Sampling from empirical data can't be compare to a phone poll, man.  I'm sorry, but it can't!  Exit polls on Election Day are far more accurate than cold calls over a wide geo area...I'm no statistician, but I’ve seen enough statistics to know that GI's "survey" was completely wonky…what were their controls?!?  Square trade is not polling, they are reporting empirical data, and you can extrapolate as you like from that.

Dude, you're the one that said to take it with a grain of salt!

I'm not saying either one is better than another. I'm saying there are big problems with each. People lie on polls and companies shift data around. They are equally fallible.

One thing I know from experience, is that most warranty companies tell customers to send the product to the manufacturer first. This most happens when warranty costs start eating all the profit. They sell the warranty to provide ease and help, but when all is said and done, warranties are devised to make pure profit margin with no product. What kingofwales said could very well be true. At first ST was taking in the repairs, but then got in a crunch.

There ARE companies that use cold calling very effectively. I'm pretty sure gallup does. Anyways, like I said, neither is reliable information.
This graph, in fact, is meant to show how jaspers are better than previous chipsets. Knowing that, this graph can't in good measure, be used to show anything else. Especially when the chipset performance is shown on the graph as a projection.

EDIT: just noticed that the performance is actual, not a projection. That was meant to be another nail in the coffin and is not really necessary to my point.

I'm confused, they just said that the Xbox 360 never went over 10% failure rate?

That's what I gathered.

no, they said, it never went over 10% failure rate depending on the time when they bought the system. There's a big difference. A difference why these graphs aren't very dependable.

Not more than 10% came from any specific quarter, except maybe that spiked one. This graph doesn't differentiate total repaired from total bought. It differentiates failure rates from other failure rates.

As an example,
If you sell 20 360s each day of the week and 5 crash from each of those days, then only
5 out of 140 came from monday, or 3.5%. However, on that monday, and everyday, the rate is actually 25% (5/20)

EDIT: or you can subtract all repaired consoles from the pool 140-30 (leaving out any specific days value of 5)=110, to get 5 out of 110 or 4.5%

But like I said, we don't know what method they used exactly.

No...it was 23.7% failure rate for their warranty customers (the total systems repaired were 16k), but the also pointed out that some of their customer when directly to M$, so the rate could be as high as 35%...based on exchanges with them.  BTW, I have a 3 year warranty for my 360...M$ only covers RROD and E74 now...not disc failure or others, so I got it. 

You've got 3 variables...failure rates...and 16000 systems.  One of you guys who are still in college do the math...better yet...high schoolers should be able to figure it out.

eh, that's going to be hard to figure out without knowing exactly how many were sent to MS.

SO then 23.7%=16k? So they cover roughly 70k units out of 10M? It also doesn't take into account people who simply sent in their consoles (to MS) without consulting ST first.

And yes, none of us college folk are going to do this work. We have more than enough to do as it is lol. Most high schoolers don't know how to deal with people of the opposite sex yet, so they should have more free time to do stuff like this,

Pssst...hey...it's 3 VARIABLES...16000 units total...so it can be partially solved with Calculus..set the equation to zero...etc...

I don't wanna do any more maaattthhhhh *crying*

I'm going to bed..

we have the number fixed :16k
we have the total number insured: 70k

I think setting the equation to zero is more algebraic though, not cal. Although cal is algebra. It doesn't require cal knowledge to do. So yes, let's get a high schooler to do this, or we can just use their estimated number of ~34% and find a distribution. Although this 34% might already be one or two standard deviations up. I think they know their math, so I'm going to say they've seen something like a 29% total failure rate over those years with a standard error 2-3%, and a certainty (or whatever it's called) of 67%+ (based on the sample size vs total size). But there is probably a lot more we can find from that. I'm not willing to do 10 minutes of work to figure it all out though, especially if they don't themselves know the total failures.