Simply select any text on the page to have your computer read it to you.
Choose the preferred Text to Speech accent from your available system voices.
Jose Luis Cervera, Plan Eval; DevStat; Jose Vila, DevStat; Monica Luzure, Therese El Gemayel, UNEP; Souleiman Kacha, DevStat
Andrea Hinwood, UNEP; Joseph E. Flotemersch, U.S. Environmental Protection Agency; Ludgarde Coppens, UNEP; Maria Schade, UN-Water; Nada Matta, UNEP
UNEP’s Measuring Progress reports aim to improve the understanding of the progress made by SDG indicators by analysing their trends, including through analysis of interlinkages between the indicators of the SDG framework using statistical methodologies. While the Measuring Progress: Environment and the SDGs (2021) report focused on understanding the interlinkages through a statistical correlation analysis, the proposed methodology herein will further develop the statistical correlation analysis of one thematic area into a multivariable analysis by including additional socioeconomic and environmental factors that might impact the relationship between the analysed indicators.
The main objective is to use the driver-pressure-state-impactresponse (DPSIR) framework to investigate the relationship between actions taken by countries and their impacts on freshwater- and marine-related ecosystems as well as secondary impacts on human well being through the lens of the SDG indicators.
The methodology enables analysis of the potential impacts on freshwater- and marine-related ecosystems separately and includes different geographical levels, based on data availability. For freshwater-related ecosystems, the analysis is conducted at the global, national (Colombia and Mongolia) and basin levels (China). For marine-related ecosystems, analyses at the global and national (Sri Lanka) levels are conducted.
The statistical model features four types of components. Indicators within the first three components (drivers of change, state of the ecosystem and state of human well-being) are all selected from the SDGs indicator framework.
After defining the components of the statistical analysis, identifying and classifying SDG indicators into the above-identified categories was imperative (Figure 3.1). The selection of the indicators was done at the indicator level and separately for each type of ecosystem, considering the specificities of each ecosystem type according to experts’ knowledge.
Potential synergies were then identified between the selected indicators for further analysis (see Annex E). Potential synergies are based on the potential impact that one indicator (independent variable) might have on another indicator (dependent variable). An initial identification of potential synergies was complemented by online experts’ consultation. Theoretical models were then developed for the analysis of the state of freshwater- and marinerelated ecosystems and their impact on human well-being.
The next step entailed setting the criteria for data to be used in the statistical analysis. For global and national data, the number of observations was set to a minimum of 20 observations, no gaps in the time series and using available data to date. For global-level data, the criteria were set to have data from at least 50 per cent of the countries for a global aggregate to be generated. This aligns with UNEP’s methodology to aggregate data to the global level. Global aggregates (i.e. one value representing global value for one year) were used in the statistical analysis, in comparison with country data. Specifically for marine-related ecosystems, the 45 landlocked countries were excluded from the analysis due to their lack of direct accessibility to and benefit from marine and coastal areas.
Multiple sources of data were used for this analysis. For SDGrelated indicators, data were extracted from the SDG Indicators Database on 14 June 2022. Subsequent updates of the database were not considered in this analysis. Other sources were used, such as the Department of Economic and Social Affairs, Population Division[7] and the World Bank.[8]
For basin-level analysis, the number of observations was set to 15, with the time series starting from 2004 until the most recent available data without data gaps in the time series. Basin-level data were provided by the Government of China (received on 22 June 2022).
One of the main challenges of this analysis is related to the lack of data at the global and national levels. This point is very relevant, because only a partial study of the theoretical models has been done. Only some of the proposed sub-indicators had available data due to the numerous missing data existing in the time series. This limitation requires the design of an ad hoc measurement methodology, achieving an equilibrium between (i) the demanding requirements of the theoretical models and actual policy impact dissemination mechanisms and (ii) the data that can be used to measure such impacts. The following approaches were used to achieve this equilibrium:
To avoid the restriction of linear impact diffusion and the comparability of the impact being measured, and to homogenize the estimated impact (indicators’ units vary, that is, by per cent, age, km2, US$ and so on), all indicators’ data were transformed and standardized before the estimation of the instrumental models by using the logarithmic transformation.
The number of available variables in the theoretical model is larger than the number of available observations of each variable. Therefore, only partial instrumental models (i.e. models including a small subset of explicative variables) can be estimated.
To produce the general measurement model, all potential models including a strategy of three explanatory variables were estimated using ordinary least squares. The number of explanatory variables has been selected as the minimum number to estimate partial correlations instead of total correlations. The methodology opts for the minimum number of explanatory variables to maximize the degrees of freedom of the model and avoid model overfitting. Hence, instrumental models for which less than two coefficients are found to be statistically significant are dismissed. The regression models for each state variable are estimated independently. Formally, the instrumental regression for each ecosystem can be written as
After the estimation, the t-test is applied to test the null hypothesis for each individual impact coefficient being null at a significance level of 0.05. Models are re-estimated after the elimination of one non-significant coefficient, providing a measure of the intensity of impacts that can be considered as non-null. This impact measure is given by the corresponding estimated coefficient (β for impact in the state of the ecosystem and ϒ for impact in the state of human well-being). The value of the estimated coefficients that are significantly non-null for all the instrumental models for each response variable is presented below and used to conduct the following analysis of the results.