Skip to content
Prev 4266 / 20628 Next

LRT tests in lmer

Hi
The ordinal response are categorical - different levels of threat, however they can be successfully used as a continuous response ( Purvis, Mace etc) they are not associated with any of the fixed factors, i am trying to use the fixed factors (life history traits) to predict the ordinal response. 

I shall have a play with the priors as
Has improved things, but not greatly

Thanks

Chris

Intercept)              -0.23325 -2.89744  2.83429    793.1 0.884  
STOStorage organ         -0.04486 -0.28088  0.23706   1306.4 0.722  
BSUnisexual flower        0.21329 -0.11396  0.52257    861.1 0.206  
BSUnisexual plant         0.33547 -0.04818  0.75086    806.5 0.122  
PDBiotic                  0.28292 -0.13199  0.63020    599.1 0.184  
PDMammalia               -0.46017 -2.15330  1.44028    862.8 0.640  
FRNon_fleshy_fruit       -0.22784 -0.54680  0.10850    764.5 0.192  
ENDNon_endospermous       0.44173  0.10830  0.74418    747.8 0.016 *
WOWoody                  -0.22039 -0.59506  0.11227    631.0 0.252  
RGTwo+                   -0.04816 -0.24944  0.15221    816.4 0.666  
SEAHapaxanthic           -1.53904 -4.55702  1.67797    688.6 0.330  
SEAHapaxanthic            0.18037 -1.72087  2.27258    796.5 0.800  
SEAPerennial             -0.07601 -0.44810  0.33258    926.0 0.712  
SEAPleonanthic           -0.14699 -1.14695  0.81452    723.9 0.748  
ALTHigh                  -0.13191 -0.46780  0.22911    725.0 0.452  
ALTLow                   -0.17699 -0.51173  0.10969    772.8 0.292  
ALTMid                    0.06855 -0.21312  0.41342    882.1 0.684  
BIOBoreal                 1.74800 -1.18782  4.72759    782.0 0.242  
BIOMediterranean-type     2.08074 -0.62533  5.05527    780.1 0.140  
BIOSubantarctic           2.17686 -1.13669  5.24883    806.7 0.180  
BIOSubarctic              2.39551 -0.91077  5.41454    839.1 0.138  
BIOSubtropical/Tropical   2.31132 -0.36795  5.24304    791.5 0.110  
BIOTemperate              2.29529 -0.41744  5.18185    795.5 0.104  
SEFew-Several             1.86331 -0.57544  4.01647    732.1 0.106  
SENumerous                0.20823 -0.14937  0.57547    851.4 0.226  
SESeveral                 0.66868 -0.13298  1.45685    894.6 0.102  
SESingle                  0.42408  0.07265  0.80295    872.5 0.022 *
FSZygomorphic             0.01505 -0.22554  0.27481    760.5 0.908
On 11 Aug 2010, at 17:34, Jarrod Hadfield wrote:
Hi Chris,


The model syntax looks reasonable but there seems to be some large posterior means (outside of the 95% credible range). I bet plot(model$VCV) looks pretty horrible too. You need to  consider using proper priors in this instance because the chain is getting stuck at zero for long periods of time and generating numerical problems. I tend to use parameter expanded priors more and more as they improve mixing and seem to be only weakly informative. For example: G1=list(V=1, nu=1, alpha.mu=0, alpha.V=1000) ....  There is also the possibility that you have complete separation as you have a lot of fixed effects and many levels in the ordinal response - are all 5's for example associated with a single fixed factor, or something like this?

Jarrod
On 11 Aug 2010, at 17:20, Chris Mcowen wrote: