Wednesday, March 19, 2008

Sydney, statistics and the need for caution

There are nuts and then there are statistical nuts. I may belong to the first class. I certainly belong to the second!

It's not just that I like numbers and patterns. I rely on them for evidence to support arguments. So I get upset when I find that I have been wrong, that I have misinterpreted data.

A key problem in interpreting statistics, especially when looking at change over time, is the way that definitions change. Unemployment is a case in point. It is very hard to compare present and past unemployment statistics because the way employment is defined has changed.

A second problem is the way in which changes made by Governments for other reasons affects statistics. As an example, local government changes - boundary changes, mergers - make it hard to compare population changes over time in specific areas.

I was sharply reminded of all this when trawling through statistical data on Sydney and NSW checking facts for my part completed series on the SMH's campaign.

Accoding to the census stats, the resident population of the Sydney statistical division at the time of the 2006 census was 4,119,190 out of a total state population of 6,549,174. This is the number usually quoted when people talk about Sydney.

Then browsing through the recent Australian Bureau of Statistics publication, Sydney - A Social Atlas - 2006, I read that at the 2006 census there were 3,645,153 usual residents in Sydney, 55.7% of the NSW population. Now this is a very different number, so I was forced to investigate.

The social atlas defines Sydney as the area bounded by the suburbs of Palm Beach and Berowra in the north, Cronulla and Heathcote in the south, Camden in the south-west, Faulconbridge in the west, and Riverstone in the north-west. In broad terms, this is what most people today would think of as Sydney.

Now compare this to the Sydney statistical division. You will find all the Australian statistical divisions here. In addition to Sydney as defined in the social atlas, this adds a big additional swag of territory including all the Blue Mountains and Gosford-Wyong, areas that most people would not think of as being part of Sydney.

Does this matter? Well, it depends on what you want to use the statistics for.

At a macro level, I am interested in Sydney's population relative to the rest of NSW because shifts in Sydney's share of the state population over time have been a key political driver. If we are going to talk, as the SMH has done, about Sydney's relative decline, then we have to be able to measure it.

Much of my recent analysis has used the Sydney statistical division as the source of numbers on Sydney. This is clearly invalid for at least some of my arguments. Among other things, the use of the broader number conceals the extent of population shifts. Without checking the data, I think that it is a very long while since Sydney as normally defined has only had 55.7% of the NSW population.

At micro level, the inclusion of the Blue Mountains and Gosford-Wyong with their very different demographic structures affects the averages and percentages often used in discussions about Sydney. In particular, it conceals the extent of change within Sydney as normally defined.

I do not think that all this affects the arguments that I have begun developing. If anything, is strengthens them. However, it does mean that I have to re-check data.

In the meantime, I have been almost totally sidetracked by the detail in the Sydney Social Atlas. It is an absolute gold-mine for someone like me interested in patterns of social change.

No comments: