Harmonising international migration flow statistics in Africa

今天，非洲有54个国家（不包括属地或其他领土），人口达12.6亿（美国人口普查局2019年）。国际移徙在人口变化和人口地理再分配中发挥着关键作用。根据盖洛普（Gallup）的调查结果（2017年），非洲渴望移民的人口比例正在增加。2013年至2016年期间，有13个非洲国家的潜在移民超过了其人口的30%。其中3个甚至超过50%，而塞拉利昂潜在移民的比例高达62%。尽管最近非洲的移民形象主要指向欧洲，并受到冲突和气候变化的推动，但它主要是一个大陆内移民，以及海湾国家和美洲的大陆（Flahaux和De Haas 2016）。根据全球双边移徙数据库，2000年大陆内移徙人数达1 050万人，而在非洲以外生活的非洲出生的移徙人数为870万人。此外，尽管人们普遍认为，难民和处于“类似难民状况”的人占非洲国际移民的240万或14%（联合国难民署2011年）。根据一项全球盖洛普调查，世界上有14%的成年人（15岁以上）表示，如果可以的话，他们希望离开自己的国家。然而，只有3%的有迁移意向的人真正开始为他们的叶子做准备（Esipova et al. 2011）。虽然有相当大比例的人希望移民，但只有少数人这样做，因为个人移民的决定很复杂，取决于他们自己的情况，例如，人们是否有足够的资源来支持他们的移民行为。如一个多阶段的过程不能在宏观层面上反映，这严重限制了对迁移的认识，导致总体迁移统计的不准确。到目前为止，估计非洲移民流动的文献是基于宏观模型（Abel and Sander 2014）。然而，缺失值严重限制了估计值的准确性。而现有的迁移数据由于测量不准确，往往是不准确的。如果减少不确定度，则可以提高精度。在本文中，微观模拟方法将用于估计双边国际移民流动以及非洲国家内部的缺失流动，并减少不确定性。本文的贡献在于通过微观模拟的方法减少不确定性，对缺失的双边国际移民流进行估算，提高移民数据的测量精度。为了预测国际移民，测量和建模不确定性是必要的。为了考虑到这种不确定性，主要的方法是贝叶斯推理，它将多个不确定性连贯地计算在内（Raymer et.al 2013, Wisniowski 2017）。测量不确定度是必要的，但降低不确定度更为重要。正如Willekens（2018）所指出的，应该通过考虑移民的异质性（即个人的特征，如年龄、性别、教育水平等）来减少不确定性，并理解这些人为什么决定离开或留在自己的国家。因此，有必要在微观层面上研究迁移行为。模型应该是一个基于个体的模型，agent（个体）由于其特征的不同，应该有不同的生命历程。

There are 54 countries in Africa today (excluding the dependencies or other territories), which amounts to 1.26 billion people (US Census Bureau 2019). International migration plays a key role in population change and population geographical redistribution. According to Gallup's finding (2017), the percentage of people in Africa who desire to migrate is increasing. There are 13 African countries which potential migrants are above 30% of their population between 2013 and 2016. And 3 of them are even higher than 50%, while the percentage of potential migrants in Sierra Leone is up to 62%. Despite recent image of migration in Africa being directed mainly towards Europe and fuelled by conflict and climate change, it is a continent of predominantly intra-continental migration, as well as towards Gulf states and Americas (Flahaux and De Haas 2016). According to the Global Bilateral Migration Database, intra-continental migrant stocks amounted to 10.5 million people in 2000, while stocks of African-born migrants living outside Africa were 8.7 million. Also, despite a common belief, refugees and people in a 'refugee-like situations' represented 2.4 million or 14% of international migrants in Africa (UNHCR 2011).According to a worldwide Gallup survey, there are 14% of the world's adults (aged over 15) said that they would like to leave their own country if they could. However, only 3% of them who have migration intentions actually started to make preparations for their leaves (Esipova et al. 2011). Although there are a considerable proportion people desire to emigrate, there is only few persons made it because the decision for an individual to emigrate is complex and depends on their own situations, such as, if people have enough resource to support their emigration behaviour. Such as a multistage process cannot be reflected in the macro level, which severely limits the understanding of migration and leads to the inaccurate of aggregate migration statistics. So far, the literature on estimating African migration flows are based on macro models (Abel and Sander 2014). However, the missing values severely limits the accuracy of estimation values. And the existing migration data is usually inaccurate due to the inaccurate measurement. The accuracy can be improved if uncertainty is reduced. In this thesis, the microsimulation approach will be utilised to estimate bilateral international migration flows as well as the missing flows within African countries and reduce the uncertainty. The contribution of this thesis is to impute missing bilateral international migration flows and improve the measurement accuracy of migration data by reducing the uncertainty by microsimulation approach. To predict the international migration, measuring and modelling the uncertainty is necessary. To take this uncertainty into consideration, the leading method is the Bayesian inference that counts for the multiple uncertainty coherently (Raymer et.al 2013, Wisniowski 2017). Measure uncertainty is necessary, but reduce the uncertainty is more important. As indicated in Willekens (2018), the uncertainties should be reduced by taking the heterogeneity of migrants in consideration, (i.e. individuals' characteristics such as age, gender, education level, etc) and understand why these people decide to leave or stay in their own country. Thus, there is a need to study migration behaviour in a micro level. The model should be an individual-based model, the agents (individuals) should have different life courses due to the difference of their characteristics.