Northern Ireland local group: When Official Statistics don’t cut the mustard

The Northern Ireland group held a joint meeting with the Northern Ireland Statistical Research Agency (NISRA) on Thursday, 31 October 2019 at 1pm, in the Offices of the Registrar General in Belfast. The speaker was Dr Alexandre Noyvirt of the Data Science Campus at the UK Office for National Statistics (ONS).

Alexandre’s talk was about ongoing work in the Data Science Campus of ONS in the areas of generating faster economic indicators from ship movements as well as rapid evaluation of the amount of surface water for early warning system using satellite images. He said that he would concentrate on the ship movement area today. This was important in the context of International trade where currently Government data were published relatively infrequently and there was perhaps the potential of making reliable estimates in near, real time, using the methods proposed below.

The main objectives were to (a) study the links between ship movement and official economic indicators; (b) extract the most informative features (indicators) from the shipping data; (c) model the trends (up/down) in specific economic sectors as a function of these indicators; and (c) forecast (earlier) parts of the economy using the indicators. 

Two main indicators had been developed:

  1. Time in port – aggregated time in seconds spent by ships in UK ports
  2. Total traffic – number of unique ships entering UK ports

The approach relies on the routine live transmission of data from vessels (greater than 300 tonnes and passenger vessels) using the Automatic Identification System (AIS). The AIS data include: GPS position, speed, bearing, ROT (Rate of Turn) and heading, over time, to a base station or satellite. Transmission occurs every two seconds for moving vessels and up to every three minutes for stationary vessels (at anchor or moored). Globally there are live stream, 500 messages/sec, in near real time, to satellite and base receivers. The data are available from a commercial provider and constitute 'big data' (at 15-25TB/year). There are also static data on destination and type of vessel, etc. These data allow spatial tracking and graphical representation of each ship’s position and movement.

There were a number of data quality issues which Alexandre described in detail. Processing the data, to extract indicators, requires a sophisticated cluster computer system and specialised software.

The features were then compared, over a two-year period (September 2016 to July 2018), with a number of regularly published indices by UK Official Statistics: GVA, Trade Imports and Trade Exports (Sources: ONS and HMRC). Some of the graphical time series comparisons were similar: the maximum Pearson correlation (r=0.64) was between the Port Traffic indicator and Imports (ONS); the next highest (r=0.56) was between the Port Traffic indicator and Imports (HMRC). In relation to the Time in Port indicator the maximum value of the
correlation (r=0.45) which occurred with Imports (HMRC).

Alexandre explained that these were early days in the analysis of these new indicators and more detailed statistical analysis was required. Looking to the future, he also explained that they were refining the current approach by considering a classification by type of ship focusing on the larger ships. They were also deriving additional indicators such as the number of port visits which was currently being added. There was, he assured the audience, ample scope for extension.

The talk was received with acclaim. Alexandre had taken several questions as the talk unfolded. These included questions about satellite tracking and extent of missing data in the AIS data and in the static data. In relation to the derived indicators he agreed that basically we were dealing with a bivariate time series and that methods to enrich the current analysis would need to be developed. He was thanked for a very stimulating talk.

Load more