A. Full dataset (labelled = first number, ordinary = second number)
H1a MC identical: 63/177 (35.6%) vs 133/858 (15.5%) fisherOR=3.01 p1=4.5e-09 | GEE OR=2.38 [1.69-3.34] p2=5.8e-07 n=1035 k=152
Da Xu fanqie identical: 68/204 (33.3%) vs 128/947 (13.5%) fisherOR=3.20 p1=1.6e-10 | GEE OR=2.43 [1.74-3.39] p2=1.8e-07 n=1151 k=164
fanqie coverage (0 = ordinary, 1 = labelled): {0: 947, 1: 204}
H1c member departing | alternation: 16/33 (48.5%) vs 43/150 (28.7%) fisherOR=2.34 p1=0.025 | GEE OR=2.44 [1.34-4.48] p2=0.0038 n=183 k=66
departing tone involved | alternation: 21/33 (63.6%) vs 58/150 (38.7%) fisherOR=2.78 p1=0.0077 | GEE OR=2.42 [1.32-4.46] p2=0.0044 n=183 k=66
member departing | departing tone involved: 16/21 (76.2%) vs 43/58 (74.1%) fisherOR=1.12 p1=0.55 | GEE OR=0.27 [0.03-2.78] p2=0.27 n=79 k=39
departing-tone alternation among all MC pairs: 21/177 (11.9%) vs 58/858 (6.8%) fisherOR=1.86 p1=0.018 | GEE OR=1.64 [1.04-2.57] p2=0.032 n=1035 k=152
member-departing among all MC pairs: 16/177 (9.0%) vs 43/858 (5.0%) fisherOR=1.88 p1=0.032 | GEE OR=1.91 [1.13-3.22] p2=0.016 n=1035 k=152
head-departing among all MC pairs: 5/177 (2.8%) vs 15/858 (1.7%) fisherOR=1.63 p1=0.25 | GEE OR=0.86 [0.41-1.78] p2=0.68 n=1035 k=152
Tone/voicing alternations without a departing tone, by tone of head and member:
group                        labelled  ordinary
mc_tone_head mc_tone_member                    
上            上                      3         4
             平                      1        11
入            入                      3        10
去            去                      0         1
平            上                      5        46
             平                      0        20
Voicing differs among tone/voicing alternations:
group               labelled  ordinary
mc_voicing_differs                    
False                     23        92
True                      10        58
identity or member-departing: 79/177 (44.6%) vs 176/858 (20.5%) fisherOR=3.12 p1=9.3e-11 | GEE OR=2.53 [1.87-3.42] p2=1.6e-09 n=1035 k=152
MC match confidence (member), by group:
group                   labelled  ordinary
mc_confidence                             
identical_fanqie             121       571
initial+rhyme+tone            30       172
initial_only                  15        56
no_match_first_reading         3        23
rhyme+tone_only                8        36
MC match confidence (head), by group:
group                   labelled  ordinary
head_mc_confidence                        
identical_fanqie             121       534
initial+rhyme+tone            37       250
initial_only                  12        63
no_match_first_reading         5         4
rhyme+tone_only                2         7

B. First sample, 100 items
Non-blind pass (instance that built the dataset) vs blind pass: n=100  four categories (Y/E/N/X) kappa=0.773 (agreement 86.0%)  Y vs not-Y kappa=0.684 (agreement 93.0%)
Y-N disagreements between the two passes: 0
  disagreement (召, 詔): non-blind Y, blind E
  disagreement (成, 盛): non-blind E, blind N
  disagreement (豈, 螘): non-blind N, blind X
  disagreement (合, 洽): non-blind E, blind N
  disagreement (象, 像): non-blind Y, blind E
  disagreement (丩, 句): non-blind Y, blind E
  disagreement (賈, 價): non-blind Y, blind E
  disagreement (羊, 羌): non-blind Y, blind X
  disagreement (龍, 隴): non-blind N, blind X
  disagreement (舛, 舜): non-blind N, blind X
  disagreement (反, 返): non-blind Y, blind E
  disagreement (吅, 患): non-blind E, blind N
  disagreement (生, 姓): non-blind Y, blind E
  disagreement (平, 苹): non-blind N, blind X
items 100; group in the dataset: {'ordinary': 60, 'labelled': 39, nan: 1}; in analysis set: {'True': 94, 'False': 5, '': 1}
items whose group in the coding sheet differs from the dataset: []
analysis-set items 94 {'ordinary': 59, 'labelled': 35}
Blind pass all: Y+E 25/33 vs 6/50 OR=22.9 p=3.9e-09 | Y only 8/33 vs 0/50 OR=inf p=0.00035
Blind pass no paronomastic gloss: Y+E 18/26 vs 6/50 OR=16.5 p=7e-07 | Y only 6/26 vs 0/50 OR=inf p=0.0011
Non-blind pass all: Y+E 27/35 vs 8/53 OR=19.0 p=5.2e-09 | Y only 13/35 vs 1/53 OR=30.7 p=1.3e-05
Non-blind pass no paronomastic gloss: Y+E 19/27 vs 8/53 OR=13.4 p=1.4e-06 | Y only 8/27 vs 1/53 OR=21.9 p=0.00053
Cross-table (group, coded related) x MC relation:
mc_relation      identical  other  tone_voicing_alt
grp      semrel                                    
labelled 0               0      8                 0
         1               9     10                 6
ordinary 0               3     35                 6
         1               1      4                 1
logit ident ~ label + semrel
              Coef.     P>|z|
Intercept -2.851302  0.000004
label      0.333545  0.691764
semrel     1.837127  0.037984
logit semrel ~ label + ident
              Coef.         P>|z|
Intercept -2.245737  3.115781e-06
label      3.053070  8.851619e-07
ident      1.837127  3.798380e-02
phonetics in the sample 63
Paronomastic gloss among the sample, by group:
paronomastic_gloss  False  True
grp                            
labelled               27     8
ordinary               59     0
H2 GEE (blind pass): n=83, phonetics=58, OR=22.012 [7.122, 68.032], one-sided p=3.94e-08
H2 GEE without paronomastic glosses: n=76, OR=16.361 [5.065, 52.851], one-sided p=1.49e-06
n in the identity-semantic model: 83

Note: the sensitivity analysis without Wang Yun's rejected labels is in scripts/sensitivity_wang_yun.py (output/yisheng_models_wangyun_sensitivity.csv).
