There is no doubt Iraq is a dangerous place and probably getting more deadly by the month. But how many Iraqis have really died there since the United States invaded the country in 2003 and toppled President Saddam Hussein?

Is it roughly 49,000 civilian deaths, the upper estimate of an independent group called Iraq Body Count, which takes its numbers from local media reports and triple-checks them, when it can, with calls to hospitals and morgues?

Or is it 12 times that estimate — 601,000 violent deaths since the war began, over and above what might have occurred anyway.

That is the newest figure just published by a team of Iraqi and U.S. doctors, as well as statisticians working out of the Bloomberg School of Public Health at Johns Hopkins University in Maryland.

Both approaches have their detractors. The Body Count number is often said to underestimate the true death toll of the war because ordinary Iraqis have become so inured to the conflict they are simply burying their dead where they can, without resorting to hospitals, morgues or official death certificates.

The Johns Hopkins estimate (this is their second one) is even more controversial in some circles. That is because it is so at odds with official estimates and because it is based on a random survey of 50 neighbourhoods scattered throughout Iraq, with the results extrapolated to the country as a whole.

Its proponents argue that such epidemiological studies have historically proven to be much more exact than the so-called passive surveillance methods (like that of the Body Count group).

This one was extensively peer-reviewed and was published in the prestigious British medical journal The Lancet. (See The Human Cost of the War in Iraq, a Mortality Study in The Lancet)

But it is still based on a number of assumptions, the chief one being: Can you do a truly random "household survey" in a country as chaotic as Iraq?

Fun with figures

This is the second survey from the doctors and statisticians at Johns Hopkins and al Mustansiriya University in Baghdad. The first, published two years earlier, suggested nearly 100,000 Iraqi civilians had died as a result of the war.

This one, using the same methodology but a slightly larger sample, basically restates the earlier study and comes out with an estimate that is six times greater. One of the reasons for this is that the death rate in the past year, the study claims, is nearly three times what it was in the first year of fighting.

What the researchers did was send teams to 50 randomly selected sites in 16 of Iraq's 18 regions over a two-month period this summer.

At each site researchers would choose a household at random, interview its occupants about anyone who had lived there and died over the past four years, going back to January 2002, almost 1½ years prior to the invasion. They would then repeat the process in 39 of the nearest homes.

The researchers said they were presented with death certificates in 92 per cent of the cases they inquired about. And in the end they ended up with data on 12,801 people in 1,849 homes in which there had been 629 deaths and 1,474 births between January 2002 and June 2006.

The sample - 12,801 people - represents slightly less than .05 per cent of the Iraqi population. Though the actual "polling sample" is much smaller, proabably fewer than 2,000, as the researchers took their information from the heads of households in most cases.

From these numbers, the researchers extrapolated the results to the entire 26.1 million Iraqi citizens and concluded that there were somewhere between 426,369 and 795,663 violent deaths in the country since 2003, over and above what would have happened under Iraq's pre-war death rate.

The most statistically probable number, it said, was 601,027, the one that made all the headlines. It represents almost 2.5 per cent of the Iraqi population.

The confidence factor

Extrapolating from small random surveys like this to entire populations is an accepted form of statistical study and is often used by health scientists to track the incidence of disease or such things as childhood obesity in developed countries.

It is the technique used by researchers to estimate deaths in war-torn countries such as the Congo or Sudan. Small representative groups are surveyed and the results are extrapolated to the larger population.

The United Nations is testing a pilot project of using small random surveys like this, with very similar methodology, to help it gauge the success of some of its humanitarian efforts.

Still, these techniques rely on a number of assumptions and adjustments to account for important variables, which in Iraq would include a highly mobile and fearful population, widespread sectarian violence, and a conflict that has been much more intense in certain parts of the country than in others.

The first issue here: Iraq's pre-war mortality rate. The first Johns Hopkins study from 2004 pegged it at five per every 1,000 population, based on what those interviewed recalled. This one was 5.5/1,000.

But UN reports had suggested Iraq's crude death rate was higher than this in the 1980s and '90s. It was in at least the 6.8/1,000 range and rising, which would make the difference between normal deaths and what the researchers called "excess deaths" brought about by the war quite a bit smaller.

The other issue here is what statisticians call the "confidence interval."

In their first study in 2004, the confidence interval was huge: The Johns Hopkins team basically concluded that it had 95-per-cent confidence the war had caused somewhere between 8,000 and 194,000 extra deaths by that point. Its 100,000 figure was the most probable number on this large continuum, which of course assumes violent deaths in Iraq can be plotted on the same bell curve as, say, breast cancer rates in North America.

That earlier study interviewed 30 households in each of 33 neighbourhoods. This one roughly 40 households in each of 50 sites and as a result the confidence continuum has narrowed considerably to between 426,369 and 795,663 — which is still quite a range.

If this was a political poll, it would be like saying a prime minister's popularity was anywhere between 35 and 65 per cent.

Random versus actual

So which is more accurate? The random, detailed survey extrapolated to a population at large? Or the "passive" actual account attempted, in the face of considerable odds, by groups like Body Count?

The full story of civilian deaths in this war will probably never be known until there is stable government in Iraq and someone can do a proper census.

But whatever you might think of this latest Johns Hopkins study (which, coincidentally, has turned up twice now in the midst of U.S. election campaigns), it has at least helped put a face on the type of violence confounding Iraq.

Of those surveyed, researchers found 56 per cent of those who died violently were killed by guns. Slightly more than 27 per cent were killed by explosions, including car bombs, and 13 per cent lost their lives because of American air strikes (which of course were heavily concentrated in key areas).

Not surprisingly perhaps, the largest group killed were young males under 30. There was not an unusual number of women killed, the study found: The death rate for women in Iraq was about the same as that for other countries.

But among female deaths, the largest single group was, quite noticeably, those under 15.