Re: Ratios of normal samples and BOOTSTRAP
- From: "\"Luis A. Afonso\"" <licas_@xxxxxxxxxxx>
- Date: Sun, 26 Mar 2006 15:24:42 EST
Follow-up concerning Z=X/Y
The fractions limited by the bounds of BIAS show that the Source sets and the Bootstrap ones (means) have very different characteristics about their <spans>. The weakness of the latter is flagrant.
____________>0.005_____>0.010_____>0.015
___X, Y = N(100, 5):10_____
Source bias____0.026______0.016______0.014
Bootstrap_____ 0.800______0.612______0.486
S____________0.038______0.020______0.016
B____________0.822______0.668______0.526
S____________0.052______0.022______0.006
B____________0.812______0.656______0.508
____________>0.005_____>0.010_____>0.015
___X, Y = N(100, 5):20_____
Source bias____0.012______0.000______0.000
Bootstrap_____ 0.784______0.510______0.330
S____________0.018______0.008______0.000
B____________0.760______0.550______0.340
____________>0.005_____>0.010_____>0.015
___X, Y = N(100, 5):30_____
Source bias____0.004______0.000______0.000
Bootstrap_____ 0.730______0.486______0.290
S____________0.112______0.002______0.000
B____________0.720______0.494______0.316
For this last example I report the means
______S = 0.003 and <B>0.012
Each line of results concerns 500 source samples and 2000 Boostraps/sample.
Well
_____Facta, non verba
or FACTS cannot be manipulated to fit a Theory (whatever it is). This is an intentionally free translation from Latin.
______licas (Luis A. Afonso)
REM "ratio3"
CLS
DIM x(102), y(102)
PRINT " w<1 "
pi = 4 * ATN(1)
mx = 100: sx = 5: my = 100: sy = 5
nx = 20: ny = 30
true = mx / my
sources = 500: boots = 2000: fails = 0
FOR sc = 1 TO sources
meanb = 0
RANDOMIZE TIMER
FOR ix = 1 TO nx: a = SQR(-2 * LOG(RND))
x(ix) = mx + sx * a * COS(2 * pi * RND)
aa = aa + x(ix) / nx
NEXT ix
FOR iy = 1 TO ny: a = SQR(-2 * LOG(RND))
y(iy) = my + sy * a * COS(2 * pi * RND)
bb = bb + y(iy) / ny
NEXT iy
src = aa / bb: dsource = ABS(src - true)
IF dsource > .005 THEN odd = odd + 1
IF dsource > .01 THEN odd1 = odd1 + 1
IF dsource > .015 THEN odd2 = odd2 + 1
FOR i = 1 TO boots
sumx = 0
FOR iix = 1 TO nx: g = INT(RND * nx) + 1
sumx = sumx + x(g) / nx: NEXT iix
sumy = 0
FOR iiy = 1 TO ny: g = INT(RND * ny) + 1
sumy = sumy + y(g) / ny: NEXT iiy
meanb = meanb + sumx / sumy
NEXT i
b = meanb / boots: db = ABS(b - true)
SUMDB = SUMDB + db
IF db > .005 THEN yes = yes + 1
IF db > .01 THEN yes1 = yes1 + 1
IF db > .015 THEN yes2 = yes2 + 1
sumsource = sumsource + dsource
w = dsource / db
IF w < 1 THEN fails = fails + 1
LOCATE 5, 50
PRINT USING "###### #.###"; sources - sc; fails / sc
NEXT sc
PRINT " MEAN OF BOOTSTRAP BIAS ";
PRINT USING "#.### "; SUMDB / sources
PRINT " MEAN OF SOURCE BIAS ";
PRINT USING "#.### "; sumsource / sources
PRINT " DSOURCE > 0.005 , > 0.010 , > 0.015 ";
PRINT USING "#.### ";
odd / sources; odd1 / sources; odd2 / sources
PRINT " Dboots > 0.005 , > 0.010 , > 0.015 ";
PRINT USING "#.### ";
yes / sources; yes1 / sources; yes2 / sources
END
.
- References:
- Ratios of normal samples and BOOTSTRAP
- From: \"Luis A. Afonso\"
- Ratios of normal samples and BOOTSTRAP
- Prev by Date: Re: Working out accuracy
- Next by Date: Re: Survival Data Simulation: Number of Iterations
- Previous by thread: Ratios of normal samples and BOOTSTRAP
- Next by thread: PhD Student Position in Switzerland on Bayesian Methods for Integrated Genomics
- Index(es):
Relevant Pages
|
|
