1

Like the title says. I'm using the latest versions of both packages. My model is fairly simple to begin with:

model <- glmer(binary~factor1*continuous + (1|factor2), data=my.data, family=binomial)

Both packages should be using a Laplace estimator, but I get a convergence warning for the lme4 package that I don't get in glmmADMB. Estimates are very similar for both packages.

lme4 output

BT20.glmer <- glmer(survival ~ tree * pctrans + (1|trayid), data=BT20.trimmed, family=binomial)
Warning message:
In checkConv(attr(opt, "derivs"), opt$par, ctrl = control$checkConv,  :
  Model failed to converge with max|grad| = 0.00103723 (tol = 0.001, component 1)

summary(BT20.glmer)
Generalized linear mixed model fit by maximum likelihood (Laplace Approximation) ['glmerMod']
 Family: binomial  ( logit )
Formula: survival ~ tree * pctrans + (1 | trayid)
   Data: BT20.trimmed

     AIC      BIC   logLik deviance df.resid 
   775.7    833.1   -374.8    749.7      598 

Scaled residuals: 
    Min      1Q  Median      3Q     Max 
-2.0280 -0.7618 -0.4364  0.8399  2.2683 

Random effects:
 Groups Name        Variance Std.Dev.
 trayid (Intercept) 0.1577   0.3971  
Number of obs: 611, groups:  trayid, 42

Fixed effects:
                   Estimate Std. Error z value Pr(>|z|)
(Intercept)        -0.39192    0.57742  -0.679    0.497
treeBC3F3           1.26660    0.86633   1.462    0.144
treeD54            -0.23165    0.83908  -0.276    0.782
treeD58            -1.19841    0.88656  -1.352    0.176
treeEllis1         -1.28773    0.87973  -1.464    0.143
treeQing           -0.62360    0.74321  -0.839    0.401
pctrans             1.58330    1.45625   1.087    0.277
treeBC3F3:pctrans  -1.00729    2.02795  -0.497    0.619
treeD54:pctrans     0.09817    2.12836   0.046    0.963
treeD58:pctrans     0.43424    2.42337   0.179    0.858
treeEllis1:pctrans -0.57459    2.30508  -0.249    0.803
treeQing:pctrans    0.47065    1.64346   0.286    0.775

Correlation of Fixed Effects:
            (Intr)    tr1     tr2    tr3    tr4    tr5 pctrns   tr1:    tr2:   tr3:    tr4:
tree1       -0.666                                                                        
tree2       -0.687  0.459                                                                 
tree3       -0.651  0.434   0.448                                                         
tree4       -0.657  0.437   0.450  0.427                                                  
tree5       -0.776  0.518   0.536  0.506  0.509                                           
pctrans     -0.896  0.597   0.615  0.583  0.589  0.695                                    
tr1:pct      0.643 -0.894  -0.442 -0.419 -0.422 -0.499 -0.718                             
tr2:pctrn    0.612 -0.409  -0.887 -0.399 -0.401 -0.477 -0.683  0.491                      
tr3:pctrn    0.537 -0.359  -0.372 -0.896 -0.352 -0.420 -0.599  0.431   0.412              
tr4:pct      0.565 -0.377  -0.391 -0.369 -0.897 -0.441 -0.630  0.453   0.433  0.381       
tr5:pctrn    0.793 -0.529  -0.548 -0.517 -0.520 -0.871 -0.885  0.635   0.607  0.533  0.560
convergence code: 0
Model failed to converge with max|grad| = 0.00103723 (tol = 0.001, component 1)

glmmADMB

> BT20.admb <- glmmadmb(survival ~ tree*pctrans + (1|trayid), data=BT20.trimmed, family="binomial")
> summary(BT20.admb)

Call:
glmmadmb(formula = survival ~ tree * pctrans + (1 | trayid), 
    data = BT20.trimmed, family = "binomial")

AIC: 775.7 

Coefficients:
                   Estimate Std. Error z value Pr(>|z|)
(Intercept)         -0.3915     0.5777   -0.68     0.50
tree1                1.2660     0.8675    1.46     0.14
tree2               -0.2320     0.8394   -0.28     0.78
tree3               -1.1993     0.8873   -1.35     0.18
tree4               -1.2882     0.8808   -1.46     0.14
tree5               -0.6242     0.7435   -0.84     0.40
pctrans              1.5821     1.4572    1.09     0.28
tree1:pctrans       -1.0056     2.0310   -0.50     0.62
tree2:pctrans        0.0995     2.1292    0.05     0.96
tree3:pctrans        0.4365     2.4250    0.18     0.86
tree4:pctrans       -0.5735     2.3078   -0.25     0.80
tree5:pctrans        0.4722     1.6445    0.29     0.77

Number of observations: total=611, trayid=42 
Random effect variance(s):
Group=trayid
            Variance StdDev
(Intercept)   0.1577 0.3971


Log-likelihood: -374.835

lme4 optimizer output

start par. =  1 fn =  761.1452 
At return
eval:  17 fn:      749.70460 par: 0.394784
(NM) 20: f = 749.704 at  0.394784  -0.38221   1.23536  -0.22496   -1.1676  -1.25786 -0.611638   1.54437 -0.981101 0.0945286  0.422298 -0.559069  0.466482
(NM) 40: f = 749.704 at  0.394784  -0.38221   1.23536  -0.22496   -1.1676  -1.25786 -0.611638   1.54437 -0.981101 0.0945286  0.422298 -0.559069  0.466482
(NM) 60: f = 749.704 at  0.394784  -0.38221   1.23536  -0.22496   -1.1676  -1.25786 -0.611638   1.54437 -0.981101 0.0945286  0.422298 -0.559069  0.466482
(NM) 80: f = 749.693 at  0.403214 -0.398286   1.27238 -0.229772   -1.1895  -1.28617 -0.611556    1.5764 -0.934195  0.131065  0.427062 -0.577178  0.503229
(NM) 100: f = 749.683 at  0.398067 -0.386183   1.25362 -0.224924  -1.17371  -1.26664 -0.607996   1.53317 -0.956612  0.117117  0.431422 -0.552243   0.48632
(NM) 120: f = 749.677 at  0.403184 -0.384828    1.2495 -0.237432   -1.1928  -1.27768 -0.630575   1.56274 -0.962548  0.140994  0.409612 -0.558911  0.489161
(NM) 140: f = 749.676 at  0.401892 -0.391983   1.25148 -0.231979  -1.17635  -1.28301 -0.621995   1.55619 -0.955253  0.148741  0.419699 -0.551909   0.49057
(NM) 160: f = 749.673 at  0.401711 -0.389823   1.24776 -0.248815  -1.18684  -1.29189 -0.625875    1.5646 -0.957093  0.149736   0.40804 -0.553094  0.488138
(NM) 180: f = 749.672 at  0.400601 -0.385288   1.25068 -0.242765  -1.19045  -1.29892 -0.629416   1.56929 -0.974233  0.122587  0.398086 -0.542567  0.483422
(NM) 200: f = 749.671 at  0.400013 -0.385172   1.25169 -0.243985  -1.19194  -1.30175 -0.630152    1.5716 -0.975104  0.120825  0.396965 -0.542863  0.482651
(NM) 220: f = 749.671 at  0.398195 -0.385741    1.2534 -0.245356  -1.19137  -1.30133  -0.62962   1.57661  -0.98824  0.119274  0.404967 -0.548629  0.478899
(NM) 240: f = 749.671 at  0.396846 -0.386089   1.25599  -0.23783  -1.18838  -1.29705 -0.627903   1.56999 -0.983097  0.112611  0.412207 -0.550918  0.482465
(NM) 260: f = 749.671 at  0.396791 -0.386088    1.2563 -0.237798  -1.18869  -1.29718 -0.627977   1.57024 -0.983812  0.112462  0.412037 -0.550764  0.482539
(NM) 280: f = 749.67 at  0.396502 -0.385236   1.25722 -0.238627  -1.19152  -1.29885 -0.629215   1.56901 -0.989994  0.110833  0.409519 -0.545798  0.484243
(NM) 300: f = 749.67 at  0.396416 -0.386106   1.26028 -0.236847  -1.19112  -1.29556  -0.62914   1.56895 -0.992995  0.107319  0.413889 -0.551313  0.484151
(NM) 320: f = 749.67 at  0.395913 -0.387481   1.25938 -0.237382  -1.19242  -1.29511 -0.628587   1.57013  -0.98866  0.110182  0.414504 -0.548281  0.484826
(NM) 340: f = 749.67 at  0.395808 -0.387955   1.26093 -0.236321  -1.19224  -1.29443 -0.629328   1.57052 -0.993145  0.107512  0.419233 -0.551661  0.486564
(NM) 360: f = 749.67 at  0.396856 -0.388637   1.26064 -0.236805  -1.19136  -1.29248 -0.629241   1.57061 -0.990447  0.115044  0.418231 -0.554314  0.487459
(NM) 380: f = 749.67 at  0.396118 -0.389248   1.26173 -0.236795  -1.19326   -1.2919 -0.629308   1.57276 -0.993918  0.110452  0.422509 -0.555278  0.487294
(NM) 400: f = 749.67 at  0.396621 -0.389002   1.26128 -0.235907  -1.19246  -1.29135  -0.63008   1.57381 -0.995446  0.110692  0.422069 -0.558286  0.486561
(NM) 420: f = 749.67 at  0.396519 -0.388862    1.2613  -0.23713  -1.19317  -1.29216 -0.629661   1.57408 -0.994968  0.112797  0.419094 -0.557443  0.486028
(NM) 440: f = 749.67 at  0.396248  -0.38993   1.26353 -0.235109  -1.19359   -1.2903 -0.630125   1.57579  -1.00009  0.108874  0.425007 -0.561882   0.48679
(NM) 460: f = 749.67 at  0.396277 -0.389883   1.26476 -0.233885  -1.19395  -1.29026 -0.630761   1.57728  -1.00328  0.105159  0.424146 -0.563362  0.485087
(NM) 480: f = 749.67 at  0.396348 -0.389975   1.26565 -0.234128  -1.19478  -1.29018 -0.630783   1.57876  -1.00724  0.104176  0.423989 -0.566655  0.483861
(NM) 500: f = 749.67 at  0.396215 -0.390091   1.26585 -0.233658  -1.19439  -1.28956 -0.630044   1.57901   -1.0065  0.103831  0.423434 -0.567486  0.482324
(NM) 520: f = 749.67 at  0.396484 -0.391154   1.26807 -0.233121  -1.19536  -1.28886 -0.629592   1.58241  -1.01112   0.10188  0.423351 -0.573978  0.479046
(NM) 540: f = 749.67 at  0.396361 -0.390515   1.26651 -0.234651  -1.19448  -1.28952 -0.628941   1.58035   -1.0062  0.105582  0.422028 -0.570506   0.47979
(NM) 560: f = 749.67 at  0.396698 -0.390748   1.26668 -0.234277  -1.19455   -1.2887 -0.628991   1.58135  -1.00742  0.105631  0.421337 -0.573112   0.47857
(NM) 580: f = 749.67 at  0.396857 -0.391228   1.26664  -0.23428  -1.19452  -1.28799 -0.628566    1.5812   -1.0071  0.107476  0.422354  -0.57374  0.479248
(NM) 600: f = 749.67 at  0.396639 -0.391074   1.26688 -0.233833  -1.19452  -1.28761 -0.627606   1.58228  -1.00798  0.105107  0.421428 -0.576241  0.475988
(NM) 620: f = 749.67 at  0.396514 -0.391446   1.26679 -0.233815  -1.19457  -1.28567 -0.625137     1.583  -1.00846  0.105126   0.42157 -0.580426  0.471714
(NM) 640: f = 749.67 at  0.396445 -0.391424   1.26699 -0.233415  -1.19449  -1.28578 -0.625553   1.58222  -1.00829  0.103624  0.422954 -0.578994  0.472756
(NM) 660: f = 749.67 at  0.396452 -0.392158   1.26703 -0.233823   -1.1947  -1.28512  -0.62531   1.58277  -1.00754  0.104789  0.425069 -0.579177  0.473394
(NM) 680: f = 749.67 at  0.396455 -0.392274    1.2671  -0.23388  -1.19477  -1.28495 -0.625224     1.583  -1.00758  0.104935  0.425272 -0.579541  0.473252
(NM) 700: f = 749.67 at  0.396481 -0.392492    1.2674 -0.233511  -1.19516  -1.28439  -0.62484   1.58438  -1.00936  0.104054  0.425045 -0.582091  0.471637
(NM) 720: f = 749.67 at   0.39644 -0.392519   1.26742 -0.233029   -1.1955  -1.28441 -0.624835   1.58457  -1.01027  0.101949  0.425569 -0.582011  0.471336
(NM) 740: f = 749.67 at   0.39652 -0.392395   1.26728 -0.232842  -1.19564  -1.28487 -0.625022    1.5845  -1.01085  0.100923  0.425203 -0.581493  0.471482
(NM) 760: f = 749.67 at  0.396412 -0.392344   1.26728 -0.233176  -1.19569  -1.28522 -0.625116   1.58465     -1.01  0.101249   0.42496  -0.58113  0.471918
(NM) 780: f = 749.67 at  0.396497 -0.392331   1.26719 -0.232923  -1.19593  -1.28524 -0.624906   1.58523  -1.01089 0.0996949  0.424681 -0.582026  0.470846
(NM) 800: f = 749.67 at   0.39655 -0.392371   1.26689 -0.233195  -1.19585   -1.2853  -0.62467   1.58501  -1.00981  0.100425   0.42451 -0.581754  0.471202
(NM) 820: f = 749.67 at  0.396567 -0.392335   1.26671 -0.232965  -1.19587  -1.28512 -0.624433   1.58508  -1.00973 0.0997577  0.424518 -0.582178  0.470849
(NM) 840: f = 749.67 at  0.396732 -0.392229   1.26611 -0.233071  -1.19556   -1.2851 -0.624647   1.58446  -1.00836  0.100429  0.424319 -0.581106  0.471419
(NM) 860: f = 749.67 at  0.396703 -0.392226   1.26609 -0.232815  -1.19551  -1.28519   -0.6244   1.58497  -1.00843  0.099104  0.424076  -0.58195  0.470471
(NM) 880: f = 749.67 at  0.396746  -0.39207   1.26541 -0.232133   -1.1953  -1.28533 -0.624674   1.58472   -1.0068 0.0971924  0.424188 -0.581183  0.471429
(NM) 900: f = 749.67 at  0.396766 -0.392408   1.26572 -0.232447  -1.19564  -1.28481 -0.624072   1.58574  -1.00791 0.0975808   0.42433 -0.582941  0.469748
(NM) 920: f = 749.67 at  0.396699   -0.3925   1.26588 -0.232031  -1.19554   -1.2844 -0.624127   1.58569  -1.00812 0.0970635  0.425244  -0.58326  0.470014
(NM) 940: f = 749.67 at  0.396713 -0.392349   1.26587 -0.232277  -1.19553  -1.28479 -0.624183   1.58556  -1.00817 0.0974172  0.424577 -0.582925  0.469947
(NM) 960: f = 749.67 at  0.396687 -0.392355   1.26594 -0.232175  -1.19552  -1.28485 -0.624305    1.5855  -1.00822 0.0972216  0.424832   -0.5828   0.47036
(NM) 980: f = 749.67 at  0.396698 -0.392454   1.26589 -0.232042  -1.19551  -1.28465 -0.624227   1.58578  -1.00815 0.0968939  0.425079 -0.583228  0.470127
(NM) 1000: f = 749.67 at  0.396682 -0.392514   1.26604 -0.232122  -1.19546  -1.28455 -0.624167   1.58579  -1.00828 0.0972957  0.425222  -0.58347  0.470069
(NM) 1020: f = 749.67 at  0.396685 -0.392423   1.26596 -0.232176  -1.19541  -1.28462 -0.624197   1.58546  -1.00801  0.097574  0.425059 -0.583059  0.470341
(NM) 1040: f = 749.67 at  0.396686 -0.392564    1.2661 -0.232132  -1.19548   -1.2845 -0.624159   1.58586   -1.0083 0.0974297  0.425244 -0.583608  0.470085
(NM) 1060: f = 749.67 at  0.396728 -0.392532   1.26598 -0.232147  -1.19547  -1.28453 -0.624109   1.58582  -1.00817 0.0972738  0.425098 -0.583576  0.469964
(NM) 1080: f = 749.67 at  0.396728 -0.392496   1.26605 -0.232187  -1.19543   -1.2846 -0.624203   1.58561  -1.00813 0.0976207  0.425015 -0.583363  0.470273
(NM) 1100: f = 749.67 at   0.39677 -0.392593   1.26611 -0.232074  -1.19545  -1.28443 -0.624014   1.58572  -1.00816 0.0973341  0.425115 -0.583941  0.470047
(NM) 1120: f = 749.67 at  0.396793 -0.392563   1.26632 -0.231778  -1.19539  -1.28445 -0.624179   1.58544   -1.0081 0.0970493   0.42508 -0.584049  0.470614
(NM) 1140: f = 749.67 at    0.3968 -0.392604   1.26623 -0.231627  -1.19539  -1.28427 -0.623946   1.58554  -1.00805 0.0964774  0.425111 -0.584385  0.470109
(NM) 1160: f = 749.67 at  0.396831 -0.392569   1.26627 -0.231529  -1.19541  -1.28426 -0.624149   1.58536  -1.00792 0.0965457  0.425083 -0.584181  0.470591
(NM) 1180: f = 749.67 at  0.396825 -0.392629   1.26628  -0.23151  -1.19551  -1.28419 -0.624008   1.58565  -1.00813 0.0962724  0.425088 -0.584545  0.470113
(NM) 1200: f = 749.67 at  0.396825 -0.392629   1.26628  -0.23151  -1.19551  -1.28419 -0.624008   1.58565  -1.00813 0.0962724  0.425088 -0.584545  0.470113
(NM) 1220: f = 749.67 at  0.396817 -0.392608   1.26629 -0.231538  -1.19549  -1.28418 -0.624036   1.58555    -1.008 0.0965452  0.424986 -0.584406   0.47019
(NM) 1240: f = 749.67 at  0.396804 -0.392681   1.26634  -0.23157  -1.19547  -1.28408 -0.623942    1.5857  -1.00806  0.096697  0.425128 -0.584691   0.47001
(NM) 1260: f = 749.67 at  0.396802  -0.39264   1.26626 -0.231461  -1.19547  -1.28412 -0.623967   1.58558   -1.0079 0.0964204  0.425118 -0.584497  0.470173
(NM) 1280: f = 749.67 at  0.396807 -0.392638   1.26625 -0.231381   -1.1955   -1.2841 -0.623893    1.5856  -1.00785 0.0962084  0.425024  -0.58459  0.470034
(NM) 1300: f = 749.67 at  0.396771 -0.392722   1.26622 -0.231369  -1.19554  -1.28403 -0.623729   1.58583  -1.00779 0.0961983  0.425128 -0.584769  0.469716
(NM) 1320: f = 749.67 at  0.396768 -0.392784   1.26617 -0.231366  -1.19562  -1.28404 -0.623649    1.5859  -1.00762 0.0963004  0.425202 -0.584636  0.469723
(NM) 1340: f = 749.67 at  0.396767 -0.392806   1.26622 -0.231389   -1.1956  -1.28401 -0.623607   1.58579  -1.00759 0.0965361  0.425325 -0.584578  0.469763
(NM) 1360: f = 749.67 at  0.396866 -0.392762   1.26624  -0.23137  -1.19555  -1.28417 -0.623728   1.58533  -1.00711 0.0970542  0.425119 -0.584159  0.470329
(NM) 1380: f = 749.67 at  0.396826  -0.39282    1.2661 -0.231337  -1.19567  -1.28435 -0.623529    1.5855  -1.00691 0.0969928  0.425347 -0.583752  0.470006
(NM) 1400: f = 749.67 at  0.396828 -0.392809   1.26587 -0.230523  -1.19567  -1.28454 -0.623204   1.58518  -1.00592 0.0958208  0.425714 -0.583203  0.469847
(NM) 1420: f = 749.67 at   0.39682 -0.392772   1.26583 -0.231355  -1.19569  -1.28484 -0.623301   1.58509  -1.00597  0.097786  0.425473 -0.582301  0.469858
(NM) 1440: f = 749.67 at  0.396817 -0.392632   1.26561 -0.230757   -1.1958  -1.28514 -0.623166   1.58496  -1.00593 0.0960783  0.425831 -0.581773  0.469414
(NM) 1460: f = 749.67 at  0.396878 -0.392651   1.26552 -0.231005  -1.19575   -1.2856 -0.622971   1.58465  -1.00551 0.0972454  0.426505 -0.580689  0.469065
(NM) 1480: f = 749.67 at  0.396877  -0.39262   1.26542 -0.230338  -1.19594  -1.28565 -0.623137     1.585  -1.00615 0.0947591  0.427144 -0.580886  0.469078
(NM) 1500: f = 749.67 at  0.396912 -0.392667   1.26537 -0.230245  -1.19599  -1.28571 -0.623118   1.58511  -1.00615 0.0944749  0.427491  -0.58083  0.468978
(NM) 1520: f = 749.67 at  0.396907 -0.392819   1.26548 -0.230669  -1.19601  -1.28547 -0.623182   1.58553  -1.00624  0.095432  0.427372 -0.581363  0.469012
(NM) 1540: f = 749.67 at  0.396907 -0.392819   1.26548 -0.230669  -1.19601  -1.28547 -0.623182   1.58553  -1.00624  0.095432  0.427372 -0.581363  0.469012
(NM) 1560: f = 749.67 at  0.396846 -0.392872   1.26549 -0.230975  -1.19603  -1.28571 -0.623079   1.58562  -1.00611 0.0965022  0.427656 -0.580792  0.468764
(NM) 1580: f = 749.67 at  0.396871 -0.392731   1.26548 -0.230966  -1.19629  -1.28622 -0.623358   1.58536  -1.00663 0.0961115  0.428155 -0.579597   0.46913
(NM) 1600: f = 749.67 at   0.39686  -0.39284   1.26563 -0.231135  -1.19639  -1.28626 -0.623407    1.5856  -1.00688 0.0965733  0.428763 -0.579568  0.469227
(NM) 1620: f = 749.67 at  0.396932 -0.392808   1.26574 -0.231183  -1.19665  -1.28668 -0.623765   1.58549  -1.00711 0.0967714  0.429543 -0.578619  0.469839
(NM) 1640: f = 749.67 at  0.396889 -0.392691   1.26572 -0.230796  -1.19694  -1.28691 -0.623865    1.5856  -1.00733 0.0955276  0.429535 -0.578312  0.469878
(NM) 1660: f = 749.67 at  0.396886 -0.392516   1.26562 -0.230521  -1.19706  -1.28715 -0.623744   1.58529  -1.00658 0.0953614  0.429132  -0.57776  0.469805
(NM) 1680: f = 749.67 at  0.396993 -0.392586   1.26569 -0.230802   -1.1969  -1.28694 -0.623985   1.58509  -1.00626 0.0965312  0.429345 -0.577799  0.470392
(NM) 1700: f = 749.67 at  0.396936 -0.392258   1.26567 -0.230714  -1.19731  -1.28786 -0.624302   1.58469  -1.00631 0.0965168  0.429975 -0.575631  0.470725
(NM) 1720: f = 749.67 at  0.396981 -0.392454   1.26553 -0.230937  -1.19695  -1.28732 -0.623964   1.58477  -1.00581 0.0972308  0.429443  -0.57663  0.470304
(NM) 1740: f = 749.67 at  0.396973 -0.392387   1.26563 -0.230911  -1.19706  -1.28728 -0.624211   1.58489  -1.00612 0.0969993  0.429718 -0.576671  0.470485
(NM) 1760: f = 749.67 at  0.396943 -0.392263   1.26568 -0.230933  -1.19699  -1.28745 -0.624264   1.58456  -1.00628 0.0970891   0.42972 -0.576271  0.470648
(NM) 1780: f = 749.67 at  0.396939 -0.392186   1.26572 -0.230985    -1.197  -1.28738 -0.624367   1.58447  -1.00643 0.0970968  0.429719 -0.576269  0.470732
(NM) 1800: f = 749.67 at  0.396948 -0.392171   1.26577 -0.230764   -1.1971  -1.28732  -0.62441   1.58452  -1.00673  0.096286  0.429841 -0.576488  0.470805
(NM) 1820: f = 749.67 at  0.396971 -0.392141    1.2659 -0.230947  -1.19698  -1.28718 -0.624475   1.58422  -1.00674 0.0969716  0.429722 -0.576586  0.471092
(NM) 1840: f = 749.67 at  0.396945 -0.392139   1.26597 -0.230962  -1.19701  -1.28707  -0.62448   1.58437  -1.00708 0.0966966   0.42988 -0.576925  0.471011
(NM) 1860: f = 749.67 at  0.396972 -0.392213   1.26593 -0.231074  -1.19723  -1.28721 -0.624402   1.58453  -1.00711 0.0968097  0.430196  -0.57657   0.47081
(NM) 1880: f = 749.67 at  0.397024 -0.392183   1.26595 -0.231329  -1.19731  -1.28733 -0.624461   1.58427  -1.00698  0.097675  0.430796 -0.575895  0.470995
(NM) 1900: f = 749.67 at  0.396991 -0.392182     1.266 -0.231467  -1.19731  -1.28704 -0.624205   1.58434  -1.00729 0.0974862  0.430715 -0.576544  0.470548
(NM) 1920: f = 749.67 at  0.397039 -0.392191     1.266 -0.231734  -1.19793  -1.28731 -0.623951   1.58448   -1.0074 0.0977164  0.431918  -0.57564  0.470021
(NM) 1940: f = 749.67 at  0.397027 -0.392047   1.26626 -0.231944  -1.19794  -1.28729 -0.624103   1.58408  -1.00784 0.0982291  0.432216 -0.575468  0.470392
(NM) 1960: f = 749.67 at  0.397066 -0.392037   1.26612 -0.231914  -1.19802  -1.28724 -0.623767   1.58399   -1.0072 0.0983412  0.432213 -0.575423  0.469928
(NM) 1980: f = 749.67 at  0.397087 -0.391862   1.26629 -0.231432  -1.19829  -1.28781 -0.623692   1.58353  -1.00703 0.0975409  0.433466 -0.574501  0.470314
(NM) 2000: f = 749.67 at  0.397087 -0.391862   1.26629 -0.231432  -1.19829  -1.28781 -0.623692   1.58353  -1.00703 0.0975409  0.433466 -0.574501  0.470314
(NM) 2020: f = 749.67 at  0.397087 -0.391862   1.26629 -0.231432  -1.19829  -1.28781 -0.623692   1.58353  -1.00703 0.0975409  0.433466 -0.574501  0.470314
(NM) 2040: f = 749.67 at  0.397077 -0.391939   1.26646 -0.231486  -1.19841  -1.28789 -0.623662   1.58361  -1.00741 0.0975635  0.433883 -0.574515  0.470425
(NM) 2060: f = 749.67 at   0.39712 -0.391966   1.26646 -0.231625  -1.19838  -1.28778  -0.62354   1.58344  -1.00716 0.0980888  0.434002 -0.574528  0.470418
(NM) 2080: f = 749.67 at   0.39712 -0.391966   1.26646 -0.231625  -1.19838  -1.28778  -0.62354   1.58344  -1.00716 0.0980888  0.434002 -0.574528  0.470418
(NM) 2100: f = 749.67 at  0.397107 -0.391912   1.26662  -0.23165  -1.19843  -1.28786  -0.62364   1.58331  -1.00743 0.0981731  0.434137 -0.574402  0.470663
(NM) 2120: f = 749.67 at  0.397106 -0.391911    1.2666 -0.231627  -1.19847  -1.28781 -0.623645   1.58341  -1.00742 0.0980212  0.434218 -0.574503  0.470601
(NM) 2140: f = 749.67 at  0.397119  -0.39194   1.26659 -0.231646  -1.19842  -1.28776 -0.623553   1.58336   -1.0073 0.0981474  0.434229 -0.574597  0.470556
(NM) 2160: f = 749.67 at  0.397123 -0.391945    1.2666 -0.231625  -1.19835  -1.28768 -0.623633   1.58336  -1.00732 0.0981253  0.434178 -0.574749  0.470696
RegalPlatypus
  • 43
  • 9
  • What's the exact warning message provided by lme4? it could be that you got a warning, but things are actually o.k. (the software wasn't sure). Or it could be that the solutions from both packages are lousy, but glmmADMB didn't provide you a warning. I recommend you use the highest level of output available, and provide the output for the runs of both packages. – Mark L. Stone Jul 28 '16 at 19:36
  • The largest magnitude gradient component was just a smidge over the tolerance level. I think you can control the threshold via lmerControl. The line 'check.conv.grad = .makeCC("warning", tol = 2e-3, relTol = NULL),' makes it seem like the relevant threshold is set by default to 0.002, not 0.001, but anyhow, I know nothing of this package. If you increase the threshold a bit, I think the warning message will go away. As to whether your solution is really o.k., I'm guessing yes, but really don;'t know. it seems like a lousy setting to only have absolute, and not also relative tolerance set. – Mark L. Stone Jul 28 '16 at 19:53
  • Edited to add information – RegalPlatypus Jul 28 '16 at 19:57
  • Is it possible to set it so that you get intermediate output of the optimizer, showing its progression (convergence metrics) as it goes along? – Mark L. Stone Jul 28 '16 at 20:01
  • Yup. Editing it right now. – RegalPlatypus Jul 28 '16 at 20:09
  • Which optimization algorithm are you using? Try LBFGS_B if you can (with exact derivatives if possible).. Read pp. 13-14 of https://cran.r-project.org/web/packages/lme4/lme4.pdf . Unfortunately, this doesn't sound like a top-notch piece of software. – Mark L. Stone Jul 28 '16 at 20:35
  • By default lmer4 uses a "mix of Nelder-Mead and bobyqa." I'm already close to sinking. Changing that under the hood looks like it'd drown me. – RegalPlatypus Jul 28 '16 at 21:39
  • Is there no high level option to change the optimizer? At least go with bobyqa over Nelder-Mead. What does glmmADMB use? I see no reason not to use a derivative-based method for this problem. The 2 algorithms you mentioned for lmer4 do not use derivatives; of these, i suspect bobyqa is better. – Mark L. Stone Jul 28 '16 at 21:42
  • I think glmmADMB uses Newton-Raphson. – RegalPlatypus Jul 28 '16 at 21:48
  • 1
    Depending on the implementation, that's probably better. Do you have a reason to want to use lme4 instead of glmmADMB? Perhaps you should show intermediate solver output (every iteration, if it's not too much) for glmmADMB? If the algorithm is any good, it's probably many fewer iterations than lm4 took (its output was every 20 iterations).. – Mark L. Stone Jul 28 '16 at 21:54
  • My primary reason for wanting to use lme4 is that it seems to be far more commonplace in the literature. The literature for lme4 says that optimizers from other packages can be used, but I haven't found a good code recipe for how to do that. See [Page 14](https://cran.r-project.org/web/packages/lme4/lme4.pdf) – RegalPlatypus Jul 28 '16 at 22:04
  • At this point, I think you need to seek out more specialized help. That said, my guess is that your solution from lme4 on this problem is probably o.k., but maybe not "nailed to the wall". – Mark L. Stone Jul 28 '16 at 22:07
  • Thanks for all the help you've given, @MarkL.Stone ! I appreciate it! – RegalPlatypus Jul 28 '16 at 22:12
  • Some great comments by @MarkLStone.... I think that you should use whatever package that you are comfortable with. But convergence issues are always a tricky thing to solve... as they recommend in the lme4 package try a few different optimization routines and if they converge at the same point you are ok... as already seen with glmmADMB...Since its only a warning and not a failure to converge, which could be even more trickier.. – ashokragavendran Jul 29 '16 at 01:17
  • [This page](https://rstudio-pubs-static.s3.amazonaws.com/33653_57fc7b8e5d484c909b615d8633c01d51.html) was extremely helpful. I was able to run the regression with multiple optimizers all in one go and compare the LogLik for each of them. In this case, even those that gave convergence warnings, the LogLik was very similar for all. – RegalPlatypus Jul 29 '16 at 03:09
  • At https://rstudio-pubs-static.s3.amazonaws.com/33653_57fc7b8e5d484c909b615d8633c01d51.html , why did L-BGS-B and the other optimizers fail to converge, having max|grad| > 1? Is the gradient "exact", correct, and not horribly scaled? Does component 19 refer to the 19th component of the gradient as being the offender? – Mark L. Stone Jul 29 '16 at 21:33

1 Answers1

1

Summarizing some of the above commentary:

  • a lot of this discussion is summarized in the ?lme4::convergence help page, including 
source(system.file("utils", "allFit.R", package="lme4"))
fm1.all <- allFit(fm1)

which is doing the same thing as the Rpubs page referenced in the comments above.

  • the difference between glmmADMB and lme4 is not really that one is converging and the other isn't, but that lme4 has more checks on the convergence -- and as hinted at in ?convergence, these checks give an unfortunately large number of false positives.
  • I'm a little bit surprised that you're getting gradient-convergence warnings with max|grad| of 0.00103, since the gradient tolerance was changed to 0.002 in 2014; are you sure you're using the latest version?
  • the reason for setting absolute rather than relative tolerances is that everything is scaled; the response is on the deviance scale (where absolute, not relative differences matter), and the predictors are scaled by their second derivatives (although that doesn't always work so well ...)
  • glmer does indeed use bobyqa for the first (nAGQ=0) pass and Nelder-Mead for the second (nAGQ=1) pass ... I would indeed recommend using bobyqa throughout (control=glmerControl(method="bobyqa")), the next release will probably make that switch. AD Model Builder (the underlying engine for glmmADMB uses a quasi-Newton method (Fournier et al 2012, DOI: 10.1080/10556788.2011.597854 ), but very little documentation beyond the source code is available.
  • if you want to do further cross-checking, the (experimental) glmmTMB package offers another alternative.
  • the r-sig-mixed-models@r-project.org mailing list is an alternative venue for these questions
  • I think I'm a little hurt by @MarkL.Stone's "Unfortunately, this doesn't sound like a top-notch piece of software" ... have you ever tried to write and maintain a general-purpose GLMM-fitting package ... ?
Ben Bolker
  • 34,308
  • 2
  • 93
  • 126
  • It's great to have you in the comments, Ben! I've checked and I'm using lme4 v. 1.1-12. The allFit() function in package afurl that I linked to above seemed to work great for me at printing the results from several optimizers at once (just an FYI). While I have you here, may I shamelessly ask you whether you know it's possible to calculate VIF or a condition number for a lme4 (or glmmADMB) GLMM? (In my particular case, one variable is continuous and the other is categorical, which may complicate things). Thanks for all your hard work, @BenBolker! – RegalPlatypus Jul 29 '16 at 21:10
  • I upvoted your answer even though I'm a little hurt that you're hurt. My opinion on the quality of the optimization aspect has not changed. – Mark L. Stone Jul 29 '16 at 21:16
  • 1
    Seriously, if you have any advice at all on how to improve the optimization and/or convergence-testing capabilities of the package I'd be happy to hear it (best channel is probably to post an issue at https://github.com/lme4/lme4/issues ). See also https://github.com/Stat990-033/Timings/blob/master/inst/doc/Paper.pdf ... – Ben Bolker Jul 29 '16 at 21:29
  • @RegalPlatypus : don't know. Ask at r-sig-mixed-models ? or see http://stats.stackexchange.com/questions/82984/how-to-test-and-avoid-multicollinearity-in-mixed-linear-model – Ben Bolker Jul 29 '16 at 21:30
  • Yeah, I didn't see anything there helped in my situation. Thanks though! – RegalPlatypus Jul 29 '16 at 21:33