======================================================================== 
1. CONDUCT CATEGORIES CLUSTERED ON OUTCOME PROFILES
  k=2: Corruption, Deceit / falsification, Gender-based violence, Off-duty conduct | Discourtesy, Unnecessary force, Firearms, Information misuse, (story too thin), Investigative neglect, Property misuse, Unlawful arrest / search, Vehicle operation, Workplace harassment
  k=3: Corruption, Deceit / falsification, Gender-based violence, Off-duty conduct | Discourtesy, Unnecessary force, Unlawful arrest / search | Firearms, Information misuse, (story too thin), Investigative neglect, Property misuse, Vehicle operation, Workplace harassment
  k=4: Corruption, Deceit / falsification, Off-duty conduct | Discourtesy, Unnecessary force, Unlawful arrest / search | Firearms, Information misuse, (story too thin), Investigative neglect, Property misuse, Vehicle operation, Workplace harassment | Gender-based violence
======================================================================== 
2. SANCTION-MEASURE CO-OCCURRENCE (11 statutory measures)
  eigenvalues of the phi-correlation matrix (11 measures):
   [1.66 1.36 1.3  1.16 1.11 1.03 1.01 0.83 0.79 0.65 0.09]
  axis 1 explains 15%, axis 2 12% of variance
  axis 1 loadings (most negative / most positive):
    training -0.42; close supervision -0.41; suspension -0.35  ||  advice future conduct +0.07; verbal reprimand +0.15; dismissal +0.40
  axis 2 loadings (most negative / most positive):
    suspension -0.42; dismissal -0.20; advice future conduct -0.19  ||  training +0.12; program activity +0.15; written reprimand +0.79
  measures, k=2: suspension, written reprimand, training, advice future conduct, verbal reprimand, close supervision, counselling, program activity, reduction in rank, transfer reassignment | dismissal
  measures, k=3: suspension, written reprimand, training, advice future conduct, close supervision, counselling, program activity, reduction in rank, transfer reassignment | dismissal | verbal reprimand
  measures, k=4: suspension, training, advice future conduct, close supervision, counselling, program activity, reduction in rank, transfer reassignment | written reprimand | dismissal | verbal reprimand
======================================================================== 
3. MCA OF ALLEGATIONS (5 active variables; GBV projected on)
  outcome levels: {'lighter measure': 648, 'demoted / suspended': 410, 'fired': 177, 'no measure recorded': 8}
  raw inertia dims 1-3: [9.83568474 7.57586507 6.6004144 ]
  Benzecri-adjusted:    [64.1 19.7  8.4]
  GBV (92) supplementary point: [ 1.34 -0.11]
  strongest coordinates on dim 1:
                                     d1    d2
outcome: fired                     1.46  0.32
crim: criminal process             1.37 -0.39
conduct: Deceit / falsification    1.31  0.36
conduct: Gender-based violence     1.31 -0.08
conduct: Discourtesy              -1.05 -1.31
conduct: (story too thin)         -1.02  1.46
conduct: Unlawful arrest / search -0.85 -2.10
pathway: public complaint (PC)    -0.83 -1.45
======================================================================== 
4. RESPONSE REPERTOIRES (what was done, clustered; Jaccard)
  repertoire 1: n=   2 | fired 0% | GBV 0% | Msr_Suspension 0%, Msr_Written_Reprimand 0%, Msr_Training 0% | top conducts: Investigative neglect, Workplace harassment
  repertoire 2: n=   8 | fired 0% | GBV 12% | Msr_Counselling 100%, Msr_Suspension 12%, Msr_Program_Activity 12% | top conducts: (story too thin), Deceit / falsification
  repertoire 3: n=   2 | fired 0% | GBV 0% | Msr_Training 100%, Msr_Close_Supervision 100%, Msr_Program_Activity 100% | top conducts: Workplace harassment, Discourtesy
  repertoire 4: n= 152 | fired 0% | GBV 1% | Msr_Advice_Future_Conduct 100%, oi 44%, Msr_Training 11% | top conducts: Investigative neglect, Firearms
  repertoire 5: n=1079 | fired 16% | GBV 8% | oi 69%, Msr_Suspension 35%, Msr_Written_Reprimand 30% | top conducts: Off-duty conduct, Deceit / falsification
  GBV allocation across repertoires:  {2: 0.01, 4: 0.01, 5: 0.98}
  rest allocation across repertoires: {1: 0.0, 2: 0.01, 3: 0.0, 4: 0.13, 5: 0.86}
======================================================================== 
5. LATENT CLASS ANALYSIS (Bernoulli mixture, EM; same 15 response items)
  k=2: loglik     -6159  BIC     12539
  k=3: loglik     -5835  BIC     12004
  k=4: loglik     -5614  BIC     11678
  k=5: loglik     -5501  BIC     11565
  k=6: loglik     -5424  BIC     11525
  k=7: loglik     -5376  BIC     11543
  BIC chooses k=6; average classification certainty 0.96
  class 1: n= 177 (14%) | fired 100% | GBV 18% of class | dismissal 99%, oi 85%, criminal 45% | top conducts: Deceit / falsification, Gender-based violence
  class 2: n= 323 (26%) | fired 0% | GBV 11% of class | suspension 100%, oi 85%, had adjudicative review 22% | top conducts: Off-duty conduct, Corruption
  class 3: n= 149 (12%) | fired 0% | GBV 10% of class | training 60%, oi 50%, suspension 40% | top conducts: Workplace harassment, Gender-based violence
  class 4: n= 317 (26%) | fired 0% | GBV 3% of class | written reprimand 99%, oi 57%, training 19% | top conducts: Information misuse, Workplace harassment
  class 5: n= 133 (11%) | fired 0% | GBV 1% of class | verbal reprimand 99%, oi 50%, training 15% | top conducts: Investigative neglect, Firearms
  class 6: n= 144 (12%) | fired 0% | GBV 0% of class | advice future conduct 99%, oi 46%, had adjudicative review 8% | top conducts: Investigative neglect, Firearms
  GBV allocation across classes:  {1: 0.35, 2: 0.38, 3: 0.16, 4: 0.1, 5: 0.01}
  rest allocation across classes: {1: 0.13, 2: 0.25, 3: 0.12, 4: 0.27, 5: 0.11, 6: 0.13}
  -- page table (allocation, both columns sum to 100%) --
  class 1: GBV 32/92 = 35%  | rest  145/1151 = 13%
  class 2: GBV 35/92 = 38%  | rest  288/1151 = 25%
  class 3: GBV 15/92 = 16%  | rest  134/1151 = 12%
  class 4: GBV  9/92 = 10%  | rest  308/1151 = 27%
  class 5: GBV  1/92 = 1%  | rest  132/1151 = 11%
  class 6: GBV  0/92 = 0%  | rest  144/1151 = 13%
  heaviest two (1+2): GBV 73% vs rest 38% | lightest two (5+6): GBV 1% vs rest 24%
  chi2 on the 2x6 allocation table: p=6.1e-13
  -- class item profiles (items at 25%+ of the class) --
  class 1: dismissal 99%, oi 85%, criminal 45%
  class 2: suspension 100%, oi 85%
  class 3: training 60%, oi 50%, suspension 40%
  class 4: written reprimand 99%, oi 57%
  class 5: verbal reprimand 99%, oi 50%
  class 6: advice future conduct 99%, oi 46%
Done. Figures in exploratory/.
