Friday, 2 October 2015

Marsden Maths

Motu's released the working paper they'd presented at the NZAE meetings earlier this year. They show that Marsden grants do increase research output. But as for whether the programme is cost effective? Well, that's more fun.

The average researcher on teams receiving Marsden grants made 6 proposals and received 1.2 grants over the period 2000-2012. The Fund allocated just under $68 million in 2013; the standard grant's maximum budget is $300,000 per year while Fast Start grants are limited to $100,000 per year. Panels reject 71-84% of first round proposals. The Motu analysis looks at second round proposals, 41% of which were funded.

Results? Here's the big headline chart from their press release.

So about the biggest effect that they found was that a researcher who received two successful research grants received a 10% increase in publications and a 14% increase in citations. In the regression discontinuity design, comparing project teams whose proposals just made the cut with those that just didn't, they found no effect.

Their abstract's version:
Overall, we find that funding is associated with a 6-15% increase in publications and a 22-26% increase in citation-weighted papers for research teams. For individuals, funding is associated with a 3-5% increase in annual publications, and a 5-8% increase in citation-weighted papers for 5 years after grant; however, the lag structure and persistence of this effect post-grant is difficult to pin down. Surprisingly, we find no systematic evidence that the evaluation of proposals by the Marsden system is predictive of subsequent success. We conclude that the Marsden Fund is modestly successful in increasing scientific performance, but that the selection process does not appear to be effective in discriminating among second-round proposals in terms of their likely success.
An annual 8% increase in the number of citation-weighted papers for five years after a grant might amount to one full paper over the entire time period. A research grant costing upwards of a hundred thousand dollars per year seems a pretty costly way of generating one citation-weighted paper.

It would be ... interesting ... to compare the cost-effectiveness of Marsden with an alternative scenario in which academics who maintained a reasonable publication record received an unconditional grant and the money spent on Marsden-related administration went instead into hiring more academics. There have to be at least 30 full-time equivalent academic-years that go into Marsden grant writing and grant evaluation.

Income confusion

Today's NZ Income Survey numbers have me perplexed. I've emailed asking them what's up, but here's my puzzle. Maybe one of you can help me out.

I keep median and average weekly earnings from salaries and wages floating around in my head because they're important touchstone figures that everybody working in policy should know. They give you a sense of proportion for things. The number I had floating around in my head for average weekly earnings from salaries and wages among salary and wage earners was $991. And that is the correct 2014 figure. Median was $863.

This year's release has a population rebasing taking into account the 2013 Census and they urge that current figures not be compared to prior ones because they've rebased things. Ok, fair enough.

This year's release has median weekly income from salaries and wages at $882, which is up $19 from last year. Last year's median figure was $863, so whatever's going on in the rebasing didn't affect the median figures. $863 + $19 = $882. A 2.2% increase.

The headline figures didn't include averages though, and they usually do. For those, you have to drill over to Tab 5 of the spreadsheet. Average wage and salary income for those in paid employment was $891. And they do not provide a rebased figure for 2014.

Whatever happened in the rebasing didn't affect median earnings at all, but knocked average earnings down by over ten percent. Average earnings didn't drop: all the hourly rates and such are increasing, and the average earnings from salary and wages across the entire population (not just those in employment) is up, but they don't give the version of average earnings that would allow a straight comparison of averages among those in paid employment.

I'm probably missing something obvious. But I just don't buy that there's only a $9 difference between median and average weekly earnings. If there is, well, even I've been massively overestimating the amount of inequality in New Zealand - and I've been complaining about other folks overstating it.

Update: StatsNZ replies:
"Regarding to your query, median weekly income from wages and salaries for those receiving income from this source was at $882 in the June 2015 quarter, compared with $863 in the June 2014 quarter.
The average was at $1,021 in the June 2015 quarter (which can be found in the supplementary table 4), compared with $993 in the June 2014 quarter.
The figures above are calculated based on wage and salary earners only.
 However, figures in table 5 are calculated based on the people earning income from either wages and salaries or self-employment, or both.
As the population is different, the average of $891 is different, compared with $1,021 mentioned before.
 Hope this could help answer your question."
So it's the relatively low earnings of those reporting self-employment that drag down the Table 5 averages.

3 Strikes

It looks like New Zealand's version of three-strikes for criminal offending is working. Here's Graeme Edgeler's analysis.
You need to be careful when crafting OIA questions around this sort of comparison, and I am relying on the Ministry of Justice to have correctly understood my intention. It is not enough to compare the number of convictions before and after the law change. Almost a third of convictions for “strike” offences since three strikes was enacted haven’t attracted first warnings because they relate to offending that occurred before the law change. You need to exclude similar offending in the comparison.

Between 1 June 2005 and 31 May 2010, 6809 people received convictions for strike offences that occurred between 1 June 2005 and 31 May 2010.

Between 1 June 2010 and 31 May 2015, 5422 people received convictions for strike offences that occurred between 1 June 2010 and 31 May 2015. So strike crime is down around 20% since three strikes came into effect. Claiming cause and effect over something like that is the type of intractable debate that you get into over the effect of longer prison sentences. But what we are looking at is not the general deterrent effect of three strikes (fear of punishment in the public at large), but specific deterrence: fear of punishment by those who have a conviction for strike offending who have been personally warned by a judge that further strike offending is treated very seriously. And that is where we can check the comparison between the five years before three strikes and the five years after it.

We know there were 81 second strikes in the first five years of three strikes. These are people who have been convicted for committing a strike offence after the law came into force, and subsequent to that conviction, been convicted of a further strike offence, itself committed after their earlier conviction occurred. The pre-strike comparison therefore needs to be people convicted of an offence committed after 1 June 2005 (but before 31 May 2010), who were then convicted before 31 May 2010 of a further offence committed after that conviction. And it turn out that that number is a lot higher. Had the three strikes law been in place on 1 June 2005, the following five years would have seen 256 offenders receive second strikes.

Now, strike crime is down in general, but the ~20% fall in strike offending is dwarfed by the ~62% fall in strike recidivism.
And here's where it gets really interesting:
Because of the lack of retrospectivity in our three strikes law, two people convicted on the same day, in respect of the same charge can have different strike consequences: someone convicted for offending that occurred after the law came into force receives a strike warning, but someone convicted of an offence committed before the law was enacted receives no warning.

A comparison between these two groups may help confirm or quash the alternative hypothesis that some change in treatment is the cause of the substantial reduction in strike recidivism.

In the first 4 years and 7 months of three strikes (curse you tier one statistics!), 2437 people have been convicted of strike offending that did not result in a strike warning, and of those, 360 had subsequently earned a first warning for an offence committed after that conviction. That’s a strike recidivism rate over 1000% higher among those who didn’t receive a warning than those who did. Of course, this direct comparison is misleading, as the post-strike convictions for pre-strike offending will be front-loaded, occurring on average much earlier in the ~5 year period since three strikes was enacted, and thus allowing more time for strike-level recidivism to occur. However, it remains useful, as it provides evidence to negate the alternative explanation for the pre-strike/post-strike comparison of much improved recidivism treatment.

And that is what we are left with: in the first five years of three strikes, there were 81 second strike convictions. In the five year before three strikes, there would have been 256.

81 second strikes seemed low. Now we know it is.
It looks like everything is there for a proper full difference-in-difference analysis, but ocular least squares here seems to be pretty strong.

A full set-up would compare the year t+1, t+2, t+3 and so on recidivism rates of those convicted of a first strike offence for offending that happened before and after the regime and of other offences before and after the regime.

Were I at Canterbury, I'd be begging Graeme for his data as basis for setting a Masters thesis and putting in the OIA requests for the finer-grained data that could let it happen.

Thursday, 1 October 2015


An interesting Twitter exchange between the NZ Classification Office and Graeme Edgeler:
Some people are cut out for jobs like that; I can't imagine doing it.


The correlation between alcohol consumption and any particular disorder matters a lot less than the link between alcohol consumption and overall mortality. Anti-alcohol folks like to talk a lot about that alcohol consumption increases cancer risk (while ignoring that moderate consumption reduces stroke and heart disease); the other side downplays the cancer link while highlighting the cardiovascular benefits.

They're both real, but what matters is the net effect: the J-curve. People who drink about a standard drink per day are about 14% less likely to die than are never-drinking teetotallers,* and the benefits of moderate consumption wash out by the time you're consuming about 3-4 standard drinks per day. And the risks mount from there.

Another study is out on the cardiovascular benefits of moderate drinking. There are some oddities in there, like finding differences across different types of alcohol. Rimm & Moats' 2007 study remains the most convincing: they restrict things to a sample of healthy adults with few potential unobserved confounds and find that moderate drinkers' relative risk of chronic heart disease was 0.38. Drinkers' chances of getting chronic heart disease are less than half of abstainers' risk. They conclude:
The evidence discussed above provides substantial support for the hypothesis that moderate drinking reduces the risk of CHD. Beer, wine, and spirits all have demonstrated significant benefits. These benefits are likely mediated through strong and lasting effects of alcohol on HDL cholesterol, fibrinogen, and glycemic control. The ‘‘sick-quitter’’ hypothesis and the concern that moderate drinkers lead a healthier lifestyle may explain a small proportion of the benefit attributed to alcohol in some studies, but recent studies which have removed sick quitters, updated alcohol and covariate information on diet and lifestyle factors, and separately documented benefits of alcohol among healthy and unhealthy populations further add to the evidence that moderate alcohol consumption is causally related to a lower risk of CHD.
So it is interesting that the New Zealand Heart Foundation's Heart Age Forecast Tool doesn't ask about your alcohol consumption. If you can cut your risk of chronic heart disease by more than half by drinking moderately, shouldn't that be in the Forecast Tool? The Heart Foundation tries to downplay the effects of moderate alcohol consumption on CHD, saying that the benefits don't hold for everyone. But if most of the benefits come from drinking in middle age, and the forecast tool asks your age...

* Update: the relative mortality risk of moderate drinkers as compared to never drinkers is 0.86. So whatever your risk of dying if you're a non-drinker, moderate drinkers' risk is 86% of that.

Tuesday, 29 September 2015

Planning externalities

The Finance Minister's speech to the Victoria University Business and Investment Club last week was excellent. I was not in attendance, but the text is now available online.

After running through a litany of areas where the application of market signals through policy reform improved outcomes - from agriculture to energy, he turns to housing. He walks through how urban planning regulations have jacked up housing costs, delayed new construction, worsened housing price cycles, hurt economic growth, driven inequality and cost the government plenty in state housing and accommodation supplements.
So these are the reasons why the Government pays attention to the housing market and issues stemming from poor planning.

For those among you who are economists, I would go so far as to say that while the justification for planning is to deal with externalities, what has actually happened is that planning in New Zealand has become the externality.

It has become a welfare-reducing activity.

And as with other externalities, such as pollution, the Government has a role to intervene, working with councils to manage the externality.

We're starting to get analysis that shows planning’s costs.

Too often the discussion about how our cities are planned is couched in vague terms of general good. "A world-class city." "Quality urban design." "A liveable, walkable city."

Those are all desirable, achievable objectives. But we need to understand the costs of achieving them so that we can make the trade-offs transparent.
I love it when people rediscover Tullock's 1998 argument that policy itself generates external costs that are underappreciated - which was of course simply an extension of his 1962 work with Buchanan on external costs and decision-making costs in The Calculus of Consent.

And it gets better:
As we get more information about what actually happens, often we find planning doesn’t achieve what people think it is achieving.

Planners and councils have a very difficult job in planning our urban areas.

Cities are incredibly complex systems. They are the product of millions of individual choices.

The idea that a small group of people could understand what choices we're making is asking too much of them.

Not because they are in any way incapable. But because the task is overwhelming.

The Auckland Unitary Plan is 3,000 pages long.

It's trying to regulate everything from the size of bedrooms to biodiversity in the Waitakere Ranges. No one person could possibly understand all the trade-offs in that plan.

Which means many of its effects will be certainly be unintended.

Planners can’t know everything – so of course they can’t be perfect in making trade-offs on our behalf.

Successful planning requires an understanding of its own limitations.
And then we come to the meat:
The funding base for councils is increasingly people on low fixed incomes. That is a product of an ageing population.

So you can understand why councils are under pressure not to expand if they think an expanding city is going to push up rates for existing ratepayers.

Councils need clear funding models so that development worth having can occur and future homeowners and current renters who might want to buy are taken into account.

So that's a brief overview of how important it is that housing is regulated in a way that enables flexible supply, and I hope some indication of the progress we're making.

That progress is necessarily slow, because these issues are complex.

If we better understand the economics of what is happening we can make better choices about housing regulation.

And that depends on one of the most important parts of public policy, which is the institutional arrangements by which those decisions are made.

That means looking at the incentives confronting an individual sitting in a council when making a decision about whether to allow a new subdivision.

We need to understand the incentives councils are reacting to.

Next month the Productivity Commission will produce a further report on the regulation of land supply. It will be another input into further, ongoing improvements in this area.

And we are seeing new thinking on a range of issues affecting housing, including from councils.

Often politicians are accused of being focused on the short term. That’s one of the reasons this issue has never been dealt with properly in the past.

The Government is taking a long term view.

All of the things I've talked about today will take 10 to 15 years to sort out.

So it’s important that a broad group of people understand our single-biggest asset class – the most-important asset most of us will own – how is valued, how it is regulated, and how it can contribute to our general welfare.
The New Zealand Initiative's report on policy trial zones, coupled with improved council financial arrangements to strengthen incentives, will be coming out 19 October.

I hope to see you at the report launch.

The Canadian Leaders' Debate - that could have been

Here's Chris Selly summing things up: Here's what I'd have loved to have heard instead, from any party leader:
“Right now, Canadian dairy prices are much higher than they need to be. Mothers pay too much for infant formula; families pay too much for cheese. And the system as a whole doesn’t even benefit dairy farmers any longer: getting into the industry is expensive because buying quota eats up whatever benefits the system provides to farmers. But there is a better way."

"We are committed to protecting the quality of dairy products on store shelves – as we are with every food product sold in Canada. But we don’t protect food quality with 300% tariffs for vegetables, fruit, or thousands of other products that cross our borders each and every day. For that, we use food inspections. The dairy quota system isn’t necessary for protecting food quality.”

“Today, we are buying back all of the dairy quota and opening the borders. Farmers should not see their retirement savings wiped out by a policy decision from Ottawa. We are able to afford to do this because dairy prices, in a competitive world market, are low enough that we can fund the buyback with a levy on all dairy products sold in Canada while still keeping prices lower than they are now. And those levies will disappear when the bill is paid in full. Canadians will have better access to the world’s products, and Canadian agricultural producers will have better access to world markets.”
It's simple. It would work. But Canada gets the government it deserves, and it gets it good and hard.