The name is absent

83

tio test using the log-likelihood test statistic: ∆_o = L(λι, pλ₁, θ) — L(λ, λ, 0), where Λ
is the maximum likelihood estimator of the failure rate in a simple exponential model.
Let Λι = 1.0. The authors use simulations to get the asymptotic distribution of 2Δ₀.
Both Weibull and Gamma distribution are tested using simulations. Matthews and
Farewell suggest that this procedure should also be applicable to censored data. How-
ever, only independent censoring is considered.

Henderson (1990)

Henderson (1990) considers the same model as in Matthews and Farewell (1982).
A test of Hq : ρ = 1 against Hγ : ρ ≠ 1 is considered with θ unknown. The Monte
Carlo power and Mean Squared Error estimates are presented by simulations. Hen-
derson shows that the adjusted log-likelihood method can be used when the likelihood
ratio test is not sufficient. The adjusted method gives better power and smaller Mean
Squared Error than unadjusted log-likelihood statistic.

Loader (1991)

Loader (1991) considers the following model:

ʌɪ , t > θ.

δ = Zog(λι∕λ₀) is used for inference about θ and the size of the change. Loader
uses the log-likelihood ratio process and the score process considered by Mathews et

<<   93   94   95   96   97   98   99   >>

More intriguing information