In the frame of the Sustainable Development Goals (SDGs), the International Community has renewed its commitment to reach sustainable development by 2030 that encompass multiple aspects of human well-being. 17 goals were defined and include the environment, water and sanitation, energy, poverty, health, gender, economic growth among others [1] . This integrated view of development is not new, and has shifted from the assessment of the economic growth [2] to an integration of additional social and environmental dimensions like inequality (e.g. [3] ) education/literacy, and the WSS access (e.g. [4] or gender [5] . Composite indicators have been also built to provide synthetic measures of several aspects of development at national scale: for instance, the Inequality-adjusted Human Development Index (HDI) [6] ), the Multidimensional Poverty Index (MPI) [7] or the Employment-adjusted Human Development Index [8] . (Gain et al., 2016) developed a global water security index associated socio-economic aspects and biophysical indices to better assess the water sector, to contribute to the framework of the SDGs [9] . Holden et al. (2017) assessed and cross-analysed three aspects to measure sustainable development: “the human needs, ensuring social equity, and respecting environmental limits” [10] .

Following such trend to understand development through more than one aspect, this study extends the multi-dimensional approach conducted for year 2004 [11] . It aims at finding patterns and trends in country sustainable development over the period 1995-2010 from the point of view of the water sector. Indeed, considering the 17 sectors of the SDGs together would have been complex. The water and sanitation sector is a transversal sector that interacts with other sectors, such as education, health, agriculture, energy, environment/biodiversity and general economy, is estimated as a good entry point-proxy to assess country development. In addition, it has been included the Official Development Assistance flows (external financial support) which can be a significant amount depending on the recipient country [12] . The main objective is to analyse the country evolution considering several aspects: living conditions such as access to Water and Sanitation services, urbanisation and the housing conditions, the access to basic health care, the balance and estimation of water uses per main economic consumers, and the official development flows (general and water sector specific flows). This study focuses on developing countries where development is key to improve the living conditions of a significant part of the population.

The WatSan4dev database, developed by the Joint Research Centre in 2012 [11] , has been enhanced by adding and pre-processing the data series from 1995 to 2010 with a 5-year interval [13] . 103 developing countries are selected from Latin America, Africa and South Eastern Asia. These profiles are the framework to observe the country evolution over 1995 to 2010 and are characterised around 5 thematic factors: Human Development and Poverty (HDP), Agriculture pressure on water resources (AP), Official Development Assistance (ODA Composite Index), Water Resource (WR) and Governance (GOV).

Therefore, this paper presents the data sources and selected variables that are included in this study (Section 2), the methods and pre-processing performed to establish the coherence of the dataset (Section 3). It finally presents the results with the 5 country profiles that define different countries behaviours related to development, water resources management, governance and donors support (Section 4). The year 2005 is taken as reference for building the profiles because of being most complete and robust dataset as well as enabling a coherence validation with [11] .

The variables and data were collected from providers that are internationally mandated institutions (the World Bank, the OECD or the FAO). They delivered worldwide datasets at national scale for which validation and harmonisation processes were carried out. Validation processes performed by each data provider are described in [14] . The selected datasets were processed to set up the WatSan4tDev database, developed by the Joint Research Centre (JRC) of the European Commission (EC). From this database, 22 variables are selected related to the environment, economic and social development, governance and financial aid flows in developing countries and collected for years 1995, 2000, 2005, and 2010. The list of data gathered, their unit and source are detailed in Table 1. Note that Total Water Renewable Resources (TWRR), precipitations (Precipit), and percentage of land under risk of desertification (Desert risk) are considered as constant variables over the 15 years period.

As in [13] , no independent territories, developed countries, countries with less than 150,000 inhabitants and the countries with more than 35% missing data for the set of variables selected were not considered in this analysis. Eastern European countries and Central Asia are also excluded because focusing on Africa, Latin America and South-East Asia.

There is little information about uncertainty of data, only standards errors for Worldwide Governance Indexes (WGI) (between 4% and 7% in average) are given by observations (WGIs database). The data used take into account updates made by the data providers until May 2015.

Missing data rate is a significant constraint for several data series, therefore, an imputation process is setup to complete datasets with realistic values. Prior, the variables were processed using linear regression to adjust the outlier values. The multi-variate imputation software Amelia II which uses EMB algorithm has been used [15] . Additional and manual verifications were made based on the data available in previous or consecutive years and specific literature. Therefore, this data is to be used only for qualitative analysis, quantitative analysis should be avoided because of the heterogeneity of the data collection and computation methodologies.

A combination of methods is applied to the 22 variables, 103 countries and 4 years (1995, 2000, 2005, and 2010) as detailed in [13] . It consists in, first, performing a Principal Factor Analysis (PCA) to reduce the number of variables into coherent thematic principal factors [16] , then clustering the countries observations with 2005 as reference year to observe country change across profiles.

2005 was chosen as reference year because of being most completed and thus, robust dataset.

In more details, the PCA was first performed on the 2005 dataset with the projection as supplementary observations of the datasets for 1995, 2000, 2010 into the multidimensional 2005 space. This ensures the consistency of the comparison and the clustering of observations across time and makes the change of country profile observable and measurable against 2005. The spearman rank-order correlation is used as the dataset satisfies its two assumptions: 1) it does not required normal distributions; and, 2) the two-by-two variable relationships are monotonic. Varimax rotation is applied to maximize the variance of the loadings (Abdi, 2003). The first 5 main factors selected provide a simple and easy interpretable structure that is the data baseline of the clustering next step. The factors are selected because having an eigenvalue above 1 and at least two variables with an above 0.5 factor load [17] [18] .

The main two clustering methods used are the Agglomerative Hierarchical Clustering (AHC) and the k-means clustering (k-means) [19] [20] . The AHC is used in a first step, as a discovery tool to define the number of clusters that suits the dataset. The AHC analysis was enough stable, indicating that the appropriate number of clusters is 5 clusters [13] .

The k-means is then applied to 2005 observations setting 5 clusters because it provides more sharped cluster [21] . The observations of 1995, 2000 and 2010 were distributed, in a second step, into the 2005 clusters according to the shortest Euclidian distance with each cluster centroid (Appendix 2). In this way, it is easy to observe the progress of the countries along the period 1995-2010.

With regards to the methodology and the datasets several remarks are to be made [13] . First, the quantity and completeness of data constraint the accuracy and interpretation of this multivariate analysis. Therefore, only quantitative interpretation is enabled. Since the period is relatively short and introduces uncertainty in the duration of the change, the parameters to analyse are ODA CI and GOV factors because being the most responsive to short term events impacts (e.g. civil wars, natural disaster…). Therefore, a minimum of 30/40 years period would already improve the quality of the interpretation of country long term evolution, in particular for HDP and AP factors. The developed countries were excluded in this study. The main reason stands in the very different levels of economic development and access to basic services between developed and developing world. As a consequence, making such a global analysis will make less visible the differences within developing countries and mainly highlight the gap between developed and developing countries. Such approach would be useful in the case of a research aiming at assessing this gap.

The interest of WR factor can be improved by including water quality indicator or climate variability trends. In such case, it is expected to strength the relationship between country development and availability of water resources necessary to ensure it. This would be more interesting as the climate change affects the spatial and temporal distribution but also the amount of water available in particular in the sub-Sahelian Africa. In developed countries, the quality of freshwater is already a key issue subject to policy regulation. In the developing world, it is becoming a growing issue due to the lack of treatment in a context of demographic growth and high urbanisation [32] .

In addition, the AP is less discriminated across profiles if compared with the pilot study for 2004. In fact, Profiles 1 to 3 show similar behaviours, with a majority of water resources dedicated to Agriculture (>60%) confirming its strong pressure on these resources. This similarity is due to the missing variable “water use intensity in agriculture” that measured for 2004 the water efficiency of agriculture practices. Such variable allowed. This variable, together with irrigation variable, allowed to refine the analysis of agriculture dominance providing insights on the water-efficiency and productivity of the agricultural sector (from intensive agriculture using irrigation to less productive one mainly with rain-fed crops). This is to be noted that the irrigation practices are limited in the developing world and occurs almost exclusively in countries from profiles 1, 2 and 3 (mainly Latin American, MENA and South East Asian countries).

The AP is in this study focused on water resources uses however, a reinforcement of this factor could be done opening it toward a better evaluation of the agricultural sector as a whole. In fact agriculture sector is very linked to the water sector and the development of countries. In particular, it could be explored how to evaluate the intensivity of rainfed agriculture.

This analyse of cluster changes illustrates a key aspect: the diversity of the developing countries according to these 22 variables makes the five profiles fitting well or roughly each country situation. The reduction into 5 factors is a mean to reduce this multi-dimensional space and support the clustering of observations. The multiplicity of status and evolution is reflected by setting of several countries in between of two profiles, or distant from the average behaviour of the profile. This is, for instance, the case of Swaziland, Madagascar, South Africa, mentioned as in periphery or in between two profiles. The advantage of such clustering around few and understanding factors and profiles is to ease the overview of 103 developing countries, the identification of similar development behaviours (or on the contrary specific cases) and potential changes. This broad picture is a useful basis to orient the deepening of the analysis or to identify weakest countries according to one or several factors of interest.

Sustainable development is an international priority and countries are committed to the Sustainable Development Goals to be reached by 2030. As sustainable development is complex, the approach of this study was to analyse country development through the following factors: water resources availability and uses, the economic and social conditions, Official Development Assistance and governance context of the country.

This study performed a cross-cutting analysis on 22 indicators that represent environmental, socio-economic, governance and financial aid flows aspects. It considered a 15-year period with 4 years: 1995, 2000, 2005 and, 2010. Since [13] detailed the methodology applied to analyse this multi-variate dataset, this paper focused and analysed the trends according to 4 out of the 5 thematic axes: Human Development and Poverty (HDP), Agriculture pressure on water resources (AP), Official Development Assistance (ODA CI), a measure of Governance (GOV). Water Resources (WR) is estimated constant over this 15 years period, climate change impacts on Water Resources are generally observable on a longer period (minimum 30 years). Applying clustering methodologies, five reference profiles were identified and based on 2005 dataset which is the most complete dataset. The observations for 1995, 2000 and 2010 were in second step distributed among these profiles. Profile 4 and 5 are the poorest country profiles and profile 5 the less advantaged one.

In summary, this paper aimed at analysing the evolution of countries toward sustainable development and better water resources management. Over the 1995-2010 period, the HDP has progressed in all countries (expect Bahrain). GOV and ODACI trends show more diversity with positive and negative evolutions. AP remains globally driven by agriculture water use for the large majority of countries. AP decrement is generally due to the increase of the water dedicated to domestic-municipal sector. 16% of countries moved out a reference profile and are considered mid-term changes, while short term changes (5 cases) are due to the specific position of these countries between two profiles.

The main output of this 15-year analysis is that 22 countries changed cluster on this period and, 17 of these changes can be considered as mid-term changes. This is in line with the nature of the processes and reforms required for a country to improve its development, which correspond to long term efforts. A short majority of these changes (10 out of 17 countries) moved out of the most disadvantaged profiles to reach profile 1, 2 or 3. Out of these changes, 6 were due to the fact that the country of was already in the cluster periphery, showing divergences with its central behaviour. This highlights the diversity of country situation and therefore, the difficulty to define profiles that would fit well all countries. Most of the other observed shifts (11 countries) were due to an evolution of ODA CI and GOV factors. In fact, these factors are responsive to short term changes, such as political crisis or natural disaster that makes observable trends on this relative short period. This current limitation in the assessment of development will be overcome with an extended period (>30 years) as soon as the data will become available. The Official Development Assistance flow per recipient country fluctuates, in general, over the observation period. Note that a drop of governance is often associated to a drop of ODA [14] .

For future research, the analysis of country evolution will be complemented with variables to better assess the agricultural pressure and domestic water impact on the environment (water quality, resources efficiency of agriculture, etc.…) which will support the assessment of the sustainability of the country development. Additional variables measuring the climate variability will also reinforce the WR factor and enable the analysis of its impact on HDP. Inequality aspects and remittance flows can also be analysed to catch the quality and endogenous part of the ongoing development through, the enhancement of the measure of HDP and the complement of ODA, respectively.

The authors acknowledge that this paper is following up the methodological work done with Ms María del Rocío Rivas López, cited in reference (Rivas et al, 2016). This research benefited from the financial support of the Joint research Centre and European Commission through the project Mékrou.

Dondeynaz, C. and Carmona Moreno, C. (2018) Assessing Development Pathway Trends in Developing Countries over the 1995-2010 Period Using Multivariate Analysis. Journal of Water Resource and Protection, 10, 304-326. https://doi.org/10.4236/jwarp.2018.103018