WPS6417 Policy Research Working Paper 6417 Municipal Vulnerability to Climate Change and Climate-Related Events in Mexico Christian Borja-Vega Alejandro de la Fuente The World Bank Social Development Department Sustainable Development Network April 2013 Policy Research Working Paper 6417 Abstract A climate change vulnerability index in agriculture is which increases uncertainty for harvest periods, presented at the municipal level in Mexico. Because and for agricultural yields and outputs. The index the index is built with a multidimensional approach shows at baseline that coastal areas host some of the to vulnerability (exposure, sensitivity and adaptive most vulnerable municipalities to climate change in capacity), it represents a tool for policy makers, Mexico. However, it also shows that the Northwest academics and government alike to inform decisions and Central regions will likely experience the largest about climate change resilience and regional variations shifts in vulnerability between 2005 and 2045. Finally, within the country. The index entails baseline (2005) and vulnerability is found to vary according to specific prediction (2045) levels based on historic climate data variables: municipalities with higher vulnerability have and future-climate modeling. The results of the analysis more adverse socio-demographic conditions. With the suggest a wide variation in municipal vulnerability vast municipal data available in Mexico, further sub- across the country at baseline and prediction points. index estimations can lead to answers for specific policy The vulnerability index shows that highly vulnerable and research questions. municipalities demonstrate higher climate extremes, This paper is a product of the Social Development Department, Sustainable Development Network. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at cborjavega@worldbank.org and adelafuente@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Municipal Vulnerability to Climate Change and Climate- Related Events in Mexico † ‡ Christian Borja-Vega and Alejandro de la Fuente Keywords: Vulnerability, Climate Change, Mexico JEL Codes: I31, I38, Q15, O54, Q54 Sector Board: Social Development † This is a background paper for the study Social Dimensions of Climate Change in Mexico, coordinated by Rodrigo Serrano-Berthet. We like to thank comments to previous drafts of this document from Emmanuel Skoufias, Svetlana Edmeades, Rasmus Heltberg, Ana Elisa Bucher, Rodrigo Serrano-Berthet, Martin Henry Lenihan, John Nash, Alejandro Monterroso and Alexandra Ortiz. Victor Manuel Celaya del Toro and his team at the Ministry of Agriculture, Livestock, Rural Development, Fisheries and Food (SAGARPA) also provided valuable inputs for the selection of indicators. We are also grateful to the Agricultural and Fisheries Information Service (SIAP) from SAGARPA for providing part of the data employed in the analysis. The findings, interpretations and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the World Bank, its Executive Directors, or the countries they represent. ‡ World Bank. Email correspondence: Christian Borja-Vega (cborjavega@worldbank.org) and Alejandro de la Fuente (adelafuente@worlbank.org). Introduction Mexico is among the most exposed countries to natural hazards in the world (World Bank, 2005; de la Fuente, 2009) 1. Only last year Mexico experienced one of its worst droughts in seven decades 2, and suffered historical losses in 2010 due to hurricane Alex in northeastern Mexico, and then serious floods in various southern states. Moreover, recent evidence and predictions indicate that climate change is accelerating and will lead to wide-ranging shifts in climate variability (or indicators) (UNISDR, 2009; IPCC, 2012), with consequent increases in extreme weather events, and associated likely impacts on economic activities closely linked to climate. Agriculture is one of the sectors that climate change is expected to hit hardest. Extreme weather affects agricultural productivity, and can raise the price of staple grains important to poor households. Mexican agriculture is particularly vulnerable to climate change. The participation of agriculture in the economy has shrunk over the past decades 3, but about 3 million smallholders grow maize, mainly for subsistence. Unfortunately, they do so under very precarious conditions and have restricted ability to adapt given their low income. Rain fed maize production is a critical livelihood strategy for the poor in Mexico. It therefore makes sense to start assessing the potential vulnerability of agriculture to climate change. This paper develops a multidimensional municipal index that assesses the vulnerability (as defined by the Intergovernmental Panel on Climate Change, IPCC) of the agricultural sector in Mexico to climatic contingencies and climate change. The aim is to better understand how and why vulnerability to climate change and climate variability varies by municipality in Mexico. Akin to the marginality index4 developed in Mexico in the mid-1990s, such an index could facilitate the (re)design of new interventions for reducing the risk to the most vulnerable populations, especially small subsistence farmers who have limited ability to adapt to adverse economic and climatic events. The index can also be used to improve the targeting of sectoral plans and the current federal system of disaster compensation and 1 “Government Expenditures in Pre and Post Disaster Risk Management� Background Note for World Bank-U.N. Assessment Natural Hazards, Unnatural Disasters: Effective Prevention through the Economic Lens. November 2009. 2 According to the Mexican Government, 21 Mexican states were affected by one of the most intense droughts in the last 70 years. The states mostly affected by this drought are Chihuahua, Coahuila, Durango, San Luis Potosi, Zacatecas and Aguascalientes, which constitute the northern-western and central agricultural areas. The percent of harvest lost in beans for 2010-2011 was around 60 percent and estimated losses amount 100 USD million. (SAGARPA, 2012; ) 3 In 2010, agriculture accounted for only 3.6% of GDP, down from 7% in 1980, and 25% in 1970; Baez and Mason, 2008; INEGI, 2010. 4 The marginality index is a policy-oriented indicator, created by the National Population Council in Mexico (CONAPO) that measures the lack of basic public infrastructure, as well as education and material living conditions at the state and municipal levels. It has been based traditionally on census data, and uses the following indicators for its construction: the share of illiterate people over 15 years; the share of people over 15 years without completed primary education; the share of the employed labor force earning less than twice the minimum wage (approx. US$7 per day); the share of people living in households in localities with less than 5,000 inhabitants; the share of people without running water, electricity, sewage facilities, and solid floor materials and the share of households with some degree of overcrowding. Principal component statistical analysis is performed to construct the index which is a normalized Z- score ranging between -3 and 3 standard deviations that correspond to very low and very high marginality, respectively (CONAPO, 2006). 2 state agricultural subsidies to the most vulnerable groups and sectors. Finally, the proposed methodology and use of comparable municipal information could allow analysts to monitor the progress of new adaptation policies for climate change within the agricultural sector. Despite important advances in the understanding of vulnerability, quantitative estimates of spatial and temporal vulnerability at the sub-national level are rare, and methodologies for doing this are very much in their infancy. 5 Vulnerability indices have been computed at the national level (in Europe, World Bank, 2009) and at the regional/provincial/district level (districts in India, O’Brien et al. 2004; regional in Brazil, Fontes, 2009), but never nationwide at the municipal level of resolution. Recently climate change vulnerability indices have been constructed for Eastern European and Central Asian (ECA) countries, including Tajikistan (Fay and Patel, 2008; Heltberg and Bonch-Osmolovskiy, 2010). The Tajikistan indices combined indicators that capture each country‘s exposure, sensitivity and adaptive capacity to climate change. The Climate Change Vulnerability Index (CCVI) in ECA assesses current vulnerability to climate change, at the regional (Rayons; local municipal subdivision) or provincial (Oblast; Gorno-Badakshan Autonomous Provinces) levels and is useful to integrate and prioritize regional policies. However, due to the lack of data indices cannot be constructed at the more disaggregated municipal level. Much of the data and indicators necessary to compute a municipal level CCVI must be collected on a regular and consistent basis. Mexico represents a good case to build a CCVI given its high exposure to natural hazards (and climate change) along with a vast amount of data and indicators at state and municipal levels. In addition, Mexico routinely and consistently collects quantitative historical data on climate and temperature, and socioeconomic indicators at municipal and state levels. A geographically disaggregated picture of vulnerability can help with the preparation of adaptation strategies and allocation of financial and technical assistance to municipalities. This would happen in the same manner to poverty maps, in which Mexico has a rich experience, that support the design and financing of anti-poverty policies and programs. Moreover, Mexico has established sound climate change adaptation policies that are necessary to cope with future climate-related threats. These efforts can be complemented by constructing an analytical tool that is useful for policy makers and local governments to prioritize resources and actions necessary to minimize climate change risks in the future. 5 The literature suggests two existing approaches to assess vulnerability: As an “endpoint’, in terms of the amount of damage in a system caused by a particular climate event; and as a “starting point�, looking at the existing state of a system before facing a particular phenomenon (Kelly and Adger, 2000). In the “endpoint� approach, vulnerability is a residual of climate change impacts after adaptation; therefore, it is the net impact of climate change (Ribot, 1995; Clark et al, 2000; Luers, 2003). Most index vulnerability assessments have applied the “endpoint� approach looking at historic climatic variability, without making future projections of climate change. Studies based on the “starting point� approach would evaluate the different factors that can cause a society to become vulnerable. This research will assess the social and economic processes that underlie climate vulnerability from a “starting point� approach. We agree that the ‘adaptation deficit’—excessive vulnerability to current climate variability—is a good proxy of future vulnerability to climate change (e.g., World Bank 2009b). This has led to our main focus on understanding vulnerability to current climate variability. 3 The study employs geo-physical data on climate at baseline (2005) 6 and its projections due to climate change (2045), using nine climate models (See Annex). It also relies on household surveys and censuses of municipalities and rural producers (see Annex for a full list of these data sources). A set of indicators thought to be important for assessing agricultural vulnerability were chosen in close consultation with counterparts from the Ministry of Agriculture (SAGARPA) in the Mexican government. All data were merged into a single dataset to conduct the statistical aggregation of the index, after ruling out those variables that showed high endogeneity. Once the final list of variables was selected, these indicators were combined through Principal Components Analysis to compute a vulnerability index at baseline. Then the index was recomputed based on projected climate scenarios. Alternative indicators on climate variability 7 and socio-economic factors for PCA construction were used to verify the stability of the index (see Annex). The estimates presented here disaggregate vulnerability at the municipal level. The index allows comparisons across space and time. Our main findings suggest that the effects of climate change in Mexico will be uneven across municipalities, regardless of the model employed. Predictions point to higher vulnerability increases in central and northern Mexico; and states with the highest vulnerability at baseline are in coastal areas (Pacific coast, Yucatan peninsula and Gulf of Mexico). All models also showed that states with high poverty rates have consistently higher vulnerability at baseline and over the long-term. Overall, the index shows the highest increases in vulnerability for states such as Zacatecas, Yucatan, Guanajuato, Chiapas, and Chihuahua. Other states, such as Oaxaca, Puebla and Tlaxcala, also show important increases in the index between 2005 and 2045. These states are located in Coastal and Central-Northern regions, with relatively lower levels of human development. The states that experienced the greatest decreases in vulnerability between baseline and prediction periods are Tabasco, Sonora, Campeche, Sinaloa and Nayarit. Tabasco and Campeche are located in high vulnerability areas subject to floods and hurricanes that affect all types of farmers. However, Tabasco and Campeche have reported relatively lower agricultural losses in the presence of recent climate-related extreme events, due to their participation in Catastrophic Agricultural Insurance (ECLAC, 2008). On the other hand, Sonora, Sinaloa and Nayarit are pacific northern states with relatively high human 6 Climate and temperature data included in the analysis cover the period from 1960 to 2005. 7 A main model was estimated using Growing Degree Days (GDD-Temp) and the coefficient of variation of rain (CVR) as climate variability measures for the 1960-2005 and 2005-2045 periods. An alternative model included more specific climate variability measures such as the total number of frost days (<10 Cº), the number of days with rain above 10mm, the maximum number of consecutive dry days, and the percentage of rain above the high 95 percentile. These indicators were selected because they are well accepted and defined by the literature for Mexico (Peralta et al. 2009; Biasutti et al. 2011) 4 development and agricultural indicators. These states also diversify their crops substantially and keep a high coverage of irrigation in the agricultural sector 8. I. Conceptual Framework The framework for this paper is an adaptation of the IPCC’s vulnerability framework, which distinguishes between exposure, sensitivity, and adaptive capacity. The vulnerability of people can be reduced by decreasing the exposure and sensitivity of people, assets and livelihoods to climate risks, and by increasing the adaptive capacity of individuals, households, communities, and governments. Key terms are defined in the Glossary. Figure 1 offers a framework for understanding how exposure, an exogenous driver of vulnerability, interacts with endogenous drivers – sensitivity and adaptive capacity – to create vulnerability and its opposite, resilience. The level of a community’s vulnerability determines the frequency and severity of climate change impacts. By contrast, a resilient community will not be significantly impacted by climate change. Throughout this paper we use terms such as risk, vulnerability, exposure and hazard in very specific ways. There is an ongoing debate on the definitions of these terms, which are used to mean different things by different disciplines. Sorting out the differences in semantics is important for identifying causal relationships between climate change-related risks and human vulnerability, and for designing interventions to help people manage risk and vulnerability. This paper tries to present a coherent approach, focused on how risks associated with climate change may contribute to the vulnerability of individuals and households. In this framework, it is the interaction of exposure and sensitivity to risk, with adaptive actions that determine vulnerability 9. The IPCC definition characterizes vulnerability (to climate change) as a function of a system’s exposure and sensitivity to climatic stimuli and its capacity to adapt to their (adverse) effects (IPCC 2007), which corresponds to outcome (or end-point) vulnerability, but it does not provide a clear definition of these attributes or the relationship between them 10. 8 According to the Food and Agriculture Organization’s Aquastat, Sonora and Sinaloa concentrate over 25 percent of the total irrigated land in Mexico, for both irrigation districts and irrigation units. See http://www.fao.org/nr/water/aquastat/countries_regions/mexico/indexesp.stm 9 Vulnerability: the extent to which a natural or social system is susceptible to sustaining damage from climate change (IPCC 2001). For practical purposes, this means that a person is vulnerable to climate change risks if he/she has a high probability of becoming poor, sick, or of food insecurity due to climate change related events. 10 According to Fussel (2009) it is crucial to guide the development of any vulnerability index, or set of indicators. Given the diversity of decision contexts that can be informed by climate change vulnerability assessments and of normative preferences, the design of vulnerability indices is as much a political as a scientific task. Normative differences may strongly influence the combination of diverse information sources into an aggregated vulnerability index. Normative challenges include the aggregation of future and current climate risks. 5 As one will notice, in Figure 1 we have further adapted the IPCC vulnerability framework. In practice, it is difficult distinguish what counts as sensitivity and what is adaptive capacity, since they both deal with similar issues. For example, poverty is a good indicator of a community’s sensitivity; since poor communities are often more sensitive to impacts from climate change, however, the lack of income and access to resources are important characteristics of adaptive capacity. The same can be said for other issues. For example, forest cover prevents soil erosion and run-off, thereby increasing adaptive capacity, simultaneously, the loss of forest cover makes it so erosion and soil run-off are more likely a result of climate exposure, which means that the communities become more sensitive. So to separate issues such as poverty or forest cover into separate categories is problematic. For this reason, we have separated exogenous drivers of vulnerability (exposure) - which are not immediately impacted by human activity (excluding the role humans play in carbon emissions) – from endogenous drivers of vulnerability such as sensitivity and adaptive capacity. Sensitivity and adaptive capacity are basically two sides to the same coin in that the former refers to characteristics that increase vulnerability and the latter refers to traits that reduce it. Figure 1 - Conceptual Framework: Drivers of vulnerability and impacts from climate change Exposure (of people and assets to) Exogenous Drivers Hazard of vulnerability Climate variability Sensitivity Adaptive Capacity Endogenous  Natural resource based Risk insurance Drivers of livelihoods Access to  Coastal livelihoods vulnerability resources  Land or housing at risk from Resilient extreme weather  Poverty livelihoods Impacts on individual, household and community welfare 6 II. Data Sources The units of analysis for this study are 2,200 of the 2,454 municipalities in Mexico. The conceptual framework proposed requires variables that capture exposure, sensitivity and adaptive capacity to estimate vulnerability. The analysis uses four types of information: (i) historic and projected changes in precipitation and temperature, weather and climatic shock data, and use of specific variability indicators (frost and drought days, rain level variation and extremes); (ii) agricultural production, socio-economic conditions, infrastructure, and geographic data; (iii) poverty rates and other population-related variables; human, social and financial capitals; historical subsidies and transfers to municipalities; and (iv) climate scenario projections based on scientific climate models (see Annex). Agricultural and socio-economic data come from the Agroalimentary and Fisheries Information Service (Servicio de Información y Estadística Agroalimentaria y Pesquera— SIAP) of the Ministry of Agriculture (SAGARPA). Weather data comes from meteorological stations and the National Weather Service (Servicio Meteorológico Nacional—SMN, and the National Water Commission (CONAGUA)); and all climate models (including the projections of temperature and rainfall) are credited to the Coupled Model Intercomparison Project Phase 3 (CMIP3) of the World Climate Research Programme (WCRP) referenced in the Intergovernmental Panel on Climate Change’s (IPCC) Third and Fourth Assessment Report. Poverty rates were obtained through small area estimation techniques using data from the Income and Expenditure Household Survey (ENIGH) and the Count of Population and Housing 2005. Population data come from the National Population Council (CONAPO). Finally, important indicators were collected from the Summary Statistics of the 2007 Agricultural Census (INEGI). All data are available at the municipal (county) level. (See Annex Figure I and Table 1 for summary statistics and a detailed explanation of their construction.) The selection of variables for each component was made in consultation with officers at the Ministry of Agriculture in Mexico, and by reviewing relevant literature. A large fraction of the population in municipalities relies also on rain fed Maize production as the main economic activity. Climate-related indicators show large variability across municipalities. For instance the standard deviation of annual rainfall (mm) is almost the same as the average annual rainfall. The maximum rain levels reported in municipalities is almost ten times the average rainfall. Socio-demographic characteristics also vary considerably. There are municipalities with practically no access to services, while others have almost universal coverage. In a similar fashion, infant mortality and poverty rates show large standard deviations with respect to their means. Finally, some agricultural variables are measured in agricultural production units 11, and not necessarily relative to households or populations. 11 Concept defined by the Ministry of Agriculture (SAGARPA) and by the 2007 Agricultural Census, where a production unit refers to formal production arrangement of more than one individual to exploit individual or communal land. Therefore, there can be multiple production units headed by households or one production unit (farming companies) in leased collective land. 7 III. Methodology - PCA Method to Build Multi-Dimensional Indices Principal Components Analysis (PCA) was used to build a composite index for climate- related vulnerability. Because of the complexity of interactions between social, economic, climatic, disaster and agricultural dimensions, using the PCA method to aggregate variables into a single index is an efficient way to construct risk categories. The primary problem when constructing a PCA index is the choice of component indicators. Most indices use only a few variables, but the principal components methodology allows the use of a large number of continuous variables 12. As stated earlier, the variables were selected to capture the exposure, sensitivity and adaptive capacity of households and communities to climate-related shocks or events. Choosing appropriate component indicators minimizes errors and differences in measurement across municipalities. At the same time, the variables must fit consistently into the same general categories mentioned. PCA creates uncorrelated indices or components, where each component is a linear weighted combination of the initial variables. 𝑃𝐶1 = 𝑎11 𝑋1 + 𝑎12 𝑋2 + ⋯ + 𝑎1𝑛 𝑋𝑛 : : 𝑃𝐶𝑚 = 𝑎𝑚1 𝑋1 + 𝑎𝑚2 𝑋2 + ⋯ + 𝑎𝑚𝑛 𝑋𝑛 where amn represents the weight for the mth principal component in the nth variable. The weights for each principal component are given in the correlation matrix, or if the original data were standardized, a covariance matrix (multi-dimension correlations). In the case where multiple variables interrelate, covariance matrices are used as weights. The components are ordered so that the first component (PC1) explains the largest possible amount of variation in the original data, subject to the constraint that the sum of the squared weights (a211+ a212+…+ a21n) is equal to one. The second component (PC2) is uncorrelated with the first component, and explains additional but less variation than the first component, subject to the same constraint. Subsequent components are uncorrelated with previous components; therefore, each component captures an additional small variation with respect to other variables within the data, while explaining smaller and smaller proportions of the variation of the original 12 Data in categorical form are not suitable for PCA, as the categories are converted into a quantitative scale which does not have any meaning. To avoid this, qualitative categorical variables should be re-coded into binary variables. Another data issue is that of missing values. Cortinovis et al. (1993) excluded households with at least one missing value from their analysis to develop socio-economic groups. Gwatkin et al. (2000) replaced missing values with the mean value for that variable. Given that some indicators might have few observations by Municipalities in certain surveys, it is convenient to replace the mean value of each geographical unit. 8 variables. The higher the degree of correlation among the original variables in the data, the fewer components required. Once the specific variables have been detailed, two interrelated issues must be addressed concerning the construction of a PCA index. The underlying variables need to be converted to compatible scales so they can be combined to produce a single index. All variables were transformed into a normal standard distribution with a mean of 0 and standard deviation equal to unity. The second issue is the choice of weights for each of variable. The issue is not just to give the appropriate weight to each of the component statistics, but also to take into account any correlation between the component statistics. Ultimately, the weights calculated at baseline (2005) for the CCVI are structurally the same weights used for the predicted scenarios in 2045. The variance (λ) for each principal component is given by the eigenvalue of the corresponding eigenvector. As the sum of the eigenvalues equals the number of variables in the initial data set, the proportion of the total variation in the original data set accounted by each principal component is given by λi/n. The second component (PC2) is completely uncorrelated with the first component, and explains additional but less variation than the first component, subject to the same constraint. Subsequent components are uncorrelated with previous components; therefore, each component captures an additional dimension in the data. McKenzie (2004) highlights that a major challenge for PCA-based indices is to ensure the range of variables included have enough non-missing values to avoid problems of ‘clumping’ and ‘truncation’. In the case of our index, we used a wide variety of variables collected as administrative records, or from CENSUS data (population and agricultural). In this sense, non-missing data in each municipality are relatively small so clumping and truncation are not affected by estimation errors. In addition, according to McKenzie (2004) the problems of clumping or truncating indices can affect the variability of the index, so the first principal component needs to be constructed for each municipality relative to its standard deviation, instead of using the standard deviation of the all municipalities. Construction of Weights Weights Based on Component Variance Explained at Baseline Discriminating variables through PCA can be helpful in selecting the weights to construct the index based on the amount of variance explained for each component. The proportion of variance explained by each relevant variable is a strategy also used to weight them. The principal factors or components that explain the outcomes in the data always explain in a larger proportion the variance compared to the rest of the components. The position of each observation with the proportion of variance explained according to each component is calculated as a linear combination of the original variables. A simple regression using the 9 principal component variables and the outcome variable would reproduce almost the exact same weights as the proportion of variance explained by each component, so: 𝑌𝑘𝑟 = 𝑎𝑘1 𝑋𝑘1 + 𝑎𝑘2 𝑋2 + ⋯ + 𝑎𝑘� 𝑋𝑘� In interpreting the principal components, it is often useful to know the correlations of the original variables with the principal components. The correlation of variable Xi and principal component Yj is 𝑎2 𝑉𝑎𝑟(𝑌 𝑗 𝑟𝑖𝑗 = � 𝑖𝑗 � 𝑆𝑖𝑖 But weighting based on the percent of variance explained by each factor also involves a certain amount “rule of thumb�. One common criterion is to use principal components at the point at which the next principal component offers a large increase in the total variance explained and weights can be used at baseline and prediction points. A second criteria is to include all those PCs up to a predetermined total percent variance explained (structural weights), such as 90%. A third standard is to ignore components whose variance explained is less than 1 when a correlation matrix is used or less than the average variance explained when a covariance matrix is used 13. Estimation Procedure We ran several specifications to estimate the CCVI. The models incorporated variables with the highest explanatory power. In addition, variables were added in the models to test the stability and sensitivity of the index. This proved to be helpful in reducing the amount of variables used to construct the index without losing conceptual rigor 14. In addition, testing multiple variables for estimating the index helped to identify and remove endogenous variables and substitute them with variables that better fit the model. All variables were standardized into a normal distribution, and outliers were removed to build the index. Outliers were identified based on the method by Davies and Gather (1993). The distribution of outliers was tested by constructing cutoff points for the index. The cutoff points were then used to test each variable for each municipality. When a variable failed to pass the Bonferroni’s correction, which sets the alpha-value for the entire set of n comparisons equal to alpha, by taking the alpha-value for each comparison equal to alpha/n, it was not included in the model: when the value is half a percent point within an extreme cutoff point then the value was considered an outlier. Around 10 to 25 municipalities were withdrawn from the index estimation as outliers representing 0.1 13 The distributions of each variable should be checked for normality and transforms used where necessary to correct high degrees of skewness in particular. Outliers should be removed from the data set as they can dominate the results of a principal components analysis. 14 Annex show results for other PCA models run as robustness and index sensitivity tests. 10 percent of the total number of municipalities. There were three specifications used to estimate the index. The higher the consistency of index distribution and the ranking (of municipalities), based on relative risk, the better the model fit. In addition, models were built with and without outliers to verify the influence that outliers had on the index distribution. Endogeneity tests were carried to eliminate variables. In some cases endogenous variables were substituted with proxy ones. Once endogenous variables were identified and removed, the estimation procedure was improved by incorporating other indicators collected at the municipal level that strengthened the conceptual model and proxy for relevant characteristics. For instance, Table 2 shows the endogeneity tests in four variables, all of them endogenous. In the case of total population, the variable showed considerable endogeneity because many indicators are estimated as proportions or percentages of population. Population growth substituted total population. The index was then re- estimated without endogenous variables. Table 2 Endogeneity Tests for Some Index Variables Durbin–Wu–Hausm Endogen Variable Observations Sum of Residual an Endogeneity Test P-value ous (F-test) Proportion of Indigenous Population 2396 1.8E-08 95.49 0.000 Yes Cattle and non-farming activities 2447 -5.0E-10 38.82 0.000 Yes Non-access to Health Services 1046 -2.8E-08 80.24 0.000 Yes Total Population 2449 -2.0E-04 95.49 0.000 Yes Source: Own estimations based on CCVI dataset Sensitivity tests and different PCA specifications 15 were estimated to verify the changes in index distribution and rankings. Abrupt changes in rankings are indicative of an unstable index which may display an inadequate vulnerability risk distribution. Figure 2 shows the minimum and maximum values of the index at the state level using nine different prediction models with climatic scenarios. Consistent changes in risk are predicted across states for minimum and maximum index levels. That is, except for a few states, all models predict changes in the same direction. The index ranges from -0.78 (Very Low Vulnerability) to 1.91 (Very High Vulnerability) with a S.D. of 0.652 and an Average of 0.525. The criterion for building the 5 vulnerability cohorts was based on equal counts. Out of the 2,456 municipalities in Mexico, the PCA model kept 2,257 municipalities with valid data for the main estimation specification 16. 15 The full specification model included the following core variables: drought risk; number of reported environmental risks, yield Loss due to weather; temperature and precipitation; percentage of farmers receiving remittances; percentage of farmers that belong to organizations; percentage of agricultural production units without irrigation systems; percentage of population in agricultural activities; hectares for agricultural, forestry, and cattle activity; poverty rate; Farmers lacking credit; Federal disaster assistance per capita. Upon these variables different specifications were modeled to build the index by adding and replacing variables. The more variables included, the more restrictive the model. 16 The specifications for robustness checks and sensitivity kept 2,240 and 2,100 municipalities, respectively. 11 The difference between the minimum and maximum values of the index in all seven models is on average 0.299. Not a single state showed differences higher than 1. Only two states (Colima and Zacatecas) show changes in the vulnerability index from negative to positive, or vice versa. However, the differences in the index levels between baseline and prediction points are statistically significant for the states of Baja California, Campeche, Chiapas, Nayarit and Sonora 17. It is worth considering such shifts and heterogeneity prevailing at the municipal level to better identify vulnerability risk profiles over different periods of time. The preliminary results and rankings (state level) based on risk vulnerability are shown in the next section. Figure 2. Minimum and Maximum Values of ICCV Min Value Max Value 2.000 1.500 1.000 0.500 0.000 -0.500 -1.000 Tlaxcala Jalisco Baja California Tabasco Sinaloa Nayarit Oaxaca Hidalgo Tamaulipas Chihuahua Guanajuato Baja California Sur México Chiapas Campeche San Luis Potosí Puebla Aguascalientes Querétaro Colima Morelos Durango Zacatecas Quintana Roo Distrito Federal Yucatán Sonora Guerrero Nuevo León Coahuila Veracruz Michoacán Source: Own estimations IV. Index Results and Profiles This section presents estimates of the municipalities that are the most vulnerable to climate change and climatic disasters. This study only estimates a composite index, not its parts. 17 Based on mean differences t-test values for unequal standard deviations at 90% level. 12 Where Are the Most Vulnerable Municipalities? Overall, the study results suggest a wide variation in municipal vulnerability across the country. The most vulnerable municipalities are located along the coastlines and in many Southern areas, in line with findings from similar work in Mexico (IMTA, 2009; Martinez- Austria, 2007). The Northern and Central parts of the country are comparatively less susceptible to climate change and variability, but with some pockets of high vulnerability. Coastal areas host some of the most vulnerable municipalities to climate change in Mexico. This is likely due to the relatively high exposure of these municipalities to hurricanes and the increased risk of flooding that comes in these areas. The drier northern and central regions of Mexico also face high exposure given recurrent droughts and a lack of protective vegetation. The southern states of Mexico appear to be the most vulnerable to climatic events in the entire country. Many municipalities in the southern states of Guerrero, Oaxaca and Chiapas display the highest levels of vulnerability. With large and highly impoverished indigenous populations, it comes as no surprise that their relative capacity to manage climate risk is lower than other areas. By contrast, the tourist areas on the Yucatan Peninsula have a high capacity to adapt to climate change. The tourist industry has led to higher incomes, lower poverty rates, and thus less sensitivity and higher adaptive capacity. Again the north displays higher resilience than elsewhere, and this could be due to its better socio-economic development and higher access to remittances. But there are also pockets of high vulnerability in northern states. States like Chihuahua contain high vulnerability pockets due to prolonged droughts that are increasingly prevalent among the poorest Tarahumara territories. Recent droughts have affected mainly the north and central parts of the country –the states of Durango, Chihuahua, Coahuila, San Luis Potosí and Zacatecas– where the economy relies strongly on agricultural activity 18. The estimation of the CCVI permits mapping using baseline and prediction points. Maps 1 and 2 show the spatial distribution of the CCVI in 2005 and 2045, respectively. Coastal regions show high vulnerability persistence particularly in the pacific south and Yucatan peninsula over the next 20 years. Other high-poverty incidence municipalities in the north- west show increasing vulnerability, in part due to predicted increases in temperatures. 18 The federal government through the CONAGUA (National Commission of Water) is also taking action to provide relief to Mexicans suffering from drought. As of January 2012, CONAGUA reported to have spent nearly 60 million pesos (5.4 USD million) to support the Tarahumara men and women. Part of the government’s relief efforts is to provide temporary employment to the Tarahumara whose farming suffered significantly from the drought. Employment may include cleaning of the existing water bodies, channel and ditches dredging and building of dams. CONAGUA is also inspecting Mexico’s water systems to ensure water provision even during times of drought. It is recognized that much of Mexico’s water systems are inefficient due to leakages, and that infrastructure improvements must be made to prevent droughts from having such serious impacts on Mexico’s people in the future. 13 Map 1. CCVI by Municipality (2005) Very High Vulnerability High Vulnerability Moderate Vulnerability Low Vulnerability Very Low Vulnerability No Data Note: Darker colors imply higher vulnerability Source: Own Estimations Map 2. CCVI by Municipality (2045 Prediction) Very High Vulnerability High Vulnerability Moderate Vulnerability Low Vulnerability Very Low Vulnerability No Data Note: Darker colors imply higher vulnerability Source: Own Estimations 14 Beyond the climate change vulnerability levels in agriculture, it is relevant to identify the areas or regions where vulnerability shows the highest relative changes between 2005 and 2045. Map 3 shows that most of the municipalities with the biggest changes are concentrated in Central Mexico (Bajio). This finding is in line with previous environmental and climate change studies conducted in Mexico (Martinez, 2010; IMTA, 2009; Martinez and Fernandez, 2004; Martinez-Austria, 2007). A number of studies predict a 10 percent reduction in water availability for agriculture between 2030 and 2050 for northwest and central Mexico (Bajio). This will especially impact states such as Sonora, Guanajuato, San Luis, that will experience critical water shortages in the predicted scenarios (Martinez, 2010). In addition, Martinez and Fernandez (2004) report that the regions with highest risk of vulnerability for the next 40 years correspond to the Bajio central region (including states such as Guanajuato and San Luis Potosi). Other states located in the Bajio region (Hidalgo and Queretaro) could experience a large shift in their vulnerability risk in the absence of investments for climate change adaptation. The reasons given to explain this shift into high vulnerability vary from water availability and temperature changes, to soil degradation and poor implementation of adaptation policies. Martinez-Austria (2007) indicates that drought vulnerability risks will be a particular concern for national and regional policies in the northwestern region of the country due to the predicted change between 3 to 4 degrees (Cº) by 2040. The predicted shifts in territorial vulnerability associated with droughts in the Bajio and Northwestern regions are also confirmed in a recent study by the Mexican Institute for Water Technology (IMTA, 2009). According to this study, climate predictions for 2025 suggest risks of water shortages in northern and central regions in Mexico, where irrigated surface land will accelerate water scarcity over the years. Map 3. Vulnerability Index Changes 2005-2045 Very High Change High Change Moderate Change Low Change Very Low Change No Data Note: Darker colors imply higher change in index between baseline and prediction points Source: Own Estimations 15 Who Are the Most Vulnerable? The purpose of this paper is to identify which social groups in rural Mexico are the most vulnerable to climate change. First, we show how vulnerability profiles change across municipalities from baseline to prediction points. The purpose is to assess the probability and number of municipalities falling into different categories of vulnerability at baseline and prediction points 19. Second, the municipal vulnerability profiles relate index estimates at baseline and prediction points to three different sets of variables: 1) climate indicators, 2) farmer categories, and 3) socio economic characteristics. Changes in Vulnerability Risk Profiles The risk profile of municipalities is shown in table 3a. Overall, almost three of every four municipalities (around 1,810 out of 2,454) do not show substantial changes between baseline and prediction points. Additionally, 344 municipalities increase their vulnerability risk, compared to 300 showing reductions in vulnerability reductions. Both sets of winners and losers are profiled below. Although shifts in vulnerability risk may not appear substantial, the fact that over a third (34.6%) of municipalities maintain high vulnerability, particularly in coastal (Pacific and Gulf) areas is relevant. The conditional probability of high vulnerability municipalities at the prediction point, having shown a high vulnerability risk at baseline, is 41%. This percentage is similar for the conditional probability of municipalities being in low vulnerability risk at baseline and prediction points (39%). Some authors stress that economic impacts in agriculture from climate fluctuations are substantial if high risks prevail over time (Deschenes and Greenstone, 2007; Lobell and Asner, 2003). Table 3 Conditional Probabilities of Vulnerability Risks Changes Baseline and Prediction Points High Moderate Low Categories of Vulnerability Risk Vulnerability Vulnerability at Vulnerability at at Baseline Baseline Baseline Probability 0.406 0.099 0.012 High Vulnerability at Prediction Number of 850 141 23 Municipalities Probability 0.056 0.238 0.127 Moderate Vulnerability Number of at Prediction 89 218 180 Municipalities Probability Low Vulnerability at 0.068 0.051 0.391 Prediction Number of 140 71 742 Municipalities Note: Numbers in Italics indicates no Change in Index Category between Baseline and Prediction Source: Own estimations 19Categories are: Very High Vulnerability, High Vulnerability, Moderate Vulnerability, Low Vulnerability, Very Low Vulnerability. 16 States such as Zacatecas, Yucatan, Chiapas, Guanajuato, Chihuahua, Oaxaca and Puebla exhibit the highest increases in vulnerability over time. Other states such as Campeche, Tabasco, Sonora, Sinaloa and Nayarit showed reductions in their vulnerability risk profiles between baseline and prediction points. In general, the index predictions show that high vulnerability will prevail in southern coastal areas (gulf and pacific) with a tendency to increase vulnerability in the central-norhtern basin (Bajio) states. States shown in Table 3a have the highest increases and decreases in vulnerability index changes between baseline and prediction points. However, there are municipalities that rank highest in terms of index increases and decreases that may or may not belong to the states presented in Table 3a. For instance, Oaxaca has 124 municipalities with an increase higher than 0.25 in the index between baseline and prediction points (such increases are higher than the mean increase of 0.069 in the index), but the rest of the 570 municipalities in Oaxaca have relatively lower increases than the average. Table 3a Highest Increases and Decreases in Vulnerability (2005-2045) by State Highest Increase Index Vulnerability Vulnerability State Change BL Prediction at Baseline at Prediction Zacatecas 0.3749 -0.3273 0.0476 Very Low Low Yucatan 0.2667 0.5469 0.8136 Moderate High Guanajuato 0.1897 -0.2409 -0.0513 Very Low Low Chiapas 0.1725 1.3906 1.5631 Very High Very High Chihuahua 0.1544 0.1014 0.2558 Low Moderate Highest Decrease Index Vulnerability Vulnerability State Change BL Prediction at Baseline at Prediction Tabasco -0.4630 1.1752 0.7122 Very High High Sonora -0.4075 -0.0196 -0.4272 Low Very Low Campeche -0.4038 0.7842 0.3804 High Moderate Sinaloa -0.3407 -0.0064 -0.3471 Low Very Low Nayarit -0.3246 0.6354 0.3108 High Moderate Source: Own estimations Tables 4a and 4b present municipal vulnerability profiles in relation to key climate, social, and agricultural indicators. Risk categories of the index are divided in five cohorts (Very Low, Low, Moderate, High and Very High) of vulnerability. With municipalities arranged by categories of vulnerability at baseline, it is possible to construct socio-demographic and 17 agriculture municipal level profiles. Such profiles bring additional information about the patterns of risk in the advent of climate change In terms of climate variables, municipalities with elevated vulnerability levels show higher climate extremes as measured by frost days and consecutive dry days, in both baseline and prediction points. In addition, municipalities under most vulnerable categories show an increase in the Coefficient of Variation of Rain and the Growing Degree Days (GDD) between baseline and prediction points. The shifts in this last indicator are important for assessing the suitability of a region for producing a particular crop, and to better estimate harvest dates. Municipalities with high levels of vulnerability also have the highest ratio of increase of rain’s coefficient of variation. The larger is the rain variability, the higher is the uncertainty for harvest periods for agricultural yields and outputs. In Mexico irrigated agriculture contributes about 50% of the total value of agricultural production and accounts for about 70% of agriculture exports (CONAGUA, 2008). However, the rest of agriculture depends to a larger extent on temporal or seasonal harvesting. The risks confronted by municipalities in terms of rain and temperature changes could shape the changes in cropping patterns (planting multiple crops with different vulnerabilities to weather events), irrigation systems (to decrease the farmers dependence on precipitation), farm incomes, and financial instruments available to famers to strengthen resilience. Table 4c presents the distribution of vulnerability risk by type of agricultural producer. This table shows that larger producers are more resilient and less likely to be present in highly vulnerable municipalities. On the other hand, small and subsistence producers are more likely to live in highly vulnerable municipalities, and municipalities that will experience a high increase in vulnerability during 2005 - 2045. Low-capital intensity producers with large land sizes face the largest shifts in vulnerability between baseline and prediction points. These types of larger land-size producers often have higher rates of participation in subsidized agricultural programs. On the other hand, small land-size producers with intensive or non-intensive capital requirements, located at a higher proportion within highly rural municipalities, are more likely to be in highly vulnerable municipalities. Table 4a also shows consistently that higher vulnerability risk is associated with less favorable socio-economic conditions. Municipalities situated within the “low vulnerability� categories show substantially lower average proportions of a) indigenous populations, b) households with elderly members, and c) households with dirt floors; compared to municipalities situated within “high vulnerability� categories. The dispersion of these socio- economic indicators also increases as vulnerability risks become higher. The profiles are also shown for agricultural and income support variables (Table 4b). In this regard, the percentage of agricultural workers having liquid savings reduces considerably from 12.4 (for municipalities under the “very low vulnerability� category) to 1.8 percent (for municipalities under the “very high vulnerability� category). Moreover, the number of agricultural workers having outstanding credit debt for their economic activity increases as vulnerability risk increases. The average support of agricultural programs devoted to 18 farmers does not vary substantially, but municipalities with lower vulnerability profiles tend to receive marginally higher transfers from these programs. The profiles for these variables indicate how farmers use financial instruments and other financial mechanisms to cope with vulnerability, which brings up front information useful to improving the targeting and redistributing options of current support programs and financial products. Finally, the risk profiles are presented according to pairwise correlations between the index (at baseline and prediction points) and the socio-demographic variables in Table 4c. The results show, first, that municipalities with higher vulnerability risks have higher indigenous populations at baseline and prediction points—as shown by a positive and significant correlation. Although the correlations are not as high, there is a positive association between higher vulnerability and adverse housing conditions. Such correlations become higher as vulnerability becomes higher. The highest correlations within socio- demographic characteristics are found when households have a higher rate of elder dwellers. These living arrangements may enhance household risks to climate change through exposure, sensitivity and adaptive capacity factors. Mexico reflects high levels of family care-giving for the elderly and a high degree of continuity of parent-child co- residence over the life-course (Kanaiaupuni, 2000) fed by economic conditions and demographic patterns. Mexico will face a substantial increase in elderly populations over the next 20 years, so there may be higher vulnerability risks under these care-giving arrangements 20. And limited access to support programs or savings (for smallholder populations), is associated with higher the levels of vulnerability. Remittances show a negative correlation with the vulnerability index at both baseline and prediction points. High vulnerability is associated with lower levels of remittances influx by municipality. Access to different forms of capital “insures� families from several forms of uncertainty. The complex migration patterns found in municipalities across Mexico are usually undertaken to insure families via remittances, which is often a result of stress-induced movements (conflict) or through resource constraints (climate change) (Schreider and Knerr, 2000; Fiki and Lee, 2004). With the advent of climate variability and uncertainty, many small landholders will face risks of being forced to abandon agriculture, due to financial losses and the burden of debt. Improved financial instruments used to ease debt arising from agricultural credits, and financial support to improve farming activities, could in turn improve the adaptive capacity of exposed and climate sensitive farmers. 20Another interpretation is that a large elderly population contributes to low adaptive capacity/high sensitivity because they are not economically active, and thus more likely to be in poverty. 19 Table 4a Profiles of Vulnerability Risk (Baseline Socio-economic/Climate Variables) Index Percent of Percent of Percent of Elderly Rain Coefficient Risk Category **/ Statistics CCVI Index Prediction Indigenous Households GDD (65+) Population of Variation (2045) Population with Dirt Floors All Mean 0.349 0.557 24.463 10.263 7.896 0.281 9.385 Range 2.483 2.708 100.000 95.314 30.041 0.797 451.937 Standard Deviation 0.582 0.652 35.627 18.640 3.834 0.072 26.405 Percentile 5 -0.538 -0.448 0.204 0.000 3.632 0.190 1.884 Percentile 95 1.361 1.687 98.815 56.978 15.624 0.408 13.997 N 2433 2356 2455 2454 2454 2451 2451 Very Low Vulnerability Mean -0.315 -0.282 4.103 0.879 7.378 0.266 6.456 Range 0.717 1.406 90.772 41.043 28.882 0.461 433.954 Standard Deviation 0.168 0.254 11.528 3.048 3.338 0.067 20.380 Percentile 5 -0.610 -0.746 0.151 0.000 3.679 0.234 1.722 Percentile 95 -0.093 0.105 19.883 5.420 13.449 0.442 11.039 N 450 451 451 451 451 451 451 Low Vulnerability Mean 0.108 0.171 11.299 2.841 9.399 0.275 9.580 Range 0.348 1.225 99.448 56.501 29.251 0.480 451.724 Standard Deviation 0.104 0.203 21.790 7.465 4.852 0.068 34.682 Percentile 5 -0.055 -0.183 0.163 0.000 3.807 0.217 1.139 Percentile 95 0.260 0.501 70.433 17.177 19.069 0.422 13.758 N 474 474 474 474 474 474 474 Moderate Vulnerability Mean 0.420 0.492 25.169 8.603 8.801 0.293 11.08 Range 0.350 1.146 100.000 95.314 28.472 0.363 443.75 Standard Deviation 0.101 0.217 34.580 16.530 3.883 0.048 36.18 Percentile 5 0.296 0.094 0.217 0.000 4.094 0.201 1.00 Percentile 95 0.610 0.819 97.478 52.536 16.702 0.350 14.19 N 430 429 430 430 430 430 430 High Vulnerability Mean 0.745 0.887 35.049 14.384 7.706 0.305 10.20 Range 1.098 0.541 100.000 84.475 24.051 0.266 435.50 Standard Deviation 0.211 0.157 39.813 20.431 3.190 0.044 20.49 Percentile 5 0.384 0.648 0.293 0.025 4.142 0.194 2.51 Percentile 95 1.085 1.130 99.500 61.246 14.038 0.337 14.40 N 448 450 450 450 450 450 450 Very High Vulnerability Mean 0.788 1.521 40.396 20.449 6.690 0.369 19.59 Range 2.483 0.746 100.000 91.217 26.562 0.785 443.82 Standard Deviation 0.745 0.193 41.562 24.060 3.137 0.092 17.47 Percentile 5 -0.707 1.204 0.284 0.000 3.014 0.174 3.48 Percentile 95 1.455 1.809 99.666 67.686 12.500 0.446 14.03 N 631 552 650 649 649 646 646 ** Vulnerability cutoff points based on baseline index Source: Own estimations 20 Table 4b Profiles of Vulnerability Risk (Baseline Agricultural and income support variables) % of Average % of Agriculture % of Agriculture Agriculture Agriculture workers Risk Category **/ Statistics workers with workers with support in Pesos receiving credit Savings 2009 * remmitances All Mean 5.08 34.12 381.64 3.09 Range 40.79 100.00 999.80 37.58 Standard Deviation 2.78 28.39 311.50 4.49 Percentile 5 0 0 0.5219 0 Percentile 95 6.95 87.79 932.24 12.18 N 2447 2447 1212 2447 Very Low Vulnerability Mean 12.37 30.41 454.43 3.89 Range 40.79 98.82 999.80 25.18 Standard Deviation 3.02 22.61 303.86 4.53 Percentile 5 0.00 0.00 2.78 0.00 Percentile 95 6.73 79.57 951.92 13.04 N 451 451 229 451 Low Vulnerability Mean 8.40 31.54 440.53 3.43 Range 26.14 100.00 999.20 24.27 Standard Deviation 2.98 28.59 296.09 4.36 Percentile 5 0.00 0.00 15.17 0.00 Percentile 95 8.43 85.10 918.82 11.97 N 474 474 216 474 Moderate Vulnerability Mean 6.20 37.45 377.80 3.85 Range 33.01 100.00 997.13 37.13 Standard Deviation 3.28 28.87 287.97 5.38 Percentile 5 0.00 0.00 16.01 0.00 Percentile 95 7.62 88.15 941.41 14.55 N 430 430 184 430 High Vulnerability Mean 2.72 41.86 341.82 3.03 Range 12.48 98.99 991.09 37.58 Standard Deviation 1.96 29.56 318.44 4.55 Percentile 5 0.00 0.00 4.07 0.00 Percentile 95 5.84 89.09 922.48 12.36 N 450 450 181 450 Very High Vulnerability Mean 1.81 45.75 340.55 1.80 Range 22.39 100.00 997.49 34.43 Standard Deviation 2.50 30.40 322.52 3.45 Percentile 5 0.00 0.00 0.00 0.00 Percentile 95 6.35 89.14 927.52 7.26 N 642 642 402 642 * Averages computed for those municipalities that received support; some municipalities don't receive support in that year, but they are still beneficiaries. ** Vulnerability cutoff points based on baseline index Source: Own estimations 21 Conclusion Mexico is in constant threat of experiencing natural disasters, and is among the most exposed to climatic hazards in the world. Recent evidence and predictions indicate that climate changes are accelerating and will lead to wide-ranging shifts in climate variables. Agriculture is one of the sectors where climate change is expected to hit hardest. Little quantitative evidence has been produced to aggregate multidimensional aspects of livelihoods, socio-demographic and economic characteristics, and climate change historic and predicted scenarios. With rich data available for most (2,200 of 2,450) municipalities in Mexico, a statistical technique (Principal Components Analysis) was applied to estimate a Vulnerability Risk Index in Agriculture for baseline (2005) and prediction points (2045). The aim of this analytical tool is to better understand how and why vulnerability to climate change and climate variability varies by municipality in Mexico. The index can be used to better target federal and state level adaptation programs to local conditions, and to inform the design of municipality adaptation strategies. The conceptual framework used for the vulnerability analysis and the index construction is based on an adaptation of the IPCC’s vulnerability framework, which distinguishes between exposure, sensitivity, and adaptive capacity. The results of the analysis suggest a wide variation in municipal vulnerability across the country at baseline and prediction points. Currently, Coastal areas host some of the municipalities most vulnerable to climate change in Mexico. This is likely due to the relatively high exposure of these municipalities to hurricanes and the ensuing flood risk. However, Northwest and Central regions will likely experience the largest shifts in vulnerability between 2005 and 2045, in the advent of temperature increases and water scarcity for agricultural activities. Recent environmental and climate change studies conducted in Mexico [Martinez, 2010; IMTA, 2009; Martinez and Fernandez, 2004; Martinez-Austria, 2007] support these claims and trends. The analysis presented here provides municipal estimates of agriculture vulnerability associated with temperature and rainfall changes, but it is also necessary to assess the distributional impact of climate change across urban and rural areas and population groups. The profiles of municipalities show that the shifts in vulnerability across municipalities, between 2005 and 2045, are quite heterogeneous because of differences in socio-economic, climate and agricultural variables. Highly vulnerable municipalities demonstrate higher climate extremes, which increase the uncertainty for harvest periods, and for agricultural yields and outputs. Also, municipalities with higher vulnerability have more adverse socio- demographic conditions. The profile also shows a positive correspondence between the percentage of people lacking support programs or savings and vulnerability. Finally, smallholders display higher vulnerability to climate change at baseline (2005) and prediction (2045) points. 22 References Achuta Rao, K. and Sperber, K. R. 2002. Simulation of the El Nino Southern Oscillation: results from the coupled model intercomparison project. Climate Dyn. 19, 191–209 AGROASEMEX, S.A. “Sistema de Información sobre Riesgos Agrícolas. Guía de Interpretación�. Septiembre, 2009. http://www.agroasemex.gob.mx/index.php/es/actualizate2/mapariesgos Antony, G. M. & Rao, K. V. (2007). A composite index to explain variations in poverty, health, nutritional status and standard of living: Use of multivariate statistical methods. Public Health, 121, 578-587. Biasutti, M. et al. 2011. Projected Changes in the Physical Climate of the Mexican Gulf Coast and Caribbean. To be Published in Climate Change. Vol 1 2012. Draft available at : http://www.ldeo.columbia.edu/~biasutti/papers/ClimaticChange.pdf Boelhouwer, J. & Stoop, I. (1999). Measuring well-being in the Netherlands: The SCP index from 1974 to 1997. Social Indicators Research, 48(1), 51-75. CONAGUA. 2008. "National Water Program". SEMARNAT. Retrieved March, Mexico. CONEVAL. (2010). Social Gap Index (Indice de Rezago Social). Consejo Nacional de Evaluacion. Mexico. Connolley, W. and Bracegirdle, T., 2007. An Antarctic assessment of IPCC AR4 coupled models. GEOPHYSICAL RESEARCH LETTERS, VOL. 34, L22505 Davies L., Gather U., �The identi�cation of multiple outliers,� Journal of the American Statistical Association, 88(423), 782-792, 1993. Davey, M. K., Huddleston, M., Sperber, K. R., Braconnot, P., Bryan, F. and co-authors. 2002. STOIC: a study of coupled model climatology and variability in tropical ocean regions. Climate Dyn. 18, 403– 420 Deschenes, O., and M. Greenstone. 2007. “The Economic Impacts of Climate Change: Evidence from Agricultural Output and Random Fluctuations inWeather.� American Economic Review 97(1):354– 85. ECLAC, 2008. Perdidas en los Sectores Economicos de Mexico. UN Economic Commission for Latin America and the Caribbean. Santiago, Chile. http://www.eclac.cl/publicaciones/xml/3/33373/L864_parte_6_de_8.pdf Fay, Marianne and Hrishi Patel. 2008. ―A simple index of vulnerability to climate change. Background paper prepared for World Bank report. Washington, DC. Fiki, O.C. and Lee, B. 2004. Conflict Generation, Conflict Management, and Self-organizing capabilities under drought-prone rural communities in NE Nigeria. Journal of Social Development in Africa. 19 25-48. Fotso, J. & Kuate-defo, B. (2005). Measuring socioeconomic status in health research in developing countries: Should we be focusing on households, communities, or both? Social Indicators Research, 72, 189-237. Fukuda, Y., Nakamura, K., & Takano, T (2007). Higher mortality in areas of lower socioeconomic position measured by a single index of deprivation in Japan. Public Health, 121, 163-173. 23 Fussel, HM. (2009) Development and Climate Change. Background note to the World Development Report 2010 Gjolberg, M. (2009). Measuring the immeasurable? Constructing an index of CSR practices and CSR performance in 20 countries. Scandinavian Journal of Management, 25, 10-22. Havard, S., Deguen, S., Bodin, J., Louis, K., & Laurent, O. (2008). A small-area index of socioeconomic deprivation to capture health inequalities in France. Social Science & Medicine, 67, 2007-2016. Heltberg, R. and Bonch-Osmolovskiy, M. (2010) A climate vulnerability index for Tajikistan, World Bank. Hotelling, H. (1933) Analysis of a complex of statistical variables into principal components. Journal of Educational Psychology, 24, 417-441. IMTA, 2009. Perspectivas de la Gestión del Agua en México al año 2025. IMTA-Estudios Prospectivos. http://semarnat.janium.net/janium/Documentos/48133.pdf INE. 2007. Análisis de la precipitación histórica de la zona Norte de México. In Esquivel, E. Gaceta Ecológica No. 65 Mexico. IPCC (2001) Climate Change 2001: The Scienti�c Basis. Contribution of Working Group I to the Third Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press IPCC (2007) Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the IPCC. Cambridge University Press Kanaiaupuni, S.M. 2000. Leaving Parents Behind: Migration and Elderly Living Arrangements in Mexico. Center for Demography and Ecology (# 99-16). University of Wisconsin. Lai, D. (2003). Principal component analysis on human development indicators of China. Social Indicators Research, 61(3), 319-330. Latif, M., Sperber, K., Arblaster, J., Braconnot, P., Chen, D. and coauthors. 2001. ENSIP: the El Ni˜no simulation intercomparison project. Climate Dyn. 18, 255–276. Lobell, D.B. and G.P. Asner. 2003. “Climate and Management Contributions to Recent Trends in U.S. Agricultural Yields.� Science 299: 1032. Manly BFJ. Multivariate statistical methods. A primer. 2nd Edition. London: Chapman and Hall; 1994. Markides, K. S. & McFarland, C. (1982). A note on recent trends in the infant mortality-socioeconomic status relationship. Social Forces, 61(1), 268-276. Martínez-Austria, P. F., Efectos del cambio climático en los recursos hídricos de México. Jiutepec, Mor IMTA 2007 http://www.imta.mx/gaceta/anteriores/g07-11-2007/gaceta-imta-07.pdf Martinez, J. 2010. “ABC de Cambio Climático: Impactos y Acciones en México� Vulnerabilidad y Adaptación Disponibilidad de agua: Proyecciones al 2030. INE-SEMARNAT, Mexico. Martinez, J. and Fernandez, A. (2004). Cambio Climatico: Una Vision desde Mexico. INE-SEMARNAT, Mexico. McFarlane, N.A., J. F. Scinocca, M. Lazare, R. Harvey, D. Verseghy, and J. Li, 2005: The CCCma third generation atmospheric general circulation model. CCCma Internal Rep., 25 pp. 24 McCarl, B.A., X. Villavicencio, and X. Wu. 2008. “Climate Change and Future Analysis: Is Stationary Dying?� American Journal of Agricultural Economics 90(5): 1241-1247. McKenzie DJ. 2003. Measure inequality with asset indicators. BREAD Working Paper No. 042; Cambridge, MA: Bureau for Research and Economic Analysis of Development, Center for International Development, Harvard University. Messer, L. C., Vinikoor, L. C., Laraia, B. A., Kaufman, J. S., Eyster, J., Holzman, C., Culhane, J., Elo, I., Burke, J. G., & O‟Campo, P. (2008). Socioeconomic domains and associations with preterm birth. Social Science & Medicine, 67, 1247-1257. Milliken KT, Anderson JB, Rodriguez AB (2008) A new composite Holocene sea-level curve for the northern Gulf of Mexico. In: Response of Upper Gulf Coast estuaries to Holocene climate change and sea-level rise, Special Paper 443, Geological Society of America, Boulder, CO, pp 1–11, DOI 10.1130/2008.2443(01) Morris, R. & Castairs, V. (1991). Which deprivation? A comparison of selected deprivation indices. Journal of Public Health Medicine, 13, 318-326. Nakicenvoic et al., 2000. Special Report on Emissions Scenarios. A Special Report of Working Group III of the Intergovernmental Panel on Climate Change. Cambridge University Press: Cambridge. 599 pp. Onwujekwe O, Hanson K, Fox-Rushby J. Some indicators of socio-economic status may not be reliable and use of indices with these data could worsen equity. Health Economics 2006;15:639-44. Pearson, K. (1901). On Lines and Planes of Closest Fit to Systems of Points in Space Philosophical Magazine 2 (6),559–572. Peralta-Hernandez, A.R. et al. (2009) Comparative Analysis of indices of extreme rainfall events: variations and trends from southern Mexico. Atmosfera 22 (2) pp. 219-228. Polsky, C. and Neff, R. and Yarnal, B. (2007) Building comparable global change vulnerability assessments: The vulnerability scoping diagram. Global Environmental Change 17 (2007):472-485 Ribot, J. 2010 Vulnerability Does Not Fall from the Sky: Toward Multiscale, Pro-Poor Climate Policy. In: Means, R., Norton, A. (Eds.), Social Dimmensions of Climate Change: Equity and Vulnerability in a Warming World. The World Bank, Washington, DC. Ruiz-Corral J.A., Medina-García G., González-Acuña I.J., Ortiz-Trejo C., Flores-López H.E., Martínez- Parra R.A., y Byerly-Murphy K.F. 1999. Requerimientos Agroecológicos de cultivos. SAGAR. INIFAP. Rygel, L., O‟Sullivan, D. & Yarnal, B. (2006). A method for constructing a social vulnerability index: An application to hurricane storm surges in a developed country. Migration and Adaptation Strategies for Global Change, 11, 741-764. Saltelli, A., Nardo, M., Saisana, M., & Tarantola, S. (2004). Composite indicators-The controversy and the way forward, OECD World Forum on Key Indicators, Palermo, 10-13 November. Saurral, R. and Barros, V. 2009. THE HYDROLOGICAL CYCLE IN SOUTHERN SOUTH AMERICA IN THREE GENERAL CIRCULATION MODELS: CNRM-CM3, ECHAM5/MPI-OM AND GFDL-CM2.0. CIMA and DCAO, Int. Guiraldes 2160, Ciudad Universitaria, Pab. II, C1428EGA, Buenos Aires 25 Servicio de Información Agroalimentaria y Pesquera (SIAP). Anuarios Estadísticos a Nivel Municipal 2001-2009. Servicio Meteorológico Nacional (SMN). Base de Datos del CLICOM. Julio, 2010. Shaw, L.H. 1964. “The Effect of Weather on Agricultural Output: A Look at Methodology.� Journal of Farm Economics 46 (1): 218-230. U.S. Census Bureau. 2009. Population Estimates, http://www.census.gov/popest/archives/, release date: March 19, 2009. Schlenker, W. and M.J. Roberts. 2009. “Nonlinear Temperature Effects Indicate Severe Damages to U.S. Crop Yields under Climate Change.� PNAS 106(37): 15594-15598. Schreider, G. and Knerr, B. 2000. Labor Migration as a Social Security Mechanism for Smallholder households. Oxford Development Studies. 28. 223-236 Sekhar, C. C., Indrayan, A., & Gupta, S. M. (1991). International Journal of Epidemiology, 20(1), 246- 250. Shevky, E. & Bell, W. (1955). Social Area Analysis. Stanford: Stanford University Press. Tata, R. J. & Schultz, R. R. (1988). World variation in human welfare: A new index of development status. Annals of the Association of American Geographers, 78(4), 580-593. USDA/NASS (2009) U.S. & All States County Data - Crops, http://www.nass.usda.gov/QuickStats/Create_County_All.jsp. Vörösmarty, C.J., P. Green, J. Salisbury and R.B. Lammers. 2000. ―Global Water Resources: Vulnerability from Climate Change and Population Growth. Science 289: 284–288. Vyas, S. & Kumaranayake, L. (2006). Constructing socioeconomic status indices: How to use principal components analysis. Advance Access Publication, 9, 459-468. Wallace, H.A. 1920. “Mathematical Inquiry into the Effect of Weather on Corn Yield in the Eight Corn Belt States.� Monthly Weather Review 48 (8): 439-446. Westphal, Michael I, 2008. ―Summary of the Climate Science in the Europe and Central Asia Region: Historical Trends and Future Projections. Background paper prepared for World Bank report. Washington, DC 26 Figure I Core Variables Used in MAIN Model (Sources and Definitions) for the Climate Change Vulnerability Index in Agriculture Core Variables UNIT DEFINITION SOURCE EXPOSURE Total Agricultural Surface ha Total agricultural area within municipality (all crops included, Agroalimentary and Fisheries Information Service (Servicio de Información y Estadística Area subsistence and non-subsistence agriculture. Agroalimentaria y Pesquera—SIAP) of the Ministry of Agriculture (SAGARPA). (www.siap.gob.mx) Average temperatura (past °C Average temperature between May-August for the period 1950-2000. Digital Climatic Atlas for Mexico produced by the Informatics Unit for Atmospheric and 1960-2005; and predicted Environmental Sciences (UNIATMOS in Spanish), at the Center for Atmospheric Science at UNAM. 2005-2065 (www.uniatmos.atmosfera.unam.mx) Average precipitation (past mm Average precipitation between May-August for the period 1950-2000. Digital Climatic Atlas for Mexico produced by the Informatics Unit for Atmospheric and 1960-2005; and predicted Environmental Sciences (UNIATMOS in Spanish), at the Center for Atmospheric Science at UNAM. 2005-2065) (www.uniatmos.atmosfera.unam.mx) Past and Future temperature °C Includes variables with average, min and max temperature for the Digital Climatic Atlas for Mexico produced by the Informatics Unit for Atmospheric and variability indicators period (1961-2005) and variables with predictions averages using Environmental Sciences (UNIATMOS in Spanish), at the Center for Atmospheric Science at UNAM. temperature models (Echam, Hadgem (2030) and 9 models) [ projected (www.uniatmos.atmosfera.unam.mx). Measurements from National Water Commission(CNA) and temperature (°C) between May-August under scenario A2 for 2030 and the Institute for Water Technology (1961-2005) (IMTA). Estimations based on World Bank’s 2045-2065 for 9 models, respectively]. Includes indicators that Environment Unit Predictions. measure variability for historic and predicted periods (GDD, Frost days, Consecutive Drought days) Past and Future mm Includes variables with average, min and max precipitation for the Interpolated through models Hadgem1 y MPIEcham5, A2 for 2030. Digital Climatic Atlas for Mexico precipitation variability period (1961-2005) and variables with predictions averages using produced by the Informatics Unit for Atmospheric and Environmental Sciences (UNIATMOS in indicators precipitation models (Echam, Hadgem (2030) and 9 models) [ projected Spanish), at the Center for Atmospheric Science at UNAM. (www.uniatmos.atmosfera.unam.mx). precipitation (mm) between May-August under scenario for 2030 and Measurements from National Water Commission(CNA) and the Institute for Water Technology 2045-2065 for 9 models, respectively]. Includes indicators that (1961-2005) (IMTA). 9 models Estimations based on World Bank’s Environment Unit Predictions. measure variability for historic and predicted periods (Variation Coefficient Rain, Number of days with precipitation>10mm, percentage of days with rain above 95 percentile of rain) SENSITIVITY Food poverty % Households in municipality where its member’s income falls below the National Council for Evaluation of Social Development Policy in Mexico (CONEVAL, 2008) lowest income necessary to afford a minimum basket of food. Percentage of Maize % Production units that are under irrigated systems and do not depend on Agroalimentary and Fisheries Information Service (Servicio de Información y Estadística production under irrigated seasonal precipitation for crop production Agroalimentaria y Pesquera—SIAP) of the Ministry of Agriculture (SAGARPA). (www.siap.gob.mx) areas % of Population in % National Agricultural and Farmin Census (2007) INEGI. Agricultural Activities ADAPTIVE CAPACITY Farmers that belong to % Production units that belong to any producers association, especially to National Agricultural Census, 2007 INEGI. organizations access credit. Farmers receiving % Production units that self-reported receiving remittances. Censo Agrícola, Ganadero y Forestal 2007 de INEGI. remittances Distance from Municipality km Distance from Municipal Government location (or in its case Agroalimentary and Fisheries Information Service (Servicio de Información y Estadística Center to Road geographical centroid for highly rural municipalities) to the main road Agroalimentaria y Pesquera—SIAP) of the Ministry of Agriculture (SAGARPA). (www.siap.gob.mx) (dirt or paved) Federal disaster assistance $ Sum of monetary transfers per cápita (population in the primary sector Temporary Employmet Program (Programa de Empleo Temporal – PET) per capita in municipality) from various federal programs (PROCAMPO, PET y Weather-Indexed Insurance (Programa de Atención a Contingencias Climatológicas –PACC). PACC) between 2002 y 2009. 27 Figure II Variables Used for Robustness Checks(Sources and Definitions) for the Climate Change Vulnerability Index in Agriculture Complementary Variables UNIT DEFINITION SOURCE EXPOSURE Total number of reported # Sum of environmental problems (illegal logging, fires, pests, loss of Encuesta Nacional de Gobiernos Municipales (SEDESOL, 2004-2005) environmental risks biodiversity, water pollution), self-reported by municipality. Reforestation Rate % Contains the rate of all reforested area from fires, drought and Agricultural Census (INEGI) data (2007) and National Institute of Ecology (2008) decertification per municipality SENSITIVITY Migration Rate % Average net migration flow per municipality Census (1960-2005) INEGI Average corn yield tons/ Average rain fed maize yields during the spring-summer cycle, 2005. Agroalimentary and Fisheries Information Service (Servicio de Información y Estadística ha Agroalimentaria y Pesquera—SIAP) of the Ministry of Agriculture (SAGARPA). (www.siap.gob.mx) ADAPTIVE CAPACITY Total population growth by % Municipal population growth rate between 1960 and 2005. Count of Population and Housing 2005 (INEGI) and National Population Council (CONAPO) municipality Population Density Inhab Degree of aglommeration/urbanization of municipality Census data 2005 (INEGI) /km2 Farmers reporting climate- % Refers to the proportion of agricultural production units that declare Censo Agrícola, Ganadero y Forestal 2007 de INEGI. related losses losses due to weather contingencies within each municipality Population lacking access to % Percentage of population per municipality that has access to health care Census data 2005 (INEGI) health care services (public or private). Population 65 and older % The percentage of households with at least one elderly dweller. Average Census data 2005 (INEGI) living within household aggregated by municipality Indigenous Population per % Percentage of indigenous population relative to all population within Census data 2005 (INEGI) Municipality Municipality. Definition of Indigenous is self-reported and corroborated with language spoken at home 28 Statistical Annex: Table 1 CCVI Dataset Summary Statistics Topic Indicator Mean Std. dev. Min Max Climate-Related Variables Mean accumulated rain (Tons) 1,042.51 633.79 61.57 4,552.01 Average Yearly Rain (mm) 2.88 2.09 0 37.13 Average Maximum Rain Yearly (mm) 27.66 3.98 4 43.50 Maximum Consecutive days of Dryness (agricultural year) 71.12 48.50 0 365.00 Heat Wave Duration Index (DAYS) 13.84 27.38 0 365.00 Total number of Frost days in Agr. Year 0.73 5.87 0 180.00 Average number of days per year with rain greater than 10 mm (international standard) 33.16 29.98 0 365.00 Average aquifer extension 3,084.90 3,288.55 60.41 12,616.61 Proportion of Overexploited aquifers 0.20 0.40 0 1.00 Average extraction aquifer (lt/seg) 63.52 149.87 0 930.92 Sociodemographic Percent non-literate population (15 and older) 16.31 11.03 0.81 70.96 Percent population without sewage 9.95 12.29 0 82.87 Percent population without electricity 5.28 7.87 0 70.30 Percent households with overcrowding 50.26 13.97 10.67 90.67 Percent Population that live with dirt floors 24.24 22.25 0.12 95.60 Marginalization Index (CONEVAL) -0.05 1.02 -2.37 3.36 Infant mortality rate (per 1000) 22.65 8.22 3.02 78.83 Poverty Rate 32.06 19.07 0.11 84.01 Other Variables Proportion of Rural Households 0.49 0.50 0 1 Average size of households 4.08 2.10 1 25 Average Number of children per household 1.77 3.51 0 10 Average Number of elderly per household 0.38 0.66 0 5 Average years of education (all HH) 7.95 4.56 1 21 Proportion of agricultural dependent HH 0.20 0.40 0 1 Agricultural Average agr. prod. Units per mun. 1,714 2,058 3 17,949 Average surface with agricultural prod. Per mun (ha) 13,407 23,085 0 258,679 Average surface non-arable Per mun (ha) 33,575 119,026 0 2,132,465 Average surface pastures Per mun (ha) 13,985 65,293 0 1,188,921 Average surface forest Per mun (ha) 1,690 6,522 0 106,353 Average surface non-vegetation Per mun (ha) 1,081 8,383 0 197,879 Average number of males economically dependent from agriculture per mun 1,963 2,704 0 33,579 Average number of women economically dependent from agriculture per mun 2,900 3,816 0 43,335 Average prod units per mun with piped water 1,283 1,538 0 14,729 Average prod units per mun with sewage 498 715 0 6,829 Average prod units per mun with energy 1,544 1,874 3 17,321 Average prod units per mun with gas 998 1,318 0 16,075 Average prod units per mun with irrigation system 333 630 3 9,770 Source: CCVI dataset 29 Variable Mean St.dev. Min Max Drought Ri sk 0.190 0.297 0 1 Total number of reported 2.332 1.529 0 5 envi ronmental ri sks Yi el d Loss due to weather (kg/ha) Percentage of popul ati on 3.089 4.485 0 37.57 recei vi ng remi ttances Percentage of Farmers that bel ong 97.273 4.694 41.4384 100 to organi zati ons Percentage of Agri cul tural Producti on Uni ts wi thout 88.735 15.254 14.557 100 i rri gati on systems Percentage of popul ati on i n 12.773 7.839 0 55.32 agri cul tural acti vi ti es Percentage of Popul ati on wi th 34.121 28.388 0 100 access to credi t for agr. Acti vi ti es Average temp (Hadgem, 2030) 23.662 4.481 12.025 32.325 Average preci p. (Hadgem, 2030) 150.380 80.648 2 609 Average temp Echam 2030) 2.232 0.522 -0.5 3.2 Average preci p. Echam 2030) -32.230 14.404 -120 30 Proporti on of pop that mi grated 0.134 0.032 0.024911 0.519 between 2000 and 2005 Average surface of reforested 108.211 394.238 0 8646 area Average popul ati on growth -0.041 0.433 -0.6928298 4.403 Average pop densi ty (hab./sq km) 258.302 1122.609 0.1248199 17893.44 Average Total Indi genous 4015.608 10154.610 0 200002 popul ati on Mai ze Ri sk Hi gh 29.720 2.392 0 1 Low 36.380 1.542 0 1 Medi um 33.900 3.544 0 1 Hurri cane Ri sk Hi gh 25.270 5.347 0 1 Low 52.740 1.375 0 1 Medi um 21.990 7.137 0 1 Fl oodi ng Ri sk Hi gh 30.100 10.463 0 1 Low 43.37 4.462 0 1 Medi um 26.54 6.111 0 1 Type agri cul ture (%) Very Intensi ve Agri cul ture (Hi gh 0.16 Producti on) Intensi ve Agri cul ture (Hi gh 5.01 Producti on) Medi um Intensi ty Agri cul ture 1.1 (Hi gh Producti on) Low i ntensi ty Agri cul ture (Hi gh 4.89 Producti on) Transi ti onal extensi ve Agri cul ture 28.59 Subsi stence agri cul ture i ntensi ve 29.2 Subsi stance agri cul ture non- 23.67 i ntensi ve Other type 7.38 Source: CCVI dataset 30 Steps for Index Construction The algorithm used to construct indices of vulnerability in this paper follows similar applications as in Cutter, Boruff, and Shirley (2003), and Schmidtlein et al (2007). First it relies on the inclusion of data standardization for the input variables and the final index scores. The computations were carried out using the following steps: 1. Standardize all input variables to mean 0 and standard deviation 1 2. Perform the PCA with the standardized input variables with the following main/core variables (all variables aggregated at the municipal level): Total Agricultural Surface Area (ha), Average temperatura (past 1960-2005; and predicted 2005-2045, Average precipitation (past 1960-2005; and predicted 2005-2045) , Past and Future temperature variability scenarios (9 climate models, see Annex Background Paper), Past and Future precipitation variability indicators (9 climate models, see Annex Background Paper), Food poverty, Percentage of Maize production under irrigated areas, % of Population in Agricultural Activities, % Farmers that belong to organizations, % Farmers receiving remittances, Distance from Municipality Center to Road (km), Federal disaster assistance per capita ($). 3. Rotate (varimax) the initial solution and build weights matrix. [Weights are kept at baseline to allow structural relationship for predictions]. 4. Order and select in matrix main components resulting from how they may influence vulnerability in three dimensions and assign eigen values to the components accordingly. [An output of the loadings of each variable on each factor was used to determine if high levels of a given factor tend to increase or decrease vulnerability. 5. Because PCA is sensitive to the values of the input variables, the data standardization step is necessary so that all variables have the same magnitude. With the standardized data set the PCA can be performed in the second step. It returns a set of orthogonal components which are all linear combinations of the original variables. By construction the first component is the linear combination that explains the greatest variation among the original variables, the second component the greatest remaining variation, and so on. 6. Based on the results of the performed PCA, select a parsimonious subset of components that explain the underlying data set as closely as possible. [the index was not bounded with upper and lower limits to allow full vulnerability assessment] 7. Perform sensitivity using Varimax rotation and the interpreted components were summed with equal weights to verify that index does not fluctuate substantially. 8. Perform same steps for predictions using Climate Change unit prediction data (with structural weights from baseline) 9. Sensitivity of this approach to creating vulnerability indices was carried out in two main phases. a. Change variables included in PCA with other proxy variables that can provide similar results in terms of levels and distribution of index. b. The correlation between the county level indices was calculated to determine how closely the index constructed with the subset of variables matched the index with the full set of social variables. 31 Scenarios and Climate Models Used GCMs (Global Climate Models) are widely applied for weather forecasting, understanding the climate, and projecting climate change. Models are designed for decade to century time scale climate predictions, containing a number of prognostic equations that are stepped forward in time (typically winds, temperature, moisture, and surface pressure) together with a number of diagnostic equations that are evaluated from the simultaneous values of the variables. Predictions are also based on resolutions from globe sections. In the case of Mexico, where INEGI builds higher resolution grids, compared to other countries, HadGEM1 and ECHAM models use an grids with higher resolution in the tropics to help resolve processes transformation between spectral and grid-point space (higher local accuracy). The most widely accepted models in Mexico for climatic prediction are ECHAM and HADGEM (2030) (UNAM, 2010), which were used to estimate the CCVI, and subsequently compare results to the 9 climatic model predictions for robustness and calibration purposes. The Index reported in this document contains the 9 prediction models (2045- 2065) because calibration and robustness checks showed only slight differences in the distribution of the index across municipalities. Yet, the 9 prediction models offered more detailed climatic prediction scenarios. For that reason, we report only the index built under the 9 prediction models. For the emissions scenarios change in 2045 used the A2 scenario because is at the higher end of the SRES, and it better captures changes in adaptation and climate change. The tradeoff of using other type of scenario lies on the ability to capture a smaller climate change shifts of the lower end scenarios which is computationally intensive and provides little value added to the Index. A low emissions scenario potentially gives less information from an impacts and adaptation point of view. In addition, the current actual trajectory of emissions (1990 to present) corresponds to a relatively high emissions scenario 21. 21 This scenario considers the following emission levels that alter climate (temperature and precipitation). Cumulative CO2 emissions by the middle and end of the 21st century are projected to be about 600 and 1850 GtC respectively, and expected CO2 concentrations (in parts per million, ppm) for the middle and end of the 21st century in this scenario are about 575 and 870 ppm, respectively. The current concentration of CO2 is about 380 ppm. Methane and nitrous oxide increases grow rapidly in the 21st century. Sulfur dioxide increases to a maximum value just before 2050 (105 MtS/yr) and then decreases in the second half of the century (60 MtS/yr by 2100). 32 Source: http://www.narccap.ucar.edu/about/emissions.html For the climatic predictions, there were several models used 22: Nine Models used for Index construction CGCM3.1 (2045-2065): CGCM3.1 is run at two different resolutions, with two levels of accuracy of predictions. The T47 version (used in our estimates) has a surface grid whose spatial resolution is roughly 3.75 degrees lat/lon and 31 levels in the vertical. This has a good fit into Mexico’s littoral areas, but limited accuracy in central regions. The ocean grid shares the same land mask as the atmosphere, but has four ocean grid cells underlying every atmospheric grid cell. The ocean resolution in this case is roughly 1.85 degrees, with 29 levels in the vertical. The T63 version has a surface grid whose spatial resolution is roughly 2.8 degrees lat/lon and 31 levels in the vertical. As before the ocean grid shares the same land mask as the atmosphere, but in this case there are 6 ocean grids underlying every atmospheric grid cell. The ocean resolution is therefore approximately 1.4 degrees in longitude and 0.94 degrees in latitude. This provides slightly better resolution of zonal currents in the southern Tropics, more nearly isotropic resolution at mid latitudes, and somewhat reduced problems with converging meridians in the Arctic. CNRM-CM3 (2045-2065): This model provides similar resolutions from the above mentioned models but presents bias to the cold side in most of the tropics. This model has proven to overestimate the stream flows in summer, with the opposite occurring during the winter in the Americas (Saurral and Barros, 2009). Although for the American continent the model shows some deficiencies in the representation of the water cycle across the region, validations of temperature and precipitation fields are relatively accurate for the northern hemisphere of the Americas. 22 Scenarios used with these models: 20c3m SRESa2 SRESb1 (IPSL does not have data for the far future under SRESB1 experiment). 33 CSIRO-Mk3.5 (2045-2065): Created by the Centre for Australian Weather and Climate Research, this model uses a dynamical framework of the atmospheric model is based upon the spectral model with the equations cast in the flux form that conserves predicted variables. The application of this model is vastly used over long-term climate change simulations. The most significant improvements result from the use of a more physically realistic set of parameters to represent the transport of heat and freshwater by oceanic eddies. It also features considerably more realistic circulation and stratification in the Southern Ocean, affecting precision in temperature and precipitation estimates over the fall and winter. GFDL-CM2.0 & GFDL-CM2.1 (2 models) (2045-2065): This is a coupled atmosphere-ocean general circulation model (AOGCM) developed at the NOAA Geophysical Fluid Dynamics Laboratory in the United States. It is one of the leading climate models used in the Fourth Assessment Report of the IPCC. The atmospheric component of the CM2.X models is a 24- level atmosphere run at a resolution of 2 degrees in the east-west and 2.5 degrees in the north-south direction. This resolution is sufficient to resolve the large mid-latitude cyclones responsible for weather variability. It is too coarse, however, to resolve processes such as hurricanes or intense thunderstorm outbreaks. The inclusion of this model as part of the 9 model-prediction estimations is useful to incorporate intense outbreaks. IPSL-CM4 (2045-2065): One of the goals of the IPSL modeling is to study how these different couplings can modulate climate and climate variability, and to determine how feedbacks in the Earth system control the response of climate to a perturbation such as the anthropogenic emissions of greenhouse gases. This is a relatively simple modeling that comprises four atmospheric prognostic variables: a) northward and eastward wind components, b) temperature, c) water availability, d) surface pressure. The data used in this model requires the time period between 1961 and 1990, for precipitation and temperature, which is data that is contained in our dataset for each municipality in Mexico on a weekly basis. ECHO-G: Is a hybrid coupled model, using ECHAM4 atmosphere and HOPE ocean models. The model contains a control simulation, allowing 1000-year simulation with constant external forcing. The model is capable of simulating unconventional climatology, which is consistent with other similar models with flux-adjusted modulation on climate and gradients, although the flux adjustment does not guarantee a more accurate simulation (Latif et al., 2001; AchutaRao and Sperber, 2002; Davey et al., 2002). ECHAM5/MPI-OM: This is the latest version of the ECHAM model. ECHAM5 may host submodels going beyond the meteorological processes of a GCM. The model can be used in special modes. This model perform best globally, with some biases in certain artic regions, which makes it one of the strongest models to be used in tropical and sub-tropical areas (Connolley, W. and Bracegirdle, T., 2007) 34 MRI-CGCM2.3.2: Meteorological Research Institute (MRI) Coupled Global Climate Model (CGCM; version 2.3.2a), produce realistic rainfall patterns at low latitudes. This model can be applied globally and regionally with the feature of permitting the partitioning of the total variance of precipitation among intra-seasonal, seasonal, and longer time scales. This is reproduced by the model, except over the western Pacific where the models fail to capture the large intra-seasonal variations. Models used for Robustness Checks ECHAM4 (2030): This was created by modifying global forecast models default configuration of the model resolves the atmosphere (primarily used to study the lower atmosphere), targeting arid, semi-arid, sub-tropical and tropical areas. Given this climate distribution, Mexico’s climates fit this model. This model has been used extensively to study the climate of the troposphere in Mexico, allowing to include also the middle atmosphere. HADGEM (2030): Is the most recent atmospheric model (precipitation and temperature) atmospheric component has 38 levels extending to ~40km height, with a horizontal resolution of 1.25 degrees of latitude by 1.875 degrees of longitude, which produces a global grid of 192 x 145 grid cells. These grid cells are similar in size to those reported by the geographical unit of INEGI and the Autonomous National University in Mexico (UNAM). One of the main differences between this climate configuration and previous versions is the use of the New Dynamics core which is a non-hydrostatic (assumption of precipitation changes), fully compressible (ability to be disaggregated spatially), with a semi-implicit semi-Lagrangian time integration scheme (longer prediction periods). CCVI Eigen Values for Baseline and Prediction (2005/2045) Scree plot of eigenvalues after pca 4 3 Eigenvalues 2 1 0 0 5 10 15 20 Number 95% CI Eigenvalues 35 Scree plot after PCA for MainPrediction Model 4 3 Eigenvalues 21 0 0 5 10 15 Number 95% CI Eigenvalues Distribution of Agricultural Vulnerability Risk to Climate Change 1 Higher Risk .8 .6 Density .4 .2 0 -1 0 1 2 3 CCVI CCVI for Municipalities with Poverty above 50% CCVI for Municipalities with Poverty below 50% kernel = epanechnikov, bandwidth = 0.2800 36 State Level CCVI (2005-2045) Main Model Baseline (2005) Main Model Prediction (2045) Change State Index S.d. Mun Index S.d. Mun Index Vulnerability Aguascalientes -0.5009 0.1243 11 -0.4061 0.1395 11 0.0948 (+) Baja California -0.2540 0.1428 3 -0.3823 0.1353 5 -0.1282 (-) Baja California Sur -0.8512 0.1929 3 -0.7969 0.1958 5 0.0543 (+) Campeche 0.7842 0.2973 11 0.3804 0.3055 11 -0.4038 (-) Chiapas 1.3906 0.3856 112 1.5631 0.2786 117 0.1725 (+) Chihuahua 0.1014 0.4552 37 0.2558 0.4139 67 0.1544 (+) Coahuila -0.3650 0.1360 25 -0.6504 0.1700 38 -0.2854 (-) Colima 0.1803 0.3522 9 -0.0288 0.2696 10 -0.2091 (-) Distrito Federal 0.4160 0.2679 7 0.2436 0.1443 10 -0.1724 (-) Durango -0.1372 0.4382 37 -0.1825 0.4744 39 -0.0453 (-) Guanajuato -0.2409 0.2064 46 -0.0513 0.2085 46 0.1897 (+) Guerrero 0.9046 0.3753 76 0.8003 0.3178 76 -0.1043 (-) Hidalgo 0.1700 0.7793 84 0.2691 0.5747 84 0.0991 (+) Jalisco 0.2546 0.3200 121 0.2048 0.2689 124 -0.0497 (-) Michoacán 0.4081 0.3588 112 0.4033 0.3149 113 -0.0048 (=) Morelos 0.4146 0.2216 31 0.3182 0.2237 33 -0.0964 (-) México 0.2441 0.4412 121 0.2501 0.3701 122 0.0060 (=) Nayarit 0.6354 0.2261 20 0.3108 0.2511 20 -0.3246 (-) Nuevo León -0.1046 0.1926 32 -0.3072 0.2121 49 -0.2026 (-) Oaxaca 0.7378 0.5899 557 0.8766 0.4383 570 0.1388 (+) Puebla 0.4391 0.6530 214 0.5706 0.4504 217 0.1315 (+) Querétaro -0.2415 0.3416 18 -0.1551 0.3522 18 0.0864 (+) Quintana Roo 0.5705 0.3686 8 0.6398 0.3477 8 0.0694 (+) San Luis Potosí 0.2549 0.7717 57 0.0932 0.5224 57 -0.1617 (-) Sinaloa -0.0064 0.3300 15 -0.3471 0.3459 18 -0.3407 (-) Sonora -0.0196 0.2967 10 -0.4272 0.2403 72 -0.4075 (-) Tabasco 1.1752 0.3408 17 0.7122 0.2461 17 -0.4630 (-) Tamaulipas 0.0156 0.3312 38 -0.2791 0.3585 43 -0.2948 (-) Tlaxcala 0.0301 0.1664 59 0.1221 0.2110 60 0.0920 (+) Veracruz 1.1434 0.4075 208 0.8737 0.3701 210 -0.2697 (-) Yucatán 0.5469 0.3142 106 0.8136 0.3116 106 0.2667 (+) Zacatecas -0.3273 0.3087 56 0.0476 0.2742 57 0.3749 (+) Source: Own estimations 37 Main Model Mean Significant State Index 2005 S.d. 2005 Index 2045 S.d. 2045 Difference t- Difference value Aguascalientes -0.5009 0.1243 -0.4061 0.1395 1.369 No Baja California -0.2540 0.1428 -0.3823 0.1353 -1.744 Yes Baja California Sur -0.8512 0.1929 -0.7969 0.1958 0.571 No Campeche 0.7842 0.2973 0.3804 0.3055 -1.702 Yes Chiapas 1.3906 0.3856 1.5631 0.2786 2.039 Yes Chihuahua 0.1014 0.4552 0.2558 0.4139 0.407 No Coahuila -0.3650 0.1360 -0.6504 0.1700 -1.206 No Colima 0.1803 0.3522 -0.0288 0.2696 -0.774 No Distrito Federal 0.4160 0.2679 0.2436 0.1443 0.169 No Durango -0.1372 0.4382 -0.1825 0.4744 -0.075 No Guanajuato -0.2409 0.2064 -0.0513 0.2085 0.963 No Guerrero 0.9046 0.3753 0.8003 0.3178 0.111 No Hidalgo 0.1700 0.7793 0.2691 0.5747 0.256 No Jalisco 0.2546 0.3200 0.2048 0.2689 -0.034 No Michoacán 0.4081 0.3588 0.4033 0.3149 0.146 No Morelos 0.4146 0.2216 0.3182 0.2237 -0.477 No México 0.2441 0.4412 0.2501 0.3701 0.124 No Nayarit 0.6354 0.2261 0.3108 0.2511 -1.747 Yes Nuevo León -0.1046 0.1926 -0.3072 0.2121 -0.944 No Oaxaca 0.7378 0.5899 0.8766 0.4383 0.752 No Puebla 0.4391 0.6530 0.5706 0.4504 0.600 No Querétaro -0.2415 0.3416 -0.1551 0.3522 0.300 No Quintana Roo 0.5705 0.3686 0.6398 0.3477 0.390 No San Luis Potosí 0.2549 0.7717 0.0932 0.5224 -0.157 No Sinaloa -0.0064 0.3300 -0.3471 0.3459 -1.107 No Sonora -0.0196 0.2967 -0.4272 0.2403 -1.761 Yes Tabasco 1.1752 0.3408 0.7122 0.2461 -0.628 No Tamaulipas 0.0156 0.3312 -0.2791 0.3585 -0.866 No Tlaxcala 0.0301 0.1664 0.1221 0.2110 0.412 No Veracruz 1.1434 0.4075 0.8737 0.3701 -0.449 No Yucatán 0.5469 0.3142 0.8136 0.3116 0.887 No Zacatecas -0.3273 0.3087 0.0476 0.2742 1.278 No Source: Own estimations 38 Table 3b Characteristics of States with Highest Vulnerability Shifts Highest Vulnerability Decrease % of Average Index Percent of Percent of Percent of Rain % of Agriculture % of Agriculture Agriculture Agriculture Indicator CCVI Index Prediction Indigenous Households with Elderly (65+) Coefficient GDD workers with workers receiving workers with support in Pesos (2045) Population Dirt Floors Population of Variation Savings remmitances credit 2009 * mean 0.562 -0.105 7.02 1.94 7.96 0.35 10.63 3.97 37.34 230.85 1.86 range 2.331 2.032 88.62 52.39 14.95 0.71 12.94 22.39 88.57 987.53 16.90 sd 0.544 0.507 15.48 5.57 3.37 0.10 2.71 4.36 19.40 290.23 3.06 p5 -0.452 -0.799 0.00 0.00 3.81 0.22 6.24 0.00 12.40 0.00 0.00 p95 1.482 0.831 40.30 11.40 14.92 0.54 14.30 14.10 75.90 839.86 8.77 N 138 138 138 138 138 138 138 138 138 97 138 Highest Vulnerability Increase % of Average Index Percent of Percent of Percent of Rain % of Agriculture % of Agriculture Agriculture Agriculture Indicator CCVI Index Prediction Indigenous Households with Elderly (65+) Coefficient GDD workers with workers receiving workers with support in Pesos (2045) Population Dirt Floors Population of Variation Savings remmitances credit 2009 * mean 0.277 0.488 16.41 7.91 6.78 0.27 10.46 1.82 27.75 439.16 3.65 range 2.367 2.657 99.82 69.73 20.97 0.33 444.91 10.19 100.00 999.80 25.18 sd 0.724 0.887 29.47 15.26 3.47 0.06 36.13 1.56 22.83 330.08 4.40 p5 -0.701 -0.615 0.18 0.00 2.60 0.17 2.68 0.00 0.00 0.00 0.00 p95 1.327 1.756 94.01 45.46 13.45 0.36 14.58 4.82 69.80 948.68 11.96 N 289 251 289 289 289 289 289 289 289 102 289 States with highest vulnerability Decrease: Tabasco, Sonora, Campeche, Sinaloa, Nayarit. States with highest vulnerability Increase: Zacatecas, Yucatan, Guanajuato, Chiapas, Chihuahua Source: Own estimations 39 Table 4c Correlations Selected Variables and Vulnerability Risk Category Frost Days (<10 C) Consecutive Dry Days Growing Degree Days Coefficient of Variation Rain Risk Category * Baseline Prediction Baseline Prediction Baseline Prediction Baseline Prediction Very High Vulnerability 1.83 3.06 88.37 88.26 9.59 15.72 0.32 0.46 Risk High Vulnerability Risk 0.92 1.25 83.03 82.49 10.20 14.06 0.29 0.37 Moderate Vulnerability 0.45 0.53 81.09 85.21 11.08 13.68 0.26 0.36 Risk Low Vulnerability Risk 0.13 0.15 68.21 72.84 9.58 12.41 0.27 0.34 Very Low Vulnerability 0.38 0.01 53.89 40.29 6.46 4.33 0.27 0.29 Risk Low Capital intensity Transitional extensive Subsistence agriculture Subsistance agriculture non- Other (Small Farms) Risk Category ** Agriculture Agriculture Capital intensive intensive Baseline Prediction Baseline Prediction Baseline Prediction Baseline Prediction Baseline Prediction Very High Vulnerability 0.39 4.77 14.15 17.85 38.15 45.97 33.4 41.54 5.23 5.11 Risk High Vulnerability Risk 2.22 1.85 17.08 24.44 36.89 39.51 31.78 35.19 4.00 5.97 Moderate Vulnerability 2.09 4.11 30.7 32.24 27.44 27.52 31.63 26.69 6.51 7.80 Risk Low Vulnerability Risk 5.06 5.13 32.7 39.63 20.94 23.63 16.43 23.63 12.03 13.96 Very Low Vulnerability 10.2 13.17 41.91 40.53 12.86 8.23 11.09 6.17 13.08 7.20 Risk % of Agriculture % of Agriculture % of HH in Mun. w/ % of Agriculture workers % of Agriculture % of Indigenous pop. by Mun. % of HH by Mun w/ Dirt floors workers with support workers receiving Risk Category *** dwellers above 65 yo with Savings workers with credit programs remmitances Baseline Prediction Baseline Prediction Baseline Prediction Baseline Prediction Baseline Prediction Baseline Prediction Baseline Prediction Very High Vulnerability 0.141 0.155 0.153 0.098 0.182 0.246 -0.177 -0.153 0.090 0.139 -0.103 -0.171 -0.153 -0.270 Risk High Vulnerability Risk 0.058 0.053 0.069 0.074 0.048 0.083 -0.117 -0.106 0.099 0.103 -0.080 -0.269 -0.140 -0.183 Moderate Vulnerability 0.089 -0.034 0.089 0.067 -0.026 -0.072 -0.044 0.034 0.009 0.039 0.041 0.105 -0.022 -0.131 Risk Low Vulnerability Risk 0.055 -0.026 0.056 -0.014 -0.049 -0.022 -0.048 0.079 -0.046 -0.044 0.059 0.112 0.011 -0.017 Very Low Vulnerability 0.044 -0.068 0.057 -0.054 -0.094 -0.031 -0.011 -0.010 -0.124 -0.054 0.050 0.135 0.065 0.065 Risk * Consecutive dry days based the number of days below 2 standard deviations from Monthly average or no rain at all reported. GDD are calculated by taking the average of the daily maximum and minimum temperatures compared to a base temperature. The coefficient of variation (CV) is defined as the ratio of the standard deviation to the mean. ** Percent of Municipalities under risk categories, figures don't add up to 100 horizontally because 3 categories of Capital Intensive Agriculture production units not included. *** Pairwise Correlations between Index and Variable in question. Bold indicate significant at 10% level. Source: Own estimations 40 Annex II Literature Review on Applications of PCA to build Multidimensional (small area) Indices For many years the statistical literature lacked a uniform approach to combine indicators that result in a composite index from multidimensional data. A number of indices were devised over the years, including Duncan’s index that combined labor and income data of individuals, or the Townsend’s index designed to explain variation in health in terms of material deprivation (Morris & Castairs, 1991). However, a major problem facing researchers when constructing indexes is determining an appropriate aggregation strategy to combine multidimensional variables into a composite index. For years, researchers built aggregated indices from multidimensional variables using simple Summation of Standardized Variables (SSV). This approach initially developed by Shevky & Bell (1955) and applied by Markides & McFarland (1982), used statistical standardization of variables to add them up and test variability of the index according to different development outcomes applied to infant mortality. However, many statistical experts found that such methods rely on applying weights to the constituent variables that make up individual as well as composite indices, which rely on subjective factors, thus raising questions about internal coherence and robustness of such methods (Gjolberg, 2009). Despite that the PCA technique is not new its application to develop composite weighted indices is relatively recent. The PCA technique developed by Pearson (1901), though it is often attributed to Hotelling (1933), is useful for transforming a large number of variables in a data set into a smaller and more coherent set of uncorrelated (orthogonal) factors, the principal components. The principal components account for much of the variance among the set of original variables. Each component is a linear weighted combination of the initial variables 23. The components are ordered so that the first component accounts for the largest possible amount of variation in the original variables. The second component is completely uncorrelated with the first component, and accounts for the maximum variation that is not accounted for the first. The third accounts for the maximum that the first and the second not accounted for and so on. PCA was first used to combine socioeconomic indicators into a single index (Boelhouwer & Stoop, 1999). Acknowledging the inappropriateness of simple aggregation procedures, Lai (2003) modified the UNDP Human Development Index by using PCA to create a linear combination of indicators of development. Several 23 The weights for each principal component are given by the eigenvectors of the correlation matrix or the covariance matrix, if the data were standardized. The variance for each principal component is represented by the eigenvalue of the corresponding eigenvector. 41 researchers have used PCA, especially since late 1990s, to compute area socioeconomic indices (Antony & Rao, 2007; Fukuda, Nakamura, & Takano, 2007; Fotso & Kuate-defo, 2005; Havard, Deguen, Bodin, Louis, & Laurent, 2008; Messer, Vinikoor, Laraia, Kaufman, Eyster, Holzman, Culhane, Elo, Burke, & O’Campo, 2008; Rygel, O’Sullivan, & Yarnal, 2006; Tata & Schultz, 1988; Sekhar, Indrayan, & Gupta, 1991; Vyas & Kumaranayake, 2006; Zagorski, 1985). Finally, the PCA is computationally easy and also avoids many of the problems associated with the traditional methods, such as aggregation, standardization, and nonlinear relationships of variables affecting socioeconomic inequalities (refer Vyas & Kumaranayake, 2006, for an assessment of advantages and disadvantages of PCA and Saltelli, Nardo, Saisana, & Tarantola, 2004, for the pros and cons of composite indicators, in general). Graphically the steps to conduct a PCA computation are based on the following diagram: PCA Algorithm Procedure Source: Based on Krishnan, 2010 Annex III Examples of Multidimensional Indices built for Mexico using Principal Components Analysis Mexico has a history in building important municipal indices that capture multidimensional aspects of social and economic variables. In 2005 the United Nations Development Program (UNDP) supported the government of Mexico to build a Human Development Index at the municipal index. This indicator was build using Principal Components Analysis (PCA) combining life expectancy, literacy rates, school enrollment rates, GDP per capita, inequality and ethnic composition. The index was used to rank municipalities in order to prioritize public spending to 42 those municipalities and regions with lowest levels of human development (IDH 2005, UNDP). Also, the index constructed at baseline (2000) and at a follow up (2005) periods, assessed changes in human development at the state and municipal levels (Graph 1a). Graph 1a Mexico’s Inequality in Human Development Index by State, 2000-2005 Source: IDH, UNDP 2005. With a precedent in building a human development index for municipalities in Mexico, the National Population Council (CONAPO) in Mexico, embarked in the task of building a more refined index that incorporated other dimensions of social well- being beyond human development. In 2000 and 2005 CONAPO used PCA analysis to build a socioeconomic index that measured the level of marginalization by municipality based on three dimensions. The first dimension measured education- related indicators (years of schooling, level and type of education, literacy rates), mostly captured in CENSUS data. The second dimension of the index measured household conditions and access to public services (household physical characteristics, access to water and sanitation, and energy) collected from two sources: CENSUS data and two large sample surveys (ENOE and ENIGH). The last dimension to measure marginality incorporated variables related to municipal characteristics in terms of population size, labor occupancy rates, and urbanization collected from CENSUS and large sample data as well. This index was build based on the above-mentioned indicators including only those with highest explanatory power over the covariance of all indicators. PCA was used then to aggregate all three dimensions to build the index that categorized municipalities in five levels of marginality: very low, low, average, high and very high. The index helped to rank states in order to prospectively plan the allocation of resources from the programmatic plans elaborated by the Ministry of Finance, where high priority of funding was given to states and municipalities with high and very high marginalization (Graph 2a). 43 Graph 2a Marginality Index by State (Mexico 2005) These examples illustrate previous efforts to build indices used for important policy decisions. Other indices have been built to assess multiple dimensions of well-being. In 2010, the National Evaluation Council (CONEVAL) built a composite index using PCA analysis that measured the Social Gaps prevailing across municipalities (Graph 3a). Graph 3a. Social Gap Index by Municipality Mexico 2010. Source: CONEVAL, 2011 44 The main purpose of the social gap index is to prioritize specific policies and programs that target multiple social development interventions. This index ranks municipalities based on human development, access to social services and household conditions. The index is helping to reshape social policies and priorities at the municipal level and it is used to assess social inequality as well. With this tool state and national governments have evidence to allocate federalize funds into municipalities that show highest social gaps. Recently, other indices have been built to assess specific inequalities in the distribution of risk against climate change. The Mexican Institute of Water Technology (IMTA) built a Municipal Index for Water Scarcity Risk from Climate Change. This index is completely submerged in the climate change agenda and has the advantage of incorporating multiple dimensions to assess Water Scarcity risks. These dimensions include health, education, household conditions, employment, population and family structure, gender, adaptive capacity and risk perception. Although this index is still under review, it conceptually measures an important challenge that municipalities will face in the future: the risk of water resources reduction and their allocation. These examples illustrate the importance of using rich data and statistical tools to assess various aspects of economic, social and sustainability issues at the local level. 45