I don’t really have time for this today, but….
I wrote again yesterday about how getting rid of departure cards seems set to degrade the quality of our timely net migration data (currently some of the best available anywhere in the world, which we need since our net migration flows are large and volatile). SNZ has previously promised that future PLT estimates
will be generated through a probabilistic predictive model of traveller type (ie short-term traveller, or long-term migrant), based on available characteristics of travellers. Such a model will provide a provisional estimate of migration, which we can then revise (if required) as sufficient time passes for us to apply the outcomes-based measure.
In media commentary yesterday, the Minister of Immigration was heard to suggest that under the new system the data will be better than what we’ve had now.
That seemed unlikely, but later yesterday morning SNZ put out a media release including this
Moving to the new methodology means it will be 17 months before final migration estimates are available. That’s because someone has to be in the country for 12 months out of 16 before they can be classified as a long-term migrant.
“A delay of that length would have been unacceptable to those who rely on migration data for planning and analysis, so we are developing a statistical model that will provide a provisional estimate of migration. A first look at provisional external migration estimates will be released tomorrow,” said Mrs Theyers.
In future, statistics for New Zealanders travelling overseas will be largely based on when they return. Some variables – including occupation and country of next residence – will no longer be available.
That statement itself confirmed one of my points – some important data is going to be lost altogether (eg data on net outflows to Australia will in future have to be inferred, rather than available directly – and while I’m sure that isn’t the motivation, that will be convenient for governments). But there was a promise that they would reveal more today. I was hopeful we might get a proper discussion paper, with details of their modelling techniques, and the results of backtesting, and (for example) the identification of key periods (especially around turning points – a key focus of macroeconomic analysts) where the new procedure worked well and when it hadn’t.
What was released this morning was three charts and a page of text. There is nothing about methodology, nothing about backtesting, nothing about the identification of turning points, in fact nothing that any serious analyst is likely to find useful.
We are told
To mitigate the impacts of such a delay, we are developing a statistical model that gives provisional estimates of migration to give a timelier statistic. The first provisional migration estimates are now available.
“Preliminary data presented today gives our customers their first glimpse of what migration statistics will look like once the outcomes-based approach becomes the official way we measure migration in New Zealand,” population insights senior manager Brooke Theyers said today.
But nothing at all about the model.
But here are results they are happy to show us
(I presume that these numbers are not seasonally adjusted, which probably accounts for some of the jumping around in the median estimates from month to month).
Recall that under the 12/16 methodology, the numbers from 17 months ago become final (and are, in many – but not all – respects better quality than the current PLT numbers). But the latest monthly data has huge margins of errors – even a 50 per cent confidence interval looks to be about 3000 people wide (on a monthly basis – and bearing in mind that the average monthly inflow in recent years has been about 6000 people).
But to repeat:
- no model,
- no series as to how the estimates have evolved over time with the addition more data,
- no backtesting,
- no analysis of turning point information
Almost nothing at all. And none of this is being consulted on, instead the government and SNZ are simply junking one of our best high frequency sets of economic data, about a variable which adds considerable volatility to the New Zealand economy. We should expect a lot more, especially from a notionally independent national statistics agency.