Several weeks in Graz

I realised it’s been over a month since I started my secondment in Graz and thus I should get a blog post done for the TEAM blog. So my first several weeks here have been fun so far, with initially rather British style weather giving way to constant sunshine and temperatures over 30 degrees C! I’ve explored Graz (see photos here and here) and discovered new beers (Murauer being my favourite so far) and everyone has made me feel very welcome here. I’ve even learned a small amount of German, but mainly for doing things like ordering food or saying “ich spreche ein kleines bissche Deutsch”…

I’m working on deduplication of bibliographic records, with the aim of this work feeding back to Mendeley for an improved version of the catalogue. Specifically I’m developing ground truth datasets that can be used to train machine learners to deduplicate and also developing tools to analyse the clusters produced by the deduplication system, e.g. to enable analyses of the mistakes the system makes. I will also finish off work started at Mendeley on improving the efficiency of the deduplication system.

So far the main achievements I’ve made in this work are:

• Generating a deduplication dataset consisting of 400K documents, with 500K duplicate pairings and 500K non duplicate pairings derived from the arXiv data we imported into Mendeley and user added documents carrying arXiv identifiers. The code will later be adapted to generate another ground truth data set derived from the PubMed data we’ve imported and user documents carrying PubMed IDs.
• Obtaining the distribution of similarities between user added documents carrying arXiv or PubMed identifiers and the imported metadata from these sources in Mendeley’s catalogue. These distributions suggest that when user documents carry these identifiers the metadata is usually very similar to the imported metadata, and this is particularly true for the PubMed data.
• Creating tools for analysing the results of clustering such as evaluating how well documents carrying arXiv or PubMed identifiers have been clustered (based on assuming the arXiv or PubMed identifier is correct, or at least correct when the metadata is validated against the imported metadata from arXiv/PubMed), inter cluster similarities and cluster purities.

I’ll be moving on to performing experiments using the arXiv data set, whilst constructing a PubMed dataset to work with as well. Also, I intend to swap data with my fellow TEAM project member Ago who is doing deduplication in a different field.

Does AV make some votes count more than others?

But, in many ways, your vote alone is not enough. In the days and weeks ahead, you also need to challenge your friends and colleagues who are thinking of voting “Yes”. Ask them why they are voting for it. I bet you none of their arguments will stack up – and you need to take them on.So when they say: “AV will make every vote count”, tell them it won’t. It will actually make some people’s votes -especially those who vote for extremist parties – get counted more than others. [David Cameron arguing against AV in the London Evening Standard].

Is the Prime Minister correct with this line of argument? I don’t believe so. A clue to the argument I’m about to advance that Cameron is incorrect lies in the fact that AV is also referred to Instant Runoff Voting.

In full (i.e. non instant) Runoff Voting, what happens is that if in the first round of voting, no candidate obtains a majority of the vote, the lowest placed candidate is eliminated and another round of voting occurs. This process repeats until a candidate obtains more than 50% of the vote.

AV/Instant Runoff Voting aims to approximate this process by replacing multiple rounds of voting with multiple rounds of counting. If you assume that voters who voted for a candidate in an earlier round will vote for the same candidate in later rounds unless that candidate is eliminated and that voters for eliminated candidates will redistribute votes according to their preferences, then by asking at the outset what those preferences are, and then performing multiple rounds of counting until a majority winner is found, you can eliminate the need for several rounds of voting but can still retain the essential point of Runoff Voting, namely trying to achieve majority support for the winner (admittedly the majority is only of those votes making it through to the final round).

Note that in full Runoff Voting, every voter gets a chance to vote in every round, and all votes count equally in each round of voting. By comparison, under AV, every voter gets a chance for their vote to count in each round (by stating their preferences should their preferred candidates get eliminated) and each vote counts equally in each round of counting, it’s just that some votes are first preference and some votes are later preferences, just as is the case under Runoff Voting.

Cameron’s argument implicitly ignores the fact that the votes obtained for a candidate on first preference in the first round get carried through to the next round unless and until that candidate is eliminated, which is the equivalent of voters sticking with the same candidate through multiple rounds in run-off voting. These votes are in fact counted once in each round just like every other vote is.

It thus seems to me that the argument Cameron advances is flawed, equivalent to claiming that full Runoff Voting makes the votes for the lowest placed candidates count more than higher placed candidates when it clearly doesn’t. Each round of voting/counting treats each vote equally under both systems.

The main thing missing from AV that you get in full Runoff Voting is actually the opportunity for voters to change their minds from round to round, i.e. vote for A in round one, but for B in round 2 despite A still being in the running. Runoff Voting thus allows voters to alter their preferences from round to round, but this opportunity is not available either in AV or FPTP, so is irrelevant to making a choice between AV and FPTP.

Will Alternative Vote lead to more coalitions?

British Prime Minister David Cameron claims that Alternative Vote (AV) could lead to coalition governments becoming the norm in British politics:

But there’s another argument I want to make – and one, again, you might find surprising coming from a Prime Minister leading the first coalition government since the Second World War: we shouldn’t vote for a system that could make coalitions the norm rather than the exception.

Cameron’s recent speech also claimed that AV would make coalition government much more likely:

The more people see a clear link between the pledges in a manifesto and the action taken in government, the greater the sense of accountability.

And the real, unavoidable truth about AV is that it would damage that chain of accountability…

…because it makes coalitions much, much more likely.

Furthermore, on Have I Got News For You this week, Tory MP Louise Bagshawe (about 16 mins in) claimed we’d end up with coalition governments under AV “every single time”.

The problem with this claim, is that the evidence for AV making coalition more likely than FPTP is inconclusive, but does suggest that a strong 2 party system makes coalition under AV unlikely, whilst under a multi-party system it will still tend to produce a majority for one party, even if coalitions occur more often in such cases. For example,  according to Channel 4′s Fact Check blog:

• Australia has been using AV since 1919 and, according to the IPPR, has only had 2 coalition governments since then (compared to 5 for Britain since 1900, and 12 for Canada since 1900 under FPTP). However it should be noted that this comparison is complicated by the fact that the Australian National Party and Australian Liberal Party have for decades fought general elections under a coalition agreement both in and out of power, but this agreement has operated to such an extent that they’re effectively treated as one party for general elections. If you treat them as 2 separate parties, then they had coalition governments from 1949 to 1972, plus after the 1980, 1998, 2001 and 2010 elections. Whilst this amounts to a sizeable period of coalition government, it’s not the norm, not as much as you’d get with PR, and certainly isn’t a coalition “every single time”. Also, it’s not clear whether FPTP would have led to fewer coalitions under circumstances where you have a multi-party system with 2 parties having such an entrenched agreement — they could equally have had a pact not to oppose each other in constituencies where this would “split the vote”.  It does suggest that coalitions and even long-running cross-party agreements may become quite common under AV when voters move decisively away from voting for one of two big parties, but then consider how often Britain had national governments and cross-party agreements in the 1920s and 1930s when the Liberals were split and in decline but Labour had yet to decisively supplant them — that was all under FPTP.
• The British Election Survey(BES) has run simulations of how British general elections would have turned out under AV since 1983, assuming voters preferences were the same as they were at each election. Only in 2010 does a coalition government definitely result (and we know FPTP delivered one then), though possibly we might have had one in 1992.

Of course assessing the likely outcome of switch to AV is inherently speculative, e.g.  the BES assumes voters preferences would be unchanged under AV, whilst Australia has a different political culture to that of Britain and the form of AV used there is different to that proposed for Britain making extrapolation from their experience rather uncertain.

It’s worth noting though that much speculation that we’ll end up with hung parliaments rests on the idea that the Lib Dems will benefit by being the second preference of both Labour and Tory voters, however given their rapid drop in the polls since forming the coalition with the Tories in 2010, that does not seem a safe bet anymore. Furthermore, under the projections of the BES, where the Lib Dems were often a second preference,  the Lib Dems did gain more seats but coalition rarely resulted partly because, in the simulation, AV actually increased the government’s majority in landslide elections such as that of 1987, 1997 and 2001.

In the end, given that AV will not produce a proportional outcome, and can result in amplified majorities in landslide elections in a similar manner to the way FPTP does, it seems to me that it is unlikely to make coalition government the norm, but it is possible it’ll make coalition occur more often if voters continue to vote in sizeable numbers for parties other than the 2 main parties. It’s hard to say whether FPTP would avoid coalition in such circumstances.

So what difference would Alternative Vote make?

On May 5th, voters in Britain will vote in a referendum to choose between sticking with the existing First Past The Post (FPTP) voting system or switch to using the Alternative Vote(AV).

Under the AV system being proposed, you vote by ranking the candidates in order of preference, and you’ll be able to rank as few or as many of the candidates you like. So, in a constituency where the Tories, Labour, Greens and Lib Dems were standing, a voter could put their first preference for say the Green Party candidate, their second for the Lib Dem candidate and ignore the Tory and Labour candidate.  When votes are counted, one or more rounds of counting are required:

• First of all the first preference votes are counted. If a candidate receives a majority of votes then that candidate wins.
• Otherwise, the lowest placed candidate is eliminated and their votes restributed according to the second preference for a second round of counting. This process repeats until a candidate receives a majority of the votes for a given round.
• Should someone who’s candidate has been eliminated not have expressed a preference to allow redistribution in a given round, their vote gets eliminated in that round.

The aim is to try and ensure that the winning candidate has over 50% of the votes cast, but in practice this can only be guaranteed if every voter gives a preference to every candidate. Australian elections often require this but the proposed British system will not. Thus it will be possible for someone to win without a majority of all votes cast. An extreme example is if all voters only give a first preference then a mere plurality of first preferences will suffice. More probable is someone’s first and second preference candidates being eliminated and their vote being set aside.

Nevertheless it can  be argued that AV gives the following advantages over FPTP:

• Only those who get a large majority of first preference votes can consider their seats “safe”.
• Voters can vote for their favourite party on the first preference and use their second preference to avoid “splitting the vote’. E.g. a UKIP voter could vote UKIP for first preference with the Tory candidate as a second preference to ensure Labour don’t get in, where under FPTP they might let Labour in by voting UKIP rather than Tory.
• Would-be MPs would need to broaden their appeal since simply obtaining a purality of first preference votes is unlikely to be sufficient to win, they will need to campaign for second and third preferences. This would reward MPs who develop strong constituency backing beyond their party boundaries and thus encourage more independence of MPs from parties.

For these reasons, I believe AV will be an improvement over FPTP and will give voters more power over the final outcome.

However it can be argued that AV might not really make much difference, for example John Curtice argues that the following points suggest AV might not make much difference:

• most of the time people are likely to put their first preference down as the same party they would previously have marked X against.
• even where 2nd and subsequent preferences are counted, because only the preferences of the lowest place parties are redistributed, there’s a limit on the impact of such redistribution. E.g. the two most popular candidates get 70% of first preferences between them, that only leaves 30% for redistribution.
• in the end AV is not proportional and will still tend to reward large parties as FPTP does.

Whilst in the short-run, the impact may be modest, I think this underestimates the long-run impact of a switch to AV for two reasons:

• Curtice bases much of his argument on studies of voter behaviour in mock AV elections. However as voters get used to how AV works they may well alter their behaviour, making more effective use of the second and subsequent preferences over time. Extrapolating behaviour from voters in mock exercises does not cater for this possibility.
• The incentives for politicians will change, they will have to appeal to a broader range of their constituents than before in order to win and keep winning. They will also be disincentivised from attacking a section of their electorate lest they need the second preferences to win.

Thus, I think AV will lead to a long term evolutionary change in how British elections are fought towards more broad based appeals and less divisive politics, albeit still within a majoritarian framework where large blocks of MPs form a government after each election.

That said, AV does have many flaws, e.g.:

• It’s only the second and subsequent preferences of eliminated candidates that get taken into account in each round and the second and subsequent preferences of the two most popular candidates on first preference will only rarely get used.
• It is also quite possible that a larger majority would exist for e.g. the candidate placed first on first preferences if you took all second preferences into account than for the eventual winner.
• Finally, tactical voting is not eliminated under AV, where there may be occasions where it makes sense to put your first preference tactically against a party in order to increase the chances of defeating your least favoured party or even to increase the chances of your favoured party winning. That said, it’s difficult to predict how such tactical voting will work out in practice and there’s less need of it under AV than under FPTP.

 

 

Eulogy for an old friend

This week I found myself thinking about a friend I met during my PhD at the University of Birmingham, Dr Jonathan Maxwell. I met Jon (as he preferred to be called) when he was still an undergraduate via the karate club I joined there. For several years we’d been friends, but a few years after I’d left Birmingham, around 2002 or 2003, we lost touch: the email I had was no longer valid and he’d moved on from Portsmouth University where we’d last been in contact.

On Thursday I decided to try and get back in touch. A few years back I’d come across one of his papers (he’d gained his PhD in Sports Psychology, studying implicit learning) during a search for papers on machine learning. This paper indicated he was based at the Institute of Human Performance at the University of Hong Kong. I looked up the web site of the institute but he wasn’t in the staff list. I double checked the paper and he’d definitely been there.

I figured he must have moved on again, but searching for him or his papers or looking at his co-authors’ institutions and checking to see if he’d moved to one of them shed no light on the matter. Eventually I did an image search and I found a picture of him! His hair had turned grey and had receded a bit but it was definitely him. So I clicked on the picture…

The link took me to his obituary at the Institute of Human Performance. He’d died in January 2009 at the age of 39, leaving behind his pregnant wife and two sons. I was, of course, shocked and saddened to hear this. Someone only a few years older than me, who’d been a good friend of mine at Uni and who kept himself in good shape, had died suddenly and unexpectedly. So many thoughts and questions have gone through my head since finding out, but in the end here is what I have to say:

Jon, I’m sorry we lost touch and I’m sad you’re gone.

I’d like to tell you about the things I’ve been up to since we last saw each other, when I stayed for a weekend and we watched Gladiator on DVD at your flat in Birmingham with a meal (pizza or curry, I can’t quite remember) and, of course, a few beers. Could that really have been 10 years ago?!

I’d love to tell you about my 2 years in Groningen, the Netherlands, using neural networks to model an aspect of language acquisition, whilst in my spare time researching Dutch and Belgian beers, and even learning some Dutch!

I’d love to tell you about my return (for 5 years) to my home town of Glasgow, including buying a flat there and getting involved in a political campaign, NO2ID.

And of course,  I’d love to tell you about my present life in London working for an exciting start-up company.

I’d love to hear about your life and work in Hong Kong.

I’d love to have met your family and friends there.

In other words, I’d love for us to catch up and share some beers over a curry again…

However, now it’s too late, and I am sorry we don’t have that chance now.

I’m sorry it was over 2 years before I found out you’d passed away and I’m sad you’re not around for your wife and children.

I am sorry you passed way due to a rare medical condition, one which physical fitness and general good health are no protection against. It was a condition you didn’t even know you had, one that is difficult to diagnose. Sometimes  life  really  sucks! But on the other hand…

I enjoyed the curries, the beers and the bottles of wine (you chose the latter particularly well!) we shared.

I enjoyed the nights out we shared at pubs and clubs in Birmingham and watching the 5 nations (for it was 5 back then!) with you, even when you teased me about how England would beat Scotland (I hate to admit you were usually right!)

I apologise again for the black eye I accidentally gave you when we were mock sparring on the dance floor at the Guild! (How drunk were we?) Black-eye aside, that was a good night, one of many…

I was also happy to have been a subject in one of your PhD experiments on implicit learning and found the work you were doing very interesting.

So thankyou Jon, for some great times and good memories.

I am glad to see that you established a strong academic career with a solid record of publications in your field.

I am glad you found someone to settle down and start a family with, and had the chance to experience the sheer joy I’ve seen in the friends and relatives of mine who have also started families.

I’m glad you and your wife made a big impression with the people you met in Hong Kong, and I’m glad that you made good friends there who, I was so moved to learn, looked after your wife and family in their time of need after you were gone.

I am glad that, in the time you had, you made a good life for yourself and your family in Hong Kong and that it looks like you were very happy there.

For if there is a lesson to be learned from your life, it is that we should strive to make a good life for ourselves and those we care about and to make the best use of our time in this world not just because that time is limited and we don’t know when it will be up, but also because we can have a good life and we can be happy if we work at it.

So finally, it’s time to say goodbye Jon. You were a thoughtful, generous person and it was my privilege to have been your friend.

The impact of the coalition’s plans on Britain’s debt

When considering the impact of the coalition’s spending plans it’s worth noting the impact they claim it will have on the Britain’s debt. There are 3 key figures to consider:

• The total public debt. I.e. the total amount the government owes. According to the June budget, the official public sector net debt stood at 53.5% of GDP in 2009/2010 and was projected to rise to 74.4% of GDP by 2014/2015, had we stuck to the previous government’s plans. Under the coalition’s plans, the figure for 2014/2015 is just under 70% of GDP by 2014/2015, peaking at just 70.3% the previous year and falling to 67.4% by 2015/2016. In other words, even with the additional reductions in planned spending and increases in tax, the UK’s debt as a proportion of GDP will still grow for the next 4 years.
• The budget deficit. This is the shortfall between total public spending and tax revenues for any given year. Again according to the June budget, the public sector net borrowing stood at 11% of GDP in 2009/2010, and was projected to fall to 4% of GDP in 5 years time under the previous government’s plans. Under the coalition’s plans, the deficit is projected fall to 1.1% of GDP by 2014/2015, and the “cyclically adjusted” deficit is project to be eliminated that year and go into surplus in 2015/2016.
• The interest paid. This the minimum amount the government has to spend to service the debt (more needs to be spent to reduce it). This figure was projected to rise from £30.3 billion in 2009/2010 to £67 billion by 2014/2015 under the previous government’s plans. It is now project to rise to £63 billion in 2014/2015 and £66.5 billion in 2015/2016.

Thus the difference between the coalition’s plans and the previous government’s plans is that the former would see the deficit eliminated and the debt as a percentage of GDP starting to fall by 2015/2016, where both would still be increasing under the latter. And these figures are all dependent on economic growth projections being realistic. If growth is slower than expected, then eliminating the deficit within the current timescales will require even larger reductions in planned spending or bigger increases in taxation than we’ve seen thus far.

On the size of the coalition’s spending cuts

The home page of the TUC‘s campaign against the coalition’s spending cuts describes the campaign as being “against the Government’s deep, rapid cuts in public spending” (emphasis added). A blog called “A Thousand Cuts”, dedicated to critiquing the coalition’s plans has “The slashing and burning of Britain’s public services” as it’s sub-title. The BBC claims the Institute of Fiscal Studies described the cuts as “longest, deepest, sustained period of cuts to public services spending at least since World War II”.

In other words, we’re being told by various sources that the cuts to public spending are going to be larger and more severe than we’ve seen in recent decades. But what are the numbers behind these claims?

The Adam Smith Institute (ASI) summarises the figures on total managed expenditure as follows:

Year	TME (£ bn)	Nominal change (%)	Real change (%)
2010-11	696.8
2011-12	699.8	+0.43	-1.54
2012-13	711.0	+1.60	-0.39
2013-14	722.0	+1.55	-0.44
2014-15	737.5	+2.15	+0.14
2015-16	757.5	+2.71	+0.70

Meanwhile they summarise current spending (covering the vast bulk of “front line services”) as in the following table:

Year	Current Ex. (£ bn)	Nominal change (%)	Real change (%)
2010-11	637.3
2011-12	651.1	+2.17	+0.16
2012-13	664.5	+2.06	+0.06
2013-14	678.6	+2.12	+0.12
2014-15	692.7	+2.08	+0.08
2015-16	711.4	+2.70	+0.69

From these tables we see that under both measures of spending, the budgets will increase in nominal terms every year, and current spending will increase overall in real terms. Note that these figures come from the budget documents themselves and assume inflation of 2%. The ASI conclude:

Now, OK, these are not exactly big rises – but nor are they swingeing cuts that will (a) have any significant effect on the economy or (b) on the public services-using population at large. What the coalition’s spending plans really amount to is a five-year, real terms freeze of current expenditure, combined with three years of significant falls in capital expenditure. The overall impact of that is a a very small, real terms drop in TME (roundabout 1.5%) between now and 2015-16.

I’m not so certain the cuts will be quite as small as that in real terms. Current official measures of inflation are significantly higher than 2%. According to the BBC, the Consumer Price Index was last measured at 3.1% whilst the Retail Price Index (which used to be the main official measure of inflation) was measured at 4.7%. At 3.1% inflation, the current expenditure of £637.3 billion would rise to £742.40 billion by 2015-16 in real terms. The projected £711.4 billion would thus represent a 4.18% cut in spending in real terms rather than a 0.69% increase. At 4.7%, the £637.3 billion would rise to £801.82 billion and the projected £711.4 billion would represent a more significant 11.27% cut.

All of these figures are a far cry from the picture of “deep, rapid” cuts or “the slashing and burning of Britain’s public services” offered by some. Of course, a freeze or small reduction in spending represents a major shift compared to the constant increases in spending under the previous government. Also, it may be that within individual departments there will be more severe cuts than this picture represents since e.g. health spending is being protected.

Nevertheless what we’re actually getting is something in between a freeze and a modest reduction in total spending in real terms.

On the impact of public spending cuts

Via this comment on a post at Liberal Conspiracy, I came across this argument from Anthony Trew against cutting quangoes that equally applies to cutting public spending in general:

With every quango abolished there will be a number of their employees losing their jobs. With every person losing their jobs there is an inevitable impact on their families. With every family affected there will be a lessening of the spending power of that family. With every family affected spending less money, local businesses will suffer. With each local businesses suffering, economies in whole areas and regions will be affected. With every local business affected every national business that supplies them will be affected too. With every business suffering, their profits will inevitably fall. With the profits of every business falling, tax receipts will fall too.

Given the above, how the hell will we not end up with a double dip recession?

The problem with this argument is that it considers only one part of the picture. One could equally validly argue against increases in public spending (and by implication against the existing public spending we have) as follows:

With every increase in public spending there will be a number of people paying more in taxes(*). With every person paying more in taxes there is an inevitable impact on their families. With every family affected there will be a lessening of the spending power of that family. With every family affected spending less money, local businesses will suffer. With each local businesses suffering, economies in whole areas and regions will be affected. With every local business affected every national business that supplies them will be affected too. With every business suffering, their profits will inevitably fall. With the profits of every business falling, tax receipts will fall too.

Given the above, how the hell will we not end up with a double dip recession?

(*) The government could fund things via borrowing, but that is really just deferred taxation.

In both arguments, the loss of spending power of those affected is claimed to undermine the country’s economic performance. In reality, assessing the impact of spending cuts on the economy will be a matter of balancing the positive impact of the reduced need for taxation  against the negative effect of making public sector workers unemployed and reducing or eliminating the services those workers were providing. If the former is greater than the latter, then the cuts will be a boost to the economy overall. If the latter is greater than the former, the cuts will undermine the economy.

Now I haven’t shown here that the government’s cuts will be either harmful or beneficial to the economy, my point is that simply looking at the reduced spending power of those made redundant via the cuts involves focusing solely on the negative impacts of the cuts.

The DNA database and Mark Dixie

Update: The Times also has a useful look at Labour’s claims on this issue.

The Labour Party continue to portray Tory plans to restrict the retention of DNA of those charged, but never convicted, of a crime as somehow being “soft” on criminals, citing the case of Sally Anne Bowman who was killed and raped by Mark Dixie:

Gordon Brown MP, Prime Minister and Leader of the Labour Party, and Alan Johnson MP, Labour’s Home Secretary, will today make a campaign visit to highlight the vital role that DNA plays in tackling violent crime and why Labour has been fighting Tory plans to downgrade the DNA database.

At the visit in Stevenage the Prime Minister and Home Secretary will be joined by Linda Bowman, whose daughter Sally Anne was murdered in 2005.

Sally Anne’s murderer Mark Dixie was convicted through use of the DNA register, having been arrested but not convicted in a pub brawl.

The problem with this line of attack is that the Tories’ plans would not have made any difference in Mark Dixie’s case. When he was arrested his DNA was taken and compared to samples from the Sally Anne Bowman case. There was no need to have his DNA on the database to do that. There was thus no need to retain DNA for those never charged or never convicted to solve similar such cases. All that’s needed is to have a database of DNA collected from crime scenes and to have a policy of checking arrestees’ DNA against that of old crime scenes.

Such an approach is surely a far more proportionate use of DNA, far more respectful of privacy whilst at the same time more focussed on solving crime than retention of the DNA of those never charged with a crime in the first place, or those who have charges dropped or are acquitted.

Meanwhile, Genewatch point out that many of Labour’s other claims about the DNA database have to be taken with a large dose of salt.

How to shoot yourself in the foot in election advertising

The Labour party are trying to put voters off from voting Tory with this:

Being compared to Gene Hunt, one of the most popular characters from recent TV shows, is actually likely to boost Cameron’s chances, not hinder them. It has also opened up an obvious Tory response:

Ouch!