Sarah Bosely writes in the Guardian:
The critics argued that the Lancet paper does not indicate that the researchers moved far enough away from the main street. "The further away you get, the further you are from the convoys that roll down the streets and the car bombs and the general violence," said Sean Gourley. "By sampling only cross streets which are more accessible, you get an over-estimation of deaths."
But Prof Burnham said the researchers penetrated much further into residential areas than was clear from the Lancet paper.
The notion "that we avoided back alleys was totally untrue". He added that 28% of households were in rural areas - which matches the population spread.
Others had suggested that it was impossible for 40 households to be surveyed in one day - but in fact the researchers were split into two teams and conducted 20 household interviews each, he said. "Our point is that it is a big number, we set out to prove that we had the power to say the mortality rate has more than doubled."
My previous post on this alleged "main street bias" is here
Also, I emailed Burnham about this matter and he replied:
In the response to John Bonhannon I paste in below what I wrote to him, and
which he chose to ignore, claiming I was unable to describe the sampling
methods:"As far as selection of the start houses, in areas where there were
residential streets that did not cross the main avenues in the area
selected, these were included in the random street selection process, in
an effort to reduce the selection bias that more busy streets would
have." Nowhere have we said that we avoided back alleys--as we did not.
Somewhere it has gotten lost that this was the process for finding the
start house--not for the whole sample. The Iraqi team indicates that the
survey of the 40 houses involved 2 and frequently 3 contiguous
streets--following the "nearest front door" approach. To me a major
verification factor is that first time we used GPS to locate the start
houses, and this time we used the random sampling methods--deemed by the
teams now to be safer--and we got results that very closely paralleled
each other. Les will probably tell you the same.
Thanks for that. I think the man's name is Bohannon, not Bonhannon, so you might want to put a "sic" in there.
Could you please e-mail Burnham on this line from the report:
How often did this happen? If the random start location was only changed once or twice then, obviously, the tenor of the results is unlikely to be effected. But if it were changed more, then the sampling is now longer vary random. (Of course, this is just as likely (?) to bias the results down (safer neighborhoods) as up, but I just want to get these details correct.
Unfortunately, Burnham does not reply to my e-mails and, AFAIK, has not released the underlying data.
Of course, the procedures Burnham describes don't eliminate the bias, they just make it easier to understand. It's rather similar to the bias from GPS sampling in their first study. By sampling addresses with equal probability, they undersampled people living in multi-unit structures.
Are there a lot of people living in multi-unit structures, or just a few? Well, most Iraqis live in a "housh," so it depends on just how the typical housh is laid out.
David Kane: "Unfortunately, Burnham does not reply to my e-mails...."
Regular viewers won't be surprised. Given that you accuse the authors of being fraudsters, what do you hope to learn from his replies? Are your interrogation skills so good that you can hope to get a confession by e-mail?
David Kane: "Unfortunately, Burnham does not reply to my e-mails...."
That was kind of funny. Unfortunate too, since I think it was David that asked them for the detailed data on individual clusters in the previous paper, which resulted in them releasing it to him and Robert here more or less rederiving their CI. Maybe someone else could ask? I was thinking of emailing them and asking for their confidence interval on the coalition-caused deaths, it being an interest of mine (being American and therefore somewhat responsible, but they probably get a lot of email and might be more likely to respond to someone with statistical credentials.
You shouldn't burn bridges like that, David. Or launch air strikes on them or park car bombs on them, or whatever.
On another blog, David Kane wrote: What will the mortality rate in Iraq be in, say, 2008 if the US were to leave in 6 months? What would it be if the US were to stay? I really don't know.
David, the US government doesn't care about the mortality rate in Iraq other than its bad PR for an invasion that was a serial violation of international law. Many policy experts pointed out in 2002 that an invasion of the country might lead to a societal implosion, but the Washington DC planners had other things on their minds e.g. Iraq was a strategic and economic 'prize'. The people of Iraq just happened to be in the way. Moreover, your argument assumes that the US-UK are now in Iraq in the role of 'peacekeepers'. This is a cunning new form of mainstream media propoganda aimed at legitimizing the occupation through the age-old trick of 'normalization'. The occupation is the problem, not the solution. Poll after poll reveals that the Iraqis want the occupation to end, irrespective of the consequences to 'security'. In fact, given the sordid agenda of the invaders, the 'security' question plays straight into the hands of the Bush-Cheney cabal - it suggests that as long as there is rampant insecurity, the occupation must continue. But the occupation IS the problem. Moreover, it shows what utter contempt the Bush regime has for 'democracy'.
What I see in these threads time and time again are feeble attempts by some contributors to suggest that the unfolding tragedy and carnage were unfortunate by-products of imperfect liberation. They were in fact the products of rapacious conquest. The country's civilian infrastructure was devastated by the first Gulf War and 12 years of sanctions before the latest invasion. I'd like to ask David this: how many Iraqi civilians do you think (make it a guess from the gut) have died as a result of the (1) Gulf War (1991), (2) the sanctions regime, and (3) the current 'war'? I ask this because for many pundit supporters of the slaughter, most of the crimes committed by the US and its proxies have been airbrushed from history. Few imperial supporters here discuss current events within a historical perspective. That the US took Iraq off of the list of states that officially sponser 'terrorism' in 1982 because it wanted to arm Saddam in fighting a war with Iran. That the US continued to arm the tyrant through the worst period of his crimes, in the 1980s', in spite of full knowledge of them. That the same people in the Reagan and Bush I governments who armed, aided and abetted Saddam in the 1980's are now recycled in the current administration. Why don't the defenders of US preventive aggression ever discuss these relevant points? Why are the Lancet results also so far-fetched in light of recent historical precedents?
I'm not sure that would be particularly informative, because the real problem here is not statistical but interpretative. While most of the deaths were documented by death certificates, whether or not the death was coalition-caused is likely to be a matter of hearsay. So this figure probably tells us more about who the Iraqis are blaming for deaths than about the actual causes
I don't see how this follows. They picked a starting address at random, and then sampled other adjacent residences in the same vicinity until they reached 40 households. If they happened to hit a multiunit structure, presumably several of those households would be in the same structure.
Oxford Physicist Sean Gourley: "By sampling only cross streets which are more accessible, you get an over-estimation of deaths."
Burnham: The notion "that we avoided back alleys was totally untrue"
I'd say Gourley is pretty uninformed about the Lancet study. He should really stick with physics.
Funny that you don't see any of world's the top physicists (eg, Nobel Laureates from MIT, Caltech and Harvard) dabbling in mortality analysis. They are smart enough to know better.
If the random start location was only changed once or twice then, obviously, the tenor of the results is unlikely to be effected. But if it were changed more, then the sampling is now longer vary random
Yes. We should also ask them how many times the interviewers dropped their data in the mud, or misplaced it in a coffee shop, or simply went to the bathroom while someone else might have altered their data. After all, these people behind the Lancet article are such amateurs that you can expect they would hire a bunch of hacks to do a survey like this, and you can be almost certain that they would simply ignore changes in methodology done by the hacks after their original design.
As our Epidemiologist in Chief, George W. said: the methodology is pretty much discredited.
the procedures Burnham describes don't eliminate the bias
OK, so there's a potential bias. Maybe big, maybe small, maybe upward, maybe downward, we don't know.
But surely no survey uses a perfectly unbiased sample. Telephone surveys in developed countries typically have much lower response rates than this one, right? -- introducing an unknown but possibly large bias. And yet we rely on them and find that they are reasonably accurate.
In the absence of specific evidence to the contrary, isn't the most reasonable assumption that residents of multiple-unit buildings have had the same mortality experience as residents of single-family buildings?
Unlike a lot of the Lancet critics, Ragout really seems to be arguing in good faith. But what he seems to be saying is that if there is any possible source of bias in a sample, we have to throw out the survey. Is this survey really worse than others done under comparable conditions? Are there any actual examples of a survey with biases similar to this one that yielded demonstrably different results from surveys without them?
trrll wrote--
"I'm not sure that would be particularly informative, because the real problem here is not statistical but interpretative. While most of the deaths were documented by death certificates, whether or not the death was coalition-caused is likely to be a matter of hearsay. So this figure probably tells us more about who the Iraqis are blaming for deaths than about the actual causes"
I'd bet that Iraqis often have a good idea about whether their relatives were killed by the coalition. Not always, of course. It is interesting and possibly significant that so many Iraqis favor attacks on coalition forces--if there is a lot of "collateral damage" going on that would make sense. It makes less sense if coalition-caused deaths are extremely rare (the impression one gets from media accounts), though I suppose xenophobia and resentment over how badly the occupation is going could explain it.
To answer my own question, on page 5 the paper states that the 95 percent confidence interval for the percentage of coalition-caused violent deaths is 26-37 %, centered on 31 percent.
As trrll says, that doesn't mean the figure is actually correct. (I'd guess the percentage is probably right, but that's my much-maligned gut speaking.) But even if Iraqis just think the US is causing such a large fraction of the deaths, it would explain why so many favor attacks on coalition forces.
Yes, I think that this is an important practical issue that seems to have been neglected in the face of the admittedly important issue of how the higher casualty estimates impact the question of the morality of the war itself. The people who are most likely to take up arms against the US, whether as insurrectionists or as terrorists, are those who perceive themselves as having suffered a personal loss of family or friends as the result of coalition activities. These statistics argue that the number of people in Iraq who fit that category is far larger than previously supposed.
"Funny that you don't see any of world's the top physicists (eg, Nobel Laureates from MIT, Caltech and Harvard) dabbling in mortality analysis. They are smart enough to know better."
I'm waitign for an epidemiologist to write a physics paper denouncing the basic theories and experiemental methodologies of the entire field.
I'm sure newspapers will line up to print their criticisms without a word of skepticism.
Lubros will be here in a moment.
I am willing to bet that the average epidemiologist could write a better paper about string theory than a physicist like Gourney writes about epidemiology.
Who knows, perhaps an epidemiologist might even bring string theory into contact with the real world, something physicists have yet to do after 20 years.
trll,
They randomly chose a block and then chose to start as some random point on the block. But a multi-unit building might have 10 times more people than a single-family house, but only twice the street frontage. So, the apartment building is only 2 times as likely to be chosen as a starting point. But sampling "proportional to population size" means that it should be 10 times as likely to be sampled.
Sure, they could sometimes have started at a single-family house and gone to the building next door, but big buildings tend to be located next to other big buildings, so that's hardly likely.
Lemuel,
I don't think the Lancet survey tells us much that we couldn't have guessed from other sources and back-of-the-envelope calculations. So, I do think they might as well throw out the survey.
The Lancet estimate of non-violent excess deaths (16,000) is about 1/6 of what I would have guessed, and their estimate of violent deaths is maybe is 2 to 4 times what I'd guess. But I don't feel remotely persuaded to revise my non-violent death estimate down, or my violent death estimate up.
Do you? Despite all evidence of the collapse of electricity, sewage, hospitals, and so on, do you really believe the Lancet estimate that there's been essentially no increase in non-violent mortality? If you don't believe that, why should you believe their estimate of violent mortality?
Lemuel,
Yes, this survey is worse than others done under comparable conditions. Many's the time I've pointed to better surveys on Tim's blog. Here's one, a study about Darfur published in the Lancet.
Looking back at it, I can't believe I failed to mention the recall period.
Darfur: 193 days or less
Iraq: 4 1/2 years.
I highly doubt that you can find a single crisis mortality study like Burnham's that has such a long recall period, or even a recall period as longs as 2 years. Nor do I think you will find a single published study that used non-PPS (proportional to population size) sampling in such a large area.
Yeah, because people are likely to forget a death in the family after three years. Or maybe the death certificates disintegrate after that amount of time.
Nope Tim, I think the reason that these surveys always use short recall periods is that the definition of a household becomes more and more ambiguous over time, as people join and leave.
Why do *you* think that these surveys never use recall periods much longer than a year?
Jeff Harvey,
I know you just want to rant, but you've hit on one of my pet peeves. The US never sold Saddam any arms. Not one single bullet. The worst you can say is that we sold him some dual-use chemicals and gave him some satellite photos.
"In the absence of specific evidence to the contrary, isn't the most reasonable assumption that residents of multiple-unit buildings have had the same mortality experience as residents of single-family buildings?"
In the case where an invididual is hunted dwon and attacked because of who he is, I'd have to say it probably makes no difference whether he lives in a large building or single family dwelling.
When it comes to attacks meant to terrorise a community, however, I'd say that if anything, living in a large building biases the mortality upward, since large buildings are more likely to be the target of attacks precisely because they harbor more people.
A bomb attack on a large building is almost certainly going to injure and kill more people than a bomb attack on a single family dwelling. particularly if the explosive charge is tailored to the building size, which one would expect to be true in many cases. As we have seen, terrorists certainly know this.
Main street bias is still real wiuth the Lancet study. Here is a response by one of the Oxford team to a writer on Medialens, giving an estimate of the effects:
http://www.medialens.org/forum/viewtopic.php?t=1949&start=0
Remember the main street bias says nothing about the non-violent
death rates, it is only concerned with the death rates from violence.
Thus we need to compare L1 and L2 from this point of view. It is not
right to say that the two surveys agree on total deaths, we need to
know if they agree on violent deaths.
Here's the latest version of the violent death rate comparison
between L1 and L2
from L1 - (excluding Falluja due to main street bias)
_____________
total violent deaths/1000
pre - 1/7438 or 0.134/1000 per 14.6 months
post - 21/7868 or 2.67/1000 per 17.8 months
normalising the pre-invasion rate to 17.8 months we have
pre = 0.164
or a difference of 2.50/1000 per 17.8 months (or 1.68/1000/year)
thus with a population of 24.4 million as estimated in the L1 paper
we get 61,000
from L2
_____________
violent death rate per 1000 per year for the first year = 3.2 (or is
this for 13 months? Doesn't make a large difference as it is a rate/
year that we are measuring)
violent death rate per 1000 per year for the second year = 6.6
this is an average rate for the whole year, so at month six -
assuming a linear rate of increase (reasonable given the data) - we
have a death rate of 6.6/1000/year.
At the start of the second period you can measure the death rate
(from the graph in the Lancet paper) to be 4.95/1000/year.
This corresponds to an increase over the 5.8 month period of ~1.6
Thus the average death rate for the 5.8 month period is 4.95+0.8=5.75
per 1000/year
Correct for the 5.8 month period, gives us 2.77/1000/5.8 months
total violent deaths = year 1 deaths + second 5.8 months deaths
= 78,080 + 67,588
= 145,668
where we assume the population to be 24.4 million as we did for L1.
this difference is higher than the excess deaths figure - but that is
because the non-violent count drops over the first two years of the
war when compared with the pre-war calculation.
So for the same survey, but one measured using gps and the other
using the main street + cross street algorithm we have a 61/145.6=2.4
fold increase in measured violent deaths.
____________
This is a pretty significant result I think, strongly points towards
the main street bias being a significant effect.
Sean
Here's also an email I recieved back from Mike Spagat when I suggested the Lancet may have thought Iraq has a grid system of roads like they do in the US:
Thank you very much for your message. We have studied a variety of maps of Iraqi cities and they definitely depart hugely from a grid system. There are plenty of places tucked away where they cannot be reached from following the methodology in the Lancet paper. There is no doubt about that.
Whether or not the authors thought things were on a grid I couldn't say. My guess is that they simply didn't think about this. I think that someone else probably suggested that procedure and they accepted it without thinking. It is clear that they had no idea where the survey teams were going and what they were doing. So I think they just didn't care and didn't pay attention to methodology, field work and implementation.
That said, based on my personal experience there is a pretty big difference between European and US roads as you suggest. While I doubt that this difference had an influence on the design of the study I think this is having a big influence on discussions of our main street bias idea. It seems that many people in the US are thinking in terms of perfect grids.
Mike Spagat
Given this overwhelming evidence, it is time for Tim Lambert to come clean and admit to the world that the Lancet study is fatally flawed.
Thanks Ragout, very clarifying.
It seems to me that the best way of evaluating this study is by comparing it to ones done elsewhere. The Darfur link is a good starting point.
So Tim, do you know of similar studies of mortality that used a lookback period of several years? And Ragout, how does cluster sampling (which is used precisely in cases where a random sample of individuals is infeasible) usually avoid the problem of multi-family dwellings?
(I don't have strong views here -- just trying to edcuate myself and hopefully others.)
Given this overwhelming evidence, it is time for Tim Lambert to come clean and admit to the world that the Lancet study is fatally flawed.
You can't get away from that logical fallacy, can you? At least you've swapped 'devastating' out of your lexicon for now.
JB, the Lancet paper gives no indication that they went to non-cross streets - Sean Gourley is commenting on the basis of the published information. Burnham is giving extra information not in the paper - as Tim's post says:
I'd say Gourley is informed about the Lancet study, as published.
Except that, *as published* it states that they avoided main streets.
Ragout has provided no convincing argument why the single family vs multifamily dwelling issue (if it indeed exists in the Lancet study), invalidates the study and in particular, no reason to believe that it would bias the Lancet estimate of excess mortality upward.
Indeed, one might expect that preferentially selecting (for the survey) those in single family dwellings over those in a multifamily dwelling would lead to an underestimate of mortality because of lower likelihood of terror attacks on single family houses than on large buildings and lower likelihood of mortality from attacks on buildings harboring fewer people.
I don't think the Lancet survey tells us much that we couldn't have guessed from other sources and back-of-the-envelope calculations.
I'm not convinced on this point. Anyone paying attention knows that an enormous number of Iraqis have died as a reult of the invasion, but is it 200,000, 500,000, one million? And it is important to have a number, even a rough one, for all sorts of reasons.
You've provided good reasons to think we should treat the study as having a larger confidence interval, but not to reject its point estimate as the best one available. Or more practically, you haven't convinced me not to use the 600,000 number, altho you have convinced me to be a little less certain of it.
As for the violent vs. non-violent mortality, (1) any subsample is going to involve greater uncertainty (2) the cause of death is going to be less clear than the fact of it, and (3) it doesn't seem wildly implausible that infrastructure has not deteriorated much from where it was after a decade of bombings and sanctions.
Anyway, I hope you continue comenting on this, and I hope Tim responds to you more seriously than he has so far.
Rgaout said: "I highly doubt that you can find a single crisis mortality study like Burnham's that has such a long recall period, or even a recall period as longs as 2 years."
Whether any survey has ever before used such a long period is largely bside the point.
Burnham et al address that issue in "Human Cost
of the War in Iraq A Mortality Study, 2002-2006"
http://www.jhsph.edu/refugee/research/iraq/Human_Cost_of_War
"Perhaps the greatest potential limitation to this type of survey is the problem people have recalling the
date of specific events, especially over several years. Again, the close similarities between the 2004 and
the 2006 data suggest this was not a major problem."
Rgaout: "Despite all evidence of the collapse of electricity, sewage, hospitals, and so on, do you really believe the Lancet estimate that there's been essentially no increase in non-violent mortality? If you don't believe that, why should you believe their estimate of violent mortality?"
Here's what Burnham et al actually said about nonviolent death rates since the invasion:
http://www.jhsph.edu/refugee/research/iraq/Human_Cost_of_WarFORMATTED.p…
"Non-violent death rates
The deaths recorded for the pre-invasion period in both the 2004 and the 2006 surveys were almost
entirely non-violent deaths. (We define non-violent deaths as not due to intentional violence--that is,
our non-violent deaths include deaths in "accidents," such a traffic fatalities.) Immediately post-invasion,
the death rate due to non-violent causes dropped slightly, then stayed level for the next period, but began
to rise in the period from June 2005 until June 2006. The excess death rate due to non-violent causes is
estimated to be 1.2 deaths/1,000/year for this most recent period of time, and 2.0 deaths/1,000/year for
the first six months of 2006. It is not possible to say that this number is a statistically significant increase
over the pre-invasion baseline death rate. However, this may represent the beginning of a trend toward
increasing deaths from deterioration in the health services and stagnation in efforts to improve environ-mental
health in Iraq.
Ragout wrote:
I'm not familiar with all crisis mortality studies, but it is certainly easy to find both fertility and mortality studies that use recall periods that extend over the length of a respondent's reproductive span, i.e., 35 years or so. There are standard methods to handle data such as those (look up some of the Brass methods from the early 1970s), and it's generally recognized that the recall period length isn't the biggest problem -- it's the age of the decedant (infant deaths, and to some extent, child deaths, are less likely to be counted both in the numerator and the denominator; that doesn't appear to be so much of a problem for adult deaths). I suspect, but emphasize that I do not know, that frequent changes in the size of the household or family will cause problems; I don't know how frequently household size changes occurred in the Burnham study, though it's possible to make that a rough estimate for the Roberts study. It appears from the person-months of exposure and the starting and ending populations that household size was actually quite stable, i.e., it appears that there wasn't a lot of short-term moving in and out, at least on net.
The UNDP/ILCS study wasn't sampled proportional to population size.
As for the geographical bias you've been talking about, it sounds like you're suggsting that the Burnham study would be less likely to sample small dwellings and more likely to sample large dwellings than the Roberts study. If so, what would your guess be about average household size in the Burnham vs. Roberts studies?
Not that I'm saying the UNDP/ILCS sampling was done poorly; I was just pointing out that it wasn't proportional to population size. As long as one has the proper weights it's not a problem. (I don't know if anyone has ever scrutinized the UNDP/ILCS study to see if they used the proper weights. I suspect not, or at least, not to the extent that the Roberts and Burnham studies have been scrutinized).
hmmmm, but the whole main street issue is still problematic, isnt it? I mean picking a main street, finding a side street, and picking a random number on that side street, and surveying around the picked house does not give you a uniform distribution over the inhabitants of the city. Its unclear that it biases the number in any one direction, but its a possible source of error, no?
No, this is not what it means. Street frontage is irrelevant. They numbered all of the households (whether separate homes or apartments), chose a random household number to start and then proceeded to nearby homes/apartments until they had sampled 40 units. So if most of the households are in multifamily units, then there is greater likelihood that the starting household will be in such a unit, and also that the neighboring addresses/apartments will be in the same building. Thus, the overall fraction of people sampled from multifamily units will correspond to the frequency of people living in such units.
Hi,
I actually have been a correspondent for the Lancet - mostly political stuff, from Brussels. I'm proud of em. I remember I used to get mobbed by bearded types and get bought drinks in pubs, following the last report.
I didn't pay much attention to it cos my area was Europe but now I am writing a piece for another magazine on this.
Here is a reply I received from Sean Gourley
"The violence does not occur in places where no one lives - the violence occurs largely on the secondary roads off the major motorways and highways. These are likely to include many of the cross streets that the survey team exclusively sampled. These are the same streets that have the road blocks, the car bombs, the military and supply convoys and the associated insurgent ambushes. These cross roads are likely to be major battlefields of the war in Iraq. But they are also home to people who live on them.
People who do not live on these streets do travel along these roads going to mosques and markets. But they only spend a fraction of their day in these areas - they do not live there.
Question: Who is the person most likely to be in front of their house at any given time? We believe that the answer is: The person living in the house.
This statement is especially significant when travel is limited in war situations like Iraq. We also know from the survey the >99% of the households surveyed had someone present at the time of the interview.
end quote
Now my query is: women and children are massively under represented in the car bomb and airstrike statistics, which either suggest that carbombs and airstrikes deliberately hit areas where adult males circulate - perhaps main roads. But not the home. Women are far less mobile and surely you would expect more of a gender balance of carbombs were placed within short distance of front doors.
Okay, so suppose men are ten times more likely to be on the street at any given time than a woman, cloistered in the home. But children?
The low deathtoll of under 15s (10% of deaths, 40% of pop) is also striking. Does that mean they never spend time on the street when they return home from school? The picture is in the Gourley analysis is of one where men lounge on the streets outside their homes all day, women stay inside the home, and children are completely absent. Even after school.
Surely the gender imbalance sugggests that bombings take place beyond those areas where women circulate - and for sure, they are not, cannot be seen, surely as much as men in city centres where men are seen more around the home area.
Another point: doesn't all this obscure the fact that non violent deaths which are not subject to main street bias - or are they - account for half the 600,000 putative excess. So we are still talking hundreds of thousands of dead here.
I suppose you can get heart attacks from being close to violence...but I have gut feeling it's more to do with failure of general care and comfort.
Also, the UNDP figures - someone said they were discredited. Interestingly, Gourley didn't reply to my point that the epidemiological survey was sufficient - and it tallied with the CIA factbook mortality rate of 5 or whatever.
Another thing:
were main streets included or not?
Gourley above says cross streets, Spagat says
The study suffers from "main street bias" by only surveying houses that are located on cross streets next to main roads or on the main road itself. However many Iraqi households do not satisfy this strict criterion and had no chance of being surveyed.
Main roads
and the Lancet says:
he third stage consisted of random selection of a main street within the administrative unit from a list of all main streets. A residential street was then randomly selected from a list of residential streets crossing the main street. On the residential street, houses were numbered and a start household was randomly selected. From this start household, the team proceeded to the adjacent residence until 40 households were surveyed
And does this matter? Are the anti-Lancet people singing from the same hymn sheet?
Pelle,
You'll find some material in other threads, especially Burnham's response to Tim Lambert on the "main street bias" question.
Personally I think it's an interesting quibble but nothing more. I'm sure most of the deaths took place near main roads but that's not the point. Where were the victims' homes? My guess is that the really dangerous places to live are:
(a) neighbourhoods where there is a "frontier" between rival militias and their communities, and
(b) the "bandit country" - places where Coalition forces get ambushed.
Really, it's up to the critics to come up with some evidence to support their theory. Scientific debates aren't supposed to be conducted by firing off press releases.
Whoops, that's a link to this thread; sorry.
BTW, note Burnham's comment that this discussion all relates to where a cluster starts. AFAIK it can then creep around a corner onto a main street or onto a street which does not connect directly to a main street.
Having said that, I'd like a much more detailed descripyion of the process.
One last thing Pelle: you misread the non-violent excess deaths: 54,000 which is surprisingly small and not statistically significant, though it may be a warning of things to come.
They numbered all of the households (whether separate homes or apartments), chose a random household number to start and then proceeded to nearby homes/apartments until they had sampled 40 units.
If this is correct, there is no bias against residents of multifamily buildings. Ragout (or any other critics), do you disagree with trrll's description here?