Home // Archive by category "statistics"

Beating the Swing Flu?

A PSA from Herräng Dance CampWith the Northern Hemisphere arrival of summer it’s swing camp season and that means one thing – completely out of sync with the rest of the population it’s swing flu season. And with the swing flu season comes all sorts of remedies to ward off the lergy from Herräng’s anti-cold juice to bottles of military-grade Nyquil. But does any of this actually work?

I’m a science and evidence based policy junkie (which explains my love for the work of Ben Goldacre and also the Youtube Channel Healthcare Triage, who actually have an episode on cold remedies), so I’ve been aware of the Cochrane Collaboration for some time. This non-profit organisation scours the literature and does systematic reviews of all sorts of different medical treatments (including plenty of CAM ones too). Fortunately Cochrane has reviewed the evidence on multiple remedies for the prevention and treatment of colds, flu and flu-like illness. I thought I’d summarise their publications to just get a very brief snapshot of what the evidence actually says.

Firstly let me get this out of the way: This is not medical advice! Making health decisions based on what you saw on some blog on the internet is an incredibly stupid idea. Before you decide to take any medication, supplement or start any sort of regimen you should – talk – to – your – doctor! Ask them about the evidence, or even take along some of the reports from Cochrane.

The one thing that stands out from looking through all these reports is how flimsy the evidence base is in many of these studies. They read like a veritable index of poor trial design, suffering from inadequate blinding, small sample sizes and a host of other issues – and that’s before we even get to issues of publication bias. So even where a small effect was found in the systematic review there’s a reasonable chance it’s no more effective than placebo. In fact the only thing that appears to work well is good hygiene – so wash those hands!

Note also that these are all reports regarding healthy adults. Others examine the evidence base for those with particular conditions or in children.

Once again, in case I wasn’t clear before This is not medical advice! Talk to your doctor about what’s right for you.

Remedy Prevention Treatment
Acetaminophen/Paracetamol Although widely included in over the counter cold/flu medication the research evidence for its effectiveness is poor quality. It may help reduce some nasal symptoms of common cold but does not appear effective on other symptoms.
Antibiotics Antibiotics have no benefit for common cold and can cause significant side effects.
Atrovent Nasal Spray (ipratropium bromide) This product may relieve some cold symptoms, but the existing evidence has some limitations.
Chinese Herbs There is very weak evidence that these may have similar effects to antiviral drugs at treating and preventing influenza
Drink Plenty of Fluids There is no evidence for or against the common recommendation to increase fluid intake during acute respiratory infections.
Echinacea There is no good evidence for use of Echinacea in treating colds, but it is possible there may be a small benefit.
Flu Vaccine The evidence demonstrates a modest impact on reducing symptoms and working days lost in the general population due to flu (NNT 40-70). No evidence of association between vaccination and serious adverse reactions was found.
Garlic There is some evidence to suggest that regular daily garlic can have a protective effect against the common cold, however the evidence is insufficient and of poor quality. Also we’re dancers – why would you want to smell like Garlic?
Handwashing Spread of respiratory viruses can be reduced by simple handwashing.

Learn how to do it properly!

Nasal Saline Use of nasal saline may relieve some symptoms of URTIs but the quality of the research is low.
NSAIDs (Aspirin, Ibuprofen etc.) Although the research evidence is somewhat limited, it appears that these types of drugs are somewhat effective at relieving discomfort, but do not appear to help with other cold symptoms.
Oscillococcinum (a homeopathic preparation) No evidence for effectiveness in influenza or flu-like illnesses No evidence for effectiveness in influenza or flu-like illnesses
Probiotics There is some evidence that probiotics are effective at preventing and minimising the impact of Upper-Respiratory-Tract-Infections, but the quality of the research is poor.
Steam Inhalation Studies on this treatment have had inconsistent results and the research quality is poor.
Steroid Nasal Spray There is no research evidence to support their use to relieve common cold symptoms, however there have only been a small number of studies.
Tamiflu and Relenza There is evidence of a small protective effect from prophylactic use of these drugs on influenza only, but side effects are a concern. There is evidence of a small reduction in duration of influenza and flu-like illnesses, but side effects are a concern.
Typical Cold and Flu preparations (containing antihistamines, decongestants and analgesics) These medications appear to be generally beneficial in treating the symptoms of common cold, but many people report side effects.
Typical Cough Medications There is no good evidence for or against the use of cough preparations in treating cough and the quality of the research in this area is poor.
Umckaloabo This may be effective at relieving some symptoms of the common cold but the quality of the evidence is low to very low.
Vitamin C Regular supplementation does not reduce the incidence of common cold, but it does have a small effect on reducing the length of common cold. Regular supplementation has been shown to reduce incidence in people exposed to high physical stress (e.g. skiers and marathon runners), but it is unclear whether this would extend to swing dancers. The evidence on therapeutic use of Vitamin C is limited and doesn’t suggest any benefit, but it may be worthwhile to consider on an individual basis.

How Many Lindy Hoppers are There

I’ve been wondering about this question for a while but with a little free time I finally got around to having a decent crack at working it out with some GIS and statistics.The short answer: by my estimation about 120,000. Read on for the longer answer.

Where are people Lindy Hopping?

This is actually a somewhat difficult question. Scenes are constantly starting and folding – such as scenes in college towns in the US and those where expats are the primary drivers and consumers.* Plus the advent of the travelling lindy hopper has led to many camps being held in places that don’t have a regular scene. For this purpose a “scene” is a location that actually has some form of regular Lindy hop be that classes, social dancing or some other organising activity.

Herrang only has dancing 5 weeks out of the year – that doesn’t count. (photo by Rikomatic)

My starting point was the World Lindy Hop Map which I supplemented with maps for countries and smaller scenes including the LA Lindy Hop Map, The Lindy Hop Map Australia, Lindy Hop and Swing Dance in Italy, and the UK Lindy Map.

Next was quality control.

I ended up removing a whole bunch of points. There were plenty with incorrect geocoding (e.g. suburbs of cities that were coded to small towns, country entries sitting in the middle of nowhere etc.), I also did a fair bit of checking to ensure currency of schools, events, etc.** Finally I did a bunch of research to add new venues, website links and cover as broad a geography as possible.

The final result was 827 organisations, dance schools or other evidence of regular lindy hop activity in a particular location. I’m sure I’ve missed plenty of organisations in local areas – but my mission was to see if a location had lindy hop, not how many dance schools/societies were there.

Guess the map projection and you get a gold star.

As I don’t plan on keeping this updated I’m not going to put it on google maps. However here is the data in kmz and shapefile format for those who wish to use the data in your own projects and maps. I’ve made my best effort with this data, but it’s necessarily incomplete and the locations are rather approximate – don’t use this for driving directions or holiday planning without further research!

How many Lindy Hop scenes are there?

Obviously, these locations aren’t all individual scenes. Recognising that some scenes are supported by 1 large organisation and others by many smaller ones I wanted to get at the number of geographically independent scenes.

First I gathered venues to the nearest urban locality (using data from Geonames) – all those that were within about 15-20km I considered to be part of that locality. I merged a number of these together where the localities were separated by less than 30km – considering that this is probably the maximum distance (as the crow flies) for there to be enough mixing amongst venues for them to be

The answer – 463 scenes spread across 58 countries (or 59 depending on how you count Taiwan) and a range of self governing territories (e.g. New Caledonia, Hong Kong and the Åland Islands).

From this we can see that Lindy Hop is an activity for the relatively wealthy in the world – here is a chart of lindy hop prescene graphed against the Human Development Index***:

How many lindy hoppers are there?

Given the data I’ve generated here’s a related question: How many people could lindy hop if they wanted to? To assess geographic access (leaving aside demographic and cultural factors that affect access) to the above listed venues I applied the World Population Layer to determine the number of people living within 15km of a lindy hop venue. The result: about 308 million people.

Now onto the number of lindy hoppers. Rather than try and guess the average scene size (which can be tricky) I’ve applied some fancier statistics and a Monte Carlo simulation to get a bounded guess. There’s a more complete explanation and the code I used in a short R script I wrote here.

The assumptions I made are:

  • The size of lindy hop scenes are lognormally distributed. (Not bad – but without any data on actual scene sizes is untested)
  • The largest scene size is 5000. (This has been oft quoted in relation to the size of Seoul‘s lindy hop scene but London and LA/Orange County could also have sizes somewhere in this vicinity)
  • The median scene size is somewhere between 50 and 150.
  • The total number of scenes are 463.

The Monte Carlo simulation generated a stochastic set of 50,000 international lindy hop communities by randomly sampling the median scene size (from a uniform distribution) and then randomly sampling individual scene sizes (from the lognormal distribution) to get a total population of each. This then allowed statistics to be generated

This gives a median of 118,000 and a 90% chance of the “true” number being being between 82,000 and 153,000.****

Only 2% of lindy hoppers made it to Frankie 100. (Photo by hoptothebeat)

So now it’s over to you. Can anyone else come up with a better answer?

* South Asia is a great example. I know of Lindy Hop being taught at one stage or another in Kathmandu, Mumbai, Dhaka, Chittagong, Bangalore, Dharamsala and Delhi at one time or another but combining short ex-pat contracts with a culture where partnered dancing is highly unusual it typically hasn’t stuck around.

**It doesn’t help that there are literally hundreds of dance schools and other organisations out there whose websites seem to have been created back when geocities was popular and left unchanged (except for content updates). Seriously people, if wordpress is too hard for you spend the money on a web designer.

*** And in those countries with Medium Human Development the lindy hop tends to be located in wealthier cities. This is probably also the case with lindy hop in countries with High and Very High Human Development.

**** The average and the mode were also about 118,000.

Who’s on first? Performance order and judging bias in Lindy Hop contests

Making decisions is hard. Judging in contests is a hard form of decision making – there’s so much to consider. What’s worse is that your own brain will try and prevent you from making the best decision. For example every time someone has to decide on information being presented to them, the order of the presentation of that information can change their decision.

A variety of studies have found this ‘performance order’ bias in everything from the Idol series, Eurovision and figure skating and music competitions. This bias isn’t inconsequential either, in events where careers can rest on the outcome performance order bias can have a significant impacts.

So what about Lindy Hop competitions? Does the order of appearance matter?

Judging – an unenviable task (photo by Jerry Almonte)

In the spirit of my previous post on electoral methods I’ve delved into the data so you don’t have to. My source: results from the 2013 and 2011 European Swing Dance Championships (most of the other major contests I’ve looked at don’t provide the performance order with the official results – if anyone has the data and wants to send it to me to include in the analysis I’d be happy to expand it). This provides a dataset of 30 contests with between 4 and 15 individuals, couples or groups in each contest.

For analysis I’ve converted all the placings into standardised scores to enable comparison between contests of varying lengths.

Here’s what the data looks like:

You can see that the number of competitors in gets rather smaller for the larger events, but you can visually see a slight rising trend even when you discount the results in larger divisions.

To test whether this was just a statistical quirk I set up a multi-linear regression model with an order variable and two additional variables to see if there were any effect of appearing first or last in addition to to the performance order effect.

Here’s the model statistics:

Estimate Standard Error T Statistic Pr(>|t|)
(Intercept) -0.35908 0.15729 -2.283 0.0234
First 0.29989 0.22564 1.329 0.1852
Last 0.08663 0.21062 0.411 0.6813
Order 0.06666 0.02848 2.34 0.0202
This table is only for statistics nerds – otherwise you can ignore it.

These data show that there appears to be a small performance order bias in these results – but though it’s at the level of significance that could get you published in certain social science journals I’d be hesitant to say that it exists for sure.

On the other hand, beyond the performance order bias there appears to be nothing special about appearing first or last.

Converting out of standardised scores let’s look at what these results could mean in an actual contest or let’s say 12 people. If the differences between everyone’s score is similar (which is not really realistic, unless there’s a tight contest) – this effect could mean the difference two places if you appeared first instead of last.

From xkcd – full comic here

What could be going on here? Firstly it could very well be nothing. The order bias is only just within the normal 95% p value – which is not nearly as impressive as you might think. More data could cause the effect to disappear.

EDIT: But, not so fast, remember how there were only a handful of divisions with large numbers in them? There was only a single division with 15 competitors and only a single one with 12. What happens if we repeat the analysis with them removed? (and thanks to one of my FB friends for suggesting this analysis) Here’s the results:

Max size Estimate Standard Error T Statistic Pr(>|t|)
15 0.06666 0.02848 2.34 0.0202
12 0.085237 0.034704 2.456 0.0149
11 0.07006 0.03885 1.803 0.0729
10 0.06518 0.05556 1.173 0.243
9 0.08808 0.06351 1.387 0.168
8 0.06139 0.07081 0.867 0.388
6 0.005641 0.124636 0.045 0.964
5 0.333 0.2568 1.297 0.202
 Again – tune out of this table if you’re not a statistics nerd

Remove the two largest divisions and the order bias effect quickly disappears. So we can be reasonably confident that there is no performance order bias for competitions smaller than 10 or so entrants. For larger divisions we’ve really only got two contests to go off – this is really not enough data to be able to say one way or the other.
It’s reasonable that if judges are affected by a recency bias then this may only appear in larger contests. But there are also other explanations besides a recency bias on the part of the judges. Audience reaction could change throughout an individual contest and this could subtly bias the evaluations of the judges.

So how could we find out if there is an order bias? More data on the largest divisions could help but the ultimate would be a controlled experiment. Judges would watch performances on video and in different orders for each group of judges. You could also implement this in actual competitions which would effectively control for any performance order bias present, but such a radical change to judging practices is unlikely to catch on.

If you’re a competition organiser and you don’t randomise the performance order of your competitors, you really should. If an order bias does exist then it’s easily something that could build up over time and give some people (for example with names at the end of the alphabet) an edge in landing that all elusive international teaching gig. Releasing judge scores may also help – some research suggests that increased transparency, the idea of someone looking over your shoulder, can help reduce unconscious biases.

Sadly the major international lindy competitions aren’t consistent in this practice, both ILHC and ESDC have posted individual judges scores previously but not for every year (and the ILHC results do not contain performance order – and appear to have been taken down), the National Jitterbug Championships, the US Open, the American Lindy Hop Championships, the Canadian Swing Dance Championships, The Snowball and Lindy Shock do not.

I’m not talking about that sort of bias

There are potentially a range of other biases that could be relevant to Lindy Hop competitions such as reference bias (where a judge gives a higher score to a person they are familiar with – though it’s unlikely to occur at the highest levels where the competitors are all familiar to the judges), difficulty bias (where more difficult routines are scored higher, even if difficulty is scored separately), in-group bias (a close relative of reference bias – where a person is judged more favourably because they share the same dance school/city/country/race etc. with the judge), the halo effect (where one element of a persons character, such as attractiveness, influences assessments of other elements of their character), memory-influenced bias (where past performance influences current assessments) and groupthink (where a group of judges reaches a false consensus – this isn’t a problem when each judge scores independently, which appears to be the practice for most international comps).

Finally I want to emphasise that I’m not suggesting that judges are behaving unprofessionally. The problem with cognitive biases is that we all have them, they can’t be switched off and thus require a lot of cognitive energy to overcome. Awareness and acknowledgement of potential bias can go a fair way to moderating their influence.

For further reading check out this article “Natural Bias, the Hidden Controversy in Judging Sports” or if you’re interested in cognitive biases and the psychology of decision making grab a copy of Daniel Kahneman’s “Thinking Fast and Slow”.

If you’ve read this far congratulations! Have a picture of an echidna:

What is the new ‘Middle Class’?

The post-budget articles in the newspapers are making me sick. Making out families on $150,000 a year as though they’re on struggle street is pretty dirty, when folk in that situation are better off than most of us.

Matt Cowgill has written an excellent piece trying to find out what the middle class (middle meaning median, not average) actually is. It’s been picked up by ABC’s The Drum as well.

I’ve looked at how big the supposed ‘working families’ demographic is before, but let’s explore what this ‘middle class’, that Government budget savings will affect, is really like.

My source – “Household Wealth and Wealth Distribution, Australia, 2005-06” from the Australian Bureau of Statistics (I know it’s a little old but it’s the latest issue of the product).

I’ll be looking at the top 20% of households by gross income. These are households making more than $100,000 a year –  including plenty of families making substantially less than those examined in the papers.

Share of Wealth and Income
Now this top 20% of households controls 45% of the income in Australia and 60% of the wealth. This is nearly double the income and triple the wealth of the next highest 20% and 10 times the income and 60 times the wealth of the lowest 20% of households. This may sound bad but it’s better than the US where the top 20% controls 61% of the income and a whopping 85% of the wealth (and the top 1% of households 34% of the wealth).

The median income of this group is $130K a year, more than double the median income for all households and nearly 10 times that of the lowest group.

The median net worth of this group is $635K, double the median for all households and three times that of the lowest group.

Income Source
This group is fairly similar to the next highest 20% in that they both get their primary income from wages and salaries, although there appears to be a very small number that claim their primary source of income is government pensions and allowances – perhaps The Australian found them. However about a quarter of this group gets 1-20% of their income from pensions and allowances. Unfortunately the stats don’t break down further but I’ll assume the contribution will mostly be closer to 1% than 20%.

Tenure Type
Most of this group, 57%, has a mortgage the highest proportion of any income group. Makes sense – mortgages are expensive so the highest income bracket will be able to have more of them.

Family Type
Nearly half of this group (and more than twice the average) have kids under 15, the highest proportion for all income groups. Makes sense – kids are expensive so the highest income bracket will be able to have more of them.

Also less than 4% of this group are lone person households, close to seven times less than the average. To get into this income bracket you need a combined income. This is also borne out in the stats – these households have an average of 2.3 employed persons and one child under 15 (again, almost twice the average).

Three quarters of these households live in capital cities which is higher than the average (about 63%).

So this group of top 20% in household earnings does fit the postcard. They’re likely to have a mortgage and kids with both parents in work, get the majority of their income from work and live in a capital city. But this is a group that as a whole controls a majority of the household wealth and close to a majority of the household income.

This group does get some government benefits, less than that of the next highest income bracket, but their biggest advantage is that they earn more and possess more. If this group actually depends on government assistance then the rest of us are screwed.

The Canadian Election: Why they need preferential voting

I like Canada and I like elections (it’s my largest tag word at the moment, which I should really do something about). Canada had an election a couple of days ago hence I shall discuss.

For a country considered fairly liberal the political system is somewhat backward. The Senate is not elected (and there have been some moves to abolish it) and they use a first-past-the-post voting system. In a country with three main parties on the left and one on the right (at the moment anyway) this is a rather daft way to go about things. I have long maintained that the Canadian electoral system gives the Conservative party more seats (or ridings as they are known) than they deserve.

This election gave me an opportunity to put that theory to the test. At the time I scraped the data off Elections Canada most of the vote had been counted and this was the results:

BQ 4
Conservatives 167
Liberals 34
NDP 102
Greens 1
Total 308

(A note on parties. BQ is the Bloc Québécois who campaign only in Quebec on a secessionist agenda with left leaning policies otherwise, the Liberals are liberals with a little l and formerly Canada’s primary left leaning party. The NDP are the New Democrat Party, social democrats and the new opposition. The Greens also won their first seat)

This has given the Conservatives a majority government for the first time in a while (they went into the election with a minority government). This is also something a lot of my Canadian friends are unhappy about.

So would preferential voting have made a difference? I spent a couple of nights crunching numbers to find out. I’m going to assume the introduction of the Alternative Vote (AV), a form of preferential voting. There is currently a referendum on this in the UK on which there is excellent Australian coverage.

The alternative vote is also known as optional preferential voting which is used in most Australian states and territories. The BBC explains it well here. It aims to elect the most preferred candidate. Basically it works by allowing voters to express preferences for more than one candidate. You mark 1 on the ballot paper to vote for your most preferred candidate and then you can add as many more numbers as you wish (or just leave it at one).

The papers are counted on the one votes. If a candidate has more than 50% of the vote they are elected. If not then the candidate with the least number of votes is eliminated and their papers redistributed according to preferences (if no further preferences are indicated then those votes ‘exhaust’). If a candidate still doesn’t have 50% of the remaining votes (the exhausted votes are not counted in this total) then then next lowest candidate is excluded and so on.

Most of the time AV will elect the same candidate as FPTP. Where it can have an impact is where votes are split between two similar candidates, as is the case in many ridings in Canada between the NDP and Liberals. Using the data I undertook a simulated distribution of preferences. First I counted up all those ridings where the candidate had won with more than 50% of the vote. In these ridings AV would not make any difference.

BQ 0
Conservatives 107
Liberals 2
NDP 36
Greens 0
Total 145

For the remaining 163 ridings I distributed preferences on the following assumptions:

  1. Roughly 50% of the votes would exhaust (which is the usual rate in Australia)
  2. Preferences from the Greens would flow to the NDP
  3. Preferences from the Liberals would flow to the NDP and vice versa
  4. Preferences from the Conservatives would flow to the Liberals
  5. Preferences from the BQ would flow to either the NDP or the Liberals.

For 112 of these ridings the use of AV would not make a difference to the result. However for 51 ridings there could be. The results of my simulated distribution are below:

BQ 0
Conservatives 129
Liberals 50
NDP 104
Greens 1
Uncertain 2
Total 308

Under AV the conservatives would only get 129 ridings with a further 12 possible. The Liberals and NDP would have at least 160 between them. 4 ridings could go to either the BQ, Libs or NDP and would be decided on Conservative preferences. 2 ridings were influenced by high polling minor parties or independents making them impossible to predict. The 24 uncertain ridings would depend on the rate preferences are exhausted (particularly in the left vs right contests) and how preferences from the Conservatives or BQ flow to the Liberals and NDP.

Under AV the Conservatives would not have won a majority. They could still have governed from minority, however unlike in the previous parliament the NDP and Liberals would have the numbers to form a coalition. Here’s a list of the ridings I’ve predicted that would be different under AV:

Riding Province FPTP AV
Edmonton – Sherwood Park AB CON ?
Vancouver South BC CON ?CON/LIB
Nanaimo – Alberni BC CON ?CON/NDP
Newton – North Delta BC NDP LIB
Vancouver Island North BC CON ?CON/NDP
Elmwood – Transcona MB CON NDP
Winnipeg North MB LIB ?LIB/NDP
Winnipeg South Centre MB CON LIB
Madawaska – Restigouche NB CON LIB
Moncton – Riverview – Dieppe NB CON LIB
Labrador NL CON LIB
Dartmouth – Cole Harbour NS NDP LIB
South Shore – St. Margaret’s NS CON NDP
Ajax–Pickering Ontario CON LIB
Bramalea–Gore–Malton Ontario CON NDP
Brampton West Ontario CON ?CON/LIB
Don Valley East Ontario CON LIB
Don Valley West Ontario CON LIB
Eglinton – Lawrence Ontario CON ?CON/LIB
Kitchener Centre Ontario CON ?CON/LIB
Etobicoke – Lakeshore Ontario CON LIB
Etobicoke Centre Ontario CON LIB
Mississauga – Brampton South Ontario CON ?CON/LIB
Mississauga – Streetsville Ontario CON ?CON/LIB
Kitchener – Waterloo Ontario CON LIB
Richmond Hill Ontario CON ?CON/LIB
London North Centre Ontario CON LIB
Mississauga East – Cooksville Ontario CON LIB
Nipissing – Timiskaming Ontario CON LIB
Ottawa – Orléans Ontario CON LIB
Pickering – Scarborough East Ontario CON LIB
Sault Ste. Marie Ontario CON NDP
Scarborough Centre Ontario CON LIB
Willowdale Ontario CON LIB
Portneuf – Jacques-Cartier Quebec NDP ?
Ahuntsic Quebec BQ ?BQ/LIB/NDP
Lotbinière – Chutes-de-la-Chaudière Quebec CON NDP
Lévis – Bellechasse Quebec CON ?CON/NDP
Montmagny – L’Islet – Kamouraska – Rivière-du-Loup Quebec CON NDP
Haute-Gaspésie – La Mitis – Matane – Matapédia Quebec BQ ?LIB/BQ
Honoré – Mercier Quebec NDP ?LIB/NDP
Lac-Saint-Louis Quebec LIB ?LIB/NDP
Papineau Quebec LIB ?LIB/NDP
Pierrefonds – Dollard Quebec NDP ?LIB/NDP
Westmount – Ville-Marie Quebec LIB ?LIB/NDP
Bas-Richelieu–Nicolet–Bécancour Quebec BQ ?NDP/BQ
Richmond – Arthabaska Quebec BQ ?NDP/BQ
Desnethé – Missinippi – Churchill River SK CON ?CON/NDP
Saskatoon – Rosetown – Biggar SK CON ?CON/NDP
Palliser SK CON NDP

Would this have been the actual result? It’s very hard to predict what would have happened in reality. It’s likely that the rate of exhausted votes would be higher for Conservatives and lower for voters on the left, depending on the how-to-vote campaigns of the parties. The reduction in strategic voting would also have different consequences (the Greens and other minor parties would probably have gotten higher votes) and could have led to an increase in votes for the left as these candidates would have been less concerned about vote splitting.

What does seem to be the case is that the Liberals should have the most interest in AV given the numbers of seats they’d pick up. First Past the Post would seem to be preferred by the Conservatives, while there’s still only one party on the right.

What are the prospects for AV in Canada?

Some of the provinces used it in the period between 1920s-1950s, however none have used it since.  British Columbia tried twice to introduce a complicated form of preferential voting similar to what is used in Tasmania, but the referenda weren’t successful.

There will certainly be some discussion about it after this election but the reformists argument will be split between some form of AV and some form of proportional representation. Just like the vote on the left….

NSW State Election Contests: The Swing

This is the final in my series of NSW State Election Contests posts, previewing the interesting contests for election watchers beyond the obvious conclusion of the poll. I’ll be back after March 26 to make more commentary. In this post I’ll examine the swing.

What is the swing?
For those Lindy Hoppers out there I’m not talking about this, I’m talking about what is mooted to be one of the largest swings in Australian electoral history. The swing is a crude measure of the change in the two-party preferred vote since the last election. It is supposed to measure the change in preference of the electorate for one side of politics over the other. However it does have a number of flaws. It’s not an entirely useful concept as in a multiseat parliament where each seat elects one candidate, the swing may not necessarily predict the outcome of the election particularly for close elections. It also doesn’t reveal complexities in voter trends, such as the recent rise of the Greens, or shed much light on local contests with independents.

However as a single number to tell you how the election fares up for the major parties it’s the best figure out there.

A brief comparison of past swings
Antony Green has an excellent post on his blog about past swings in Australian electoral history. Here’s a few records since 1950 (prior to 1950 it is difficult to determine a 2 party preferred vote):

  • Last NSW Election – 3.7% against Labor
  • Record at a State Election – 14.6% against Labor in Victoria in 1955
  • Record at NSW Election – 9.1% to Labor in the 1978 Wranslide
  • Record at NSW Election for change of Government – 8.3% against Labor in 1988
  • Record at Federal Election – 7.4% against the (dismissed) Labor Government in 1975


There are four main polling companies that gather data for the NSW Election (although there are others). Two of these are commissioned by the two major newspaper organisations in Australia (Nielsen by Fairfax,  published in the Sydney Morning Herald and The Age, and Newspoll by Newscorp published in The Australian), another poll is done by Essential Media a social marketing and communications firm and the fourth by Galaxy Research another PR and communications research firm.

Here are the last four polls done by each firm. They all differ in their timing and methodology and the additional questions asked. I’ve listed the primary votes of Labor, the Coalition and the Greens and then the two-party preferred vote (2PP) as well as the swing.

Nielsen – Published 16 February – Swing 18%

Labor 22%
LNP 53%
Green 13%

Labor 2PP 34%
LNP 2PP 66%

Galaxy Poll – Published March 4 – Swing – 16%

Labor 23%
LNP 50%
Green 14%

Labor 2PP 36%
LNP 2PP 64%

Newspoll – Published 14 March – Swing 15%

Labor 26%
LNP 50%
Green 11%

Labor 2PP 37%
LNP 2PP 63%

Essential Poll – Published 18 March – Swing 17%

Labor 24%
LNP 54%
Green 12%

Labor 2PP 35%
LNP 2PP 65%

This sort of polling generally has sampling errors of a few percent so there’s no clear trend up or down in the data. What is clear is that even the lowest swing is pointing towards what will almost certain be a record in NSW and likely a record nationally too.

Applying these results broadly suggest a result along the lines of:
Coalition: 65-71 (including 16 Nationals)
Labor: 13-19
Greens: 2
Independents: 7
Though as I outlined in my previous post, some of the independents are in for a tough fight whilst some Labor seats could be had by independents. A statewide swing doesn’t tell you much about these local contests.

Some might suggest that at least part of the swing is due to other factors. I’ll talk about two that are sometimes discussed:

It is suggested that incumbent candidates receive an advantage in elections. It’s difficult to determine what the average advantage is, but it’s usually assumed to be between 1 and 2%. With so many Labor members retiring at this election (18 vs 7 Coalition MPs) you might assume that this could affect the magnitude of the swing. However in a 93 member house the effect would be less than a half of a percent and not influence the result. It could lead to the loss of a seat that Labor would otherwise be able to retain if the seat with a retiring member is held by a margin close to the predicted swing, though only Campbelltown would fit this description.

An increase in the Green vote
An increase in the Green vote (assuming it’s coming from what would otherwise be Labor supporters) would reduce Labor’s primary vote and potentially its two party preferred vote through exhaustion of preferences. However given that the Green vote appears to have collapsed almost back to it’s 2007 figure of about 9% (possibly from left voters switching back to Labor in a futile attempt to boost its primary vote) it’s unlikely that the small increase in the Green vote will increase the swing by much.

Aside from a record breaker, there’s nothing terribly exciting about the swing in this election. The ‘Bazslide’ (you heard it here first) is going to go down in the history books. I’ll know what I’ll be watching on election night – the Legislative Council and the local side contests.

Working Families?

With a state election in NSW in the next few weeks and a federal election one by-election away we’re again being bombarded with political commercials. Here’s a few for sampling:

NSW Labor’s “Fairness for Families”


NSW Liberal’s spoof of the above (which gets points for comedy)

And (because it features Zombies) my favourite from The Greens

Although you’d only know it from the Labor video “families” appear to be a strong theme for the campaigners, as has been the case in most recent elections. But how many families are there out there for these policies to target? Although many policies will benefit the whole population the benefits will vary.

Take Labor’s Fairness for Families promise of an extended and increased energy rebate. People on a variety of welfare payments (including the aged and disability pension) currently receive an annual rebate which in 2011/12 will reach $161. The policy promises a household with combined income of $150,000 a rebate of $250 a year. $150,000 is more than double the median household income in NSW. This means that a significant proportion of the better-off half of the population will receive $250 they didn’t have whilst someone on the aged pension get’s $89 more. That’s barely a third. So it’s quite clear who this policy is aimed at from a monetary perspective.

Looking at the campaign material targetted at ‘families’ generally shows mum, dad and the kids. How common are these households?

Let’s take a look at houshold composition. The source for this is the 2006 Census Basic Community Profile for NSW. Of the 2.3 million households in NSW they are made up of:

Single person households – 24%
Non-family group households – 4%
Couples without children – 26%
Couples with children – 34%
Single parents with children – 12%

So couples with kids seem to come out on top making up a third of all households. As the biggest group it probably makes sense to market more to them.

Hold on a minute though. How many campaign ads do you see showing the 23 year old uni student still living at home or a group of run-amok teenagers in the house? Most advertising material shows young children, very rarely older siblings.

Fortunately the census data breaks down further. Of those 34% of couple with children households only about half have all their children under 15. Let’s redo that table with this new information – I’ll call the new group “Working Families”:

Single people – 24%
Non-family group households – 4%
Couples without children – 26%
Working Families – 18%
Other couples with children – 16%
Single parents with children – 12%

The largest group of households is now Couples without children – and you don’t see many ads targetting them. Admittingly this is a fairly diverse group ranging from newlyweds to empty-nesters to retirees. Also couple households should have roughly double the number of votes of the singleton households. Unfortunately the Census doesn’t collect data on voting habits so we can’t know whether working families are swinging voters or not.

In any case the stereotypical ‘working families’ bloc is not nearly as large as most people think. With another five and a half weeks to the NSW election there’ll be plenty more fodder aimed at them though.