Modelling and forecasting the spatial and temporal patterns of bilateral international migration flows

1.导论现有的关于移民流动的数据不完整/不可比较，这限制了对各种类型的移民及其影响的理解(Abel&Sanders，2014；Willekens等人，2016；Wisniowski，2017)。这一限制源于这样一个事实，即移徙没有唯一的定义，因此衡量不力(BilsBorrow等人，1997：15-28)。其他衡量问题，包括人口覆盖率、系统性偏差和缺乏准确性(Wisniowski等人，2013：585；迪士尼等人，2016：11)。由于这些问题，需要有一个一致的框架来估计双边移徙流量，以便能够作出可靠的预测。建议：制定一套统计模型，以估计和预测双边国际移徙流动的空间和时间模式。这些模型允许按迁移原因细分移民(De Beer，2008：292-302)。近年来，在制定衡量和估计国际移民流动的方法方面做出了重大努力。还开发了预测移民的方法(Bijak，2010；Bijak&Wisniowski，2010；Wisniowski等人，2015；迪斯尼等人，2015：29)。2.范围和数据这项研究将侧重于南美国家包括独立和附属领土的数据。将使用1990-2017年的数据，包括人口普查、国际调查、全国住户调查、政府移徙办公室。根据来源地/地区的不同，可以获得关于移民人口和流动的数据。这些数据将按可获得这些数据的国家的迁移理由进行分类。数据稀疏和缺失的问题可以通过扩展/应用下一节中描述的统计方法来缓解。3.建议的方法南美国家的数据要稀疏得多，而且由于是派生的，可能质量较低(例如)来自调查。这就需要一个定制的统计模型，能够结合来自各种数据来源的稀疏数据，这些数据原则上不以衡量移民为目标。推广Raymer等人的S模型(2013年：803)，来自源k的多个迁移数据流的表可以组合在校正数据不足的测量模型中。我可以为时间t的真实(未观察到的)移民流动指定一个模型，以允许对人口、社会和经济因素对流动的作用进行参数估计(Abel，2010：802；Raymer等人，2013：817)。这种方法通常基于重力型模型(Sen&Smith，2012：49-53)，该模型假设流量与发送国和接收国的人口规模成正比，与发送国和接收国之间的距离成反比，参数可估计。在时间t的真实(未观察到的)迁移流的后验分布将代表用作模型输出的最终合成数据，该模型用于估计具有相关不确定性的迁移流。通过包括特定于时间的参数，可以生成对未来流量的预测。引入特定协变量可以将模型扩展到估计特定类型的移徙。在只有移徙人口数据而不是流动数据的情况下，可以使用Abel(2013：508-524)和Abel&Sander‘s(2014：1521)方法。该模型需要移民股票数据作为输入，可以使用最大似然法进行估计。在拟议的项目中，我将考虑到对Dennet(2016)的批评来改进模型，并应用贝叶斯推理为流量估计提供不确定性度量，这些度量将被输入到所描述的流量模型中。

1.IntroductionThe existing data on migration flows are incomplete/incomparable which limits the understanding of various types of migration and its impact (Abel &Sanders,2014; Willekens et.al.,2016; Wisniowski,2017). This limitation stems from the fact that migration does not have a unique definition and is, consequently, poorly measured (Bilsborrow et.al.,1997:15-28). Other measurement problems, including population coverage, systematic bias and lack of accuracy (Wisniowski et.al.,2013:585; Disney et.al.,2016:11). Due to these issues, there is the need for a consistent framework for estimating bilateral migration flows enabling the production of reliable forecasting. The proposal: to develop a set of statistical models for estimating and forecasting the spatial and temporal patterns of bilateral international migration flows. These models allow for the breakdown of migration by reasons for moving (De Beer,2008:292-302).In recent years, there have been significant efforts in developing methods for measuring and estimating international migration flows. Methods have also been developed to forecasting migration (Bijak,2010; Bijak &Wisniowski,2010; Wisniowski et.al.,2015; Disney et.al.,2015:29).2.Scope and dataThis study will focus on data from South American countries including independent and dependent territories. 1990-2017 data will be used, including censuses, international surveys, national household surveys, governmental offices of migration. Depending on the source/territory, data on migrant stocks and flows are available. These data will be disaggregated by reason for moving for countries where such data are available. This issue of sparse and missing data can be mitigated by extending/applying statistical methods described in the next section.3.Proposed methodologyThe data available for South American countries are much more sparse and potentially of lower quality due to being derived (e.g.) from surveys. This calls for a bespoke statistical model that would be able to combine sparse data from various data sources which, in principle, do not aim at measuring migration. Generalising the Raymer et.al.'s model (2013:803), multiple tables of migration data flows from source k can be combined in a measurement model that corrects for data inadequacies. I can specify a model for the true (unobserved) migration flow at time t to allow parameter estimates for the role of demographic, social and economic factors on movements (Abel,2010:802; Raymer et.al.,2013:817). This approach is usually based on a gravity-type model (Sen & Smith,2012:49-53) which assumes that the flows are proportional to the population size of the sending and receiving countries and inversely proportional to the distance between them with estimable parameter. The posterior distribution of the true (unobserved) migration flow at time t, will represent the final synthetic data used as outputs for the model for estimates of migration flows over time with associated uncertainty. By including a time-specific parameter, forecasts of future flows can be produced. Introduction of specific covariates can allow extending the model to estimate particular types of migration.Where only data on migrant stocks rather than flows are available, the Abel (2013: 508-524) andAbel &Sander's (2014:1521) methods can be used. The model requires migrant stocks data as input and can be estimated using maximum likelihood. In the proposed project, I will refine the model considering the criticism of Dennet (2016) and apply Bayesian inference to provide flow estimates with measures of uncertainty that will be fed into the flows model described.