy on CO2
We conduct the standard linear regression of the pooled-OLS, fixed-effects and random-effects models as a preliminary investigation. Table 5 contains estimates of these models. Overall, renewable energy use negatively affects emissions, while ICT diffusion positively influences it. Except for pooled-OLS, the interaction between renewable energy use and ICT diffusion negatively affects carbon emissions. While GDP and total energy usage have a statistically significant positive effect in all the three models, trade openness has a statistically significant negative impact in the first model. Education has a statistically significant positive impact in all three models, while urbanization is not statistically significant in any estimated three models.
Table 5
OLS estimates (Dependent variable: emissions)
Variable
|
Pooled-OLS
|
Fixed effects
|
Random effects
|
Renewable energy
|
-0.0318***
(0.0042)
|
-0.0218***
(0.0045)
|
-0.0294***
(0.0070)
|
ICT diffusion
|
0.0414***
(0.0536)
|
0.0487***
(0.0271)
|
0.0292***
(0.0007)
|
Renewable energy*ICT diffusion
|
0.0092
(0.0053)
|
-0.0116***
(0.0032)
|
-0.0072***
(0.0023)
|
GDP
|
0.1230**
(0.0666)
|
0.0607**
(0.1738)
|
0.0789***
(0.0189)
|
Energy use
|
0.2895***
(0.0839)
|
0.8839***
(0.1973)
|
0.5126***
(0.0423)
|
Trade openness
|
-0.0016 *
(0.0032)
|
-0.0025
(0.0010)
|
-0.0037
(0.0008)
|
Education
|
0.0017**
(0.0039
|
0.0046***
(0.0036)
|
0.0012***
(0.0005)
|
Urbanisation
|
0.0115
0.0052
|
0.0333
(0.0149)
|
0.0422
(0.0109)
|
Observation
|
667
|
667
|
667
|
R-squared
|
0.873
|
0.787
|
-
|
Note: the parenthesis contains the robust standard errors. ***p < 0.01, **p < 0.05, *p < 0.1 |
Any strong correlation between the explanatory variable and the disturbance terms amidst complicated causal links between independent and dependent variables may result in biased estimates. Thus, we must appropriately deal with the potential endogeneity problem to obtain more reasonable and robust estimates.
Table 6 contains estimates of the model specified in Eqs. (1 & 2). Columns (1) and (3) show one-step GMM estimates, while columns (2) and (4) present estimates of two-step GMM. The underlined distinction between the two is that for their moment conditions, one-step estimators employ a random weighting matrix and two-step employ an optimal weighting matrix. The latter is more robust and efficient based on its inherent strength to handle heteroskedasticity and autocorrelation issues.
Table 6
Dynamic panel model estimation results
Variable
|
One-step GMM
Diff-GMM
(1)
|
Two-step GMM
Diff-GMM
(2)
|
One-step GMM
System GMM
(3)
|
Two-step GMM
System GMM
(4)
|
System GMM
(5)
|
Long-run System GMM
(6)
|
\(\:ln{CO}_{2}\) emissions (lag)
|
0.436
(0.080)
|
0.435**
(0.171)
|
0.888***
(0.104)
|
0.909***
(0.135)
|
0.626***
(0.093)
|
|
Renewable energy
|
-0.022***
(0.005)
|
-0.020*** (0.008)
|
-0.008***
(0.005)
|
-0.007***
(0.005)
|
-0.011***
(0.002)
|
-0.029***
(0.008)
|
ICT diffusion
|
0.009***
(0.008)
|
−0.004***
0.017
|
0.026***
(0.016)
|
0.020**
(0.019)
|
0.005***
(0.001)
|
0.013***
(0.004)
|
Renewable energy*ICT
|
|
|
|
|
-0.002***
(0.0007)
|
-0.004***
(0.001)
|
Energy use
|
0.658***
(0.182)
|
0.783**
(0.316)
|
0.157
0.115
|
0.1692***
(0.138)
|
0.014*
(0.044)
|
0.039*
(0.013)
|
GDP
|
0.066
(0.104)
|
0.294*
(0.316)
|
0.065
(0.064)
|
0.053***
(0.067)
|
0.021***
(0.006)
|
0.057***
(0.017)
|
GDP2
|
-0.039
0.008
|
-0.159***
(0.036)
|
-0.029**
(0.007)
|
-0.029**.
(0.006)
|
-0.004***
(0.001)
|
-0.028***
(0.005)
|
Trade openness
|
0.002
(0.000)
|
0.001.
(0.001)
|
0.001***
(0.001)
|
-0.001**
(0.002)
|
-0.008***
(0.008)
|
-0.022***
(0.004)
|
Education
|
0.001***
(0.002)
|
0.004***
(0.004)
|
0.002***
(0.001)
|
0.002*** (0.002)
|
0.005***
(0.001)
|
0.014***
(0.004)
|
Urbanisation
|
0.008
(0.012)
|
0.009***
(0.022)
|
-0.001*
(0.008)
|
0.002
(0.007)
|
0.001
(0.000)
|
0.005
(0.001)
|
Constant
|
-2.714
(3.022)
|
2.300
(6.448)
|
-0.360
(1.496)
|
-0.033 (1.432)
|
1.673
(0.365)
|
0.965
(0.075)
|
Observations
|
346
|
346
|
346
|
346
|
346
|
346
|
Hansen-Sargan test
|
0.021
|
0.006
|
0.175
|
0.747
|
0.483
|
0.537
|
AR(2)
|
0.216
|
0.037
|
0.268
|
0.367
|
0.625
|
0.293
|
AR(1)
|
0.000
|
0.000
|
0.092
|
0.017
|
0.007
|
0.002
|
Kleibergen-Paap LM stat.
|
0.079
|
0.140
|
0.051
|
0.008
|
0.000
|
|
Kleibergen-Paap Wald stat
|
5.30.
|
21.392
|
4.127
|
23.532
|
27.734
|
|
Stock-Yogo critical values:
|
|
|
|
|
|
|
10% maximal IV relative bias
|
13.911
|
13.317
|
12.639
|
13.511
|
12.963
|
|
20% maximal IV relative bias
|
7.683
|
7.303
|
7.514
|
7.640
|
7.715
|
|
30% maximal IV relative bias
|
5.512
|
5.257
|
5.798
|
5.834
|
5.990
|
|
Note: the parenthesis contains the robust standard errors. ***p < 0.01, **p < 0.05, *p < 0.1 |
Extant literature suggests that GMM takes account of unobserved heterogeneity and keeps away from biased estimates when the dependent variable contains a time lag [53]. Hence, columns (1) and (2) of Table 6 present estimates of difference GMM [54], while columns (3) to (5) give estimates of system GMM ([55]; [56]). We report the long-run coefficients of column (5) in column (6). Individual fixed effects banish indifference GMM as it uses the difference between t and t-1 to modify all the regressors. The bottom side of Table 6 shows a serial correlation of order (AR (1)) based on a p-value which is 0.000 in columns (1) and (2). Based on a p-value of 0.2163, there is no AR (2) serial correlation in column (1). But column (2) shows AR (2) serial correlation with a p-value less than 0.1. More so, the Hansen-Sargan test results establish that the lagged exogenous and endogenous variables are not sufficient instruments as both columns (1) and (2) have p-values less than 0.1 (0.0210 & 0.0062). Therefore, the results of difference GMM suggest a system GMM.
Columns (3), (4) and (5) all pass our model specifications and identification test. However, column (3), a one-step system GMM, suffers from weak instruments. Columns (4) and (5) depict valid and pertinent tools for the two-step system GMM model. The instruments are valid based on less than 0.01 p-value of the L.M. statistic of Kleibergen-Paap. Therefore, we accept a null hypothesis that the model is under-identified at a 1% significance level. Similarly, the Kleibergen-Paap F-statistic is more than the critical value proposed by Stock & Yogo (2005) based on the null hypothesis that the 10 per cent OLS bias is less than the bias that IV estimates embody.
We can observe a negative relationship between renewable energy consumption and carbon emission levels in column (4), while a positive relationship exists between ICT diffusion* renewable energy interaction term and the emission levels. Specifically, a 1% rise in renewable energy use results in a -0.008% reduction in carbon emission at 1% level of significance. The findings are in line with our a-priori expectations, and the extant literature underpins these (see [25]; [57]; [19]). The results suggest a rising level of renewable energy use in the SSA region, where most of the countries have begun to embrace the energy transition trend and respond to the devastating effect of environmental degradation in their respective countries. Although the region contributes less than 3% of the whole global pollution, it is very much vulnerable to its adverse effects.
The empirical evidence also shows that a 1% per cent rise in ICT diffusion induces carbon emissions by 0.02% at a 5% significance level. This adverse impact of ICT on environmental quality is supported by several studies, including that of [25], [58], [1], and [9], who find that ICT use significantly raises carbon emissions level. We attribute the ICT's detrimental effect on environmental quality to the growing penetration of ICT tools, a growing exposure of people (especially the urban dwellers) to ICT, and how these ICTs cause inefficient energy use in Sub-Saharan Africa.
Our findings support this expectation as we expect that energy consumption deteriorates environmental quality. The same column (4) shows that a 1% rise in energy use raises emission levels by 0.17%. Extant literature establishes that fossil fuels consumption degrades the environment. The perpetual and unregulated consumption of these fossil fuels undermines the environmental quality as it raises carbon emissions and other greenhouse gas emissions. Our findings are supported by related research works such as [25], [59] and [60]. The region's growing dependence and inefficient utilization of energy resources could further elucidate the positive effect of energy use on carbon emissions. Close to 89% of the total electricity used in the region comes from nonrenewable energy sources [33], while SSA scores poorly in energy efficiency.
The association between trade openness and carbon emissions is a negative one. Our results show that a 1% rise in trade openness is associated with a -0.008% decline in carbon emission at a 1% significance level. This finding is justified by the factor endowment theory, which proposes that the intensity of capital and labour in economies determines the link between trade openness and carbon emission. As it is well blessed with natural endowments and a huge workforce, SSA is likely to produce and trade environmentally friendly goods comparatively. In contrast, it is the reverse case for developed nations. Previous studies that underpin our findings include [1]; [61], and [62].
Consistent with the Environmental Kuznets hypothesis (EKC) is the effect of economic growth on carbon emission in this study. GDP exerts a positive influence on carbon emissions. A 1% rise in GDP leads to a rise in carbon emission by 0.053% at a 1% significance level. The same column (4) shows a negative but statistically significant coefficient of GDP's square term. A 1% boost in GDP results in a -0.029 reduction in carbon emission. This suggests the EKC hypothesis is supported by an inverted U-shape curve relationship between the two variables. This finding is also fully supported by [63] and [1].
Education is positively related to carbon emission. A 1% improvement in education translates to a 0.002% rise in carbon emission. This implies that the educational curriculum in the region does not significantly capture the need for environmental sensitization. This is consistent with the finding of [40], [64], [65], and [66].
Column (5) shows the findings of a sub-model that considers the interaction of ICT diffusion and the use of renewable energy sources. The findings suggest that renewable energy and C.T. jointly reduce carbon emission levels. The coefficient of the interaction term is found to be negative and highly significant at a 1% level, implying that renewable energy and ICT diffusion could jointly reduce emissions by -0.002%. The results suggest that renewable energy use dampens the impact of ICT diffusion on the quality of the environment. This result is further explained by the fact that the ICT sector continues to introduce renewable energy in its own domain and promote smarter technologies to reduce environmental emissions. However, column (5) results are qualitatively the same as the results of column (4). But quantitatively, the coefficients in column (4) carry more magnitude than those in column (5).
Column (6) presents the long-run results. The results conform to the ones contained in column (5) qualitatively. Besides the coefficients having more magnitude in column (6), the long-run results indicate that trade openness has a long-term favourable impact on carbon emissions. A 1% rise in trade openness is expected to raise emissions by 0.02%. This is unsurprising given the prevailing economic and institutional measures aimed at boosting economic activities and diversifying the region away from fossil fuel-based economies to promote green growth and industrialization while decisively tackling the energy poverty and crisis in the region.