Helle Dale attacks the Lancet study

Just when you thought you had seen all the different possible attacks on the Lancet study, Helle Dale, writing in the Washington Times, comes up with a new one: the study's authors are having second thoughts. Dale writes

As the Financial Times reported on Nov. 19, even the Lancet study's authors are now having second thoughts. Iraq's Health Ministry estimates by comparison that all told, 3,853 Iraqis have been killed and 155,167 wounded.

Gee, did the Financial Times really report that the authors were having second thoughts? Let's check. The report (subscription required) says:

Les Roberts, one of the paper's authors, said that including more clusters would have improved precision, but would have increased to unacceptable levels the risks faced by interviewers. With hindsight, he would have liked interviewers to visit two or three points in Falluja to get an idea of variations between neighbourhoods there.

One other important weakness of such surveys is associated with retrospective recall, which the authors play down. Respondents may have lied or been confused about when deaths took place.

One question that could have thrown light on the precision of the mortality figure was to seek information on the wounded. Dr Roberts said no data were gathered on this: "Injuries are more subjective."

Roberts did not say he had second thoughts. Dale is just making stuff up.

Now compare Dale's second sentence with what the Iraqi Health Minister actually reported:

"Every hospital reports daily the number of civilians (which may include insurgents) who have been killed or injured in terrorist incidents or as a result of military action. All casualties are likely to be taken to hospital in these circumstances except for some insurgents (who may fear arrest) and those with minor injuries. The figures show that between 5 April 2004 and 5 October 2004, 3,853 civilians were killed and 15,517 were injured. I am satisfied that this information is the most reliable available."

Those 3,853 deaths are not "all told" as Dale claimed, but occurred in just six months. And the Lancet study's 100,000 excess deaths includes those that resulted from the increase in infant mortality and homicide that followed the invasion. If we triple the Health Minister's number to cover the whole period we get 12,000. The Lancet number for military and terrorist deaths is very roughly 40,000. Since the Health Minister's number only covers what was reported to hospitals, the real number could easily be twice as much. A number like 25,000 military/terrorist deaths seems to be compatible with both the Lancet study and the hospital reports.

Tags

More like this

In an earlier post on the IBC I wrote: Sloboda says: We've always said our work is an undercount, you can't possibly expect that a media-based analysis will get all the deaths. Our best estimate is that we've got about half the deaths that are out there. OK, then why does the IBC page say "Iraq…
Nature has published a letter (subscription required) from Roberts and Burnham responding to this story. In our opinion, your News story about our Lancet paper "Death toll in Iraq: survey team takes on new critics" (Nature 446, 6-7; 2007) has confused the matter rather than clarified it. You…
The biggest limitation of the Lancet study is the small sample size. We can be reasonably confident that deaths have increased in Iraq since the invasion, but the 100,000 estimate is a very rough one. The sample from Falluja found an alarming number of deaths from air…
John Allen Paulos writes about Iraqi war deaths: Another figure in the news recently has been the number of Iraqis killed in the war. President Bush mentioned last month that in addition to the more than 2,100 American soldiers killed so far in Iraq, that there were approximately 30,000 Iraqis…

Unrelated to the point, but "155,167 wounded" from Helle Dale and "15,517 were injured" from the Iraqi Health Minister differ significantly by the addition of the digit "6".

Jack Straw's summary is recommended reading. As he correctly points out the figure of 3853 includes the months of August and April this year, which were the most violent so far. It also includes deaths caused by terrorists, and of some terrorists (though a smaller section, as a lot of those would not have been reported to hospitals).

Basically, the official line of the coalition and of the Iraqi government is that all is being done to prevent civilian casualties, they are all the fault of the terrorists, even if not directly caused by them, and the number of innocent civilians killed by the coalition, while protecting the population from the terrorist menace, is most credibly estimated in the low thousands (while the number of "insurgents"/Baathists/terrorists killed is likely in the ten thousands).

By Heiko Gerhauser (not verified) on 01 Dec 2004 #permalink

Someone, I think in these comments, said that a medialens article stated the same thing: the authors were having second thoughts.
It seems likely to me that 'the authors were having second thoughts' could be a disseminated talking point and Dale picked it up and ran with it. But that's my conclusion after reading so many canned-seeming talking points here in Tim's comments [Heiko excepted].

D

It's kinda hard to deny that the security situation in April deteriorated, because of Fallujah and Moqtadr. Moqtadr got dealt with in August, and Fallujah in November.

I think the security situation is now probably at its best since February.

By Heiko Gerhauser (not verified) on 01 Dec 2004 #permalink

I think the security situation is now probably at its best since February.

'Best' as in it's good that American soldier's deaths are near record numbers and that means...what, exactly? Or 'best' as in the violence moved from Falluja to Mosul when the insurgents left F for M and that's best for F residents?

D

What Fallujah residents. The place is flattened.

By Eli Rabett (not verified) on 01 Dec 2004 #permalink

Jack Straw's summary is recommended reading.

That summary (from 17 Nov.) would be

here.

The Foreign Secretary did a good job of presenting his government's side without trashing the researchers as individuals or misrepresenting their findings too badly. A few mad dogs of the blogosphere could learn from his example.
The following, though, struck me as pure spin:

This limited precision is reflected in the very large range which they use for their estimate of excess mortality (8,000 - 194,000). Although the levels of probability vary across its range, any figure within this range is consistent with the data.

True, so far as it goes, but Mr. Straw is gliding over the fact that the most likely value is indeed near the center of that interval. I figure that the PDF for excess deaths is not really Gaussian, but slightly skewed to the low side (help me out here, somebody). But it's probably not far from Gaussian, so use a normal distribution with mean of 98,000 and stdev of 47,355. That implies there's an even money chance that the true value for excess deaths lies between 66,000 and 130,000.

That is still a lot more than estimated by the Iraqi Ministry of Health, a discrepancy readily explained by the fact that not everyone dies in a hospital.

Here's a better way to deal with the problem. Every time you refer to Iraqis fighting the Americans and British, refer to them as terrorists. If you must use the word insurgents, put it between quotation marks to avoid any respectability which might adhere to that term. Then a lot of the problem of excess deaths goes away. Many of those deaths were caused by terrorists (not our fault!) or the deceased were terrorists themselves, in which case they had it coming.

Well, the security situation has improved markedly over the last two weeks. And of course last time it was this good, in February, it didn't last long.

Mosul was really bad about three to four weeks ago, it seems nearly back to "normal" now.

Which gets me on the subject of trigger happy American soldiers:

http://cbftw.blogspot.com/

I understand that soldiers are frightened for their life, and I also understand that, often enough, terrorists are breaking all the rules of war by not openly wearing arms, and therefore distinguishing themselves from civilians.

The soldier reporting in the above blog decided to stop posting shortly after he reported on an incident that involved hundreds of dead insurgents.

What worried me about his report was that he didn't talk about the dangers to civilians. This may sound strange to you, but it seemed very wrong to me, maybe just my perception, and I know that he is a likeable guy all right from what else he wrote.

What also worries me, is that there doesn't seem to be enough of a push to encourage learning of Arabic, which would help so much in communication, and therefore understanding, and greater empathy/concern, which seemed so lacking to me in this instant.

http://astarfrommosul.blogspot.com/

Najma is a Sunni girl posting from Mosul. Aya is her niece, and you can see a sweet picture of her there at the moment.

Aya's grandfather was caught in the cross-fire between American troops and "insurgents", and wounded. Due the strict enforcement of the curfew, he didn't make it to hospital and needlessly bled to death.

I understand the necessity of the curfew, and the unavoidability of stray bullets hitting innocents. Yet, communication goes both ways. The US military is trying to help Iraqis, but they could and should do better. This particular death was avoidable.

By Heiko Gerhauser (not verified) on 01 Dec 2004 #permalink

From jack straw, courtesy of jre's link.
"The figures derived from the survey's data on Fallujah would have resulted in an estimated 200,000 excess deaths within Fallujah alone over the past 18 months. This would amount to almost two-thirds of the total population of the town - which is just not credible. The authors of the study understandably discounted the data."
Is this true ? If so, it would completely undermine the credibility of the study.
jre may also have missed the following statement:
"The article estimates that between 8,000 and 194,000 more people died following the invasion of Iraq than previous rates of mortality would have predicted, with the "most likely" figure being 98,000 extra deaths."
j

By James Brown (not verified) on 01 Dec 2004 #permalink

http://antiwar.com/casualties/list.php

Heiko, unfortunately I'm not sure the security situation is improving. Between the 18th of November (and the tiem US spokesmen claimed Fallujah was "100% secure" and November 29th, 35 American servicemen died in Iraq. during that same period, four private security guards were killed by a rocket attack on the Green Zonei n Baghdad, supposedly the most secure location in the netire country and the British Embassy formally told its personnel not ot use the road from Baghdad to Baghdad airport due to the extreme danger.

By Ian Gould (not verified) on 02 Dec 2004 #permalink

Heiko, a hospital report covering six months in 2004 is fine enough, but it won't contain any fatalities as a result of civilian casualties of the war itself, which would obviously have been a big part of the Lancet count.

Heiko's nonsense continues. Yesterday, the UN's Integrated Regional Information Network reported of Fallujah:
"Approximately 70 percent of the houses and shops were destroyed in the city and those still standing are riddled with bullets." ('Fallujah still needs more supplies despite aid arrival', http://www.irinnews.org, November 30, 2004). You would not know from the media coverage that a vast war crime has taken place in Fallujah. If Saddam Hussein had demolished 70% of Kuwait in 1990, it would surely have been declared one of the great atrocities of the twentieth century. Yet our media have continued to channel official lies or to ignore this latest atrocity completely. Heiko quotes Jack Straw as if he is an honest broker and part of a legitimate invasion. But Straw is as much of a liar as the rest of Blair's cabal. It is indeed amusing to watch the war mongers trying desperately to downplay the Lancet study, clutching at any argument that they can, no matter how superficial. But civilians have always been a target of western bombs: the entire civilian infrastructure of Iraq was targeted during the first Gulf War. The U.S. and its allies destroyed Iraq's water, sewage and water purification systems and its electrical grid. Nearly every bridge across the Tigris and Euphrates was demolished. Twenty-eight hospitals were struck and thirty-eight schools were destroyed by coalition bombs. All eight of Iraq's large hydropower dams were hit, and grain silos and irrigation systems were also badly damaged. Farmlands near Basra were inundated with saltwater as a result of allied attacks. More than 95% of Iraq's poultry farms were destroyed, as were 3.5 million sheep and 2 million or more cattle. The U.S./coalition bombed textile plants, cement factories and oil refineries, which helped to contribute to an environmental nightmare. When an American general was confronted with press reports of Iraqi women carting home buckets of filthy water from the Tigris River, he shrugged his shoulders and said, 'People say, "You didn't recognize that the bombing was going to have an effect on water and sewage". Well, what were we trying to do with the sanctions: help out the Iraqi people? What we were doing with the attacks on the infrastructure was to accelerate the effect of the sanctions'. In other words, civilians were the targets. What has changed? The only thing Bush, Blair and their minions fear is public opinion. But with the media amplifying their lies, and concealing the truths of the barbarity of this invasion, they need not have worried. The mainstream media at the very worst has presented the attack and occupation of Iraq as a flawed but well-intentioned act of 'liberation' and 'rebuilding'. People like Heiko swallow this garbage whole - in spite of the fact that civilians have been targeted by U.S./U.K. bombs in many previous conflicts. I have said it before and I will say t again (until in Heiko's mind it finally registers): Human rights does not enter into western policy (to be fair, it rarely enters into the policies of most nation states, only the stakes now are much, much higher). The people in Iraq, like those in Afghanistan before them, as well as in Nicaragua, Guatemala, Viet Nam, Korea, Iran, Haiti, Cuba etc. etc. etc. are, in the warped minds of western planners, UNPEOPLE. They are often an impediment to the prime objectives of our foreign policy, which is that the resources of other countries are "ours" by right and that we should pursue policies aimed at ensuring this "law" is enforced.

By Jeff Harvey (not verified) on 02 Dec 2004 #permalink

Hi dsquared,

If I read their graph right, they recorded 1 violent death in April 2003 and 3 or 4 in March.

And Ian, things don't always go downhill. The situation three weeks ago was really bad, particularly in Mosul, Fallujah of course, and also Baghdad.

Some of the soldiers you are counting up here died in accidents, from injuries received earlier or were actually stationed in Afghanistan. The rest got killed by improvised explosive devices, a pretty permanent scourge, and in Anbar province, which may now be under full Iraqi government control, but is still very dangerous.

Still being under full government control is progress compared to the unacceptable situation prevalent between April and November, when the terrorists were free to have torture chambers in Fallujah, cut people's heads off and send propaganda videos off to Al-Jazeera showing them slowly sawing while their victim first screams and then continues to suffer as his vocal chords are cut (you know the word beheading is not quite adequate to describe what they actually do, it sounds much too clinical).

By Heiko Gerhauser (not verified) on 02 Dec 2004 #permalink

....its strange that these terrorists weren't in Iraq until after the invasion, and this is exactly what many critics of the invasion said would happen beforehand. Moreover, lets not forget that many of these terrorists were trained, funded, and made by the CIA in its initial campaign to support opponents of Russia during the Afghan war in the 1980's. Lastly, the U.S. armed, aided and abetted remnants of the Somosza guard (the notorious 'contras') who did the same kinds of things to civilians in Nicaragua during the 1980's that the terrorists Heiko describes are doing to civilians and foreigners in Iraq. But this does not register on Heiko's blinkered morality screen, which states: whatever the U.S. and U.K. do is good (after all, we are the good guys, the noble defenders of freedom and democracy); whatever "they" (the bad guys) do is evil. Moreover, Heiko loathes to discuss the terrorism of states, which makes groups like Al Queda look like lethal fleas (at least in terms of killing potential). Why? Because no matter how many people "we" slaughter, its always for a good cause. Thus, in Heiko's twisted lexicon, when "we" attack "them", its business-as-usual; when "they" attack "us", the world is coming to an end. The sheer hypocrisy of Heiko's views should preclude them from any rational discourse.

By Jeff Harvey (not verified) on 02 Dec 2004 #permalink

Hi Jeff,

Zarqawi was in Iraq all right, and so were plenty of Baathists spreading terror rather legally. Zarqawi was Saddam's guest, and ally of convenience, and the Baathists weren't terrorists, because they were acting as part of the government structure.

Some additional terrorists have likely headed towards Iraq since the liberation, because they see it as a good place to humiliate the Americans and thwart their plans.

Bin Laden has said that he has never received any aid from the US. The US did support "resistance" against Soviet forces in Afghanistan.

The US also supported the Soviets in WWII. Hitler was the greater danger, and realistically, the US couldn't fight both together.

On "state terrorism", this is just a matter of definitional clarity. I'd refer to war crimes, or crimes against humanity. While terrorism is a term restricted to sub-state groups.

Who is "we" and "them" in your argumentation, Jeff?

If "them" is terrorists sawing off heads in a manner that ensures a minute's worth of hard to describe pain and terror in the victim, and "us" is anybody really, like you or myself, then, I would indeed describe "they" attack "us" as unambiguously evil.

We are somewhat talking past each other, because I don't think you quite get my view of history and of the way the world is developping.

In short, life used to be short and brutish. 90% of people were subsistence farmers. Wars of conquest, discrimination and dictatorship were normal. Child mortality was 50%. There were few "democracies" and most of those were really more like oligarchies, ie the rule of the few (free men, or wealthy free men) over the many (every body else, ie slaves and women).

This describes most of history until very recently. The vast majority of the change is confined to the last 150 years. Life expectancy in Africa today is ten years or so higher than it was for blacks in the US 40 years after the abolition of slavery. In 1800, the US, in many respects, was less free and tolerant than Afghanistan under the Taleban. Blacks were slaves with few rights. Most of the population was subsistence farmers, literacy and urbanisation were extremely low. Native Americans were systematically persecuted in genocidal fashion. Women didn't have the vote, and of course, if they were so unfortunate to be black, their owners were free to legally rape them.

Even just going back 20 years, you find the spread of democrcacy remarkable. It has expanded in wildfire fashion. Just 60 years ago, WWII resulted in the deaths of 60 million, and Europe had few democracies and a history of near permanent conflict. Today, war between France and Germany has become unthinkable. My grandmother was driven out of France for no fault other than being German in 1917. Now I can live in France as a citizen with equal rights.

Wars of conquest (wars aimed at the outright annexation of territory prosecuted by one established souvereign state against another) have virtually ceased. Over the last 30 years, I am only aware of two examples, both incidentally perpetrated by Saddam Hussein (Iran-Iraq war and the invasion of Kuwait).

Why do Norwegians, who are the third largest exporters of crude oil in the world and who are virtually defenseless, have nothing to fear of the United States or the European Union?

It's not because Germans today are better people than 60 odd years ago, when Hitler did invade Norway.

The difference is the "system" in place, freedom of the press, education, communications technology, independent courts, prosperity and so forth.

I believe that in a 100 years time, there will be no more wars. The whole world will be where the European Union, Japan, the US etc. are today, with war unthinkable and armies abolished.

That's far from far fetched, when you look at the last 100 years and how Europe has developed over that period. No established democracy has ever attacked another. Democracies only attack dictatorships.

By Heiko Gerhauser (not verified) on 02 Dec 2004 #permalink

Heiko's claim that Zarqawi was in Iraq as Saddam's guest is contradicted by the CIA which admits there's no conclusive evidence he was ever in Iraq. http://www.msnbc.msn.com/id/6189795/. Saddam was vicious murdering thug, however he wasn't the ominpresent force of evil supporters of the invasion like to portray him as.

By Ian Gould (not verified) on 03 Dec 2004 #permalink

A small addendum to my previous post: there's no evidence Zarqawi was in Iraq prior to the invasion.

By Ian Gould (not verified) on 03 Dec 2004 #permalink

Same link:

"The CIA report concludes Zarqawi was in and out of Baghdad, but cast doubt on reports that Zarqawi had been given official approval for medical treatment there"

And the CIA isn't admitting anything here. It's an unnamed official being quoted.

Zarqawi has produced a video, where he is personally seen doing a beheading. He's got family in Jordan that have been interviewed etc.. He's definitely a real person, and he was in Baghdad for medical treatment before the invasion and after the US overthrew the Taliban.

I don't care how "official" his guest status was. Saddam had every reason not to advertise this particular link to terrorism too hard, and also every reason to accept Zarqawi's alliance.

By Heiko Gerhauser (not verified) on 03 Dec 2004 #permalink

I haven't seen anyone address this, yet. I would be grateful.
From jack straw, courtesy of jre's link.
"The figures derived from the survey's data on Fallujah would have resulted in an estimated 200,000 excess deaths within Fallujah alone over the past 18 months. This would amount to almost two-thirds of the total population of the town - which is just not credible. The authors of the study understandably discounted the data."
Is this true ?

By James Brown (not verified) on 03 Dec 2004 #permalink

Heiko Gerhauser wrote, Wars of conquest (wars aimed at the outright annexation of territory prosecuted by one established souvereign state against another) have virtually ceased. Over the last 30 years, I am only aware of two examples, both incidentally perpetrated by Saddam Hussein (Iran-Iraq war and the invasion of Kuwait).

You forgot Indonesia's annexation (with the blessings of Washington) of East Timor, which resulted in the slaughter of 1/6 to 1/3 of the population.

And don't forget that the US leaned heavily towards Iraq in the Iran-Iraq conflict.

The intelligence that place Zarqawi in Baghdad also stated that Iraq had and was developing WMD and that Zarqawi had a leg amputated while in Baghdad.

By Ken Miles (not verified) on 04 Dec 2004 #permalink

True, so far as it goes, but Mr. Straw is gliding over the fact that the most likely value is indeed near the center of that interval.

Could someone explain to me the reasoning behind treating the confidence interval as if it were a probability distribution? At least, that's what appears to be being done here. All my memories of graduate statistics classes tell me that this is not the correct way to interpret confidence intervals.

By David Fleck (not verified) on 09 Dec 2004 #permalink

David, a 67% confidence interval is half as wide, going from 50,000 to 150,000 for 2/3 of the confidence p value. Alternatively you could compute the probabilities for some reaonable prior distribution and you'll get a peak at around 100,000.

Nope, sorry, still not getting it. Here is how I've always understood the interpretation of confidence intervals:

A confidence interval is not a probability distribution. A confidence interval (CI) is a statement of, well, confidence in your estimate of a value; the standard 95% interval means that 95% of the time the CI range you have calculated will contain the real value (assuming your sample, and the population as a whole, meet the assumptions of whatever technique was applied).

It does not mean that the real value is more likely to be near the center.

Yes, a 67% CI will be smaller than a 95% CI - and a 99% CI will be wider. But that isn't saying that one part of a CI is more likely than another part. (Drags out dusty textbook: "We could have greater confidence that our interval covers the mean, but we could be much less certain about the true value of the mean because of the wider limits.[1]" - emphases mine.)

A large CI is the result of small sample size, large variability between samples, and non-normally distributed samples. You can think of the CI as a 'penalty' that is applied to an calculated value; an estimate based on small or highly variable samples will have a large 'penalty' applied, an estimate based on large samples with little variation will have little 'penalty'. When you have applied a 95% CI, you are saying (or should be saying) "My real value is somewhere in here; but based on the data I have, I can't be any more specific than that."

My grad. advisors would have whacked me upside the head if I told them, "well, my result has a really big CI, but the true value is probably close to the center."

I haven't read the study, so maybe they're doing some sort of data massaging that allows them to make this statement - but it strikes me as very unorthodox.

--

[1] Sokal,R.R, and F.J.Rohlf, Biometry. 2nd. ed., 1981, W. H. Freeman & Co. page 142.

By David Fleck (not verified) on 09 Dec 2004 #permalink

Actually, a 95% CI does not mean that there is a 95% chance that the true value lies in the interval. See here. But it doesn't really hurt to pretend that it is a 95% chance. If you do, then it follows that the middle of the interval is most likely. Consider the 67% interval. It's half as wide with 2/3 of the probability, so values in there are twice as likely as values further from the mean. Same thing applies for smaller confidence intervals. The confidence level does not shrink as fast as the width of the interval.

David, this can't be right:
When you have applied a 95% CI, you are saying (or should be saying) "My real value is somewhere in here; but based on the data I have, I can't be any more specific than that
If this were true, you would be able to have a maximum of one confidence interval. If you can say that the 95% CI is 100-120 and the 90% CI is 105-115, then you're being more specific than that.
If you're calculating confidence intervals at all, then you are assuming an implicit sampling distribution of the test statistic under consideration. The midpoint of the confidence interval is associated with the mean of the sampling distribution. Ignoring degenerate and multimodal cases, this is also the expectation of the true value.
Think about it this way; Bayesian statistics works. Maximum likelihood estimation works. Both these methods of estimation explicitly rely on the fact that the central point estimate is in some way the local maximum of some objective function of interest, which in most normal estimation situations is your best estimate of the true value.

If this were true, you would be able to have a maximum of one confidence interval. If you can say that the 95% CI is 100-120 and the 90% CI is 105-115, then you're being more specific than that.

No, you're not; you're saying the same thing in two different ways. In the one case, you're saying that 95% of sample means +/- 10 (the 95% CI) will cover the true mean; in the other, that 90% of sample means +/- 5 will cover the true mean. That's the tradeoff - you can get a tighter CI, at the cost of certainty that the true mean is indeed covered by your estimate.

You could get an even narrower range in your CI, by reducing the likelihood of it being correct - take a 50% CI, for example. Say that for the above example, the 50% CI is (sample mean +/- 2). This means, in effect, that only 50% of your sample means +/- CI will cover the true mean - a 50/50 chance. As the CI narrows, 'confidence' evaporates.

David, this can't be right

(shrug) It's what my references say, and it's what I remember being pounded into my head in graduate school.
If you can point me to a reference that describes CI's being used in this way (that is, indicating that central tendency in a CI more likely than other values), I would be most interested. My references strongly indicate otherwise.

By David Fleck (not verified) on 09 Dec 2004 #permalink

a 95% CI does not mean that there is a 95% chance that the true value lies in the interval

I realize that; that's why I didn't say it.

it follows that the middle of the interval is most likely

...but I think this is incorrect; see the Sokal & Rohlf quote above. A 67% CI means that 67% of calculated values +/- the CI will include the mean; a 50% CI means that only half of your estimates will include the mean; the CI is a measure of vagueness, not a probability distribution. Back to Sokal & Rohlf:
"By increasing the degree of confidence still further, say to 99.99%, we could be virtually certain that our confidence limits contain the population mean, but the bounds enclosing the mean are now so wide as to make our prediction far less useful than previously". (Emphases mine.)

By David Fleck (not verified) on 10 Dec 2004 #permalink

David, have a look at the rest of my post. Your references don't say what you think they say. If that's not working, think about it this way; your confidence interval is constructed *around* a central estimate. There is a reason for that. In fact, in any other than extremely recherche robust estimation situations, you calculate the point estimate first and the confidence interval second. The point estimate is (usually) the maximand of the likelihood function, which is why it's correct to call it the "most likely" value.
Also, I'm not "saying the same thing in two ways" in my example. I could have made the point differently and said that the 90% CI was 101-109. This would have implied a different shape for the sampling distribution and focused in on a different point estimate.
Your own memory of your references could lead you to the same place. As the CI narrows toward the central estimate, you finally reach a point where (starting at the 95% CI which for a symmetrical distribution implies that there is a 2.5% chance that the true value is higher than the higher bound and a 2.5% chance that the true value is lower than the lower bound), you are at the point estimate where there is a 50% chance that the true value will be greater and a 50% chance that the true value will be smaller. This is (by definition) the median of the sampling distribution; under reasonable assumptions, it will also be the expectation.

David, have a look at the rest of my post. Your references don't say what you think they say.

Maybe not. But I don't think your arguments have addressed what my references actually say; I'd still like to see a reference that indicates that the mean of a sample is more likely to be the correct value than other values within the CI. I don't see how "...as your confidence increases your statement becomes vaguer and vaguer, since the confidence interval lengthens.[1]" can be interpreted in the way you would appear to try to interpret it.

Sorry, I'll have to take my interpretation, plus what I was taught, plus what is common usage in the science I'm most familiar with (biology) over your interpretation - which I consider to be fairly idiosyncratic. Maybe not - maybe they do it that way in the medical sciences. I would be happy to be shown to be incorrect. With references.

If that's not working, think about it this way; your confidence interval is constructed *around* a central estimate.

true

There is a reason for that.

true

...you calculate the point estimate first and the confidence interval second.

generally, though you can do both simultaneously

The point estimate is (usually) the maximand of the likelihood function, which is why it's correct to call it the "most likely" value.

The most likely value for the given sample. But the reliability of that sample, as an estimate of the whole population, will depend on its size and variance. The more variance and the smaller the sample, the less reliable it is as an estimator.

--

[1] Sokal,R.R, and F.J.Rohlf, Biometry. 2nd. ed., 1981, W. H. Freeman & Co. p. 142.

By David Fleck (not verified) on 10 Dec 2004 #permalink

Tim-

Thank you. This looks like a good place to start reading.

By David Fleck (not verified) on 10 Dec 2004 #permalink

Tim-

Thank you. This looks like a good place to start reading.

By David Fleck (not verified) on 10 Dec 2004 #permalink

Whoops! P.S.: I still have a strong suspicion that you and Davies are mistaken. However, I need to do a bit of reading before responding further.

By David Fleck (not verified) on 11 Dec 2004 #permalink

OK, I've refreshed my memory on maximum likelihood estimation, thanks to Tim's URL. However, nothing I've run across has answered my initial question

"In what way is it legitimate to say that the mean of a sample is more likely to be the correct value than other values within the CI?"

...while I do have statistical manuals that warn against this practice.

Look, I suspect this is all fruitless; nothing you or Davies has brought up so far has convinced me that my interpretation is incorrect, and I suspect nothing I've said has made any impression on you. This discussion thread is dead, and I'm going away from the Internet for a few weeks anyway.

I'll just make one last point - providing a CI range [8,000 - 198,000] strikes me as a better way to summarize these data than providing a point estimate, because you encapsulate both the estimate and its precision. Without both pieces of data, you're not presenting enough information for someone to assess the result.

(For anyone interested, a very easy-on-the-layman discussion of these matters is presented here, especially 'Part II'. Question: which of the example studies here is the Lancet study most like?)

I'm off. Happy HanuRamaKwanzmass, as we say in these parts.

By David Fleck (not verified) on 18 Dec 2004 #permalink