655,000

#76
#76
CSpindizzy said:
Read the words really slow in my post.....you're killing me here.
Click to expand...
Funny how you seem to discredit Cordesman in one post, saying something to the effect of "what does he know about this issue," then you try to use him to further your argument. Sure, you found a statement from him saying that in the first half of 2006 approximately 80 Iraqi's died per day. Of course, you will no doubt discredit his number of only 50,000 civilian casualties thus far in the war.

Off the topic a little, I could honestly care less about what you think about me. Keep up the "read slowly" comments, it does nothing to further your cause. Just makes you look like an *ss.
 
#77
#77
therealUT said:
Funny how you seem to discredit Cordesman in one post, saying something to the effect of "what does he know about this issue," then you try to use him to further your argument. Sure, you found a statement from him saying that in the first half of 2006 approximately 80 Iraqi's died per day. Of course, you will no doubt discredit his number of only 50,000 civilian casualties thus far in the war.

Off the topic a little, I could honestly care less about what you think about me. Keep up the "read slowly" comments, it does nothing to further your cause. Just makes you look like an *ss.
Click to expand...

You need to venture into comedy. Let's run with your thinking since it's so easy to follow...

I did not discredit him. I actually required from you some of his credentials and his logic since you seem to like to attack anyone without presenting such info. You throw Cordesman's name out there and offer no substance. So frankly me asking you for the facts is discrediting you not him.

I can find several sources to back up the 80+ a day number. But I see nothing substantive of only 50K since the beginning of this conflict. I ask you for proof of those numbers and you have yet to offer anything to prove that.

I could care less what you think about me personally. Frankly a person who comes on here and discredits others without offering countered facts and substance looks worse than an @ss. You are quick to attack this methodology and numbers but have not backed your claim up at all. When you add some fact and substance to your argument, you'll be someone credible to debate with. Until then, people like that are usually labelled whiners.....
 
#78
#78
Okay gentlemen - to get back on topic:

First a few comments regarding sampling methodology. Appropriate sample size is determined by 2 key criteria: population size and variance in the "issue" being investigated.

The first issue (population) taps out pretty quickly. In otherwords, polling data in Iraq and the US would require roughly the same sample size eventhough the US is more than 10x the size of Iraq.

The second issue is where the rub lies. For political polling, the variance in the "issue" e.g. Bush/Kerry is relatively low. People fall into one of a few categories (for/against/undecided). The incidence rate is very high - likely voters are asked and they are likely to fall into one of the categories.

When it comes to did someone in your family get killed as a result of the war, the variance is much greater. Or put another way, the incidence rate is much lower. Most families don't have someone that was killed - at most it is 1 out of 10. More likely it is less since you might see a group wiped out by bomb.

More in the next post...
 
#79
#79
Now - onto some questionable issues in the study.

First - the baseline. Data was collected in from May to July 2006. They determined a baseline (normal deathrate) by asking back to Jan 2002. They then determined how many deaths occured in the period of Jan 2002 to Mar 2003 (14 months) and determined that to be the expected rate of death.

In total the recorded 610 deaths over the total time period of Jan 2002 to July 2006. Roughly 1/2 of these are used to compute the 600,000+ violent deaths in excess of expected deaths (or due to the war).

In all, the survey team recorded 70 deaths from 1849 households for the pre-invasion baseline. This baseline is critical since the comparison is made relative to this (eg. excess deaths). Further, the study concludes that ALL violent deaths occuring in the sample period are attributable to the war since very few were considered violent in the baseline. Are we to believe that NO violent deaths occured prior to the invasion? This study says yes since only 2 of the 70 were considered violent and one of those is attributed to pre-war coalition activities. Here is the first clue that you have an extrapolation problem. To extend these findings at the rate of the study (as the authors do) of 5.5/1000 - you reach a number of 143,000 annual deaths in Iraq - none of which are considered to be violent (non-accidental or natural cause). Do you believe that everyone that died in Iraq in 2002 died of either an accident or natural causes? That is the assumption the researchers used in their projection. Who'd have thought there were no murders in Iraq prior to the war.

Further a few household difference in sampling could yield a signficantly different number here -- especially because the incidence rate is so low.

Next issue. Cause of violent deaths. While the authors claim violent death due to coalition actions have diminished (as a percentage of total violent deaths), the numbers show that 1/2 of the violent deaths due to airstrikes occurred in the last year (20 out of 40 in the sample). So Shock and Awe, Fallujah, etc. did less damage to civilians than airstrikes in the last year? Doubtful. However, when you project off of small numbers 40 total airstrike deaths in the sample (of 289 total violent deaths) you can get some weird numbers.

It appears we do have a potential sampling problem here. Polling data of 1200 registered voters can be accurate (if chosen correctly) since all share the common characteristic of being a voter. If you get a few too many dems or reps, you will slightly skew the results but you still have many others in the sample to make up for it. When you have incidence rates as low as in this study (one average only about 2% of the households in the study had an airstrike death over the 4 year period) getting the wrong household can have a major impact. For example, if one families house was hit by an airstrike, several members might have been lost - the chance for RSE (random sampling error) is high. More likely, much less than 1% of households had such a death due to the likelihood of multiple deaths.
 
#80
#80
A bit more on sampling for polling vs. cases with low incidence rates.

Assume a poll of Presidential approval rating based on 1000 respondents (common).

Approve - 36%
Disapprove - 52%
Neither - 12%

If due to sampling error we had 10 (1%) excess (out of proportion with the population) respondents classifed as disapprove when they should have been approve, we see the following changes to the polling data

Approve 37%
Disapprove 51%
Neither 12%

No material change and well within the typical MOE of +/- 3% points.

Now lets look at a case with low incidence rates. If due to sampling error 18 (10% as in the example above) households with no deaths in 2002 were included in the sample (out of proportion with the population). Deaths per thousand go from 5.5 to 7.8 or so. The baseline is 42% higher. Excess death estimates would be cut nearly in 1/2 due to a relatively small sampling error. Add to this the complication that households most likely represented multiple deaths (e.g. the incidence rate of hitting a household with a death is lower since the deaths may cluster to some extend within households).
 
#81
#81
One last "statistic" from the study that should raise some eyebrows:

Extrapolating (in the same method used by the authors) airstrike deaths in the last year (July 05-July 06) would yield a conclusion of 41,500 deaths!

(20 recorded deaths attributed to airstrikes in 05-06/289 total violent deaths recorded) * 600,000 projected violent deaths.

Anyone think this is an accurate number?

Conversely, airstrike deaths during the first year (including Shock and Awe) = 12,500.
 
#82
#82
volinbham said:
Okay gentlemen - to get back on topic:

First a few comments regarding sampling methodology. Appropriate sample size is determined by 2 key criteria: population size and variance in the "issue" being investigated.

The first issue (population) taps out pretty quickly. In otherwords, polling data in Iraq and the US would require roughly the same sample size eventhough the US is more than 10x the size of Iraq.

The second issue is where the rub lies. For political polling, the variance in the "issue" e.g. Bush/Kerry is relatively low. People fall into one of a few categories (for/against/undecided). The incidence rate is very high - likely voters are asked and they are likely to fall into one of the categories.

When it comes to did someone in your family get killed as a result of the war, the variance is much greater. Or put another way, the incidence rate is much lower. Most families don't have someone that was killed - at most it is 1 out of 10. More likely it is less since you might see a group wiped out by bomb.

More in the next post...
Click to expand...

volinbham said:
Now - onto some questionable issues in the study.

First - the baseline. Data was collected in from May to July 2006. They determined a baseline (normal deathrate) by asking back to Jan 2002. They then determined how many deaths occured in the period of Jan 2002 to Mar 2003 (14 months) and determined that to be the expected rate of death.

In total the recorded 610 deaths over the total time period of Jan 2002 to July 2006. Roughly 1/2 of these are used to compute the 600,000+ violent deaths in excess of expected deaths (or due to the war).

In all, the survey team recorded 70 deaths from 1849 households for the pre-invasion baseline. This baseline is critical since the comparison is made relative to this (eg. excess deaths). Further, the study concludes that ALL violent deaths occuring in the sample period are attributable to the war since very few were considered violent in the baseline. Are we to believe that NO violent deaths occured prior to the invasion? This study says yes since only 2 of the 70 were considered violent and one of those is attributed to pre-war coalition activities. Here is the first clue that you have an extrapolation problem. To extend these findings at the rate of the study (as the authors do) of 5.5/1000 - you reach a number of 143,000 annual deaths in Iraq - none of which are considered to be violent (non-accidental or natural cause). Do you believe that everyone that died in Iraq in 2002 died of either an accident or natural causes? That is the assumption the researchers used in their projection. Who'd have thought there were no murders in Iraq prior to the war.

Further a few household difference in sampling could yield a signficantly different number here -- especially because the incidence rate is so low.

Next issue. Cause of violent deaths. While the authors claim violent death due to coalition actions have diminished (as a percentage of total violent deaths), the numbers show that 1/2 of the violent deaths due to airstrikes occurred in the last year (20 out of 40 in the sample). So Shock and Awe, Fallujah, etc. did less damage to civilians than airstrikes in the last year? Doubtful. However, when you project off of small numbers 40 total airstrike deaths in the sample (of 289 total violent deaths) you can get some weird numbers.

It appears we do have a potential sampling problem here. Polling data of 1200 registered voters can be accurate (if chosen correctly) since all share the common characteristic of being a voter. If you get a few too many dems or reps, you will slightly skew the results but you still have many others in the sample to make up for it. When you have incidence rates as low as in this study (one average only about 2% of the households in the study had an airstrike death over the 4 year period) getting the wrong household can have a major impact. For example, if one families house was hit by an airstrike, several members might have been lost - the chance for RSE (random sampling error) is high. More likely, much less than 1% of households had such a death due to the likelihood of multiple deaths.
Click to expand...

volinbham said:
A bit more on sampling for polling vs. cases with low incidence rates.

Assume a poll of Presidential approval rating based on 1000 respondents (common).

Approve - 36%
Disapprove - 52%
Neither - 12%

If due to sampling error we had 10 (1%) excess (out of proportion with the population) respondents classifed as disapprove when they should have been approve, we see the following changes to the polling data

Approve 37%
Disapprove 51%
Neither 12%

No material change and well within the typical MOE of +/- 3% points.

Now lets look at a case with low incidence rates. If due to sampling error 18 (10% as in the example above) households with no deaths in 2002 were included in the sample (out of proportion with the population). Deaths per thousand go from 5.5 to 7.8 or so. The baseline is 42% higher. Excess death estimates would be cut nearly in 1/2 due to a relatively small sampling error. Add to this the complication that households most likely represented multiple deaths (e.g. the incidence rate of hitting a household with a death is lower since the deaths may cluster to some extend within households).
Click to expand...

volinbham said:
One last "statistic" from the study that should raise some eyebrows:

Extrapolating (in the same method used by the authors) airstrike deaths in the last year (July 05-July 06) would yield a conclusion of 41,500 deaths!

(20 recorded deaths attributed to airstrikes in 05-06/289 total violent deaths recorded) * 600,000 projected violent deaths.

Anyone think this is an accurate number?

Conversely, airstrike deaths during the first year (including Shock and Awe) = 12,500.
Click to expand...

If you can't convince them, confuse them. :wacko:
 
#83
#83
OWB said:
If you can't convince them, confuse them. :wacko:
Click to expand...

Sorry if you don't follow it :)

Part of my job is reviewing and writing academic articles such as this. Just using the old tools of the trade so to speak.
 
#84
#84
volinbham said:
Sorry if you don't follow it :)

Part of my job is reviewing and writing academic articles such as this. Just using the old tools of the trade so to speak.
Click to expand...

I could probably understand it if I wasn't so tired and actually felt like reading it all. Maybe tomorrow. :)
 
#85
#85
Well let's see. Looking at the math, they say about 600K were killed during this 2.5 year period. This essentially means that they are figuring 240K per year are dying. If the normal rate is approximately 143K per year prior to the war, reason states that this is less than doubled. So for each 'normal' pre-war death, they are attributing one additional death or basically doubling it. They are also saying that only that percentage of 90% or close to it is attributed to violence WITHIN the ADDITIONAL DEATHS count.

Is it hard to believe that the death rate doubles when the nation goes from no conflict to constant conflict and primarily in urban areas?
 
#86
#86
CSpindizzy said:
Well let's see. Looking at the math, they say about 600K were killed during this 2.5 year period. This essentially means that they are figuring 240K per year are dying. If the normal rate is approximately 143K per year prior to the war, reason states that this is less than doubled. So for each 'normal' pre-war death, they are attributing one additional death or basically doubling it. They are also saying that only that percentage of 90% or close to it is attributed to violence WITHIN the ADDITIONAL DEATHS count.

Is it hard to believe that the death rate doubles when the nation goes from no conflict to constant conflict and primarily in urban areas?
Click to expand...
Read the study, it is 655K more deaths than normal. It is more than double the pre-war death rate in Iraq.
 
#87
#87
According to the researchers, the overall rate of mortality in Iraq since March 2003 is 13.3 deaths per 1,000 persons per year compared to 5.5 deaths per 1,000 persons per year prior to March 2003.
 
#90
#90
How is slightly doubling a large jump when you have dozens killed on numerous days just from bombings, assassinations, etc?
 
#91
#91
CSpindizzy said:
How is slightly doubling a large jump when you have dozens killed on numerous days just from bombings, assassinations, etc?
Click to expand...
Because to double it you would have to have more than 'dozens' killed each day. You would have to have 391 extra people die every single day. That is quite a large leap.
 
#92
#92
In a nation where every major urban center is seeing constant attacks and has since the beginning of this in March 2003?
 
#93
#93
CSpindizzy said:
According to the researchers, the overall rate of mortality in Iraq since March 2003 is 13.3 deaths per 1,000 persons per year compared to 5.5 deaths per 1,000 persons per year prior to March 2003.
Click to expand...

It's not linear though - they claim over 3x the "normal" death rate in the last year. During the year of the invasion with the massive bombing and march through the country, deaths were up a mere 36% (51,000). During the last year, they attribute an additional 372,000 deaths or so. Are we to believe that the death rate has gone up 7x since the first year of the war?

What's more is that they are making these projections based off of observations of roughly 300 total (attributable) deaths. That 51,000 estimate in the first year of the war is based on 19 total attributable deaths. Find one more household where 3 members got killed in a car bomb and your total estimate goes up over 8000! Likewise, pick a different house in Ramadi -- one that wasn't next to an aerial strike and drop 5 killed and your estimate drops about 11,300!
 
#94
#94
Well if the spin that our initial invasion was clean and with little casualties, that expansion AFTER "Mission Accomplished" would not be far fetched.

They polled households but attributed totals to the population numbers NOT the households.
 
#95
#95
CSpindizzy said:
They polled households but attributed totals to the population numbers NOT the households.
Click to expand...

Correct - that's how I did the calculations. The problem is that we don't have a one-one relationship. We have no way of knowing if any given household in the sample had more than one death.

Here's where a sampling problem arises. It is likely that a household has experienced multiple deaths. Say a family had a car accident - one sample element (household) yields multiple "positives" or "yesses". According to their sampling method, a household averages about 7 people. Find one house with an airstrike incident and you might have as many as 5 deaths -- one sample element yielding 13% of all incidents recorded (5/40).

It's equivalent to doing polling data by choosing a large household and asking everyone in that household who they support in an election as opposed to gaining the same number of respondents from different households. Add to that, they are trying to project the outliers (the rare incident of an excess death) and you run into real instability in your projections.
 
#97
#97
CSpindizzy said:
Do you not think they applied the death ceritificates they viewed into this as well?
Click to expand...

I'm sure they did - it doesn't impact the sampling issue though.

The issue is not did they come across households where people had died. It is the aggregation then projection of those results.

If we knew where they found the deaths and more importantly how many per household we'd know a lot more about the projectability of the findings. In the likely case of multiple deaths per household, a sampling error becomes multiplied.
 
#98
#98
One more look at sampling and MOE (margin of error).

When viewing polling data, we typically are looking at issues with slight majorities such as Presidential approval ratings:

Approve 38%
Disapprove 54%
Undecided 8%

with an MOE of +/- 3%.

That MOE is really saying:

Approve (35 - 41%)
Disapprove (51 - 57%)
Undecided (5 - 11%)

Notice what happens to the range for the Undecided - it becomes quite large.

Now, let's look at the Hopkins study. Essentially they are recording an excess death as Yes/No (one occurred or didn't occur). Based on their projection, you see the following:

Occurred 3%
Did not occur 97%

If we have an MOE of 3%, you can see the range of their estimate now goes (in actual excess deaths) from 0 to 1.3 million!
 
#99
#99
www.world-war2.info says:

Germany suffered 1.6 Million civ casulties. Considering we bombed Germany in to the stone age, I don't believe we killed 650,000 Iraqi civs. If this number is accurate, I say we go back to carpet bombing.
 
#100
#100
"The sampling is solid. The methodology is as good as it gets," said John Zogby, whose Utica, N.Y.-based polling agency, Zogby International, has done several surveys in Iraq since the war began. "It is what people in the statistics business do."
Zogby said similar survey methods have been used to estimate casualty figures in other conflicts, such as Darfur and the Democratic People's Republic of Congo.
Ronald Waldman, an epidemiologist at Columbia University who worked at the Centers for Disease Control and Prevention for many years, told the Washington Post the survey method was "tried and true." He said that "this is the best estimate of mortality we have." Frank Harrell Jr., chairman of the biostatistics department at Vanderbilt University, told the Associated Press the study incorporated "rigorous, well-justified analysis of the data."
 
New posts
You must log in or register to reply here.

VN Store



Back
Top