According to Bill Roggio editor of the the Long War Journal, “There is great emphasis put on limiting civilian casualties. This is why vehicles are often hit. And the weapons selection —smaller, more focused warheads — are also used specifically to limit civilian casualties.”

Paradoxically, however, near the locus of the attacks, the Waziristan agencies, where 95 per cent of the strikes have occurred, there has been little sign of protest, and anecdotal evidence suggests locals may actually be in favour of the strikes.

No reporter or official was willing to talk on the record about the strikes; however, the various accounts offered were strikingly similar and were corroborated by Pakistani and American analysts.

Referring to the lack of protests, a reporter claimed “Imran Khan and the (Jamaat-i-Islami) are always talking about mass civilian casualties. But in Peshawar, Charsadda, Tank, Bannu, nowhere in the areas adjacent to the tribal agencies where attacks have been common have there been protests. They simply don't exist.”

A senior journalist averred that many politicians protesting the drone strikes were guilty of dissembling “Privately all the Fata members of parliament support the strikes. They can't say it publicly because of the politics involved, but off the record they claim the strikes are great.”

There is little doubt civilians have been killed in the air campaign. Though the first attack occurred in June 2004, the vast majority of the 130 strikes have taken place over the past two years, following a change in American strategy the previous focus on 'high-value targets' was abandoned in favour of a broader campaign that includes lower-value targets. 85 strikes have occurred on President Obama's watch.

With more missiles being fired in the hunt for a wider range of targets, analysts have attempted to determine the impact on civilians. But the significant differences in the estimates of civilian fatalities indicate the difficulties involved in such assessments.

What is fairly clear is that the air campaign has grown increasingly accurate. Amir Rana, who runs the Pak Institute for Peace Studies, claimed, “Since late 2009, the targeting has improved. Before, in late 2008, early 2009, there was high collateral damage.”

The lowest estimate of civilian fatalities is 94 — less than 10 per cent of estimated total fatalities — made by the Long War Journal, a blog which tracks drone strikes.

Bill Roggio, managing editor of the LWJ, explained his methodology “The numbers I use are primarily from Pakistani press reports ... given that there is more 'incentive' to report on civilian casualties. Yet casualties have been low regardless. Oddly enough, I don't see many reports of the Taliban disputing the casualties; in fact they more than often confirm the reports.”

A widely-quoted study released by the New America Foundation in February estimates that between 830 and 1,210 civilians have been killed by drones since 2004, 30 per cent of estimated total fatalities.

Katherine Tiedemann, who co-authored the study, explained “We rely only on media outlets with deep reporting capabilities in Pakistan, like Dawn, Geo, NYT, the wires, BBC, etc. Often there are differing accounts of the numbers of people killed BBC will say 6, AFP will say 8, and so on. So I keep track of the high and the low, giving a range of possible figures. I specifically count those described in reliable press accounts as militants.”

The problem is that no one — not the news wires, not the foreign media, not even Pakistani papers or news channels — has direct access to the site of a strike.

A senior journalist said, “There are essentially four sources for information on casualties the local correspondent, the militants themselves who call up, the politicians and intelligence people.

“The local correspondent is closest to the site and best placed to give an unbiased report. But he faces various pressures. Often the Taliban call up and say, 'Put in your report that so many children and women were killed.' The guy can't refuse.”

A reporter who said he has filed reports on dozens of strikes concurred, “There are dangers we have to constantly be aware of, both to our families and ourselves. We can't always write the truth.”

However, the accuracy of the body count is only part of the problem, the other major difficulty being that of classification.

As one Pakistani official asked, “Who is a militant? A bodyguard? A driver? A cousin or guest sitting in a hujra next to a militant?”

The difficulty is epitomised by the strike which killed Baitullah Mehsud and, it is believed, his second wife and several bodyguards last August.

Khalid Aziz, a former political agent of North Waziristan and critic of the strikes, argued “There was nothing against (Baitullah's) wife and children. Going by this theory, Nixon's entire family should have been impeached.”

The differences over classifications are believed to have grown so acute that over the past year American officials have been providing data to their Pakistani counterparts on fatalities. According to an official who is privy to official communications between the US and Pakistan, “The data shows that since March 2009 there has been no civilian casualties.”

Self-servicing definition

But that only appears possible because of an increasingly elastic definition of 'combatant' being applied by the CIA, which runs the drone programme in Pakistan.

Roggio, of the Long War Journal, explained “The CIA is classifying drivers, bodyguards and such as combatants. The CIA is not losing sleep over their deaths.”

Scott Horton, a contributing editor at Harper's magazine and an expert on the law of armed conflict, highlighted the problems “Under international law, and specifically under the Geneva Conventions, those who render material aid are not so clearly legitimate targets as the US sometimes makes out.

“Bodyguards are clearly legitimate targets, and family members are not. In the proportionality analysis, however, it can be argued that family members recognise and accept risk of being targeted by remaining in the immediate proximity of their relative who is a command-and-control figure. Still, I would argue that the US view pushes the outer pockets of currently accepted law.”

Despite these problems, the reporters argued that locals are in favour of the strikes, though perhaps only given the alternatives. One said “The local leaders say, 'Better the drones because at least they are accurate. The (Pakistani) fighter jets and artillery cause so much more damage and loss.'”

Some reporters suggested that the lack of civilian casualties can be gauged from the rarity of compensation claims. A correspondent from the Waziristan agencies claimed “An NGO has been going around in South and North Waziristan for the last four months trying to document cases of civilian casualties and offer compensation. But not a single claim has been lodged.”

Two factors appear to be driving down civilian fatalities in recent months. First, the drone onslaught has led the militants to rethink their movements.

A reporter asserted, “(The militants) don't sleep among the population anymore because they are afraid of being exposed. They sleep in ditches, and in the mountains in tents.”

The second reason appears to be adjustments made by the CIA. According to Roggio, “There is great emphasis put on limiting civilian casualties. This is why vehicles are often hit. And the weapons selection —smaller, more focused warheads — are also used specifically to limit civilian casualties.”

Horton, though, opposed the very idea of CIA ownership “I strongly criticise the CIA's operation of a drone war programme ... it should be managed by the military — they have more depth in making the key judgements required.”

But on the ground, in Fata itself, these arguments appear to be largely academic. Amir Rana was emphatic “The strikes can't happen without local intelligence. Intel for such an intense campaign can only come if locals aren't opposed to the strikes.”