Extract Log-likelihood for VGLMs/VGAMs/etc.

Calculates the log-likelihood value or the element-by-element contributions of the log-likelihood.

Usage

# S3 method for class 'vlm'
logLik(object, summation = TRUE, ...)

Arguments

object

Some VGAM object, for example, having class vglmff-class.

summation

Logical, apply sum? If FALSE then a \(n\)-vector or \(n\)-row matrix (with the number of responses as the number of columns) is returned. Each element is the contribution to the log-likelihood.

...

Currently unused. In the future: other possible arguments fed into logLik in order to compute the log-likelihood.

Details

By default, this function returns the log-likelihood of the object. Thus this code relies on the log-likelihood being defined, and computed, for the object.

Value

Returns the log-likelihood of the object. If summation = FALSE then a \(n\)-vector or \(n\)-row matrix (with the number of responses as the number of columns) is returned. Each element is the contribution to the log-likelihood. The prior weights are assimulated within the answer.

Author

T. W. Yee.

Note

Not all VGAM family functions currently have the summation argument implemented.

Warning

Not all VGAM family functions have had the summation checked.

Examples

zdata <- data.frame(x2 = runif(nn <- 50))
zdata <- transform(zdata, Ps01    = logitlink(-0.5       , inverse = TRUE),
                          Ps02    = logitlink( 0.5       , inverse = TRUE),
                          lambda1 =  loglink(-0.5 + 2*x2, inverse = TRUE),
                          lambda2 =  loglink( 0.5 + 2*x2, inverse = TRUE))
zdata <- transform(zdata, y1 = rzipois(nn, lambda = lambda1, pstr0 = Ps01),
                          y2 = rzipois(nn, lambda = lambda2, pstr0 = Ps02))

with(zdata, table(y1))  # Eyeball the data
#> y1
#>  0  1  2  3  4  5 
#> 28  9  8  3  1  1 
with(zdata, table(y2))
#> y2
#>  0  1  2  3  4  5  6  7  9 11 13 
#> 33  3  3  1  2  2  1  1  2  1  1 
fit2 <- vglm(cbind(y1, y2) ~ x2, zipoisson(zero = NULL), data = zdata)

logLik(fit2)  # Summed over the two responses
#> [1] -131.0854
sum(logLik(fit2, sum = FALSE))  # For checking purposes
#> [1] -131.0854
(ll.matrix <- logLik(fit2, sum = FALSE))  # nn x 2 matrix
#>            y1         y2
#> 1  -1.7788391 -0.4551255
#> 2  -0.6068569 -4.2666074
#> 3  -0.2775615 -0.4653479
#> 4  -1.0743810 -2.5490054
#> 5  -2.1705889 -0.4550957
#> 6  -1.6022889 -0.4371667
#> 7  -1.4202158 -0.3346933
#> 8  -0.8077211 -0.4179719
#> 9  -2.6313808 -0.3737048
#> 10 -0.4640544 -0.4610976
#> 11 -0.3471433 -0.3373881
#> 12 -1.0913405 -0.4423398
#> 13 -3.4556299 -3.3103778
#> 14 -0.4074659 -0.3557256
#> 15 -1.9300326 -0.4575258
#> 16 -0.9970949 -0.4501086
#> 17 -0.3170426 -0.3265559
#> 18 -1.0319933 -0.4181447
#> 19 -0.6116250 -0.3960609
#> 20 -2.7416074 -0.3662881
#> 21 -0.4758236 -3.1213778
#> 22 -1.6266301 -3.1535223
#> 23 -0.7447630 -0.4558961
#> 24 -0.7142881 -0.4088163
#> 25 -0.5462736 -0.4595263
#> 26 -3.6788163 -4.0823170
#> 27 -1.5529848 -0.4522423
#> 28 -0.8521433 -2.9127306
#> 29 -2.3256844 -0.3934590
#> 30 -0.3548307 -0.3399595
#> 31 -0.3826318 -3.4900884
#> 32 -1.5451676 -5.4354411
#> 33 -2.1572908 -3.1876624
#> 34 -1.2919888 -0.3582631
#> 35 -0.9188985 -0.4522846
#> 36 -2.1703747 -2.9608889
#> 37 -1.0910099 -0.4422139
#> 38 -0.7967828 -0.4549033
#> 39 -1.4743800 -0.4509816
#> 40 -0.3706857 -0.3450336
#> 41 -0.2009077 -0.4677222
#> 42 -0.4079623 -2.4244236
#> 43 -0.3892025 -3.0649824
#> 44 -0.9079196 -0.4261489
#> 45 -1.1498986 -2.6999145
#> 46 -2.0139713 -6.0888741
#> 47 -0.6817177 -2.5623816
#> 48 -0.3414784 -2.8876591
#> 49 -0.4324802 -0.3622352
#> 50 -1.8489187 -0.4563923
colSums(ll.matrix)  # log-likelihood for each response
#>        y1        y2 
#> -59.21077 -71.87467