Akaike's information criterion, −2 logLik + 2 npar.
Schwarz's criterion, −2 logLik + log(n) npar.
The likelihood-ratio statistic against the row before (NaN in the first row).
−2 logLik (R's -2*log(L)).
The difference of npar from the row before (NaN in the first row).
The log-likelihood (of the ML refit, or the fit's own with refit: false).
R's messages (refitting model(s) with ML (instead of REML)).
The models' names (R's row names), in the order of the rows.
The number of parameters of each model (fixed effects, θ and σ).
R's Pr(>Chisq) (NaN in the first row and where Df is 0).
R's warnings (duplicated names, a mix of REML and ML fits without refitting).
Result of lmerAnova for several fits: lme4's comparison table, one row per model, sorted by the number of parameters.