issues setting up cumulative-link mixed model

r-sig-clmm.R

## "reduction" is the response variable (ordered factor, none < devoiced < deleted)
table(cleandata$reduction)
##    none  devoiced  deleted 
##   20776       360      420 

## "speaker" is a nuisance predictor, which I intend to model as a random effect
with(cleandata, table(reduction, speaker))
##           speaker
## reduction    F1   F2   F3   M1   M2   M3
##   none     3559 3482 3325 3569 3360 3481
##   devoiced    9   48  123   25  105   50
##   deleted    15   50  160    5  123   67

## "word" will be modeled as a random effect (1204 levels, 8-18 obs. per word, i.e. 0-3 obs. per speaker per word).
table(with(cleandata, table(word)))
##    8   11   13   14   15   16   17   18 
##    1    1    3    4    6    6   38 1145

## vowel is not exactly a nuisance predictor, but also not the chief interest
with(cleandata, table(reduction, vowel))
##           vowel
## reduction     ɐ    ɪ    ʊ    æ    ǝ    i    o    ɔ    œ    u
##   none     4129  566 3156  276 3038 1672 2342 3151  356 2090
##   devoiced  138    4   39   20   22   24   23   54    3   33
##   deleted   135    6   40    9   36   62   31   50    1   50

## preceding and following consonant type are what we really care about
with(cleandata, table(reduction, precedingCons))
##           precedingCons
## reduction   glide  liquid  nasal  fric  affric  stop 
##   none        780     322   3258  5304    3300  7812
##   devoiced     13       0     11    92     111   133
##   deleted      12       0     12   126     111   159

with(cleandata, table(reduction, precedingCons))
##           followingCons
## reduction   glide  liquid  nasal  fric  affric  stop 
##   none       4368    1963   3234  4145    5037  2029
##   devoiced     87      58    124    51      24    16
##   deleted     112      95    129    39      32    13

## following consonant has a complicating factor that we also want to model:
## whether or not it is intervocalic (i.e., is the consonant-vowel pattern
## CvCV... or CvCC... where lower-case v is the vowel that may have been reduced).
## this is called "coda" and is true for the CvCC... pattern.
ftable(with(cleandata, table(reduction, followingCons, coda)))
##                         coda FALSE TRUE
## reduction followingCons                
## none      gli                  993 1036
##           liq                 2380 2657
##           nas                 1021 3124
##           fri                 1681 1553
##           afr                 1021  942
##           stp                 1868 2500
## devoiced  gli                    1   15
##           liq                    4   20
##           nas                   18   33
##           fri                   71   53
##           afr                   29   29
##           stp                   44   43
## deleted   gli                    5    8
##           liq                    5   27
##           nas                   12   27
##           fri                   81   48
##           afr                   60   35
##           stp                   64   48