Undergraduate bickering over Iraq
God don't the MSM bang on about blogging. Deborah Hill Cone, who can spell really well, was the latest with an effort in the SST's Sunday magazine (not online). Deborah, this post is for you.First off, let's not pretend, as some like to, that coalition casualties are dropping. They go up and down but the overall trend since the start of the insurgency remains up. At the time of posting, just over half way through the month, April's tally was 49 dead at one of the highest monthly rates yet seen. And now it looks as if the US has to reconquer Baghdad.
Meanwhile, incidents like this, this and this are not helping. As for reconstruction, the oil ain't flowing to pay for it and then the contractors have to bilk their cut ...
Other interesting stories here, here, here and here.
Update: Chefen at Sir Humphrey's has done a lot of very good work on reanalysing the deaths data presented above. The weakness of the simple linear trend line I used is that it is very poor in registering recent trends (put simply, it isn't very responsive and takes a long time to change from positive to negative). He's presented a range of different analysis worthy of thoughtful consideration.
posted by Rob O'Neill at 10:32 AM
18 Comments:
Ah, but you see our Deborah won't read this...It is not "deeply trivial, ... and has nothing to do with figs"
Well, it's her loss Doddery. But she spels realy well, dontcha think?
Rob, best thing about lapsed blogger Ms Hill-Cone's story? She mentioned finding Simon Grigg's blog from Chad Taylors website, but all the other blogs she mentioned in the story were also linked from chad's site. And she got Chad's website address wrong. Oh dear.
Interesting, Peter. She (or someone) got PublicAddress's address wrong too.
And Gordon King has barely posted anyting in a year or so.
More paua to him.
(That's an in-joke. It's because Gordon is ... oh, nevermind...)
Actually I'm not sure all the other blogs are linked from Chad's site Peter. I may be blind but can't see Caribou there or Babies are Fireproof.
It seemed to me that the point of DHC's article wasn't that blogging about Iraq was wrong, unnecessary, a waste of bandwidth etc, but rather that even though political blogs get a lot of attention drawn to them by the mainstream media, there are actually interesting blogs out there that aren't about politics.
Mine was one of those blogs (and like Public Address and Chad Taylor, I also had the honour of having my URL misspelt). Yo, that's going on my CV.
Your figures are normalised to what, total deployments during the month? They are normalised aren't they. Otherwise what would happen if deployments double next month, or reduced by half?
Why do you fit a straight line to data with clearly cyclic variations? I guess you must actually be using a running average or similar to account for that.
Your null hypothesis is the slope of the line is not negative, at what confidence level do you find rejection of that hypothesis to hold?
You failed undergraduate stats, did you not? Post your raw data and the necessary extras and we can knock up some proper findings.
Thanks Chefen, raw data at Iraq Coalition Casualties at table on left.
I do not see clear "cyclic variations". I see random variations.
Deployments over the period have not changed markedly as far as I'm aware.
But you are right. I did fail undergraduate stats. Grab the data and have a go I'd be interested to see what a real stats guy will come up with.
sorry URL here
http://icasualties.org/oif/
From wikipedia on US deployments:
United States - As of March 2006 there were around 132,000 Army, Navy, Air Force and Marine Corps personnel in West, North and Central Iraq [5]; a planned reduction to 115,000 was cancelled due to losses and intense Iraqi resistance in Al Anbar province and a Shia uprising in the South of the country in late 2004. An increase to 153,000 was supposed to have taken place in early-mid 2005, but such plans never materialized.
UK deployment also stable.
I'll post some pics over at SH, for simplicity. Might take a couple of days though to fit round work, err not that I'm doing this on work time of course.
If you eyeball the data it looks like there is at least a yearly cycle, with the minima at 11,25 and 37 months and peaking in between. The four large spikes obscure it a little bit.
They sure do obscure it. When the rate can plummet over just two months in one of those cases and the low is bracketed by two large spikes. I'm not at all convinced you can posit a cycle over such a relatively short sequence.
Is there some way of analysing the data to see whether it is cyclical or what the odds are of it being cyclical?
Anyway, I'm sure the US govt have people pouring over these data and analysing them in all sorts of ways so an alternative treatment might throw something interesting up.
You can use an autocorrelation or perhaps a Fourier spectrum, the former is probably better here. Both should show a definite yearly cycle there, the presence of high spikes won't affect the result since there are few of them. I'll try and remember to do that as well. As to why there is a yearly cycle with lows in March-ish... dunno, but most human activities have such properties.
Kewell. Naturally I shall link when it's up.
Here you go, a first crack at some analysis.
No hypothesis testing or anything like that at this stage, just some graphs to show the shortcomings of trend analysis and an alternative. Also the cycles showing up (look 6 monthly as well as yearly).
Like I say at the end, if you want to do more we need to specify exactly what we are measuring.
'Anyway, I'm sure the US govt have people pouring over these data' - pouring what, Rob? Or are you having a DH Cone moment?
Kewel, Chefen. some comments on SH.
Anon, I'm clearly not having a DHC moment. She can spell!
Interesting question to ponder though. What would they be pouring over them?
Post a Comment
Links to this post:
Create a Link
<< Home