My presentation at the conference was on ethnic population projections, especially for small ethnic groups. I won't post today on the full details of that presentation (which will be a forthcoming working paper that I will talk about then), but I did want to discuss three challenges for producing ethnic population projections.
Statistics New Zealand produces ethnic population projections only for the main 'Level 1' ethnic groups in New Zealand (European or Other, Maori, Pacific, Asian, and the omnibus groups Middle Eastern/Latin American/African), as well as for the three largest 'Level 2' ethnic groups (Samoan, Chinese, and Indian). There are good reasons why they don't produce projections for smaller groups such as Dutch, Fijian, or Vietnamese (being the three groups that I presented projections for at the Pathways conference).
The first challenge is lack of data. If you want to produce population projections using the traditional 'cohort component model', you need to be able to project births, deaths, and migration for the population groups you want to project. To project future births, deaths, and migration, you create a model based on observed numbers and rates of births, deaths, and migration in the past. This is very difficult for small population groups, because the numbers of observed births, deaths, and migration events is smaller and noisier. In some (or many) years, there might be no events of that type. For example, there might be no births to Vietnamese mothers aged 15-19 in some years, which makes it difficult to project.
The second challenge is that, in addition to projecting births, deaths, and migration, you also need to project inter-ethnic mobility. Inter-ethnic mobility occurs when a person's ethnicity changes. You might think that ethnicity is static, but that isn't true at all, as this article by Carolyn Liebler and others explains:
Add something else to the list of things that seem simple but are actually complicated – the way someone reports their race or ethnicity... With over 160 million cases [from the U.S. Census] covering all U.S. race and ethnicity groups we found that 6.1% of people in the (not-nationally-representative) data had a different race or ethnic response in 2010 than they did in 2000.Rates of inter-ethnic mobility in New Zealand are similar (see the report from Statistics New Zealand here). This challenge arises because people can self-identify with any ethnicity, and their self-identification can change over time. These changes in self-identity are not common, but they provide yet another rare event that needs to be projected as part of an ethnic population projections model.
The third challenge is that people can hold more than one ethnicity. That might not sound like much of a challenge, but traditional models assume that each population group (by age, sex, location, etc.) is mutually exclusive. That is, a person cannot simultaneously belong to more than one group. But, if people can hold more than one ethnicity then they will belong to more than one group, which means that ethnic population projections models must run in a different way to traditional models.
Those challenges are the main reasons why Statistics New Zealand provides ethnic population projections for only a limited number of ethnic groups. In a future post, I'll discuss a method that Jacques Poot and I have been applying that allows us to go a little further and produce projections that are complementary to Statistics New Zealand's projections, and cover a wider number of much-smaller ethnic groups at both the national and regional levels.