Shooting for the moon

Did the gun buyback reduce gun homicides or suicides? According to a new paper in the British Journal of Criminology by Jeanine Baker and Samara McPhedran, it didn’t. (The two have been criticised for their affiliations to the gun lobby, but I don’t think that invalidates their study.)

The paper’s methodology starts from an important insight: since gun homicide and suicide rates have been trending downwards over recent decades, we should compare the efficacy of the gun laws not against the pre-1996 rates, but against the extrapolated trends.

Like the best papers, this approach can be shown with a simple graph. The dashed lines in the chart above (figure 4A in the paper) depict 95% confidence intervals for the trend in gun homicides from 1979-96, extrapolated forwards. The solid line is the actual gun homicide rate.

If the buyback cut gun homicides, we should have expected the solid line to break through the bottom dashed line. This would tell us that fewer gun homicides happened than the extrapolated trend. This didn’t happen, so Baker and McPhedran conclude that the buyback didn’t work. As Jeanine Baker said on the AM program yesterday:

In 1996 we were told that taking the… buying back those civilian firearms, off those licensed firearms owners would make society safer and it would reduce firearm deaths. The evidence isn’t there to support that.

But let’s just look at that graph again, and see what would have had to happen for the gun buyback to work. In 2004, the bottom dashed line hits zero. For the solid line to go below the dashed line, the gun homicide rate in Australia would have to be negative in 2004, and extremely low in earlier years. (Ironically, one of the reasons that their statistical test has so little power is that the mass shootings of the 1980s and 1990s made the gun homicide rate extremely volatile, and the 95% confidence band very wide.)

In other words, Baker and McPhedran have set the gun buyback an impossible test. Just because gun buybacks don’t lead to negative gun homicide rates, it doesn’t follow that they don’t work.

Update: Don Weatherburn has an oped in today’s SMH.

This entry was posted in Australian Politics, Economics Generally. Bookmark the permalink.

46 Responses to Shooting for the moon

  1. I haven’t read the paper, but does it distinguish between organised crime killings (where we can expect criminals to acquire guns, regardless of the law) and gun deaths (crimes of passion, suicide, accidents) caused by people who are generally law-abiding? The latter group would be the most plausible target of a gun buy-back.

  2. Andrew Leigh says:

    Andrew, they also look at gun suicide and gun accidents, and find a statistically significant reduction in the former but not the latter. They don’t break out organised crime killings, probably because there are so few of them in the data (the typical gun crime is husband shoots wife).

  3. derrida derider says:

    Its pretty obvious this method is extremely dodgy. Even just using the point estimates rather than setting confidence intervals, there’s no reason to assume a continuing extrapolation of the trend should be linear and every reason, if only on zero constraint grounds, to expect it not to be.

    If the paper does what you say, it is quite worthless. Given the lack of regional etc variation, the only reasonable way to do it is to estimate some structural model of the determinants of gun deaths (in which gun iownership is very likely to be an important variable, so there’s an endogeneity problem right there).

  4. Sir Humphrey says:

    At a glance, the disappearance of big spikes in the data after 1996 would seem to support the idea that the gun buyback was successful in stopping any more mass shootings. DD’s point is a fair one – who is to say that a linear extrapolation is a fair counter-factual anyway. My stats are pretty rusty, but shouldn’t we really be looking for a structural break?

  5. Sinclair Davidson says:

    While I like the result, I’m not convinced. Yes a simple before and after t-test is invalid, but the devil’s in the detail of the extrapolation. DD is quite correct, a time-varying model that explained the historical series projected into the future is the way to go.

    The interesting, if somewhat macabre, study is to compare the social costs of suicide technique. Is suicide by gunshot more or less socially costly than other techniques? If more costly, then we could argue some ‘benefit’ from gun control laws (even in terms of deterent etc). If less costly, then the murdr incidence becomes even more important.

  6. Steve Edney says:

    Surely you would plot the 95% confidence in percentage terms of the trend not in absolute terms? I’m a bit suprised by that.

    Even with some other criteria its difficult to see tht the laws have made a significant change in what was occuring anyway.

  7. Pingback: Guns and short memories « Criticality

  8. Andrew Leigh says:

    Steve, I agree. The rate must asymptote to zero. The log of the rate would be more sensible.

  9. Cathy says:

    The log of the rate would be an interesting one, although the data (as shown in the chart here doesn’t indicate it is starting to plateau). However, didn’t the authors use ARIMA, which is a time varying model explaining the historical series into the future?

  10. Robert says:

    Here is a direct link to the article, thoug it is pay-per-view.

  11. Grendel says:

    Sir Humphrey made a reference to Andrew’s observation about decreased volatility and the significance of that. The other interesting part of that graph is that before and after periods are of a different length – not really crucial from a data perspective given how the trend was derived but it does influence perception of what the data represents.

  12. As Cathy mentions, they did use ARIMA to crunch their numbers – I can retrieve this paper from behind the subscription wall, and I downloaded a copy yesterday.

    While it might be too strong to conclude from their data that the buyback didn’t work, equally there doesn’t seem to be any support in the data to suggest that the buyback did work either. The right conclusion to draw would probably be that homicide rates are too low, and jump around so much, that we simply don’t know whether it reduced overall homicide rates or not.

    My broader reaction is that the public desire for changes in gun laws was never about the overall murder rate anyway. It was about the specific fear of spree killings of strangers. It’s pretty much like terrorism or train crashes – the chances of it happening to any one individual are miniscule, but it scares the living daylights out of people. Hence governments take, with massive public support, what from a utilitarian perspective you might regard as ridiculously expensive actions to reduce already tiny risks. And it does seem to have worked in that respect.

    I happen to think that deaths that occur singly are just as tragic as deaths that occur in bunches; I suspect that $500 million spent on reducing post-operative infections, to take a well-known example, would have saved a lot more people. But most people seem to think that 10 deaths at once in a spectacular fashion is somehow worse than 100 mundane (but preventable and premature) deaths spread over a year.

  13. Christine says:

    Alright, I’ve played a bit with the data (from Table 1 in the paper). I’m not a time series econometrician, and I don’t use ARIMA models much, so treat this as speculative.

    1. I cannot get convergence of an ARIMA(1,1,1) model for the firearm homicide rate using data from 1979-1996 (estimated using: arima fh/100000 if year1997), I get very different results depending on AR/I/MA specification and other basic variables included.

    2. Why assume series are I(1)? (This is a consequence of ARIMA(1,1,1), which they say they use) V. hard to judge integration over 20 years, but over their longer data, certainly doesn’t look I(1). No formal tests of this proposition either (results below don’t back it up). Can’t see why wouldn’t go ARMA(1,1) with linear trend, though this also fails to converge if use 1979-1996.

    Here are STATA results for homicide rate: arima(1,0,1) with linear time trend (year) and post-97 dummy (break):

    ARIMA regression
    Sample: 1979 to 2004 Number of obs = 26
    Wald chi2(4) = 52.08
    Log likelihood = 29.16161 Prob > chi2 = 0.0000
    | OPG
    fhrate | Coef. Std. Err. z P>|z| [95% Conf. Interval]
    fhrate |
    year | -.0210047 .003401 -6.18 0.000 -.0276705 -.0143389
    break | -.0237764 .0789114 -0.30 0.763 -.1784399 .1308871
    _cons | 42.31227 6.768699 6.25 0.000 29.04586 55.57868
    ARMA |
    ar |
    L1 | -.205559 7.974017 -0.03 0.979 -15.83434 15.42323
    ma |
    L1 | .2326113 7.91586 0.03 0.977 -15.28219 15.74741
    /sigma | .0788227 .0124959 6.31 0.000 .0543312 .1033143

    For log firearm homicide rate:

    lfhrate | Coef. Std. Err. z P>|z| [95% Conf. Interval]
    lfhrate |
    year | -.0453633 .0109407 -4.15 0.000 -.0668067 -.0239199
    break | -.18068 .1584265 -1.14 0.254 -.4911902 .1298302
    _cons | 89.55273 21.78587 4.11 0.000 46.85321 132.2523

    How big is the break effect of -0.18, you ask? average annual change over the period up to 1997 was -0.015. To get a statistically significant (95%) result given the standard error of 0.158, we’d need break = -0.49, which would mean a reduction in deaths from the 1996 predicted value of -0.99 to -1.484 in 1997 and then down to -1.846 by 2004. In number of deaths, this would be: 45 in 1997 (actual: 75) and 31 in 2004 (actual: 32). In other words – we’re pretty much at the point where we’d have to say there was a break, but with a bit of delay in response (1997 still sucked). BUT: this is not good evidence of an effect, given possibility of other confounding trends/lack of a good control group/volatility of data.

    For firearm suicide rate:
    | OPG
    fsrate | Coef. Std. Err. z P>|z| [95% Conf. Interval]
    fsrate |
    year | -.0916758 .0143576 -6.39 0.000 -.1198161 -.0635354
    break | -.5633898 .1802771 -3.13 0.002 -.9167265 -.2100531
    _cons | 185.2159 28.51037 6.50 0.000 129.3366 241.0952
    ARMA |
    ar |
    L1 | .5524194 .6031163 0.92 0.360 -.6296667 1.734506
    ma |
    L1 | -.2843471 .7248501 -0.39 0.695 -1.705027 1.136333
    /sigma | .1769078 .0336132 5.26 0.000 .1110271 .2427884

    Log firearm suicide rate:
    | OPG
    lfsrate | Coef. Std. Err. z P>|z| [95% Conf. Interval]
    lfsrate |
    year | -.0441032 .0092716 -4.76 0.000 -.0622752 -.0259312
    break | -.284335 .0741948 -3.83 0.000 -.4297541 -.1389159
    _cons | 88.68672 18.43054 4.81 0.000 52.56352 124.8099
    ARMA |
    ar |
    L1 | .8596035 .3064051 2.81 0.005 .2590606 1.460146
    ma |
    L1 | -.4440248 .4169646 -1.06 0.287 -1.26126 .3732108

    Note: estimates on AR coefficients don’t give terribly much basis for notion of I(1) series, given short TS and likely low power.

    3. Looks to me as though their CI estimates for years after 1996 are wrong – shouldn’t they be expanding if they’re estimating an I(1) model? The CI for other murder rate (Fig 4b) does this, the others don’t.

    4. Why did they use 1979 as a start year? Looks from Fig 1 and 2 as if this is a local maximum in the firearm suicide and homicide rate, so you’re forced to conclude there’s a downward trend, which you wouldn’t if you’d used somewhat earlier data. Any of the pure TS models I’ve estimated for the 80s and 90s are overpredicting homicide and suicide rates in the 1970s and 1960s.

    5. Just scanning the data, I’m with Robert: the conclusion on homicides has to be it’s too murky to tell, which is emphatically not the paper’s conclusion: “When compared with observed values, firearm suicide was the only parameter [sic] the NFA may have influenced, although societal factors could also have influenced observed changes. The findings have profound implications for future firearm legislation policy direction.” Even less consistent with media reports: “Australia’s guns buyback has not reduced rates of gun murder or suicide, a new study says.” (The Age, over the net).

    We’re talking about 102 firearm homicides a year in 1979, down to 32 in 2004. Firearm suicides were 523 in 1979 and 168 in 2004. These are very important reductions in percentage terms, but less so in numbers. Questioning whether the policy was worth it is pretty easy (divide $500m by 600 people per year as an outside limit), even without knowing whether or not the policy was effective.

    Seriously, this paper is rubbish. Does the British Journal of Criminology have a good reputation?

    (Sorry – went a bit mad this morning. STATA dataset, built based on Table 1 in the paper, available on request, if anyone else has as great an urge to waste time as I do.)

  14. Ungrateful Troublemaker says:

    Excellent analyses.

    Just a few notes from a grumpy non-statistician …..

    Suicides: Shootings, hangings (whether vertical or assisted horizontal), leaps from high bridges and medication overdoses are rather obvious but what about death-by-motor-vehicle? An area that, for a lot of reasons, is notoriously underreported. Wonder if there has been a marked increase in this form of suicide since the gun buyback?

    The gun buyback: This was an opportunity not a response. The Port Arthur massacre was a terrible tragedy for the whole nation …. but it was an opportunity for the government and its pals to get most firearms out of the hands of ordinary citizens ….. and the extraordinary delays and peculiar manner in which it was inplemented would have given any incipient militias and would-be death squads all the time in the world to further arm themselves and to get their armories into secure hidey-holes. Just stick around; we may yet have a lot more statistics, slightly different ones, on deaths caused by firearms …..

  15. Pingback: More on the Guns Study « Criticality

  16. Pingback: John Quiggin » Blog Archive » Less than zero

  17. Cathy says:

    Well if it’s ‘no definite effect’ rather than ‘no effect’ then I’ll have to accept that the gun laws didn’t really make a difference and consider that it is time to see if those alternative programs the authors suggest are worth investigating are really worth investigating.

  18. MarkL says:

    “If the buyback cut gun homicides, we should have expected the solid line to break through the bottom dashed line.”

    This is not so, and the argument put forward here is not correct. Why “should” the actual rate have passed through the lower parameter of the 95% confidence band? What is being studied is the trend, and the trend shows that spending taxpayer’s money to purchase certain classes of firearms from registered and law-abiding sportemen and women has no impact on criminal activity – to wit, murder.

    It was pointed out at the time by the SSAA that SSAA members and registered gun owners were NOT over-represented in available statistics as committing murder by firearm.

    Oddly enough, most murders were committed by criminals or owners of unregistered firearms. And these were the people who DID NOT participate in the scheme – and that was known at the start.

    Fancy that, criminals not obeying the law!
    That is WHY they are criminals, something which would seem to have escaped the notice of the suipporters of this daft and wasteful scheme.

    The SSAA said at the time that the scheme would have no impact for the reasons above. It has now been shown to have had no impact.

    The only people surprised by this are anti-gun nutters who regard criminals as not responsible for their actions, and who seem to regard law abiding sportsmen who follow a sport the nutters do not like to be their enemy.

    The money would have been much more, INFINITELY more wisely spent on funding research into and support/treatment for teenage depression and suicide in adolescents. Instead it was wasted on a political feel-good program. Again this was one suggestion made by the SSAA, among many.

    I’m a rather bitter about that, having attended two funerals this year for teenagers who have committed depression-related suicide. Both were bright lads, doing well at school, and well regarded by their peers. Those boys may well have been aided (perhaps even saved) by additional funding put in to depression research, support and treatment.


  19. Paul G. Brown says:

    MarkL asks:

    ‘Why “should” the actual rate have passed through the lower parameter of the 95% confidence band? ‘

    I don’t think anyone is saying anything “should”. What is being pointed out is the impossibility of the test the authors propose.

    Put more formally, the proposition is “The gun buy-back changed patterns of gun-related deaths.”. If this is true (according to the authors) then the data (fat line) should bounce outside the 95% interval (dotted line). The authors point out that it does not, and claim that therefore the proposition is falsified.

    The response is to point out that there is no way, logically, for the data line to fall below zero (excepting some kind of gun-related impregnation – que Tom Wait’s story of the Civil War soldier, the musket ball, and the antebellum virgin birth). The more sophisticated response points out that of course you’re never going to get the rate of any social policy event below zero, meaning the trend line must be non-linear (asymptotic to zero). So, use the log value, rather than the absolute.

    When this is done, what you see (Christine’s analysis is pretty good) is some small support for the proposition. But when all is said and done, the only reasonable thing to say is that the data is very murky on the question.

  20. MarkL says:

    Fair enough, Paul.

    Let us hope that the next time we see a useless and wasteful ‘political feelgood program’ like that, the lessons of this one are not forgotten.

    There are much better things to spend nearly half a billion dollars on. I’d plump for youth mental health being the best recipient. (But I’m obviously boased that way right now by recent events.)


  21. Simon Chapman says:

    As you might expect, I’ve been over this paper thoroughly, with biostats colleagues from my School. It is an object lesson in (a) how to answer the wrong question (b) how to use the wrong method to answer the wrong question.

    When John Howard introduced the gun laws in 1996, he didn’t say “we are introducing these new laws to reduce the incidence of domestic violence homicides”. He didn’t say “ we are introducing these new laws to reduce the incidence of criminals shooting each other over drug deals.” He didn’t say we are introducing the laws to reduce gun suicides ((average 79% of all gun deaths)

    In fact this is what he said: .“How is it that weapons of this kind are still available and what earthly purpose is served by their free availability?”

    “There is no legitimate interest served in my view by the free availability in this country of weapons of this kind,”

    And why did he say that? He said it because Martin Bryant, who had no previous criminal record, had just used a semiautomatic weapon killed 35 people. He said it because Wade Frankum, a licensed shooter with no criminal record and no history of mental illness had killed seven and wounded another seven in Strathfield Plaza shopping centre in 1991. He did it because the Queen Street and Hoddle Street gunmen in Melbourne had killed 16 people and wounded another 24, using the same military style semiautomatic weapons.

    Howard introduced the gun laws to try and reduce the incidence and gun massacres, by outlawing civilian ownership of the rapid-firing weapons that are favoured by those wanting to kill lots of people quickly. There had been 11 mass shootings (5 or more dead) in the decade up to & including Port Arthur.

    Now, if you look at the paper by Baker and McPhedran, there is no mention whatsoever — not one sentence — examining the question of did the 1996 gun laws do anything to change the incidence of gun massacres like these. In the 10.5 years afterward, there have been no massacres. But somehow this paper by people from the gun lobby, completely fails to mention this.

    If I can use an analogy here, imagine that there have been a large number of railway crossing car smashes. Imagine the government said “we can do something about this, and reintroduce alarmed barriers on every road and rail crossing in Australia.” 10 years down the track, some motivated researchers like Baker and McPhedran come along and publish a paper where they look at whether the introduction of the barriers reduced the overall road toll throughout Australia. Now, that’s a legitimate question to ask. But — hello! — the more fundamental question is did the introduction of barriers reduced railway crossing deaths, as planned?

    People would think it very peculiar indeed, if such a paper failed to examine that question. So, am I the only one thinking that it’s very peculiar indeed that Baker & McPhedran’s paper conspicuously avoids answering the question that should be centre stage in such research?

    The road toll, like the death toll from violence, is influenced by a multitude of factors. The gun lobby has a total blind spot about the role of guns in violence. I actually think it is worse than that: they in fact in this exercise are producing research which would appear to be deliberately avoiding giving any consideration to whether the gun laws which took 640,000 semiautomatic weapons out of the community might have anything to do with the fact that we haven’t seen a gun massacre using these weapons in 10 years.

    As others have pointed out already, the method they chose to examine the secondary question of whether the gun laws reduced total gun deaths, was the wrong one.

    There are two main problems. One is that calculating mortality rates and then treating them as a number in a time series ignores the sampling variability inherent in the Poisson counts that make up the numerator of the rate. Another is that the Box-Jenkins class of models, including the ARIMA model used by Baker and McPhedran, is unable to explicitly address the effect of an intervention such as the introduction of gun laws. One is reduced to comparing the mortality rate expected under a model assuming no effect of the intervention with that observed in the post-intervention period. This is however an insensitive approach and its interpretation is not based on formal statistical inference but rather on visual inspection and qualitative interpretation of graphs which may be prone to selectivity.

    We’ve got a paper under review with a public health journal where we use negative binomial regression to look at whether there was any effect on total firearm deaths (yes), firearm suicide (yes), firearm homicide (yes, but low power didn’t allow significance).

    The most amusing part of all this is to read the gun lobby blogs lapping up this very poor study, and to Samara McPhedran trying to squirm out of having avoided the “elephant in the living room” effect of seeing gun massacres disappear in Australia over the past 10.5 years. She’s debating me & Don Weatherburn on Radio National’s Law Report this coming Tuesday. Unmissable, if for nothing more than listening to her auto-pilot voice. BTW, the British Journal of Criminology, typically described as “prestigious” by their cheer squad, has an impact factor a tad above 0.6 — meaning that the average paper gets cited by one person every two years. This one should lift it considerably, because very bad papers are often highly cited.

    Simon Chapman (Uni Sydney)

  22. Simon, in turn, you’ve ignored the point that several people here have made – to put it bluntly, was spending $360 million to avoid the occasional tragic but very rare gun massacre a worthwhile use of public funds?

    And, frankly, a lot of the weapons banned did have perfectly legitimate uses that don’t involve aiming them at people. A .22 semi-auto was a very popular rabbit gun, for instance.

  23. Michael Moriarty says:

    Any form of Gun Control is good in my humble opinion. At the risk of looking like Treckie (or in my case a Wingie) another interesting observation from The West Wing:

    Writer : Aaron Sorkin

    Spoken By : Richard Schiff (In reasonably sure) :

    “If you combine the populations of Great Britain, France, Germany, Japan, Switzerland, Sweden, Denmark and Australia, you’ll get a population roughly the size of the United States. We had 32,000 gun deaths last year. They had 112. Do you think it’s because Americans are more homicidal by nature? Or do you think it’s because those guys have gun control laws?”

    Fair Point I would have thought.

  24. Simon Chapman says:

    Hi Robert — you are entering the realm of how to value social policy — there are many aspects of government funding that I object to (eg: today’s announcement that $90m will be spend on school chaplains; $500m on the “fight against drugs”; detention centres etc etc). In a democracy, we are able to voice support or objection to all of these things, but at the end of the day gvts have to make decisions.

    Each time I pass through an airport, I get scanned & frisked as if I am a potential criminal. I have no objection at all about the government spending huge amounts to prevent what are plainly rare events, just as I (and most of the community) welcomed the gvt adding a small levy to my tax to pay for the buyback.

    There are lots of ways to control wabbits other than to allow armies of .22 armed shooters at them.


  25. M. Anko says:

    Hang on a minute. First Simon Chapman, former
    convenor of the coalition for gun control who pushed
    for the laws, says that Baker and McPhedran ask the
    “wrong” questions, but then he goes on to say that he
    has a paper under review asking the exact same
    questions about whether firearm suicide and firearm
    homicide declined more quickly post-laws.

    If Baker and McPhedran are asking the “wrong”
    questions, so too is Chapman. Alternatively, if he is
    asking the “right” questions, so too are Baker and

    More interesting though is his statement that, using
    different statistical methods, he arrives at the same
    results as Baker and McPhedran:

    1) An acceleration in the rate of decline in firearm
    2)No statistically significant change in the rate of
    decline in firearm homicides (unless I forget all of
    my statistical training, a non-significant result
    means that you didn’t find a statistically significant
    effect, Simon, so you can’t claim “yes, we got a
    result, it wasn’t significant, but please ignore that

    So…Chapman is agreeing with Baker and McPhedran’s
    key statistical findings.

    In light of all this agreement, the problem is – what,

  26. Simon Chapman says:

    You are obviously a selective reader, M.Anko. I also wrote above “Now, that’s a legitimate question to ask.” So we looked at it too. But for the reasons I explained, it’s not the central question. Baker & McPhedran couldn’t stand to look at the central question … making their paper almost farcically evasive.

  27. Simon, a government has limited funds to spend on improving welfare. Isn’t it a perfectly legitimate question to ask that, even if some expenditure resulted in desirable results, whether that money could have achieved better results if spent elsewhere? And isn’t assessing the costs and benefits of various bits of social policy precisely the kind of thing that economists like our fine host should be looking in to?

    As for your example of pre-flight security checking and terrorism prevention more generally, I would argue that much of the anti-terrorism activity being conducted by government is also a waste of money.

  28. ChrisPer says:

    Where is the official umpire on this? The Australian Institute of Criminology was appointed to do this work by John Howard. The only reason Baker and McPhedran did this paper is that properly controlled statistical work that is transparent (or even available) to non-academics has not been done.

    Professor Chapman, are you over-reacting to protect your ‘record of success’? You assert that the Government brought these laws in the prevent massacres; but they made no such claim at the time.

    John Howard told us “I don’t pretend for a minute that these laws will prevent future massacres” (Sale public meeting 1996). He also said “These laws are about making Australia a safer place”. NO explicit promises about massacres, but GENERAL safety.

    According to Professor Mullen and other reports recently published, National Coalition for Gun Control activism in the media probably helped cause the Port Arthur murders as a result of the contagion effect. I have collated their information at

    Click to access Ideas%20Kill%20-%20Science%20and%20the%20Massacre%203.pdf

  29. ChrisPer says:

    From the SMH Letters page last year:
    It may come as a surprise to Simon Chapman (Letters, October 31) but, like him, I too strongly supported the introduction of tougher gun laws after the Port Arthur massacre.

    The fact is, however, that the introduction of those laws did not result in any acceleration of the downward trend in gun homicide. They may have reduced the risk of mass shootings but we cannot be sure because no one has done the rigorous statistical work required to verify this possibility.

    It is always unpleasant to acknowledge facts that are inconsistent with your own point of view. But I thought that was what distinguished science from popular prejudice.

    Dr Don Weatherburn NSW Bureau of Crime Statistics and Research, Sydney

  30. Christine says:

    I still reckon the biggest problem with the Baker/McPhedran paper is that it provides in itself no information that would ever overturn a particular individual’s preconceptions – if that individual actually understood the analysis in the paper. This is because the statistical analysis is badly executed.

    But very unfortunately, that’s not the spin being put on it. The authors explicitly state in their paper and in interviews (unless they’re being misquoted) that they show the laws had no effect on firearm deaths. But, again, they simply do not show that, even if you accept their (weak) technique. To me, it looks as though the authors fail Don Weatherburn’s ‘test’ of science vs popular prejudice.

    It would be nice to have a good estimate of lives saved (even if that estimate = 0) in order to do a Cost Benefit Analysis, but again, you could presumably come up with a max number of lives saved per year, and decide whether the cost was worth saving that many lives. (Of course, one reason people hate economists is the need to put a value on a life, for this type of analysis: don’t see how you can avoid that and have a CBA based discussion, though. And putting a value on the rights of gun owners is even harder). But you could probably argue this even if you found that gun deaths fell to zero after ’97.

    Andrew’s initial post was basically about the shoddy research. After actually reading Andrew’s post and subsequent comments, does anyone actually believe the paper is not shoddy? Would anyone agree with the authors that their paper has ‘profound implications’ for the direction of gun control policy? (For those who think the laws are bad: do you believe the findings in the paper that firearm suicides fell after the laws were introduced?)

    BTW, I’m just interested, but are there any polls on public support of the gun laws? Obviously it’s a very emotional area for a lot of people and opinion may be polarised but I would like to know what the overall feeling is.

  31. ChrisPer says:

    Even if the paper does not meet the standards of highly qualified statisticians such as Andrew and yourself, I dispute that ‘shoddy’ is the appropriate term. As Dr Weatherburn implied, the research was not being done by the ‘legitimate’ research community.

    But given that gun control is a ‘marker issue’, ie supporting it is used to signify moral status in the same way as abortion and racism, there is in fact little chance that any evidence would be adequate to change people’s minds. But the negative evidence is strong: no particular change.

    Andrew abandoned an attempt to find good base data on public opinion last year. I don’t know what he did find, but it would be difficult to imagine that there is not still a solid support for the gun laws.

    I myself believe that some components of the laws are helpful and effective. But the ‘cultural changes’ being brought about by fashion and the moral panic that brought in these laws have both benefits and costs. It appears that the costs are mostly borne by the legitimate shooters, while the criminal class carry on largely uninterrupted.

  32. Peter W. says:

    In any scientific study, to compare the effectiveness of a particular change (in this case getting rid of semi-automatic longarms) we must have a “control group” where such change was not applied. What better example than NZ, as they declined to take perfectly good firearms from citizens who had never committed any crime. NZ citizens can still own and use the guns which in Australia are banned. NZ does not have longarm registration. (they scrapped it years ago). Last time I checked, NZ had a lower crime rate and lower gun crime rate than Australia. The conclusion therefore is that Australia’s gun bans and gun crushing had no effect.
    Even if there had been a change, there is still no proof, since many other factors since 1996, have had an influence on Public Safety.

  33. Ungrateful Troublemaker says:

    Simon Chapman:
    Digressing from statistics.

    You really do need to look at why Australia moved away from very strict controls on pistols and self-loading/semi-automatic military rifles some years back. Among the factors were a few million bales of unsaleable wool in Australia and warehouses full of AK-47 and SKS-type assault rifles in China manufactured for anti-imperialist struggles in Asia, Africa and Latin America but which had been made redundant by political change.

    You quoted John Howard. He said “How is it that weapons of this kind are still available and what earthly purpose is served by their free availability?” and “There is no legitimate interest served in my view by the free availability in this country of weapons of this kind,” Errrr, it’s called “Trade”, Mr Howard …. and ten spot points for hypocrisy too.

    As for saying that the gun buyback prevented massacres, it’s far to early
    too early to make such a wild assertion. If ALL semi-automatic military rifles in the hands of the public were known, accounted for and surrendered within days of the Port Arthur massacre then such an assertion might be credible …. anyway, take a glance at my post of 26th October above; I won’t repeat myself.

    Agree with you on money being better spent on tackling suicide prevention.

    Paradoxically, I think some money should have been allocated to SSAA (or a similar body) for public education about firearms safety and responsibility to counteract the very dangerous rubbish coming out of Hollywood, the gun-nut lobby, the anti-gun lobby and the computer games industry..

    (b.t.w. I do not own a firearm – I simply don’t need any these days).

    Anyway, back to the discussion of statistics, folks.

  34. ChrisPer says:

    Ungrateful Troublemaker, some good points there.

    My article ‘IdeasKill: Science Shines a Light on Port Arthur Deaths’ points out that the motivators of massacres are pretty well known to psychologists, and the role of media and activists in encouraging copycats. With this comes a model for why we have had fewer massacres.

    You notice that many perpetrators had ‘arsenals’ and would go out and source semi-autos on the black market (like Bryant). This is because the media focused attention on ‘arsenals’ as ingredients for doing this crime. Without going into specifics, the arsenal is not a functional requirement but a fashion statement. By making it seem as though the fashion statement ‘doesn’t happen here’, the Buyback becomes an act of information theatre – a publicity stunt sending a false meme. It was in that sense a correction to the previous message of media activism which taught the moral defectives how to benefit from massacres.

    Article at

  35. Bob Weston says:

    I have just read all of the blog. Most of the material on firearm homicides can be condensed into two areas– (1) has the NFA resulted in any reduction in firearm homicide rates ? and (2) would the money spent on the nfa have been better used elsewhere?

    It is just not possible to arrive at an unequivocal answer to the first question.That is only possible if contemporaneous control data are available for examination ,and they are not. No mathematical prediction technique can provide a reliable estimate of the contribution of the NFA to the rate of decline of these homicides in the post-nfa period, and this statement is not negotiable. It is time to stop wasting time on this question –we just can never get an unequivocal answer. The only answer is that there may or may not have been an effect of the NFA and it is really scientifically inaccurate to promote other conclusions.

    The second question is too hard for me and anyway ,it seems pointless to go over this old ground again. Perhaps I need some appropriate barrow to push.

  36. Christine says:

    Ah, Bob W, a bit too cold and logical on the first, there, aren’t you? What if we just have to have an answer, whether it’s right or not? Seriously, I know I’ve been taking the paper a bit too seriously given your non-negotiable point, but I got rather upset about the authors’ mis-statements.

    Someone made the point to me, re Cost-Benefit Analysis, that it might be that the appropriate cost to use now that the buyback’s done isn’t the $500m or whatever cost of the buyback, since that’s sunk, but the cost of maintaining the existing laws. Presumably the policy question now is should we relax the laws back to pre-buyback, which makes it look a lot better. But still horribly murky.

  37. Peter W. says:

    Christine. You are spot on re cost of maintaining the existing system, with expensive (and highly flawed) firearms registration systems having to be set up and maintained in each State. Estimates of the on costs put the total at more than $$$One Billion and climbing! So far there is no evidence that any of the firearms registries has stopped or solved any gun crime. (around 95% of gun crimes are committed using un-registered guns).
    The new Canadian Gov had no problem deciding to disband their longarm registry, set up in 1995, after a study found that they could have installed and staffed 900 MRI machines for Canada’s hospitals.
    We already have tough laws against murder with a gun, robbery with a gun, etc. so adding “registered” has no meaning

  38. ChrisPer says:

    Going back to pre-buyback laws would be stupid. For a start, they allowed people to pick up a phone in Perth and have a Queensland dealer ship any longarm with just a credit card. And a number of loopholes existed that I bitterly regret not exploiting now, when at the time I was too principled and too poor.

    It would be far better to decide uniform and evidence-based laws, and remove the capriciousness whereby any measure that someone pulls out of the air after watching ‘Die Hard III’ might be included.

  39. Christine says:

    Peter W.: Canada might be having a tougher time than you think with the idea of getting rid of the registry (which definitely cost way more than expected, and is not particularly popular with anyone much) since the recent shootings at Dawson college in Montreal, by a young man who legally owned his semi-automatic (?) weapons. After that, the Premier of Quebec, who is probably a bit more important than any Australian state premier, said outright that the federal government would have a fight on their hands if they tried to scrap the registry.

    A news story with reactions from a few senior members of the Conservative government is at:

  40. Bob Weston says:

    Christine is quite right in implying that I have been hard and logical and basically just negative in doing no more than just pointing out that there is no way we can be sure about whether or not the NFA has affected the incidence of firearm deaths. However, in any discussion, waffle can be reduced if all accept undisputable factors for then there is a sound base on which reasoning can be started.

    in the present discussion we can move foreward from the assessment inability to consider the current firearms situation . We have a gun control system and we have a low and declining firearm homicide rate. With this favourable scenario it would be most imprudent to now change the overall system given that the possible effects of various factors including the NFA cannot be established with useful accuracy.

    I did not wish to imply previously that critical examination of the math technique used by B &McP should not proceed — there is certainly scope for constructive comment

    It appears that many pro-gun persons just won’t accept the reality that
    — lack of evidence of an effect is not evidence of the lack of an effect.

  41. ChrisPer says:

    Beauty Bob!
    Thats logically true. And of course, benefit-cost analysis is irrelevant if you count even imaginary, unmeasurable benefits and ignore private costs.

    Even though I believe a set of well-chosen laws can greatly moderate problems with guns, you convince me that gun control logic is like religion – if I don’t believe enough, I must have more faith!

  42. ChrisPer says:

    Transcript of ABC Law Report with McPhedran, Chapman AND Don Weatherburn providing balance. Well worth the read.

  43. Ungrateful Troublemaker says:

    ChrisPers and Christine:
    Thanks for those articles – the more we know about causes and about what others, like the Canadians, have tried, the easier it will be to prevent further tragedies.

  44. Pingback: Andrew Leigh » Blog Archive » Did ‘96 make a difference?

  45. Ron Webb says:

    Did anyone think to think about how over 5,500 people employed in the firearms trade in Australia who now don’t have a job, survive?

    This includes those a Lithgow Small Arms, a world class design and manufacturing facility owned by all Australians.

    Those on the ammunition production lines inherited by ADI, (now UK Thales owned)?

    The pimply kids who worked in retail who now flip burgers at Maccas?

    With any ‘industry adjustment’ there is usually a scheme implimented to buffer the cold hard winds of the dole que.

    The Australian feral government kicked in nicely to keep Kodak alive in Victoria to the tune of over $48 million.

    The auto maker have now consumed some $2 billion of tax payers (that’s right u and moi) hard earned readies to keep them planted here in Oz.

    The fishing industry had their licenced cancelled, yet compensated [can’t find a value for that one yet].

    Even the insidious tobacco industry, along with textiles, footwear and forrestry all coped an equitable treatment.

    Why don’t you guys and gals from both sides pack their stats-wars up and spend a little time looking at the real human cost – to those alive, yet on the poverty line.

    All because JWH ‘hates guns’…

  46. Pingback: Deltoid

Comments are closed.