Warning: Check model convergence: log-likelihood estimates lead to negative Chi-squared!

[gmfitch]

Trying to run LLChisq() on the model below, I get the warning "Check model convergence: log-likelihood estimates lead to negative Chi-squared!" (and model output reports: Chi-Squared = NA with P-value = NA and on 58 degrees of freedom).

Two of the component models have very low random effect variance (robbing.sc and radius.ab.sc), but removing the random effect and making these LMs does not change this warning. In fact, I can make all the component GLMMs into LMs and the issue still persists. All component glmmTMBs are OK re: convergence.

In this model, I've z-scaled all variables because unscaled variables led to issues with model convergence. Data consists of observations of plants, 6-30 plants per site, plus information about vegetation characteristics of the site, with 2-3 levels per site, across 23 sites.

How might I address this issue? Thanks!

#Component models
forest.sc <- lm(Perc.forested.sc ~ Total.Value.sc + Year.Built.sc + Deed.Acreage.sc, data = drivers.sc)

radius.ab.sc <- glmmTMB(abund.r.sc ~ Total.Value.sc + Year.Built.sc + Deed.Acreage.sc + Perc.forested.sc + (1|Site), family = gaussian, data = flo.drivers.sc)

radius.rich.sc <- glmmTMB(richness.r.sc ~ Total.Value.sc + Year.Built.sc + Deed.Acreage.sc + Perc.forested.sc + (1|Site), family = gaussian, data = flo.drivers.sc)

trans.ab.sc <- glmmTMB(abund.t.sc ~ Total.Value.sc + Year.Built.sc + Deed.Acreage.sc + Perc.forested.sc + (1|Site), family = gaussian, data = flo.drivers.sc)

trans.rich.sc <- glmmTMB(richness.t.sc ~ Total.Value.sc + Year.Built.sc + Deed.Acreage.sc + Perc.forested.sc + (1|Site), family = gaussian, data = flo.drivers.sc)

robbing.sc <- glmmTMB(robYN ~ Tot.flrs.sc + Total.Value.sc + Year.Built.sc + Deed.Acreage.sc + Perc.forested.sc + abund.t.sc + richness.r.sc + richness.t.sc + (1|Site), family = binomial, data = sem.data.subset.scaled)

florivory.tmb.sc <- glmmTMB(n.Florivorized.sc ~ Tot.flrs.sc + Total.Value.sc + Year.Built.sc + Deed.Acreage.sc + Perc.forested.sc + abund.t.sc + richness.r.sc + richness.t.sc + (1|Site), data = sem.data.subset.scaled, family = gaussian)

pol.load.tmb.sc <- glmmTMB(n.Gelsemium.sc ~ Tot.flrs.sc + abund.t.sc + richness.r.sc + richness.t.sc + robYN + n.Florivorized + Morph + (1|Site), data = sem.data.morph.sc, family = gaussian)

otherpol.load.tmb.sc <- glmmTMB(n.Other.sc ~ Tot.flrs.sc + Total.Value.sc + Perc.forested.sc + Year.Built.sc + Deed.Acreage.sc + abund.t.sc + richness.r.sc + richness.t.sc + robYN + n.Florivorized + Morph + (1|Site), data = sem.data.morph.sc, family = gaussian)

fruitset.sc <- glmmTMB(fruit.set ~ Tot.flrs.sc + richness.r.sc + robYN + n.Florivorized + n.Gelsemium.sc + n.Other.sc + (1|Site), data = sem.data.scaled, family = gaussian)

seeds.sem.sc <- glmmTMB(seeds.per.fruit.sc ~ robYN + n.Florivorized + n.Other.sc + (1|Site), data = sem.data.scaled, family = gaussian)

#SEM
psem3.full.sc <- psem(
forest.sc,
radius.ab.sc,
radius.rich.sc,
trans.ab.sc,
trans.rich.sc,
robbing.sc,
florivory.tmb.sc2,
pol.load.tmb.sc,
otherpol.load.tmb.sc,
fruitset.sc,
seeds.sem.sc,
richness.t.sc %~~% abund.t.sc,
data = sem.data.subset.scaled
)