OK, a little help please.
TAPPED was nice enough to link to my latest irritated screed at the media’s poor coverage of military suicide rates.
Once the statistic’s initial shock value wears off, it’s clear that–as Winds of Change notes in its calculations–the figure is fairly misleading. Taking the national rate of suicide (about 13 per 100,000) and applying it to the 1.6 million U.S. troops that have to date served, the figure comes out to 8,409 — a little less than twice the number of U.S. casualties in Iraq. More an artifact of the comparatively low casualties the U.S. has suffered in Iraq than anything else.
They then go on:
The more compelling statistic is the one revealed in an independent CBS analysis last November, namely that veterans aged 20-24 (that is, those who’ve served in current wars) have a suicide rate up to four times higher than civilians the same age.
So we go over to the CBS analysis:
So CBS News did an investigation – asking all 50 states for their suicide data, based on death records, for veterans and non-veterans, dating back to 1995. Forty-five states sent what turned out to be a mountain of information.
And what it revealed was stunning.
In 2005, for example, in just those 45 states, there were at least 6,256 suicides among those who served in the armed forces. That’s 120 each and every week, in just one year.
Dr. Steve Rathbun is the acting head of the Epidemiology and Biostatistics Department at the University of Georgia. CBS News asked him to run a detailed analysis of the raw numbers that we obtained from state authorities for 2004 and 2005.
It found that veterans were more than twice as likely to commit suicide in 2005 than non-vets. (Veterans committed suicide at the rate of between 18.7 to 20.8 per 100,000, compared to other Americans, who did so at the rate of 8.9 per 100,000.)
One age group stood out. Veterans aged 20 through 24, those who have served during the war on terror. They had the highest suicide rate among all veterans, estimated between two and four times higher than civilians the same age. (The suicide rate for non-veterans is 8.3 per 100,000, while the rate for veterans was found to be between 22.9 and 31.9 per 100,000.)
OK, I’m pretty deeply puzzled here.
Back when I did my first post on suicide, here’s what I found:
So in 2004, there were a total of 14,328 suicides in the US in the age group 20 – 44 (the group that I think pretty well covers the population in Iraq – some are younger, some are older). the total population in 2004 in that age group was 104,259,000 – so the rate/100,000 population was 15.25.
And since the rate in the military is higher – significantly higher at 17.3/100,000 overall and 19.9 for those serving in Iraq and Afghanistan – I was darn concerned.
And then I asked one more question.
Well, the suicide rates by sex are pretty different, I recalled. I wonder what happens if I sex-norm the suicide rates in the military?
Here’s an approximation (because the of women in reserves is slightly higher, and I didn’t find the serving in Iraq).
According to the DoD, approximately 17% of US active-duty forces are women.
According to the CDC, the 20 – 44 population had 14,328 suicides in 2004. Of those, 11,460 were men, and 2,868 were women. The census gives an estimate for 2005 population from 20 – 44 as 52,513,000 men and 51,746,000 women.
By my math, this gives a suicide rate of 21.82/100,000 for men, and 5.54/100,000 for women.
If I norm the suicide rates by multiplying the sexes rate by the population in the military, I get (21.82*83%)+(5.54*17%)=19.06/100,000.
So let’s go to the CDC data, and see what the numbers for ages 20 – 24 look like.
They show 2,599 suicides in that age range in 2004. There are, per the census, 21.05 million Americans that age; that gives a raw rate of 12.35/100,000.
Note the article states that the rate is 8.9. OK, I’m puzzled – how the heck did he get that?
Now let’s take a moment and sex-norm the rate, as I did in the original post.
Per the census, the 20 – 24 age group has 10.86 million males, and 10.20 million females. The suicide number for males in this group was 2,105; the rate was this 20.22/100,000. For females, there were 404 suicides, for a rate of 3.96/100,000.
Assuming the same ratio of males/females in the military (which is for the Army, and hence somewhat high), we have 83% male, 17% female. So sex-norming the rations, we’d get 83% * 20.22 + 17% * 3.96 for a total rate of 17.46/100,000.
The actual rate, per Rathburn, is between 22.9 and 31.9/100,000. Now I’m not sure how he got such a huge variance, but I’ll also suggest that the number of veterans between 20 and 24 is pretty small. The VA says there are 287,400 veterans in that age group. This would suggest that there were between 66 and 92 suicides in 2005 in this group.
Let’s look at the overall population. There were a total of 24.5 million vets as of 2005, and the CBS study shows 6,256 suicides in 2005, for a rate of 25.51/100,000. Assuming a sex-normed overall population, the rate (using the 83%/17% ratio, which is high, but close enough) would be 15.5/100,000. So – looks like the rate is significantly higher – which means there is some damn serious work to do.
But – I’m seriously puzzled about where the doc got his statistics. I’ll look for an email for him and ask him directly. Meanwhile, here are my sources:
I’m open to sources. I know he’s a professor of epidemiology, and I’m some guy with Excel and a web browser. But his numbers make no sense to me.
Note: my son is in the military, and I’m darn concerned about his well-being. But I want to have a fact-based discussion; I don’t think vaguely-sourced or wrong numbers make for good discussion.
Look also at CDC WISQARS.
Using your example year of 2005, the suicide rate by sex, all ages, is
Males 25,907 145,973,538 17.75
Female 6,730 150,533,523 4.47
20-44 age group the rates are
Males 11,324 53,165,370 21.30
Female 2,815 51,898,142 5.42
20-24 age group the rates are
Males 2,195 10,858,108 20.22
Female 404 10,194,765 3.96
WISQARS doesn’t have a way to split this out by veterans.
Those rates have not varied much 1999-2005. For males in the 20-24 range, the rates were in the 25-27 range 1990-1995 and began dropping in 1996.
Rathbun’s testimony explains his numbers pretty well.
AL: “The actual rate, per Rathburn, is between 22.9 and 31.9/100,000. Now I’m not sure how he got such a huge variance….”
As you will see from the link in my previous comment, it’s not a variance, but two distinct point estimates. After Rathbun had done the analysis the CBS producer, Pia Malbran, was concerned that some active-duty soldiers may have been mistakenly classified as veterans in the state death records. So a second analysis was done with these deaths excluded.
Kevin thanks for the link. I’m still puzzled at how he got to a base suicide rate so different than the one I see in the CDC data.
Any suggestions out there?
A.L.
“Kevin thanks for the link. I’m still puzzled at how he got to a base suicide rate so different than the one I see in the CDC data.
Any suggestions out there?
A.L.”
I’m inclined to agree with the general gist of your analysis, but one way to undercut it would be to note that you haven’t excluded veterans from your count of the general population. So you really aren’t comparing the rate among veterans to the rate among non-veterans. If in fact the bulk of all suicides are committed by veterans, then the rate among non-veterans could be very low even while the rate in the general population came closer to matching the rate among veterans.
celebrim, I thought about that, but didn’t have the time to chunk numbers; but the number of veterans is small enough in this population – 287,000 vs a total population of 20 million – that I was comfy just eyeballing an estimate that the impact would be low.
A.L.
I agree that there seems to be no way for the CDC numbers to be consistent with Rathbun’s numbers, which CBS gathered from the states. Since Rathbun was required to erase the data CBS gave him, probably the best person to contact for clarification is Pia Malbran, the CBS producer. CBS still has the data and I can’t see why they would refuse to release the aggregate numbers for males and females (vets and non-vets) in each age-group. Obviously they can’t release numbers by state.
I doubt it accounts for the full difference, but CBS did their analysis on 45 states, whereas the CDC presumably has access to all 50 states’ data when calculating suicide rates. That has to have at least _some_ effect.
I would also be interested in progressive suicide rates. For example, are suicide rates in veterans climbing as the war goes on? Afghanistan started in 2001, and OIF in 2003. But while AL did his analysis on 2004 data, CBS seems to be using 2005 data to get the higher numbers. God forbid, but we might be seeing that rate climb along the time axis. (Although maybe it’s just a function of more troops switching to “veteran” status as time goes on.)
Eugene Volokh had his own troubles trying to back the CBS data out to compare with general population averages:
“Link”:http://www.volokh.com/posts/1195086841.shtml
ADDENDUM: Eugene Volokh from the comment thread:
Eugene Volokh: . . . I e-mailed Prof. Rathbun Wednesday to pose my questions to him; I haven’t yet heard back from him, but he may well be out of the office or occupied.
11.16.2007
Nothing further.
“That’s what reporters are supposed to do – to test public assertions against competing claims and facts where accessible – to try and allow a reader to end up better informed.
Doesn’t seem a lot to ask.”
Apparently, it is. (Having just read all of the above.)
Come on, you should know better than to expect a response. Those Ivory Towers don’t respond to the plebs.
_Note the article states that the rate is 8.9. OK, I’m puzzled – how the heck did he get that?_
_Any suggestions out there?_
The 8.9 number is not adjusted for population based on military distribution, I think. It comes from the 145/150M male/female in the general population, subtracting 20/2M male/female vets, and their resultant suicide numbers, as well as (probably?) adjusting for the missing 5 states, which make up about 8-10% of the population (census is around 8%, VA numbers around 10%). It leaves you with around 20,000 suicides/270M population, and adjusting for age groups and gender grows higher. For example, 9.5/9.9M men/women age 50-54, dropping a bit due to Vet status, and then adjusting the remaining number to be more like the median split between them before removing Vets.
Haven’t figured out the 20-24 age group yet. I was thinking it’s not normalized to military demographics, but he seems to have pretty clearly called out as using this.
I’m guessing that looking at the CDC’s numbers nationally skew your numbers. Alabama had rates of 32.96/.062 for population of 163k/161k M/F, while CA had 16.00/3.92 1.38M/1.24M M/F.
Given the VA numbers have males 20-24 as being 3653 in AL, and 22816 in CA, this gives you 2.24% of AL’s male age group as serving(ed), and 1.65% of CA.
If you just take the normal WISQARS answer, you end up with for a rate of 16.00/100000 men, age 20-24. I think VetStatus norming by state rates and relative populations would drive it lower.
I think you’re going to need a bigger spreadsheet.
_Come on, you should know better than to expect a response. Those Ivory Towers don’t respond to the plebs._
Or realize that as your name gets attached to any story with a hint of controversy, you’ll be deluged with email with people supporting, denouncing, or challenging you – and likely end up having to do a full email purge to catch up. Since his job is really research and teaching, not answering anyone who wanders in.
Wait, nevermind. Let’s just mock him for portraying information we don’t like.
Phone calls, AL, or wait until the storm has passed. Your personal reasons, and familiarity with numbers might help.
AL:
Aparently Obama’s numeracy is a little off kilter, as well.
On reflection I take back what I wrote upthread about the CDC numbers appearing inconsistent with Rathbun’s numbers. I’ve never worked with data of this kind and I’ve never used logistic regression, so I’d just as soon leave the number-crunching to someone who has relevant expertise. If the dataset is a bit empty in places – with some states having very few female suicides in the agegroup 20-24 for example – it’s plausible that some of the estimated coefficients would look peculiar.
Question for AL: are you sure you want to argue about this? After looking at what data I can find on the web, I’m struck by the fact that suicide in America is very much a white male thing. Also, there is marked geographical variation, with lower suicide-rates in the Northeast (8.1 in 2005) and higher rates in the West (12.1) and South (11.8). These facts seem consistent with a high correlation between military service and suicide. (Admittedly my limited knowledge of the US military may be playing a part here; am I wrong in thinking that the services are dominated by white males from the West and South?) However mere correlations don’t tell us much about causes. The actual experiences of veterans vary widely. Some have seen real horrors. Others spent their active years guarding airbases in Britain from rowdy CND protesters; they probably never saw any more combat than you can get in a punch-up at the local pub.
Granted the figures are disturbing, is there really any reason to find them surprising? (Of course every number in this dataset represents a tragedy for some family, but that’s as true for the civilians as it is for the veterans.) Physicians have a high suicide rate also; like soldiers, they have relevant expertise. More power to those who want more help for troubled veterans, but I can’t see anything startling about a high correlation between military service and suicide. As a rough generalisation, military service is likely to attract people who feel, more strongly than most, that some problems can only be dealt with by drastic means. Is it really surprising that people who see things that way are more likely to “solve” their own problems in a drastic manner? In short, is there really anything to see here?
That question may sound callous and I am mindful of the fact that you have a son to think about, so let me put the point in another way which I hope you don’t find offensive. Based on what you’ve written about yourself (apologies if my memory is playing me false) I assume your son is white (per CDC definition), from a Western state, and he was quite familiar with firearms even before he enlisted. That means he was already in a relatively high-risk group. Rathbun’s numbers don’t imply that he is now in a riskier group. It’s quite possible that white Californian male civilians who own guns are more at risk of suicide than their counterparts in the army.
#14 Demosophist:
Ha Ha Ha Ha! Great one! He can kiss his presidency goodbye now.
Back to reality. Linking to Michelle Malkin’s hate site does not buy much public credibility except in the deepest dark recesses of the Republican Right where the haters dwell. Secondly, the sheer ignorance in believing that people will actually think this kind of childish gotcha game actually might have an impact on others who are in possession of a functioning cortex is breathtaking really.
I hope you were posting this in jest, because it seems that Malkin’s site is serious. If you weren’t, than I can’t help but to feel sorry for you.
Kevin, the figures are certainly disturbing; one of the points that I try and make by doing some basic math and trying to put them into persepctive is – well, how disturbing? It’s also disturbing to me that many veterans die of treateble prostate cancer – and I’ll bet that since the number of men who die of prostate cancer is so much higher than the number of men who commit suicide, that for every dollar of effort, we could swing the overall death numbers further by pushing prostate screening harder.
I believe in government intervention; I just don’t like it when it’d driven by kneejerk reactions to anecdote and weak data.
So when I see data like this, my first back of the envelope was to say – well, the rates are going up (which is bad) but they are still as low or “lower than the civilian rates for soldiers in uniform”:http://www.windsofchange.net/archives/009741.php ; then when it’s suggested that the rates for veterans are two to three times the rates for their nonveteran peers, I want to know more – but I see a rate that so deeply contradicts the base rate I saw when I looked directly at the data, I get suspicious first.
So until someone can give me some information about how he did a linear regression that so flatly contradicts the raw base data (his base suicide rate for the nonvet population – either for 20 – 24 or for all vets (it’s not clear to me from reading it) is so radically different from the stats that I see in WISQARS that I can’t credit his work until I understand the difference.
A.L.
This seems like a daunting statistical problem. It’s analogous to conducting analysis of crime rates based on reported crimes, versus victim surveys. The numbers trend in the same direction (usually) but they still result in very different estimates. Moreover, logistic regression yields an odds ratio, which isn’t comparable to a rate per 100K. Finally, AL’s impulse to “norm” the figures is appropriate, but norming by gender may not be enough. You’d probably want to run a regression that included other instrumental variables, that aren’t directly collinear with military service, such as SES, race, location, etc.. But I suspect the main problem will be with the data itself. For instance, it might well be the case that suicides of former military personnel are more likely to be reported as suicides, simply because the military tends to keep tighter tabs on its former members. That’s a reporting effect, and it’s well known in crime analysis. That’s why people often use victim surveys rather than official reports for such analysis.
Sepp:
The rather embarrassing video is relevant because McCain opponents wouldn’t hesitate to use a similar gaff to argue Mac is just too old for the job. Which is ironic, because all he did was suggest that Iran might be funding Sunni insurgents (quickly corrected by Lieberman), when there’s rather unassailable evidence that Iran has been funding both sides at various times (though of course their main focus is on the Shi’ah militias).
It’s still not clear to me how Obama got so far off the beam. Even if you include the eight protectorates, and DC, that’s still only 59 “states.” (He implies there are 60.) Apparently the mainstream media is mum about the incident, so I wouldn’t worry too much. There’s a pretty effective double standard in place.
Sepp:
Actually, now that I think of it, the original story appeared in the blog section of that well-known hate rag, the LA Times.
Very good work, AL. As a math grad student, let me applaud and ask you not to be daunted by somebody’s credentials. The point of math is that it’s accessible to anybody who’s willing to think clearly and rigorously, as you have. There are no “special” tricks to looking at data and trying to draw information out of them.
What Obama must have meant to say was “47” instead of “57”. A tired man on the campaign trail. Big news really, eh?
Not at all the same thing as McCain’s multiple repeated gaffes either. (But thanks for playing another round of Wingnut False Equivalence).
And just because there are idiots at the LA Times does not make Malkin’s juvenile sneering any less idiotic, or desperate.
bq. The rather embarrassing video is relevant because McCain opponents wouldn’t hesitate to use a similar gaff to argue Mac is just too old for the job.
And this does not of course make the video “relevant”.
Plus, we already know what would have happened if McCain said these exact words. His media bbq friends would be quick to point out that he must have meant “47” because “he surely knows there are 50 states”.
Because that’s what they’ve been doing with most of his other gaffes…helping his campaign by “explaining” them to us.
So really it is not relevant, just like your contribution here.
Isn’t the substantive question whether veteran status is a necessary explanatory variable for suicide? It seems to me that the veteran population has any number of distinctive differences from the general population, from obvious differences like gender and race proportions to things like religious affiliation to geographical origin to access to firearms. Simple gender- and race-norming is not sufficient to reject the null hypothesis that veteran status makes no difference to the propensity to suicide.
I also have a question whether the focus on the 20-24 year-old age group was decided pre-test or post-test. As you know, multiple post-test t-tests are generally not valid. (I didn’t follow the links to see if there were any regressions done.)
BBB