Handout 1 under nøstede modeller

14 views
Skip to first unread message

KHI

unread,
Jun 4, 2011, 8:29:52 AM6/4/11
to ST2304 Statistical Modelling for Biologists/Biotechnologists
Utdrag fra Handout 1:

"> drop1(fullmodel,test="F")
Single term deletions
Model:
y ~ x1 + x2 + f1
Df Sum of Sq RSS AIC F value Pr(F)
<none> 1.4507 -7.3051
x1 1 0.62127 2.0720 -5.7406 1.7130 0.2607
x2 1 0.66091 2.1117 -5.5511 1.8223 0.2484
f1 3 2.73843 4.1892 -2.7008 2.5168 0.1969

The residuals sums of squares (the column named RSS) lists the
residual sums of squares un-
der the full model, and under di erent reduced models obtained by
removing x1, x2 and f1,
respectively.
Focusing on f1, the observed di erence between the sum of squares
between the reduced and
the full model (appearing in the numerator of equation (7)) becomes
4,1892 -1,4507 = 2,7384
(appearing in the column Sum of Sq. The change in number of parameter
p1 - p0 is one less
than the number of levels of the factor f1, that is, 4-1, and is
listed in the column named Df.
The residual degrees of freedom n - p1 is 10 - 6 since 6 parameters
(including the intercept)
are being estimated under the full model."

Litt tidligere i handhoutet står det at p1 er antall estimerte
parametere under H1 som her er y ~x1 + x2 + f1 og at p0 er antall
estimerte parametere under H0 : y ~x1 + x2

Jeg skjønner ikke hvordan dette blir til p1-p0 = 4-1 ? Og heller ikke
hvorfor p1 rett etterpå skrifter fra å være lik 4 til å være lik
6...??
Jeg ville satt p0 lik 3 og p1 lik 4, hvorfor er det ikke slik?

carex

unread,
Jun 4, 2011, 10:14:45 AM6/4/11
to ST2304 Statistical Modelling for Biologists/Biotechnologists
4-1 er i dette tilfellet ENDRINGA i talet på estimerte parameter.
Totalt er det 6 estimerte paramter i p1 (intercept, x1,x2 og (4-1=3)
nivå for faktoren f1), og 3 i p0.

Altså blir p1-p0 lik 6-3 = 3, og n-p1 = 10-6 = 4

Jarle Tufto

unread,
Jun 4, 2011, 4:12:31 PM6/4/11
to ST2304 Statistical Modelling for Biologists/Biotechnologists
Når vi tar en forklaringsvariabel ut av modellen endrer antall
parametere seg med antall parametere som er knyttet til
forklarringsvariabelen, i dette tilfelle f1. Siden f1 er en faktor
med 4 nivå er antall parametere knyttet til f1 4-1 (fordi kun effekten
av nivå 2, 3 og 4 i forhold til nivå 1 estimeres.

On Jun 4, 2:29 pm, KHI <kamillain...@gmail.com> wrote:
Reply all
Reply to author
Forward
0 new messages