C LIMATE AND M IGRATION : U NPACKING T HE ROLE OF S OCIAL N ETWORKS - - PDF document

CLIMATE AND MIGRATION: UNPACKING THE ROLE OF SOCIAL NETWORKS Jacqueline Meijer-Irons and Sara R. Curran Center for Studies in Demography and Ecology, University of Washington Seattle,WA USA Paper Presented to the International Union for the Scientific Study of Population Cape Town, South Africa October 29 – November 4, 2017

1

• 1. Introduction

Climate-related events, such as drought and floods, may increase in severity and frequency under current climate projections. Recent climate extremes have revealed areas of vulnerability, where a lack of consistent rainfall has impacted local food systems and agricultural productivity (IPCC 2014). These disruptions are of particular concern for farmers in the developing world who rely on rainfed agriculture for a significant part of their income. Yet, not all households experience climatic stress in the same way; the extent to which a household is impacted depends, in part on how vulnerable a household is to the economic impacts of environmental stress. This vulnerability, in turn, can be mitigated by a household member’s ability to migrate and engage in

• ff-farm employment, a form of adaptation (McLeman and Smit 2006). Migrant remittances sent

home are used to diversify household income and reduce risk (Bebbington 1999; de Haan 1999; de Haas 2010; Kniveton et al. 2008; Stark and Bloom 1985; Stark and Taylor 1989). A growing body of research argues that one consequence of increasing uncertainty will be the additional influence of environmental stressors on existing systems of internal labor migration from rural to urban areas, rather than creating new migrant streams (Black et al. 2011a; Black et al. 2011b). However, recent scholarship is mixed on the relationship between migration and the environment, demonstrating that migration response to the environment is highly contextual, depending on cropping patterns and cultural norms about who engages in migration (Gray and Mueller 2012; Kubik and Maurel 2016). In several studies, the type of migration a household member might typically engage in is altered in response to the environment; for example, timing

• f a first migration, or a shift away from international migration towards more local or internal

migration (Gray 2009; Gray 2010; Gray and Billsborrow 2013; Gray and Mueller 2012a; Gray

2

and Mueller 2012b; Findley 1994; Henry et al. 2004; Massey et al. 2010; Nawrotzki et al. 2015). In other cases, the odds of migration increase, but only after a sustained period of exposure to environmental stress, and then only for specific groups of migrants, suggesting households might adapt in place until they decide to engage in migration (Curran and Meijer-Irons 2014; Nawrotzki and DeWaard 2016). Finally, in some studies, the odds of migration declines when people in the study area experience degraded environmental conditions (Dillon et al. 2001); still

• ther works finds the odds of migration increase in response to positive environmental
• conditions. In this latter study, households rely on natural capital returns to fund labor migration

that can further supplement household income (Gray 2011; Gray and Billsborrow 2013). Recent scholarship proposes another way to conceptualize the role that the environment plays in migration decisions: consider, along with main effects of the environment, how the environment interacts more explicitly with specific drivers of migration (Black et al. 2011a). A recent review of the literature on migration in Less Developed Countries finds that while many studies report a relationship between rainfall and migration, few empirical papers analyze how changes in climatic conditions either strengthen or weaken drivers of migration, particularly economic drivers (Lilleor and van den Broeck 2011). Social networks in the origin and destination play a key role in facilitating migration, by lowering the costs of migration, and providing information to would-be migrants, but the role of social networks has not been fully elaborated in studies linking migration and the environment (Adamo and de Sherbinin 2011; Bardsley and Hugo 2010; Curran 2002). Recent work suggests that social networks can also serve as a key factor in facilitating migration in response to climate change, whereby people with access to social networks in the destination are more likely to engage in climate-induced movements (Bardsley and Hugo 2010). On the other hand, a recent article examining the

3

relationship between social networks, climate change, and international migration out of Mexico, Nawrotzki et al. considers the potential amplification effect of social networks, but the authors also consider the dampening or suppression effect that social networks might produce (2015). In this latter case, they find that community level social capital suppresses migratory response to the impact of climate shocks positing that social networks provide the resources and

• pportunities to foster resilience and adaption in place and therefore mitigate the need to migrate

(Adger 2003). This current study expands on the work conducted on international migration out of Mexico in response to climate shocks, situating the analyses within the context of access to social capital and internal migration. In this study, we investigate the role that migrant social capital plays in migrant selectivity in response to environmental shocks, net of other drivers of

• migration. We also consider the interaction of environmental shocks and social capital, measured

at the individual, household, and community level accumulated prior to the migration event. We use the Nang Rong data, a unique panel data set from NE Thailand that measures internal migration over a 16-year period. Nang Rong is a good choice for a study site because of the history of internal migration in the area, a former frontier region that has undergone considerable land use and population changes during the latter half of the twentieth century (Entwisle et al. 2008). Nang Rong has also been the focus of extensive study and much is known about the motivations and consequences of circular labor migration from the area. Considerable quantitative and qualitative data have also been collected on the environment in Nang Rong (Curran et al. 2005; Entwistle et al. 2016; Garip 2008; Van Wey 2003; Rindfuss et al. 2007). Seasonal migration is not uncommon in Nang Rong, where the rainy, monsoon season is often followed by drought-like conditions that require people to migrate in search of non-agricultural

4

• labor. Further, several studies using these data find a significant influence of migrant networks
• n migration response following economic and environmental shocks (Curran et al. 2005; Curran

and Meijer-Irons 2014; Curran et al. 2016; Entwistle et al. 2016; Garip and Curran 2009). We hypothesize that access to migrant social capital can either increase or decrease migratory response to climate shocks. In the first case, it might be that access to social networks and cumulative experience makes an additional move due more feasible and less costly to a

• household. On the other hand, established social networks might translate into access to

remittances and other resources that allow for in situ adaptation that (Adger 2003). Doing so, we add to the growing empirical literature that finds that migration in response to climate stress is complex, and often influenced by underlying migration systems already present in a given area.

• 1. Data and Methods

2.1 Nang Rong Migration Data Our migration data come from the Nang Rong Surveys, a longitudinal panel data collection effort conducted by the Carolina Population Center at the University of North Carolina and the Institute for Population and Social Research at Mahidol University in Thailand.1 We employ the first three waves of data (collected in 1984, 1994, and 2000) for our analyses. The 1984 data collection was a census of all households and individuals residing in 51 villages within Nang Rong. It included information on individual demographic data, household assets and village institutions and agricultural, natural, economic, social, and health resources. Further, village-level data were collected from all of the villages in Nang Rong district. The 1994 survey

1 The data and information about the surveys are available at http://www.cpc.unc.edu/projects/nangrong/

5

followed all 1984 respondents still living in the original village, as well as respondents from 22

• f the original 51 villages who had moved to one of the four primary destinations outside of the

district, plus any new village residents. The 1994 surveys included all questions from the 1984 survey, as well as a 10-year retrospective life history about education, work, and migration, a survey about the age and location of siblings, and a special survey of migrants’ migration experiences and histories. The 2000 round of surveys built on the previous data collection efforts by following all of the 1994 respondents and adding to the database any new residents and households in the original villages. The 1994 and 2000 surveys included a migrant follow-up component. This was conducted among persons who had resided in 22 of the original 1984 villages, and defined a migrant as someone who was a member of a 1984 household and had since left a village for more than two months to one of four destinations: the provincial capital, Buriram; the regional capital, Korat or Nakhon Ratchasima; Bangkok and the Bangkok Metropolitan Area; or Eastern Seaboard provinces. The migrant follow-up in 2000 included migrants identified and interviewed in 1994, and individuals who had lived in the village in either 1984 or 1994 but subsequently migrated to one of the four primary destinations. The retrospective recall items in the survey allow us to measure timing and sequencing of moves (outgoing and returning), migrant destination, occupation in destination, and duration of stay. The data for these analysis focus only upon villagers from the 22 villages where there was a migrant follow-up component. In these villages, the follow-up rate is fairly high (about 78%) because the survey team relied on multiple search methods (see Rindfuss et al. 2002). This means that migrant selectivity bias is minimized among this group of villagers and villages.

6

Our analysis file relies primarily on the data found in the life history modules implemented in both 1994 and 2000. With these data we construct an analysis file that is comprised of person-year-move records. For each individual we have information about their sequence of residences and moves within a year for the preceding 10 years in the case of the 1994 survey and for the preceding six years for the 2000 survey. Retrospective life histories were collected for most individuals who had ever resided in Nang Rong in any 1984, 1994 or 2000 household and who were 13-44 years old at some point during this time period. Our analyses examine individual behavior prospectively from 1984 and 1994 to 2000 and do not include individuals who newly appear in households in 2000. We measure migration as any move outside of the Nang Rong district for 2 months or

• more. There is a great deal of variation across the 22 villages there is a general trend of

increasing migration between 1990 and 1998, with drop-off after 1998. However, villages located variously across the landscape appear to follow very different trends annually with some exhibiting relatively high levels of migration in a year and others lower levels. In other studies, it has been shown that the cumulative patterns of migration are quite different across villages, with some villages exhibiting quite steep trajectories of accumulated migration experience and others exhibiting much lower rates of increase (Curran et al. 2005; Garip and Curran 2009; Garip 2008). In order to take into account and control for underlying currents of migration trends that might be explained by a host of other factors, besides environmental conditions, we also control for migration histories and migration experiences at the individual, household, and village level. While not perfect proxies for alternative explanations for migration patterns, prior migration prevalence is a well-known measure of cumulative migration and the temporal ordering partially

7

allays endogeneity concerns. Separately, we estimate the number of trips made by a person up through year t-1, the number of months experienced as a migrant by that person up through year t-l, the number the number of trips made by other community members up through year t-1, the months of experience accumulated by other community members through year t-1. The household and community migrant trips and months of migrant experience do not include the experience of the observed individual (for details please see Curran et al. 2005). We include additional controls that predict migration, including age, marital status, level of education

• btained, and land ownership. In addition, we include an asset measure and a variable that

indicates whether a household had ever received remittances, and variables that measure how remote the village is, as well as whether or not the village has electricity. Table 1 below summarizes the distribution of these descriptive statistics in 1984 and 2000, the starting and ending points of our data. Table 1 about here 2.4 NDVI Local Environmental Measure For our main predictor variable, we used the Normalized Difference Vegetation Index (NDVI) to examine how the localized changing conditions of vegetation health across Nang Rong may play a role in migratory decisions. NDVI has been used for many years to monitor the photosynthetically active biomass and growth (vigor) of plant canopies from satellite remote sensing imagery (Tucker et al. 1985), and is becoming increasingly popular as a tool to assess vegetation’s response to environmental change (Pettorelli et al. 2005). This vegetation index compares the intensity of light reflected in two regions or “bands” of the electromagnetic spectrum: 1) Red, where chlorophyll causes considerable absorption, and 2) Near-infrared, where spongy mesophyll leaf structure creates considerable reflectance (Tucker 1979). NDVI is

8

calculated as the difference between the values of the near-infrared and red bands divided by the sum of the values of these same bands. Vigorously growing healthy vegetation has low red-light reflectance and high near-infrared reflectance, and hence, high NDVI values. The long-term Global Inventory Modeling and Mapping Studies (GIMMS) NDVI dataset was chosen for this study because its historical vegetation health record completely overlaps the time span of the Nang Rong migration data. GIMMS provides 24 full years (1982 – 2006) of global bimonthly NDVI data (24 measures per year) compiled from a series of National Oceanic and Atmospheric Administration-Advanced Very High Resolution Radiometer (NOAA- AVHRR) satellites and instruments.2 This dataset has been corrected for calibration, view geometry, volcanic aerosols, and other effects not related to vegetation change (Tucker et al. 2005). The primary drawback of the GIMMS data is its coarse scale (low spatial resolution), with unique values reported for every 8km x 8km area (pixel). What GIMMS NDVI dataset lacks in spatial resolution it makes up for in temporal resolution, or measurement frequency. Most remote-sensing-based studies of landscape change employ imagery data resources as longitudinal or panel analysis data. For several dates throughout the study period satellite imagery is used to derive NDVI for each pixel or to classify each pixel into one of many land-use

• r land-cover (LULC) categories. These NDVI or LULC classification data are then used as

indicators of the landscape state at specific moments in time, or compared to one another to derive change trajectories. In contrast, the GIMMS data provides enough samples of NDVI to permit a more complete look at the yearly vegetation health cycle as demonstrated by plots showing the NDVI curve shape for several pixels in Nang Rong (Fig. 1). In Figure 1, we show the pixel coverage for NDVI for the Nang Rong district. We also show the villages captured in

2 GIMMS data and documentation available at http://www.landcover.org

9

each pixel with the red circles. And, following a transect from the southwest corner to the northeast corner of the district, a line that starts in the uplands and moves towards the lowlands we show the annual monthly trends of NDVI signals for 1994, 1995 and 1996 for each pixel. Taken together, each yearly set of NDVI values from a pixel (visualized by the shape of a plot) show a unique signature of vigor (stressed, normal, or highly productive) of the land cover type(s) at that location (rice paddy, upland crop, forest, etc.). Because there are so many points

• f data across time, we have 49 pixels and 26 years of data, in our study we used a simple

unsupervised (i.e. fully automated) clustering approach to group similar “pixel-years” of NDVI data for the entire Nang Rong District for comparison to yearly migration rates of individuals within villages. For the clustering algorithm we chose model-based clustering, specifically the finite Gaussian mixture models estimated by the MCLUST package in R (Fraley and Raftery 1999). This package uses Bayes factors to optimize the finite mixture model over the number of mixtures considered and the covariance matrix of the variables included in the model. For each of the 49 pixels in the study area, our clustering approach used the twelve monthly averages of NDVI values for an entire Thai “Water Year” (May – April of the following year) to correspond with monsoon-based seasons and crops (Crews-Meyer, 2004). We allowed a maximum of 20 clusters with BIC scores determining the optimal number of clusters. Given their location, each village could be associated with a pixel and its corresponding cluster for each year

• f the study. Figure 2 displays the full range of clusters that are derived from the data and

unsupervised modeling approach. There are nine statistically different clusters of NDVI annual patterns that appear to show quite different signals of wetness, green-up, and drought. Of these nine clusters, six clusters (1, 2, 6, 7, 8, and 9) include one or more of our study villages (see Figure 1 below). Cluster 1 is the modal cluster for most pixel-years. It shows a bump up or

10

green up during months Aug-October, an expected increase that is expected given the end of the monsoonal season and the resulting vegetation growth, particularly in rice paddies. Other clusters show steeper inclines in green-up indicating possibly strong and healthy vegetative growth, particularly clusters 6 & 9. Clusters 2 shows two periods of green-up, also indicating possibly relatively robust vegetative growth. Our interpretations of these clusters are necessarily speculative as we do not yet have land cover information to calibrate our understanding of these

• signals. The development of this measure is significantly different both substantively and

methodologically from previous uses of NDVI in predicting migration outcomes. Rather than using a single signal and interpreting plant stress, we consider the collective pattern of signals reflecting vegetation health over the entire agricultural year from land preparation to planting, harvesting and fallow. With this measure we also are attempting to capture the retrospective viewpoint of farmers assessing the livelihood risks of agricultural decisions over the past year as they might influence their subsequent decisions for the next year. We expect that these yearlong retrospections are better estimates of what influences a farmers’ intuition-driven assessment which then influences an individual’s and members of a household’s behavior. Figure 1 about here Model Specification We run logit models for panel data (xtlogit in Stata 14) to estimate the odds of an individual migrating out of Nang Rong in a given year. We estimate a random effects model to allow for the assumption that there is unobserved heterogeneity that is correlated with our independent variables. The functional form of our model is: Prob(Migit) = f(Itripsit-1, Imonthsit-1, Htrips it-1, Hmonths it-1, Vtripsit-1, Vmonthsit-1, Variantit , Invarianti).

11

To control for economic opportunities that might influence out-migration, we include controls for year-effects, but we do not include the results in our output. We run three models: Model 1,

• ur base model, includes individual-level characteristics: age, marital status, education level;

household-level characteristics: whether a household member was a temporary migrant, whether the household had ever received remittances, and an asset measure; finally, we include information about the village, namely whether it is considered somewhat or very remote, and an indicator variable for the year that the village received electricity (note that by 2000 all villages are electrified). Model 2 includes our measures for individual, household, and village social

• capital. Model 3, our full model includes our environmental measures. Next, we run six

interactive models, interacting our NDVI cluster variables with the social capital variables discussed above. Doing so, we test whether access to social capital dampens or amplifies the impact of climate change on the odds of an individual migrating out of Nang Rong. Preliminary Results and Discussion We report the results of the full model, first summarizing the results of the control variables, before we move to the primary predictor variables. The results for all three models are summarized in Table 2.

• Table 2 About Here-

Consistent with other empirical work on migration, the odds of migrating are significantly associated with age, initially increasing, then decreasing after a peak age of

• migration. Married people are significantly less likely to migrate, as are women. Relative to

people with primary education or less, people who have completed some secondary education are less likely to migrate, whereas completing secondary education significantly increases one’s

12

• dds of migrating. Individuals living in families with large landholdings are significantly less

likely to migrate, while individuals in household’s that reported ever receiving remittances are more likely to migrate. The greater the asset index of a household, the lower the odds of

• migrating. If an individual lives in a village that is very remote, she is more likely to migrate out

relative to a village that is somewhat remote. Similar to other work that examines migration in Nang Rong,, we find that previous individual migration experience through time t-1 significantly increases the odds of an additional trip by 18%, whereas the cumulative number of months an individual spent out of Nang Rong does not significantly influence additional migrations. On the

• ther hand, if other household members have migration experience, the odds of an individual

decreases by 7.5%. Similar to cumulative months away for the individual, the number of months a household member spends outside of Nang Rong does not significantly impact the likelihood that the individual will migrate in time t. Finally, migrant experience at the village level, particularly the number of trips taken by fellow villagers, increases the odds of the individual migrating in time t by 53%. The impact of variable environmental conditions on out migration differs by NDVI

• cluster. Recall that the modal cluster, cluster 1, is the dominant environmental signal in Nang
• Rong. We examine the odds of migration when the pixel-year falls outside of this modal cluster,

suggesting perturbations (either positive or negative) to the typical environmental signal experienced by residents of Nang Rong. The odds of migrating when vegetation conditions fall under clusters 2 and 6 are slightly increased, relative to the modal cluster, but not significantly. The odds of migrating out of Nang Rong increase by 29%, relative to modal environmental conditions, when a pixel-year falls in cluster 7. On the other hand, odds of migrating, relative to cluster 1, decrease by 19% for cluster 8 conditions, and 40% in cluster 9.

13

Next, we interact the migrant social capital variables with the environmental variables in

• rder to better understand the role that social capital might play in dampening or amplifying

migration in response to deviations from normal environmental conditions. Table 3 provides the complete results of these six interaction models, including the main effects. Individual migration experience appears to most significantly influence the probability of a migrant response, conditional on the environment. When considering the number of previous trips a person took

• ut of Nang Rong, as well as the number of months spent away from Nang Rong, the positive

coefficient on the interaction term for clusters 2 and 8, suggests that in these two environmental scenarios, the more previous trips an individual took out of Nang Rong, and the more months spent away from Nang Rong in the past, the more likely an individual will engage in migration, relative to individuals with no migrant experience, providing an amplification effect. It is interesting to note that our interpretation of cluster 2 and 8 is quite different. In our discussion of the clusters in the section of NDVI above, we speculated that cluster 2 indicates a potentially greener year relative to the modal cluster category. Cluster 8, on the other hand, appears to be a drier year, with a later onset of rain than the modal category. Yet in both cases greater social capital translates into an amplification of the odds of migrating when interacted with the environment; this warrants additional investigation. For the interaction of clusters 6, 7, and 9 with social capital, there appears to be a dampening effect, with higher levels of individual migrant experience decreasing the probability

• f out migration in these clusters. In other words, the difference in the probability of out

migration between someone with less migration experience and someone with more migration experience is much smaller in these clusters. Figures 3 and Figure 4 below illustrate the probability of migration, conditional on cluster type and previous migration experience. For

14

cluster 6, individuals with zero months away through time t-1 have a higher probability of migrating than individuals with more experience away from Nang Rong. Preliminary Conclusions Our study set out to investigate, more fully, the role that migrant social capital plays in

• ut-migration under variable environmental conditions. We attempt to move beyond other

empirical studies that limit their analyses of social capital as a main driver of migration, net of

0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 Cluster 1 Cluster 2 Cluster 6 Cluster 7 Cluster 8 Cluster 9

Probability of Migration Axis Title

Figure 3: Predicted Probability of Out Migration Conditional on Environment and Individual Migration Trips

0 trips 1 Trip 2 Trips 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 Cluster 1 Cluster 2 Cluster 6 Cluster 7 Cluster 8 Cluster 9

Probability of Migration Figure 4: Predicted Probability of Out Migration Conditional on Environment and Individual Months out of Nang Rong

0 Months Away 24 Months Away 36 months Away

15

• ther factors, and instead argue that a more nuanced understanding of how established migration

drivers interact with the environment can tell us more about how people utilize social capital when faced with variability in environmental conditions. On the one hand, when environmental conditions worsen, access to dense social networks can ease the cost of migration, allowing migrants to more easily move to supplement income shocks. On the other hand, individuals with prior migration experience, either their own, or within their household, may accumulate knowledge or adaptive mechanisms that allow them to stay in place during an environmental

• shock. In our main model, the odds of migrating out of Nang Rang when vegetation conditions

fall under cluster 8, relative to the modal category, are decreased. However, when we consider the interaction with social capital, the more migrant experience one has, the higher the probability of out-migration. If we interpret cluster 8 as a drier than normal year, it may be that funding a new migrant trip is costly, making it more likely that those with prior experience will make a trip. We interpret cluster 2 as a healthier than normal year, and here, too, we see that increased social capital leads to higher probabilities of migration. Even where the interaction between clusters and social capital yields reduced odds of migration, the more individual migrant experience, the higher the probability of migration. While initially puzzling, these seemingly contradictory results suggest that in good years, households with access to social capital might rely on migration as a resilience strategy, bolstering income and assets that can be relied on in leaner times; in leaner years, households might rely on migration as a more direct survival

• strategy. The next step in our research strategy is to further validate NDVI clusters by obtaining

precipitation and temperature data to corroborate our interpretations of NDVI signals, by cluster. Access to land use/land cover data will also aid in our interpretation.

16

Works Cited Adger, W. Neil. 2003. "Social Capital, Collective Action, and Adaptation to Climate Change." Economic Geography 79(4):387- 404. Bardsley, DouglasK and GraemeJ Hugo. 2010. "Migration and Climate Change: Examining Thresholds of Change to Guide Effective Adaptation Decision-Making." Population and Environment 32(2-3):238-62. doi: 10.1007/s11111-010-0126- 9. Bebbington, Anthony. 1999. "Capitals and Capabilities: A Framework for Analyzing Peasant Viability, Rural Livelihoods and Poverty." World Development 27(12):2021-44. doi: https://doi.org/10.1016/S0305-750X(99)00104-7. Black, Richard, Stephen R.G. Bennett, Sandy M. Thomas and John R. Beddington. 2011. "Migration as Adaptation." Science 478:447-49. Black, R., D. Kniveton and K. Schmidt-Verkerk. 2011. "Migration and Climate Change: Towards an Integrated Assessment of Sensitivity." Environment and Planning A 43(2):431-50. Curran, Sara R. 2002. "Migration, Social Capital, and the Environment: Considering Migrant Selectivity and Networks in Relation to Coastal Ecosystems." Pp. 89-125 in Population and Environment: Methods of Analysis, edited by W. Lutz,

• A. Prskawetz and W. C. Anderson. New York: The Population Council.

Curran, Sara R., Filiz Garip, Chang Y. Chung and Kanchana Tangchonlatip. 2005. "Gendered Migrant Social Capital: Evidence from Thailand." Social Forces 84(1):225-55. Curran, Sara and Jacqueline Meijer-Irons. 2014. "Climate Variability, Land Ownership and Migration: Evidence from Thailand About Gender Impacts." Washington Journal of Environmental Law and Policy 4(1):37 -74 Curran, Sara R, Jacqueline Meijer-Irons and Filiz Garip. 2016. "Economic Shock and Migration." Sociology of Development 2(2):119-57. de Haan, Arjan. 1999. "Livelihoods and Poverty: The Role of Migration - a Critical Review of the Migration Literature." Journal

• f Development Studies 36(2):1 - 47.

de Haas, Hein. 2010. "Migration and Development: A Theoretical Perspective." International Migration Review 44(1):227-64. doi: 10.1111/j.1747-7379.2009.00804.x. Entwisle, Barbara, Nathalie E. Williams, Ashton M. Verdery, Ronald R. Rindfuss, Stephen J. Walsh, George P. Malanson, Peter

• J. Mucha, Brian G. Frizzelle, Philip M. McDaniel, Xiaozheng Yao, Benjamin W. Heumann, Pramote Prasartkul,

Yothin Sawangdee and Aree Jampaklay. 2016. "Climate Shocks and Migration: An Agent-Based Modeling Approach." Population and Environment 38(1):47-71. doi: 10.1007/s11111-016-0254-y. Fraley C, and Raftery AE. 1999. MCLUST: Software for Model-Based Cluster Analysis. Journal of Classification 16(2):297-306. Garip, Filiz and Sara Curran. 2009. "Increasing Migration, Diverging Communities: Changing Character of Migrant Streams in Rural Thailand." Population Research and Policy Review. Gray, Clark L. 2009. "Environment, Land, and Rural out-Migration in the Southern Ecuadorian Andes." World Development 37(2):457-68. Gray, Clark and Valerie Mueller. 2012. "Drought and Population Mobility in Rural Ethiopia." World Development 40(1):134-45. Gray, Clark L and Valerie Mueller. 2012. "Natural Disasters and Population Mobility in Bangladesh." Proceedings of the National Academy of Sciences 109(16):6000-05. Gray, Clark and Richard Bilsborrow. 2013. "Environmental Influences on Human Migration in Rural Ecuador." Demography 50(4):1217-41. Henry, Sabine, Bruno Schoumaker and Cris Beauchemin. 2004. "The Impact of Rainfall on the First out-Migration: A Multi- Level Event-History Analysis in Burkina Faso." Population & Environment 25(5):423-60. Kniveton, Dominic, Kerstin Schmidt-Verkerk, Christopher Smith and Richard Black. 2008. "Climate Change and Migration: Improving Methodologies to Estimate Flows." in Migration Research Series. Geneva: International Organization for Migration. Kubik, Zaneta and Mathilde Maurel. 2016. "Weather Shocks, Agricultural Production and Migration: Evidence from Tanzania." The Journal of Development Studies 52(5):665-80. doi: 10.1080/00220388.2015.1107049. Marks, Danny. 2011. "Climate Change and Thailand: Impact and Response." Contemporary Southeast Asia: A Journal of International and Strategic Affairs 33(2):229-58. Massey, Douglas S, William G Axinn and Dirgha J Ghimire. 2010. "Environmental Change and out-Migration: Evidence from Nepal." Population and Environment 32(2-3):109-36. McLeman, R. and B. Smit. 2006. "Migration as an Adaptation to Climate Change." Climatic Change 76(1):31-53. Nawrotzki, Raphael J, Fernando Riosmena and Lori M Hunter. 2013. "Do Rainfall Deficits Predict Us-Bound Migration from Rural Mexico? Evidence from the Mexican Census." Population Research and Policy Review 32(1):129-58. Nawrotzki, Raphael J., Fernando Riosmena, Lori M. Hunter and Daniel M. Runfola. 2015. "Amplification or Suppression: Social Networks and the Climate Change—Migration Association in Rural Mexico." Global environmental change : human and policy dimensions 35:463-74. doi: 10.1016/j.gloenvcha.2015.09.002. Rindfuss, Ronald R., Toshiko Kaneda, Arpita Chattopadhyay and Chanya Sethaput. 2002. "Panel Studies and Migration." Social Science Research 36(1):374-403. Stark, Oded and David E. Bloom. 1985. "The New Economics of Labor Migration." The American Economic Review 75(2):173- 78.

17

Stark, Oded and J. Edward Taylor. 1989. "Relative Deprivation and International Migration." Demography 26(1):1-14. doi: 10.2307/2061490. Vanwey, Leah K. 2003. "Land Ownership as a Determinant of Temporary Migration in Nang Rong, Thailand." European Journal of Population / Revue Européenne de Démographie 19(2):121-45. Walsh, Stephen J, Thomas W Crawford, William F Welsh and Kelley A Crews-Meyer. 2001. "A Multiscale Analysis of Lulc and Ndvi Variation in Nang Rong District, Northeast Thailand." Agriculture, Ecosystems & Environment 85(1):47-64. Walsh, Stephen J., Ronald R. Rindfuss, Pramote Prasartkul, Barbara Entwisle and Aphichat Chamratrithirong. 2005. "Population Change and Landscape Dynamics: The Nang Rong, Thailand, Studies." in Population, Land Use, and Environment: Research Directions edited by B. Entwisle and P. C. Stern. Washington, D.C.: National Academy of Sciences.

18

Table 1: Descriptive Statistics 1984 N = 3923 2000 N = 6459 mean s.d. mean s.d. Age 19.00 4.408 29.11 6.884 Married 0.251 0.433 0.710 0.454 Sex Female 0.483 0.500 0.512 0.500 Male 0.517 0.500 0.488 0.500 Education Primary Education or Less 0.738 0.440 0.654 0.476 Some Secondary Education 0.220 0.414 0.215 0.411 Completed Secondary Education 0.043 0.202 0.131 0.338 Temporary Migrant in 1984 0.092 0.289 0.045 0.206 Household Had a Temporary Migrant in 1984 0.192 0.394 0.140 0.347 Land Ownership No land 0.097 0.295 0.089 0.285 1 to 10 Rai 0.207 0.406 0.255 0.436 11 to 25 Rai 0.263 0.440 0.308 0.462 25 + Rai 0.433 0.496 0.347 0.476 Household Received Remittance 0.133 0.339 0.517 0.500 Asset Measure 0.722 1.416 4.135 1.906 Village Somewhat Remote 0.660 0.474 0.645 0.479 Village Very Remote 0.178 0.383 0.181 0.385 Village has Electricity 0.313 0.464 1.000 0.000 Number of individual trips through t -1 2.046 2.853 Number of months spent away through t-1 49.227 55.147 Number of trips taken by HH Members (excluding the individual ) 1.023 1.261 Number of months spent away by HH Members (excluding the individual) 26.865 29.568 Number of trips taken by members of the village 1.916 0.373 Number of months spent away by members of village 47.303 9.132

19

Table 2: Odds of Migrating out of Nang Rong – Results of logistic regression Base Base + MSC Full Model VARIABLES

• dds ratio
• dds ratio
• dds ratio

Age 1.196*** 1.118*** 1.116*** (0.0222) (0.0208) (0.0208) Age-squared 0.996*** 0.997*** 0.997*** (0.000364) (0.000365) (0.000366) Married 0.482*** 0.527*** 0.524*** (0.0225) (0.0224) (0.0223) Sex 1.426*** 1.322*** 1.320*** (0.0603) (0.0457) (0.0456) Primary Education or Less (referent) Some Secondary Education 0.807*** 0.862*** 0.859*** (0.0391) (0.0349) (0.0348) Completed Secondary Education 2.517*** 2.279*** 2.279*** (0.206) (0.163) (0.164) Temporary Migrant in 1984 1.139 0.946 0.946 (0.146) (0.0984) (0.0984) HH had temporary migrant in 1984 1.211 1.115 1.114 (0.0905) (0.0683) (0.0682) Land (owns 1 to 10 Rai referent) 0 Rai 1.001 1.013 1.010 (0.0695) (0.0609) (0.0608) 11 to 25 Rai 0.894 0.910 0.910 (0.0451) (0.0396) (0.0396) 25 + Rai 0.795*** 0.808*** 0.806*** (0.0400) (0.0347) (0.0347) Household Received Remittances 1.782*** 1.614*** 1.610*** (0.0797) (0.0678) (0.0677) Asset Measure 0.887*** 0.887*** 0.893*** (0.0104) (0.00947) (0.00960) Village Somewhat Remote 1.123 1.059 1.051 (0.0646) (0.0511) (0.0509) Village Very Remote 1.599*** 1.373*** 1.353*** (0.114) (0.0896) (0.0891) Village Electrified 1.120 1.070 1.041 (0.0747) (0.0659) (0.0646) Number of individual trips through t -1 1.179*** 1.179*** (0.0101) (0.0101) Number of months spent away through t- 1 1.001 1.001 (0.000833) (0.000835) Number of trips taken by HH Members (excluding the individual ) 0.924*** 0.925*** (0.0212) (0.0213)

20

Table 2: Odds of Migrating out of Nang Rong – Results of logistic regression Number of months spent away by HH Members (excluding the individual) 1.003 1.003 (0.00117) (0.00117) Number of trips taken by members of the village 1.450*** 1.530*** (0.141) (0.152) Number of months spent away by Household in the Village 1.000 1.001 (0.00394) (0.00398) Environmental Predictor Variables (Cluster 1 referent) Cluster 2 1.015 (0.0880) Cluster 6 1.099 (0.0945) Cluster 7 1.286*** (0.0687) Cluster 8 0.812* (0.0605) Cluster 9 0.594*** (0.0563) lnsig2u Constant 0.0194*** 0.0249*** 0.0221*** (0.00471) (0.00662) (0.00596) Observations 68,431 68,431 68,431 Number of id 7,594 7,594 7,594 rho 0.331 0.176 0.176 sigma_u 1.275 0.838 0.838 chi2_c 3050 354.4 353.5 seEform in parentheses *** p<0.001, ** p<0.005, * p<0.01

21

Table 3: Interaction Effects of Clusters*Migrant Social Capital Individual Trips Individual Months Away HH Trips HH Months Away Village Trips Village Months Away VARIABLES

• dds ratio
• dds ratio
• dds ratio
• dds ratio
• dds ratio
• dds ratio

Age 1.075*** 1.104*** 1.115*** 1.117*** 1.113*** 1.114*** (0.0199) (0.0205) (0.0208) (0.0208) (0.0207) (0.0207) Age-squared 0.998*** 0.997*** 0.997*** 0.997*** 0.997*** 0.997*** (0.000365) (0.000366) (0.000366) (0.000367) (0.000366) (0.000366) Married 0.531*** 0.524*** 0.525*** 0.525*** 0.522*** 0.522*** (0.0223) (0.0223) (0.0224) (0.0224) (0.0222) (0.0223) Sex 1.303*** 1.315*** 1.321*** 1.321*** 1.315*** 1.317*** (0.0436) (0.0453) (0.0455) (0.0457) (0.0453) (0.0454) Primary Education or Less (referent) Some Secondary Education 0.865*** 0.853*** 0.859*** 0.859*** 0.861*** 0.860*** (0.0341) (0.0345) (0.0347) (0.0348) (0.0348) (0.0347) Complted Secondary Education 2.195*** 2.232*** 2.277*** 2.286*** 2.272*** 2.255*** (0.154) (0.160) (0.163) (0.164) (0.163) (0.162) Temporary Migrant in 1984 0.861 0.916 0.947 0.947 0.921 0.925 (0.0870) (0.0952) (0.0983) (0.0985) (0.0955) (0.0961) HH had temporary migrant in 1984 1.110 1.118 1.096 1.103 1.107 1.110 (0.0658) (0.0683) (0.0672) (0.0678) (0.0675) (0.0678) Land (owns 1 to 10 Rai referent) 0 Rai 1.005 1.012 1.013 1.011 1.022 1.016 (0.0592) (0.0608) (0.0609) (0.0609) (0.0614) (0.0611) 11 to 25 Rai 0.916 0.915 0.913 0.912 0.913 0.912 (0.0389) (0.0397) (0.0396) (0.0397) (0.0396) (0.0396) 25 + Rai 0.815*** 0.814*** 0.808*** 0.809*** 0.815*** 0.812*** (0.0342) (0.0350) (0.0347) (0.0348) (0.0350) (0.0349) Household Received Remittances 1.599*** 1.614*** 1.610*** 1.624*** 1.592*** 1.611*** (0.0661) (0.0678) (0.0677) (0.0684) (0.0668) (0.0677) Asset Measure 0.893*** 0.892*** 0.893*** 0.893*** 0.889*** 0.892*** (0.00943) (0.00958) (0.00959) (0.00961) (0.00957) (0.00958) Village Somewhat Remote 1.057 1.054 1.052 1.050 1.067 1.024 (0.0497) (0.0509) (0.0508) (0.0509) (0.0516) (0.0496) Village Very Remote 1.368*** 1.366*** 1.357*** 1.356*** 1.385*** 1.303*** (0.0875) (0.0898) (0.0891) (0.0893) (0.0910) (0.0860)

22

Village Electrified 1.035 1.039 1.044 1.043 1.048 1.049 (0.0639) (0.0646) (0.0648) (0.0648) (0.0654) (0.0653) Number of individual trips through t- 1 1.232*** 1.178*** 1.180*** 1.179*** 1.182*** 1.181*** (0.0116) (0.0100) (0.0101) (0.0101) (0.0101) (0.0101) Number of months spent away through t- 1 1.001 1.003* 1.001 1.001 1.001 1.001 (0.000811) (0.000935) (0.000833) (0.000835) (0.000831) (0.000832) Number of trips taken by HH members (excluding the individual) through t -1 0.937** 0.927** 0.961 0.924*** 0.927** 0.923*** (0.0213) (0.0213) (0.0254) (0.0213) (0.0213) (0.0212) Number of months spent away by HH members (excluding the individual) through t-1 1.003 1.003 1.003 1.004** 1.003 1.003 (0.00115) (0.00117) (0.00117) (0.00133) (0.00117) (0.00117) Number of trips taken by villagers through t-1 1.543*** 1.528*** 1.537*** 1.528*** 2.334*** 1.356** (0.149) (0.151) (0.152) (0.152) (0.262) (0.136) Number of months spent away by villagers through t-1 0.999 1.001 1.001 1.001 0.994 1.013** (0.00388) (0.00396) (0.00397) (0.00397) (0.00401) (0.00432) Environmental Predictor Variables Cluster 2 0.821 0.886 0.989 1.003 0.487 0.754 (0.0795) (0.0827) (0.0999) (0.0990) (0.175) (0.167) Cluster 6 1.710*** 1.606*** 1.376** 1.307 21.90*** 11.57*** (0.170) (0.166) (0.145) (0.141) (6.888) (3.157) Cluster 7 1.451*** 1.318*** 1.383*** 1.439*** 1.564 1.892*** (0.0877) (0.0810) (0.0891) (0.0925) (0.357) (0.332) Cluster 8 0.681*** 0.715*** 0.764** 0.781** 0.634 0.611 (0.0561) (0.0579) (0.0648) (0.0660) (0.175) (0.145) Cluster 9 0.929 0.658*** 0.626*** 0.517*** 3.432*** 1.984 (0.100) (0.0778) (0.0699) (0.0618) (1.182) (0.712)

23

Interaction Effects of Clusters*Migrant Social Capital Individual Trips Individual Months Away HH Trips HH Months Away Village Trips Village Months Away Cluster 2 * MSC 1.583*** 1.056*** 1.143 1.004 5.417 1.045 (0.126) (0.0117) (0.289) (0.0151) (4.196) (0.0288) Cluster 6 * MSC 0.871*** 0.987*** 0.773*** 0.992 0.175*** 0.942*** (0.0154) (0.00234) (0.0573) (0.00328) (0.0309) (0.00628) Cluster 7 * MSC 0.947*** 0.999 0.898 0.992** 0.865 0.987 (0.0131) (0.00162) (0.0495) (0.00255) (0.150) (0.00580) Cluster 8 * MSC 1.269*** 1.024*** 1.239 1.008 1.588 1.028 (0.0521) (0.00507) (0.157) (0.00736) (0.713) (0.0198) Cluster 9 * MSC 0.884*** 0.997 0.944 1.005 0.371*** 0.971*** (0.0148) (0.00194) (0.0534) (0.00271) (0.0675) (0.00767) Constant 0.0325*** 0.0246*** 0.0219*** 0.0217*** 0.0150*** 0.0181*** (0.00868) (0.00664) (0.00591) (0.00586) (0.00416) (0.00493) Observations 68,431 68,431 68,431 68,431 68,431 68,431 Number of id 7,594 7,594 7,594 7,594 7,594 7,594 chi2 2783 2584 2512 2500 2625 53 rho 0.154 0.174 0.174 0.176 0.172 9.011 sigma_u 0.773 0.832 0.832 0.838 0.828 31 chi2_c 303.9 356.6 349.3 353.2 348.2

• 22377

seEform in parentheses *** p<0.001, ** p<0.005, * p<0.01

24 Figure 1: Village Locations and NDVI Pixexls

25 Figure 2: Annual NDVI patterns, by Cluster

26 Figure 2: Annual NDVI for One Water Year in Nang Rong