212 Some Special Distributions
aspt(2.0,15), while the commandqt(.975,15)returns the 97.5th percentile of
this distribution. The R codet=seq(-4,4,.01)followed by plot(dt(t,3)~t)
yields a plot of thet-pdf with 3 degrees of freedom.
Before the age of modern computing, tables of the distribution ofTwere used.
Because the pdf ofT does depend on its degrees of freedomr, the usualt-table
gives selected quantiles versus degrees of freedom. Table III in Appendix D is such
a table. The following three lines of R code, however, produce this table.
ps = c(.9,.925,.950,.975,.99,.995,.999); df = 1:30; tab=c()
for(r in df){tab=rbind(tab,qt(ps,r))}; df=c(df,Inf);nq=qnorm(ps)
tab=rbind(tab,nq);tab=cbind(df,tab)
This code is the body of the R functionttablefound at the site listed in the
Preface. Due to the fact thatt-distribution converges to theN(0,1) distribution,
only the degrees of freedom from 1 to 30 are used in such tables. This is, also, the
reason that the last line in the table are the standard normal quantiles.
Remark 3.6.1.Thet-distribution was first discovered by W. S. Gosset when he
was working for an Irish brewery. Gosset published under the pseudonym Student.
Thus this distribution is often known asStudent’st-distribution.
Example 3.6.1(Mean and Variance of thet-Distribution).Let the random variable
Thave at-distribution withrdegrees of freedom. Then, as in (3.6.1), we can write
T=W(V/r)−^1 /^2 ,whereW has aN(0,1) distribution,Vhas aχ^2 (r) distribution,
andW andVare independent random variables. Independence ofW andVand
expression (3.3.8), provided (r/2)−(k/2)>0 (i.e.,k<r), implies the following:
E(Tk)=E
[
Wk
(
V
r
)−k/ 2 ]
=E(Wk)E
[(
V
r
)−k/ 2 ]
(3.6.3)
= E(Wk)
2 −k/^2 Γ
(r
2 −
k
2
)
Γ
(r
2
)
r−k/^2
ifk<r. (3.6.4)
BecauseE(W)=0,themeanofTis 0, as long as the degrees of freedom ofTexceed
- For the variance, usek= 2 in expression (3.6.4). In this case the conditionr>k
becomesr>2. SinceE(W^2 ) = 1, by expression (3.6.4), the variance ofTis given
by
Var(T)=E(T^2 )=
r
r− 2
. (3.6.5)
Therefore, at-distribution withr>2 degrees of freedom has a mean of 0 and a
variance ofr/(r−2).
3.6.2 TheF-distribution........................
Next consider two independent chi-square random variablesUandVhavingr 1 and
r 2 degrees of freedom, respectively. The joint pdfh(u, v)ofUandV is then
h(u, v)=
{
1
Γ(r 1 /2)Γ(r 2 /2)2(r^1 +r^2 )/^2 u
r 1 / 2 − (^1) vr 2 / 2 − (^1) e−(u+v)/ (^20) <u,v<∞
0elsewhere.