Comments on: What variables work for electorate profiling?

By: El Nino

El Nino — Sat, 10 Jan 2009 21:17:52 +0000

Sorry – got on to this a bit late – I have been having some computer-free days. I suggest splitting ‘overseas’ born between UK/South Africa and the rest of the world. Also worth having a look at parental country of birth. Median rent may also be a good one as it seems to be a bit of a proxy for median house value. Education level is also worth a look.

By: Labor Outsider

Labor Outsider — Wed, 07 Jan 2009 08:00:15 +0000

Hi Possum, interesting stuff. I’d group variables by type, and then see what is available in the census:

- demographic variables (poportions of different age, family type, age of children, etc)
- household means (average income, distribution of income, proportion dual income, iporportion in receipt of government income support, direct equity holdings, in receipt of property income, proportion self funded retirees, etc)
- household balance sheets (avarage housing wealth, financial wealth, leverage, proportion with mortgage, repayment burden, etc)
- geographic variables (remoteness, distance from CBD (for metro), coastal)
- regional economic variables (economic diversity, occupational proportions, industry proportions, educational proportions, employment/population growth since last census, in-migration/out-migration, change in unemployment rate since last census, change in average income since last census, etc)
- ethno-cultural variables (proportions of various ethnic groups, proportion NESB, proportion ATSIC, etc)
- tenure variables (proportion renting, paying off mortgage, proportion of housing stock public)
- attitudinal variables (proportion engaged in voluntary activity, proportion regularly attend church, proportion athiest, proportion members of union, etc)

As much as possible use dummy variable categories, rather than averages to capture possible non-linearities (ie, split income into quintiles, age into cohorts, etc).

Use information on changes since the last election as well as proportions at time of last census.

To choose the variables, you are best off running some simple regressions. As you say, there is so much information, you cannot tell beforehand, which are the most relevent for voting behaviour.

Before running the regressions, put together a covariance matrix so that it is very clear where the potential for multi-collinearity lies.

Move from general to specific so that you let the data tell you which variables are important, rather than pre-empting it too much.

Hopefully you will spit out 15-20 significant variables that you can then use to profile each electorate.

By: Jason Wilson

Jason Wilson — Wed, 07 Jan 2009 04:57:04 +0000

Re occupational groups – quality – looks great.

By: Possum

Possum — Wed, 07 Jan 2009 04:02:18 +0000

Paul, I think the Census has a data set on something like “voluntary work for an organisation”. I’ll run that through some regressions and see what pops out – I’ll post the details a little later.

By: Possum

Possum — Wed, 07 Jan 2009 03:59:18 +0000

NESB and first and second generation migrant numbers as a proportion of the electorate would be handy, I’ll go have a squiz and chase that up.

By: David Richards

David Richards — Wed, 07 Jan 2009 03:22:52 +0000

Being a resident of the electorate – the correlation between the higher than the rest of SA and Australia graphs with regard to the 20-34 age group and the upper income brackets would be due to the recent construction of high density CBD and environs yuppie apartment buildings (replacing now superfluous office buildings). Also, the electorate includes some other high income enclaves such as North Adelaide.

By: Swing Lowe

Swing Lowe — Wed, 07 Jan 2009 03:19:15 +0000

How about people from a Non-English Speaking Background?

I’m sure you’ll see a high correlation between the proportion of NESBs in an electorate and the Labor TPP vote…

By: Socrates

Socrates — Wed, 07 Jan 2009 02:45:17 +0000

Poss I’d suggest at least the following, which are significant for various forms of actvity in my field:
- age, sex, work status, average education level, % renting/own house/paying mortgage

Some statistics don’t mean much – eg car ownership doesn’t always include company cars. Average income is very messy – average wage & salary doesn’t include all the transfer payments or investment income. Plus it assumes people truthfully declare all income, which they don’t, even to ABS, let alone the taxman. Likewise % owning shares doesn’t discriminate betwene millionaires and those witha few hundred Tesltra shares.

An age profile would be useful – % old, % young – rather than just an average age woudl be useful. % over 70 is a very good indicator of health care needs.

By: Paul from Berwick

Paul from Berwick — Wed, 07 Jan 2009 02:21:01 +0000

Poss,

George from the OZ provided a good workout in ‘07. Try some stuff from him (can’t find the links).

Would be good to include some volunteering/community service metrics (apparently how many choirs there are is a good mesaure of a community’s vitality).

Another would be “distance to travel to work (plus transport mode)”.

And another would be along the lines of what type of studies the post-secondary students are undertaking.

Finally, there may be something of interest in the “what the people want” forum.

Regards,

Paul

By: Oz

Oz — Wed, 07 Jan 2009 01:39:14 +0000

Occupational groups, seconded.

How about ethnicity, or what percentage are immigrants?