Although migration of Muslims from the southernmost provinces of Thailand to Malaysia has a long history, research suggests that the intensity of this migration has increased in the past 10 years along with increased unrest in the provinces. This study examines how migration in the three southernmost provinces is affected by the ongoing unrest. Data are drawn from household probability surveys conducted in 2014 and 2016. An individual sample of 3,467 persons who were household residents at the 2014 survey was followed to see who remained in the household of origin or moved out two years later (2016 survey). Data on violent events from the Deep South Watch, an independent organization, were used to measure exposure to violence. Results from a multilevel analysis show that net of other characteristics at the individual, household, and village levels, individuals who live in a village in which a violent event occurred in the previous year are more likely to move out than those who live in a village with no violent event in the previous year. Findings suggest that in addition to the economic reasons that have long motivated migration from this area, violent events accelerate this migration.
Thailand has been known as both a receiving and sending setting for migration. The country has hosted almost 4 million labor migrants from its neighboring countries in the Mekong subregion, including Myanmar, Lao PDR, and Cambodia (International Labour Organization 2018). At the same time, although Thailand is not a major country for sending labor migrants abroad, such as the Philippines or Indonesia, a number of Thais of working age move each year to work in other countries, often to Taiwan and South Korea. Most studies on migration in Thailand have focused on internal or international Thai migrants from the northeast, a relatively poor region of the country. Less is known about international migration from other parts of the country, particularly in the southernmost provinces, where young people have commonly moved to work in unskilled jobs in Malaysia for several decades. The southernmost provinces of Patani, Yala, and Narathiwat, also known as the Deep South, are predominantly occupied by Thai Muslims. The difference from other areas of Thailand is that while people in other areas usually migrate to work in big cities of Thailand, especially Bangkok, people in the southernmost provinces often cross the border to work in Malaysia, the more economically developed neighboring country. Many of them, both men and women, have worked in Malaysia at least once (Tsuneda 2006). The historically close relationship between Thai Muslims in the southernmost provinces and the Malaysians sharing the same border—the Islamic religion—as well as the Malay culture has underpinned the movement from Thailand to Malaysia.
Although migration of Muslims from the southernmost provinces of Thailand to Malaysia has long been a phenomenon, past research suggested that the intensity has increased in the past 10 years (Nuansanong et al. 2009). The southernmost provinces, once an independent state, became part of Thailand in the late nineteenth century. Since then, people have experienced times of relative tension and cooperation with the Thai government (Chalk 2008; Croissant 2005; Forbes 1988). The tension has been related to assimilation to the Thai culture, autonomy and governance, development, and other issues. Consequently, in addition to the economic promise of a better income, another underlying reason for increased migration may be escaping the ongoing unrest. The violence began escalating in 2004; since then, nearly 6,000 people have died, and about 10,000 have been injured.
Although most research on migration in the Deep South of Thailand has documented migration as a response to an economic force (e.g., Kemkunasai and Pinsuwan 2009; Klanarong et al. 2009) and most is voluntary, the unrest may serve as another push factor in the case of southernmost Thailand. Indeed, focus groups earlier in our study suggested that the Deep South’s ongoing chaos that daily impends people’s well-being and threatens them with a sense of insecurity and anxiety may drive them to move out. In many other areas, long-term instability and violence often precipitate migration (Bohra-Mishra and Massey 2011; Czaika and Kis-Katos- 2009; Lozano-Garcia et al. 2010; Williams et al. 2012). Research directly addressing the relationship between migration and the conflict in the southernmost provinces of Thailand is scarce. A recent cross-sectional study found that migration was positively associated with violent events at the village level (Jampaklay et al. 2017). However, this study was cross-sectional, making causal inferences tentative. The study also had a measure of violence reported by the survey respondents, which may not measure the full extent of the violence. Therefore, whether the experience of violence has affected movement out of the place of origin at the individual level is still unclear. Understanding migration of Thai Muslims in these three southern border provinces (Patani, Yala, and Narathiwat) and its association with the unrest remains inconclusive.
As a common way of life that has been occurring for several decades, migration to Malaysia of Thai Muslims in this area would occur even in the absence of unrest. However, several potential explanations underlie the relationship between the unrest and migration to Malaysia. The stress resulting from the ongoing unrest may drive women and men to seek a safer environment. The reduced economic development with an associated decrease in job opportunities due to the unrest may also be a driving force for moving away. Thus, the need to migrate for economic well-being has increased.
In contrast, because some insurgents travel over the Malaysian border to avoid the authorities, migrants may be suspected by the authorities or by other villagers to be linked to the unrest. The travel of economic migrants, particularly young males, to Malaysia can arouse suspicion of involvement in the insurgency. This situation may discourage villagers from migrating: young men may have difficulties in crossing check points because of heightened scrutiny based on age and gender. The young migrants working in Malaysia and their families may also be watched and investigated by the authorities. Thus, the unrest could actually deter migration (Jampaklay et al. 2011). This possible factor was reported to us in interviews with key informants at the beginning of our study.
In the literature, the migration theory often highlights economic perspectives concerning three questions: (1) why migrate, (2) who migrates, and (3) what are the consequences that migration brings about for place of origin and destination (Bodvarsson and Van den Berg 2013)? In the mainstream economic theory of migration, the decision to relocate is an investment in one’s well-being. Economic theory also suggests that assets help finance out-migration and raise the likelihood of migration (Massey et al. 1998). The new economics of migration sees migration as a response to missing or imperfect capital markets, futures, credit, and insurance rather than differences in wages (Stark 1991). This new model also views migration as a household decision rather than an individual decision. Households send migrants to diverse locations to minimize risk, accumulate capital, and overcome credit constraints (Stark 1991).
Although this approach is useful in understanding purely economic migration, it is not so useful in understanding non–economically oriented migration, including that of refugees, family members accompanying other migrant members, or those forced to migrate against their will. From a sociological point of view, in contrast, explanations of why people migrate have tended to focus on a broader range of determinants and have taken important economic bases as a secondary emphasis. The human capital perspective implies that migration decisions depend upon origin-destination differences in the returns to factor supply, controlling for migration costs, skill levels, income inequality, and migration policies. Other models focus on how migration decisions are influenced by other non-economic factors, such as family considerations.
This analysis is an attempt to answer the question of how difficulties resulting from the ongoing unrest impact individual migration, in addition to economic and other factors usually considered in migration research. We use a multilevel analysis of a longitudinal data set, a household probability sample of the population, and an objective measure of violence at the village level to determine whether migration is affected by the unrest. This study involves not only southern Thailand but also international labor migration between Southeast Asian countries, which is of interest to members of multiple social science disciplines.
Earlier research on migration and conflict has looked at three main factors: (1) exposure to violence (e.g., Bohra-Mishra and Massey 2011; Czaika and Kis-Katos 2009; Lozano-Garcia et al. 2010; Williams et al. 2012); (2) traditional economic factors, including education, assets, and homeownership (e.g., Bodvarsson and Van den Berg 2013; Massey et al. 1998; Stark 1991); and (3) social networks developed through previous migration (Soe et al. 2011). As depicted in Fig. 1, we hypothesize that violence as well as economic stress in the community affect migration. The effect of violence may be mediated by the household’s socioeconomic status and social networks.
Data Set and Study Sample
Our analysis employs two rounds of survey data from a longitudinal study on migration and the unrest in the three southernmost Thai provinces (2014–2016). The overall objective of our study is to understand how migration in these provinces is associated with the ongoing unrest, using the concepts of gender, migration, and the effects of the conflict. The survey was designed to capture a representative sample of the population by using the proportional probability to size (PPS) sampling technique. Thirty villages across the three provinces are included in the survey. For more details of the project, see Jampaklay et al. (2017). Round 1 of data collection was completed in May 2014, and a follow-up survey of the same households was completed by the end of 2016. Our analysis includes individuals of working ages: 15–60. For the baseline round, 1,102 households were interviewed, covering 5,823 individuals of all ages listed in household rosters and 3,888 individuals aged 15–60. The follow-up survey successfully reinterviewed about 94% of the households included in the first round and 3,645 of the individuals aged 15–60. Excluding cases of incomplete variables in the analysis yields a final analytical sample of 3,467 individuals.
Measurement of Key Variables
The analysis focuses on how violence due to the unrest affects the likelihood of individual migration out of the village. The dependent variable measures whether individuals who were usual residents at the baseline survey remained in the village of origin (coded as 0) or moved out to other places (coded as 1) at the follow-up survey, two years after the baseline. This information was reported by household respondents in both surveys.
The key predictor is the unrest measured by whether any violent incidents occurred in the village during the year prior to the baseline survey. These data were derived from a database, available upon request, from an independent organization called the Deep South Watch. This organization records all incidents due to the unrest in the three southernmost provinces by village (https://www.deepsouthwatch.org). The incidents recorded by the Deep South Watch refer to violent events due only to unrest; thus, violent events not related to the unrest are not included in this measure. The first measure is a dummy variable for whether a violent event ever occurred (yes = 1, and no = 0).1 The second measure is categorized into three groups reflecting the number of events: (1) no event, (2) low/medium if there were one to three incidents, and (3) high if more than three incidents occurred in the 12 months prior to the baseline survey.
Other independent variables were measured at the village, household, and individual levels. At the village level, the analysis includes whether an individual lives in a household located in an urban (1) or a rural area (0). Urban residence refers to municipal areas, and rural residence includes areas outside the municipal territory. An additional variable—village economic status—measures economic status at the village level. Because the economic situation amid the unrest may deteriorate over time, poverty may act as a push factor for people to move out. The village economic status is measured using the number of households in the study sample considered to be rich. This measure is based on the wealth index.
At the household level, the analysis includes household socioeconomic status, using a household wealth index and education of household head as a proxy. The household wealth index was created using a principle component analysis, which provides an index of assets to serve as a proxy for wealth (Filmer and Pritchett 2001). The following asset items were included: bed, stove, microwave, electric pot, refrigerator, washing machine, computer, tablet, car, pick-up, and CD player. Using the wealth index score, our analysis ranks individuals into five quintiles and categorizes them into three groups of household economic status. We classify the two bottom quintiles as poor (1), the third and the fourth quintiles as moderate (2), and the fifth quintile at the top as rich (3). Another proxy for household socioeconomic status, education of the household head, has five categories: (1) none or less than primary level, (2) completed primary level, (3) lower secondary, (4) upper secondary, and (5) higher than secondary level.
Individual-level characteristics are included to control for individual determinants of migration decision-making. These include gender (male = 1, female = 0), age (interval scale ranges from 15 to 60), marital status (1 = single, 2 = currently married, 3 = previously married), relationship with household head (1 = household head, 2 = child of household head, 3 = spouse of household head, and 4 = other relationship), education (classified in the same way as education of household head), work status (1 = work only, 2 = work and study, 3 = neither work nor study, and 4 = study only), and health status (1 = very good, 2 = good, and 3 = moderate/bad/very bad). We also add Islamic education because individuals in this area commonly have a combination of secular education for a future profession and Islamic education for their religious practice. Islamic education is classified as 1 = none/nonformal; 2 = preschool, or Tadika; 3 = primary, or Ibtida-i; 4 = lower secondary, or Mutawassid; 5 = upper secondary, or Sunnaweeyah; or 6 = attended a traditional system of Islamic study called Pondok or Hafiz (memorizing Al-Quran verses). Last, prior migration experience of the individual was included to capture the individual’s social network, documented as a crucial factor for migration decision-making in migration literature (e.g., Soe et al. 2011). The variable is coded as 1 if an individual ever moved before Round 1 and 0 if not.
The logit model is used because the dependent variable is dichotomous. Our independent variables and covariates are multilevel (individual, household, and village levels). In addition, one household may have several individuals aged 15–60. Therefore, our individual observations, as unit of analysis embedded in households and village, are clustered, and correlations among observations may occur. To address the clustered data, we use a random-effect model with a random slope. Both random slopes for household and village were tested for significance. First, a random slope for household was tested, and the likelihood ratio model showed that the model was a significant improvement over basic logistic regression. When a random slope for village was added to the model, the likelihood ratio test showed that the slope was not a significant improvement over the first model. Therefore, results from the multilevel model that included a random slope for household are shown in the tables.
Among the 3,467 individuals aged 15–60 recorded as usual residents at the baseline survey (2014), about one-fifth (19.4%) had moved out of the household two years later. Table 1 describes characteristics of the study sample, measured at the baseline survey. About 46% of the sample are male. The mean age is about 34 years old. More than one-half (59%) are married, and more than one-third (36%) are single. Almost one-fourth are household heads, one-fifth are spouses of household heads, and 44% are children of household heads. About two-fifths of the study sample have secular education at the primary level. About one-fifth have lower education, one-fifth have upper secondary education, and 15% obtained higher than secondary education. Individuals with no secular education account for 5% of the sample. As for Islamic education, the highest proportion are those with lower secondary education (23%), and those with a higher level of education account for 19% of the sample. Individuals who attended a Pondok (traditional system of studying Islam in Malay culture) or Hafiz (memorizing the Al-Quran) make up about one-fifth of the total sample. Those without Islamic education account for 12%. About two-thirds of the study sample work full-time, and almost one-fifth (18%) are full-time students. Those who were reported as neither working nor studying account for 12%. Data on health status, reported by the household respondent, indicate that most of the study sample are perceived as healthy (i.e., 63% as good and 26% as very good), and only 12% were reported as having moderate, bad, or very bad health status. As for migration experience, 13% were reported as having ever moved prior to the first survey.
Several household characteristics are included in the model. A first measure of household economic status is the wealth index, classified into poor (the bottom 20%), moderate (about the middle 40%), and the rich (about the top 20%). The second measure of household economic status is the educational level of the household head. The data show that 60% of individuals are members of households headed by those with primary education; 13%, with lower secondary education; and nearly the same (9%) with upper secondary or higher education. The percentage of individuals from households headed by individuals with no formal education is substantial (15%).
At the village level, three characteristics are included: (1) area of residence; (2) whether the village experienced a violent incident in the previous year, our main independent variable; and (3) economic status. Data indicate that 15% of individuals live in an urban residence; thus, the majority (85%) live in a rural residence. The majority of the study sample live in villages that experienced a violent incident in the last year (78%). In addition, 59% reported one to three incidents, and 19% reported more than three incidents in the previous year. Data on the economic status at village level indicate that the majority of the study sample (79%) live in villages with no rich households.
Table 2 shows results from cross-tabulations of each characteristic measured at the baseline survey by whether a given individual moved out (migrant) or remained at the household (resident) two years later. Although the prevalence of migrating in the next two years after the first round of the survey for females and for males is not statistically different, it is evident that the migration rate was higher for younger adults than for older adults (32% for ages 15–24, 22% for ages 25–39, and 6% for ages 40–60). Single adults migrated more than twice as often as married or previously married adults. Household members who are household heads or spouses of household heads were much less likely to move. Results also show that individuals with more secular education (secondary level or higher) moved more often than less-educated individuals. Similar to secular education, individuals with higher Islamic education (secondary level or higher) also moved out in a higher proportion than individuals with lower Islamic education. With regard to work status, individuals who enrolled in school were most likely to move (32%), reflecting migration to obtain education rather than to work. We do not find a significant difference in the proportion moving between those with and without migration experience. We also do not find any relationship between individuals’ migration in the second round and household socioeconomic status (wealth index and household head’s education) and household residence (urban or rural).
Finally, results from the bivariate analysis indicate a significant relationship between the occurrence of a violent incident and an individual’s migration. Individuals living in a village in which a violent incident was reported in the year prior to the first survey round had a higher proportion moving out two years later (21% vs. 12%). A significant relationship between the unrest incident and migration is also found when the unrest incident is measured as none, low/medium, and high (13% vs. 22% and 20%, respectively). More persons living in villages without rich households moved out (20%) compared with those living in villages with at least one rich household (17%).
Tables 3 and 4 show the results from the multilevel model of the impact of the unrest, measured at one year prior to the baseline survey (the survey Round 1), on the likelihood of moving out of the village by the survey Round 2, controlling for selected individual, household, and village characteristics measured at the baseline survey. The dependent variable is whether a given person had moved out of the village of origin by the follow-up survey, two years after the baseline survey (1 = yes/migrant, 0 = no/resident).
Table 3 shows three models where the main predictor—the unrest—is measured as whether an incident occurred in the village in the previous year. Model 1 has only village-level variables: the occurrence of a violent event in the past year prior to the baseline survey, which is our main variable of interest; area of residence; urban or rural; and economic status. Results show that two village-level variables are significant, including the violent event. Residents who lived in a village with a violent event in the prior year were about two times more likely to have moved by the follow-up survey. Regardless of the unrest, those who lived in urban areas were less likely to move out. Village-level economic status is not significant in Model 1. The push impact of the unrest event on migration remains significant across three models when household- and individual-level characteristics are added to the model. The same is also true for the negative impact of urban residence on migration.
Model 2 controls for household characteristics, including the household wealth index and the education of the household head. If the household head had an upper secondary education, migration was less likely to occur. After we take household characteristics into account, both the positive impact of a violent incident and the negative impact of urban residence on migration remain significant. After we control for household-level variables in Model 2, village economic status becomes significant at the .001 level.
The last model, Model 3, includes all other individual characteristics, in addition to the village and household profiles. Again, we find that the unrest increases the likelihood of migration, and living in an urban residence decreases the likelihood of migration. In contrast to Model 2, after individual characteristics are added, household economic status appears negative and significant, indicating that individuals from rich households are less likely to migrate. Also, the negative impact of the household head’s education (upper secondary level) on migration of individuals is no longer significant. The full model also indicates that village economic status, significant in Model 2, is not significant after the model includes individual characteristics.
Results also confirm the bivariate analysis that gender is not related to the likelihood of migration in this context. Males and females move out at a similar rate. Each year of age reduces the likelihood of migration about 2%. Children of household heads or members with other relationships with household heads are about eight times more likely to move than the household head him/herself. Surprisingly, we do not find any significant impact of education on migration, neither secular nor Islamic. Work/study status also shows no significant effect. In contrast with the bivariate results, we find that currently married individuals are more likely to move than singles, when other characteristics are controlled. In addition, less-healthy individuals are less likely to move out. Finally, previous migration has a significant effect on a subsequent move two years later. Individuals who ever moved before the baseline survey were 1.8 times more likely to move again by the follow-up round than those with no migration experience. This migration experience variable shows no significant association in the bivariate analysis.
Table 4 shows the results of the three models shown in Table 3, with a three-category measure of exposure to violence: none, low/medium (one to three incidents), and high (more than three incidents). This measure is used as the main independent variable instead of a dummy variable indicating whether a violent incident occurred in the village in the past year. Results strengthen the results presented in Model 3 of Table 3. Living in the village with a low/medium or high level of incidents increases the likelihood of migrating by two times and almost four times, respectively. Results of most of the other variables are similar to the models in Table 3.
Interestingly, the results of all models in Table 4 indicate that individuals who live in villages with more rich households are more likely to migrate out. At the household level, living in a poor household pushes individuals to seek for work elsewhere; but at the village level, living in a rich village motivates individuals to leave their place of origin and look for opportunities away from home. These results are considered in the Discussion and Conclusion section.
We also tested the interactions between the unrest variable and a few independent variables. The interaction between the unrest and household wealth index was not statistically significant, nor was the interaction between unrest and village-level economic status (results not shown).
Discussion and Conclusion
This analysis uses a multilevel analysis to adjust for clustered data within households and includes multilevel measures of the independent variables. The longitudinal data of two survey rounds help to address the causal relationship between the unrest and migration. The results of this study confirm results from previous research in other areas that exposure to violent events can accelerate migration out of the area of residence. Findings suggest that although socioeconomic conditions may be related to migration, unrest has a direct positive impact on migration. The longitudinal results presented here are also consistent with an earlier cross-sectional analysis of migration from these southernmost Thai provinces (Jampaklay et al. 2017). If a violent event occurred in a village in the preceding year, both at low/medium and high levels, the likelihood of migration doubled, even when many socioeconomic and demographic variables were controlled for. The results, thus, support our hypothesis that violence as well as economic stress in the community affect migration. Findings are also consistent with our hypothesis that the effect of violence may be mediated by household-level socioeconomic status and social networks. As discussed later, both household- and village-level economic prospects affect the likelihood of migration and simultaneously strengthen the effects of violence.
Issues concerning security and protection as well as on plans for development need to be addressed side by side in the southernmost provinces of Thailand. The suffering of migrants and their families at home should also be considered. Future research should focus on the severity of the danger and the cost of non-economically motivated migration. Conventional theory on the costs and benefits of migration needs to be elaborated to include both economic and non-economic factors (e.g., Bodvarsson and Van den Berg 2013).
It is also important to note that urban people are less likely to migrate. Perhaps urban people are more likely to have wage income than agricultural income, and rural residents may migrate to supplement agricultural (often rubber) income. Consistent with this possibility, we find that individuals from rich households have a lower likelihood of migrating, which seems to indicate that moving out is related to fewer opportunities at the place of origin. However, the results for education may suggest otherwise: we do not find any significant effect of education, whether secular nor Islamic.
Different impacts of economic status at different levels on migration suggest that a complex explanation is needed. Poor household-level economic status may act as a push factor for individuals’ labor migration. At village level, however, high economic status plays a positive role for individuals to move out for work. Rich households in the same village may motivate other individuals to earn higher income to improve the financial well-being of their own household. This is, in fact, consistent with the relative deprivation hypothesis put forth by Oded Stark and colleagues (Stark and Taylor 1989) and supported by several previous studies (e.g., Bhandari 2004). The hypothesis posits that migration decisions are influenced not only by absolute income but also by relative income consideration. Evidence from Mexico, for example, suggests that relatively deprived households are more likely to engage in international migration than are households more favorably situated in their village’s income distribution, when absolute income is controlled for (Stark and Taylor 1991). The social networks associated with higher-income households may also facilitate migration. These higher-income households may have more assets because of the migration of household members.
We find some support for the selectivity of migrants in this area. Good health is positively associated with migration. In addition, our study also finds that individuals with previous migration experience are more likely to move. This may reflect the role of social networks in reducing the cost of migration.
We are aware that some variables might be causing both an increased propensity for migration as well as violence in a specific area. Our study of the violence data indicates that violence was related to household wealth, suggesting that violence occurs in more affluent areas. Although we should be cautious about causality, we adjust for at least some of this by including household wealth, village wealth, and education in our models.
We note some limitations of the analysis. At this stage, the analysis has not yet taken into account the fact that effects of the unrest may vary depending on whether migration is domestic or international. Nor have we distinguished reasons for migration, with which unrest and other factors may have different relationships. These limitations should be addressed in further investigations.
Migration is often a response to economic hardship or opportunities. However, migration may also be a response to the unrest. Household decisions may be based on safety, mental health, and other factors. Policies to provide welfare benefits for the family left behind should be considered because households may not benefit financially from this non-economic migration. Welfare interventions may be needed in conflict areas with substantial migration.
In conclusion, this study adds to the scant literature on migration in the context of unrest—particularly at the individual level, where migration is generally considered as a response to economic difficulties. The findings show the impact of exposure to violence on migration while controlling for selected traditional economic factors at several levels, including area of residence, education, assets, and social networks developed through previous migration. With a longitudinal data set, the causal link can be more firmly established than with cross-sectional data. It is evident in our study that the conflict leads to more migration, even when some measures of household-level socioeconomic status and social networks are taken into account.
The population survey data was collected through a grant from Mahidol University. The analysis of the data was partially funded by grants from the Thai Studies Grant Program at the Center for Southeast Asian Studies of the University of Michigan. This grant program is made possible by the Amnuay-Samonsri Viravan Endowment at the University of Michigan. The authors would also like to acknowledge the Deep South Watch (www.deepsouthwatch.org), which provided data on the unrest. We also thank all reviewers of the manuscript for their thoughtful comments and suggestions.
We estimated the models using the number of violent events in the previous year as an independent variable, but the results showed no significant effects, perhaps because of the skewness of the variable to the right. The number of incidents ranges from 0 to 19, with a mean of 2.9, a median of 1.5, and a mode of 1. Results from models using the dummy variable and a three-category variable as the main independent variable are presented.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.