Theoretical Motivations and Hypotheses

Nation-states, multilateral organizations, and funding institutions have set themselves ambitious global biodiversity conservation and zero-deforestation targets and committed to increasing international finance and aligning financial flows to reach these. They have formed a variety of bilateral, multilateral, and transnational forest governance and funding arrangements to ensure the protection of forests and biodiversity. Given that deforestation and biodiversity loss continue to take place at alarming rates and the predominant belief that permanent deforestation is primarily commodity-driven the obvious questions arise:

The study of aid allocation has identified that it tends to be allocated according to recipients’ needs and merits and donors’ self-interests. Studies have been done to explain biodiversity aid allocation drivers and environmental aid in general, however, few studies have focused on analyzing dyadic regression analyses of conservation aid and even fewer approaching the question with a social networks approach. By following such an approach this study does not assume independence of observations but can control for network structure, the interdependence of aid allocation choices by donors, and the dependence on other relational ties between countries (exp. trade).

That is why this analysis focuses on studying the effect of FRC trade network and several country-level node attributes on bilateral biodiversity aid network.

Table 1: Hypotheses of the effects on bilateral biodiversity aid allocation by countries’ biodiversity and development related attributes; dyadic FRC trade ties and non-state biodiversity aid
Variable Hypothesis Indicator(s)
Recipient biodiversity need H1: A higher level of forest ecosystem extent, biodiversity richness, deforestation, and biodiversity loss will be associated with higher degrees of bilateral aid ties Tree Extent; Key areas for biodiversity (valuable biodiversity); Tree Loss; Deforestation; BHI index (biodiversity loss index)
Recipient development need H2: Poorer and less developed countries will be associated with higher degrees of bilateral aid ties GDP/capita; HDI
Recipient merit H3: Higher biodiversity governance performance will be associated with higher degrees of bilateral aid ties Biodiversity & Habitat index (HBI)
Recipient development interest H4: Higher levels of deforestation and biodiversity loss will be associated with higher out-degree of exporter trade ties and lower GDP/capita (dependency hypothesis) Domestic commodity consumption; Commodity exports (out-degree trade tie)
Country donor interest Several potential hypotheses: H5: Higher degrees of trade ties will be associated with higher degrees of bilateral aid ties as donor countries tend to allocate aid based on their economic self-interests to strengthen economic ties; H6: Higher degrees of non-state aid ties will be associated with higher degrees of bilateral aid ties as donor countries want to be selective and free-ride, H7: Higher degrees of trade ties will be associated with lower degrees of bilateral aid ties as donor countries want to preserve the export of deforestation effects to maintain own commodity consumption Commodity trade ties; non-state aid ties
Control Variables   Land Mass; Domestic consumption of FRC

Data and Network visualization

The data collection for this network was done by the author. First, countries most active in importing and exporting FRCs as categorized by the Global Canopy and countries with the highest tree cover loss identified by the WRI were included as nodes in the first mode of the network. Secondly, non-state donors active in providing biodiversity-related funding were included as nodes in the second mode of the network based on a previous actor analysis study of the global forest regime. The network includes 27 country nodes and 8 non-state donor nodes.

The two within mode edges included in the analysis are trade ties (brown) and bilateral biodiversity-related aid ties (green). The between mode edges are non-state biodiversity-related aid ties (gray). All ties are weighted (in Million USD) and are directed, from exporters to importers and from aid donors to aid recipients. Trade ties represent export flows of FRC commodities of key drivers of deforestation of each export country in 2019. Bilateral biodiversity-related aid ties represent biodiversity rio-marked committed aid flows between countries from 2010 to 2019. Non-state biodiversity-related aid ties represent reimbursed aid flows for the biodiversity sector from non-state donors to countries from 2010 to 2019. Trade data was retrieved from the Observatory of Economic Complexity (OEC) national trade database and biodiversity-related aid data was retrieved from the OECD’s Creditor Reporting System database.

Table 2: Descriptive Statistics: Degree centrality in trade, bilateral aid and non-state aid sub-networks
Min Max Mean number of edges
indegree bilateral aid tie (recipient) 0.0 2,532.36 412.81 365
outdegree bilateral aid tie (donor) 0.0 3,460.22 412.81 365
outdegree trade tie (export) 0.0 42,699.08 6,031.03 133
indegree trade tie (import) 55.1 50,409.87 6,031.03 133
indegree non-state aid tie (recipient) 0.0 245.00 43.62 80

Several development-related, commodity-related, and biodiversity-related country node attributes were selected as dependent variables (monadic node covariates) that might have an effect on the structure of the trade and biodiversity-aid networks. They reflect the indicators chosen to test the hypotheses (table 1).

Table 3: Descriptive Statistics: Actor Attributes: Total of 27 Countries
unit Min Max Mean
GDP/capita constant US$ in 2019 489.00 60,837.00 18,346.26
HDI index in 2019 with range 0-1 0.48 0.95 0.77
Domestic Consumption million USD in 2019 9,888.01 5,549,965.31 996,283.54
land mass million km2 0.04 17.10 2.91
Tree Extent MHa in 2000 0.58 761.20 113.54
Tree Loss Mha from 2001-2020 0.27 69.50 12.38
deforestation Mha from 2001-2020 0.00 42.46 3.95
Key areas for biodiversity km2 in 2020 17,586.00 1,677,831.00 348,387.74
BHI index in 2020 with range 0-100 39.30 82.40 57.06
B&H index in 2020 with range 0-100 19.00 88.80 63.24
Domestic Funding Biodiversity-related government expenditure in million USD in 2010-2019 120.00 159,273.00 27,579.77

A detailed description of the data, data collection, and its limitations can be found in the metadata report of this analysis.

The combination of this data gave rise to a multiplex, 2-mode, directed, and weighted network.

Data Modelling and Analysis

As the network to be analyzed is weighted and directed there are considerable limitations in the choice of modelling and analysis techniques. That is why this study employs a quadratic assignment procedure (QAP) analysis. (MR)QAP helps account for network dependencies, however it is unable to analyse them.

First, the QAP correlation analysis is employed to investigate the simple correlation between bilateral biodiversity aid and trade networks and the monadic node attributes expressed in dyadic relationships through sender and receiver attributes matrices.

Second, the QAP regression analysis (MRQAP) examines how the two potentially contrasting bilateral biodiversity aid and trade ties, along with country attributes, affect the bilateral aid network structure.

Originally, the goal was to study the plotted multilevel network including the second mode of non-state actors and aid ties. Unfortunately, due to limits in the sna QAP analysis package, only a one-mode multiplex network could be analyzed. This was done by transforming the non-state aid ties into a node-level attribute of the received volume of non-state funding losing so the two-mode network structure.

QAP Correlation

QAP Correlation between bilateral biodiversity aid and trade network

## [1] -0.2093123
## 
## QAP Test Results
## 
## Estimated p-values:
##  p(f(perm) >= f(d)): 0.964 
##  p(f(perm) <= f(d)): 0.039

As predicted by H7, trade and bilateral aid have a negative correlation, however, it is barely insignificant(net p-value 0.081).

QAP Correlation between bilateral aid network and country attributes

Table 5: QAP correlations between bilateral aid network and monadic covariates of aid sender and receiver countries
Coefficient p(f(perm) >= f(d)) p(f(perm) <= f(d) two sided p-value
GDP/capita_R -0.330 1.000 0.000 0.000
GDP/capita_S 0.640 0.000 1.000 0.000
HDI_R -0.310 1.000 0.000 0.000
HDI_S 0.550 0.000 1.000 0.000
Land_Mass_R -0.100 0.920 0.080 0.160
Land_Mass_S 0.070 0.350 0.650 0.700
Tree_Loss_R -0.050 0.740 0.260 0.520
Tree_Loss_S -0.006 0.510 0.490 0.980
Tree_Extent_R -0.050 0.750 0.250 0.500
Tree_Extent_S -0.050 0.600 0.400 0.800
Deforestation_R 0.140 0.000 1.000 0.000
Deforestation_S -0.190 0.980 0.020 0.040
Area_Biodiversity_R -0.070 0.800 0.200 0.400
Area_Biodiversity_S 0.090 0.240 0.760 0.480
BHI_R 0.030 0.340 0.660 0.680
BHI_S -0.100 0.750 0.250 0.500
B&H_R -0.180 0.996 0.004 0.008
B&H_S 0.340 0.005 0.995 0.010
Domestic_Biod._Funding_R -0.030 0.640 0.360 0.720
Domestic_Biod._Funding_S 0.090 0.260 0.750 0.510
Multilateral_aid_R 0.220 0.000 1.000 0.000
Multilateral_aid_S -0.340 1.000 0.000 0.000
Domestic_Consumption_R -0.090 0.840 0.160 0.320
Domestic_Consumption_S 0.390 0.017 0.983 0.034

None of the variables have a specially high correlation with the bilateral aid network. The highest correlation is a positive correlation (0.64) between the GDP per capita of the sender and the network. This might also be given by the weight unit of the bilateral aid tie (Million USD).

Only 6 of the tested monadic covariates of the sender and 5 of the receiver were statistically significant (marked in green). These results give us a hint that hypotheses H2, H1 (only by forest-related biodiversity conservation goals) could be correct and H3 wrong.

QAP Correlation between trade network and country attributes

Table 6: QAP correlations between trade network and monadic covariates of commodity exporter(sender) and importing(receiver) countries
Coefficient p(f(perm) >= f(d)) p(f(perm) <= f(d) two sided p-value
GDP/capita_R 0.141 1.000 0.000 0.000
GDP/capita_S -0.132 0.000 1.000 0.000
HDI_R 0.199 1.000 0.000 0.000
HDI_S -0.091 0.000 1.000 0.000
Land_Mass_R 0.072 0.915 0.085 0.170
Land_Mass_S 0.446 0.329 0.671 0.658
Tree_Loss_R 0.025 0.741 0.259 0.518
Tree_Loss_S 0.471 0.492 0.508 0.984
Tree_Extent_R 0.020 0.766 0.234 0.468
Tree_Extent_S 0.438 0.608 0.392 0.784
Deforestation_R -0.014 0.000 1.000 0.000
Deforestation_S 0.326 0.989 0.011 0.022
Area_Biodiversity_R 0.089 0.816 0.184 0.368
Area_Biodiversity_S 0.440 0.274 0.726 0.548
BHI_R -0.080 0.298 0.702 0.596
BHI_S 0.581 0.767 0.233 0.466
B&H_R 0.011 0.997 0.003 0.006
B&H_S -0.092 0.002 0.998 0.004
Domestic_Biod._Funding_R 0.117 0.630 0.370 0.740
Domestic_Biod._Funding_S 0.127 0.276 0.724 0.552
Multilateral_aid_R -0.019 0.000 1.000 0.000
Multilateral_aid_S 0.317 1.000 0.000 0.000
Domestic_Consumption_R 0.166 0.845 0.155 0.310
Domestic_Consumption_S 0.289 0.013 0.987 0.026

Here too, none of the variables have a specially high correlation with the trade network. The highest is positive correlation(0.58) between the BHI index of the commodity exporter and the network. However this relationship is not statistically significant. The highest statistical significant relationship is a positive correlation between deforestation of the commodity exporter and the network (0.33).

Only 6 of the tested monadic covariates of the exporter and 5 of the importer were statistically significant (marked in green). These results give us a hint that hypothesis 4 (at least for commodity-driven deforestation) could be correct.

MRQAP

Several models were constructed to test the hypotheses in table 1. Dependent variables were integrated into the models based on their indication of the hypotheses and maximizing the models’ adjusted R-squared values.

Table 7: QAP multiple regression (MRQAP) results of 9 models based on hypotheses in table 1
Model 1: H1 Model 2: H2 Model 3: H3 Model 4: H1&2 Model 5: H1&3 Model 6: H1,2,3 Model 7: H1,H2,H6 Model 8: H4 Model 9: H4, H5, H7
Intercept 0.2523 0.2141 -0.0233 0.0606 0.0535 0.0543 0.0284 0.3876*** -0.0091
trade_tie               0.0000 -0.0621*
GDP/capita_R   -0.0000***   -0.0000***   0.0000** 0.0000, 0.0000** 0.0000*
GDP/capita_S   0.0000***   0.0000**   0.0000** 0.0000 -0.0194* 0.0000***
HDI_R   -0.2216   -0.1750   -0.1526 -0.1771    
HDI_S   0.0113   0.0841   0.1144 0.0841    
Land_Mass_R -0.0015   -0.0117* 0.0127* -0.0122 0.0097 0.0067    
Land_Mass_S 0.0100   0.0109 -0.0183 0.0296 -0.0224 -0.0183    
Tree_Loss_R -0.0131     0.0009 -0.0094* 0.0001 0.0036    
Tree_Loss_S 0.0317*     0.0005 0.0248 -0.0004 0.0005 0.0020** -0.0031
Tree_Extent_R 0.0008     -0.0006 0.0006 -0.0005 -0.0008    
Tree_Extent_S -0.0029*     0.0002 -0.0026* 0.0004 0.0002 0.0135** 0.0001
Deforestation_R 0.0120**     0.0054* 0.0110*** 0.0058* 0.0020    
Deforestation_S -0.0173*     -0.0026 -0.0155* -0.0020 -0.0026 0.0000** 0.0026
Area_Biodiversity_R                  
Area_Biodiversity_S                  
BHI_R 0.0046     0.0017 0.0053** 0.0021 0.0020    
BHI_S -0.0055     0.0000 -0.0068 0.0005 0.0000    
B&H_R     -0.0038**   -0.0042*** -0.0007      
B&H_S     0.0072*   0.0078** -0.0009      
Domestic_Biod._Funding_R                  
Domestic_Biod._Funding_S                  
Non-state_aid_R             0.0005    
Non-state_aid_S                  
Domestic_Consumption_R               0.0000*** 0.0000*
Domestic_Consumption_S               0.0000*** 0.0000
adjusted R-squared 0.2037 0.5007 0.1675 0.5221 0.3781 0.5218 0.5239 0.2527 0.715
model p-value 0 0 0 0 0 0 0 0 0

1

Model 9 has the highest adjusted R-squared and hence explains the most variation of the observed network. It lets us reject the H7 null-hypothesis and the negative, albeit small, coefficient, for the trade tie implies that there is a tendency to allocate less aid to countries from whom donors import FRC.

Additionally, all models that integrate recipient development need variables (H2) have a high adjusted R-squared. The coefficients for GDP/capita of aid senders and receivers are significant, however there value is almost 0, which surprisingly suggests that they have no effect on biodiversity aid allocation.

Model 5 suggests that H1 is correct and H3 is wrong. There is hence a tendency to allocate aid to countries with higher biodiversity loss and deforestation and lower biodiversity governance performance. However, the significant coefficients are quite small.

Finally, the analysis of the effect of non-state aid was not significant (model 7).

Conclusion

This analysis suggests that CFR trade ties and hence donor country’s economic interests, recipients biodiversity needs, and recipients merit have a small effect on bilateral biodiversity aid allocation and that recipients development needs do not.

Further research can be done:

  1. The dependent variable for this analysis is the bilateral biodiversity aid network. An exception represents model 8 (H4) as it has the trade network instead of the bilateral aid network as dependent variable.↩︎