reference interval for bootstrapped data

[Nick Jeffery]

Hello everyone - new on here today

I've been trying to estimate the reference interval for a set of data that has 86 datapoints in it. It seems the best way of doing this is to bootstrap - which works fine in general on the dataset.

But the problem i'm having is getting the results for the 95% reference interval (ie from p2.5 to p97.5). I can get the 90% reference interval (ie from p5 to p95) but not the 95% level.

is there a way to get the 95% reference interval using bootstrap?

I'd also like to be able to calculate the 90% CI for the upper and lower bounds if possible (rather than 95%CI).

Nick Jeffery

[Mike Lacy]

The documentation given by -help bootstrap- reveals an option that's relevant to you:
" level(#) set confidence level; default is level(95)"

(By the way, I'm a little confused by your saying you can get the 90% and not the 95%, given that 95% is the default. I'm assuming this is just a a typographical error.)

[Nick Jeffery]

Hmmm - thank you ... but perhaps i didn't state the problem properly.

What i can run is:

bootstrap r(p95) r(p5), reps(1000) seed(12345): summarize day21 , detail


Which is great .... but the output:


------------------------------------------------------------------------------
| Observed Bootstrap Normal-based
| Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
_bs_1 | 15 .4954727 30.27 0.000 14.02889 15.97111
_bs_2 | 5.5 1.370173 4.01 0.000 2.814509 8.185491
------------------------------------------------------------------------------

Is not quite what i really need. What i really need is the 2.5% lower bound and the 97.5% upper bound for the population (rather than the 5% and 95% as listed above) .... each estimated with a 90% CI (just because that's the convention for my field - biomedicine).

When i try to run the bootstrap with p(97.5) and p(2.5) i get an error message:

p97.5 invalid name
error in expression: r(p97.5)

[Nick Jeffery]

But thank you for the level advice - that's sorted that part of the problem out (so i now can get 90%CI for the upper and lower bounds ... but not the correct (95%) reference interval (reference range)!)

[Mike Lacy]

Now that I see your actual command I understand something different: By lower and upper bounds, you intend the 2.5% and 97.5% percentiles for the population, so I presume you want bootstrapped confidence intervals derived from the sample estimates of those values. -summarize- doesn't offer those percentiles, but -centile- does.

centile day21, centile(2.5 97.5)
return list

shows that centile returns the estimated percentile scores in r(c_1) and r(c_2)
so I'm thinking you want:

bootstrap p025 = r(c_1) p0975 = r(c_2), reps(1000) seed(12345) level(90): centile day21, centile(2.5 97.5)

[Nick Jeffery]

Wow - thank you!! That's exactly what i wanted ... you've really made my day ... and it's my birthday today too! Please have a large daquiri on me ...