Clarketest: Clarke's test for non-nested model comparison
In spatcounts: Spatial count regression

Description Usage Arguments References See Also Examples

'Clarketest' suggests the better of two (not necessarily nested) models according to Clarke's statistic for the parameters in each of the iterations.

1 2	Clarketest(LogLike1, LogLike2, alpha = 0.05, p = NULL, q = NULL, correction = TRUE)

`LogLike1, LogLike2`	the output of two model fits obtained by using 'LogLike'.
`alpha`	significance level, defaults to 0.05.
`p, q`	the number of estimated coefficients in models LogLike1 and Loglike2, respectively.
`correction`	boolean, if TRUE (default), the Schwarz correction will be used on the differences of log-likelihoods.

Clarke, Kevin A. (2003). Nonparametric Model Discrimination in International Relations. Journal of Conflict Resolution 47(1), 72-93.

Schwarz, G. (1978). Estimating the Dimension of a Model. Annals of Statistics 6, 461-464.

Vuongtest

data(sim.Yin)
data(sim.fm.X)
data(sim.region)
data(sim.gmat)
data(sim.nmat)

poi <- est.sc(sim.Yin, ~1+sim.fm.X[,1]+sim.fm.X[,2], sim.region, 
model="Poi", sim.gmat, sim.nmat, 3)
nb <- est.sc(sim.Yin, ~1+sim.fm.X[,1]+sim.fm.X[,2], sim.region, 
model="NB", sim.gmat, sim.nmat, 3)

DIC.poi <- DIC(sim.Yin, ~1+sim.fm.X[,1]+sim.fm.X[,2], sim.region, poi)
DIC.nb <- DIC(sim.Yin, ~1+sim.fm.X[,1]+sim.fm.X[,2], sim.region, nb)

ll.poi <- LogLike(sim.Yin, ~1+sim.fm.X[,1]+sim.fm.X[,2], sim.region, poi)
ll.nb <- LogLike(sim.Yin, ~1+sim.fm.X[,1]+sim.fm.X[,2], sim.region, nb)

Clarke.poi.nb <- Clarketest(ll.poi, ll.nb, alpha = 0.05, p = DIC.poi$p.D, 
q = DIC.nb$p.D, correction = TRUE)

acceptb/(i+1)  0.8 0.8 0.6
acceptga1/i acceptga2/(i+1)   0.4 0.8
acceptpsi/(i+1)   0.4
acceptb/(i+1)   0.8 0.8 0.8
acceptga1/i acceptga2/(i+1)   0.6 0.8
acceptr/(i+1)  0.6
acceptpsi/(i+1)   0.8
DIC             9449.463
mean deviance   9041.958
p.D             407.5048
DIC             9435.57
mean deviance   9190.586
p.D             244.984
Favour model 1   0
No decision      0
Favour model 2   1