Equity Review & Disparity Monitoring (Pipeline Analytics)

Pipeline 07A

Transparency note: equity metrics can be distorted by small cell sizes or definitional changes. When rate ratios are based on fewer than 20 cases, the values are pooled or suppressed, and confidence intervals are displayed in source dashboards.

Overview

Disparity monitoring identifies unequal outcomes across demographic or geographic groups at each stage of the youth justice pipeline. It is not only a statistical exercise but also a governance function—helping stakeholders assess fairness, consistency, and the impact of policy changes. This page outlines how such analyses are structured and interpreted responsibly.

What We Track

Rate Ratios

Relative likelihood of a group reaching a stage vs. reference group
Normalization choice documented:
- Population-based denominator
- Prior-stage denominator (conditional rate)

Comparative Frames

Cohort definitions (intake, referral, adjudication, release cohorts)
Trend intervals (monthly/quarterly/annual) and smoothing rules

Control Variables

Age banding and offense severity normalization
Prior history controls (referrals, adjudications, placements)
Geography (urban/rural, site fixed effects)

Visualization & Communication

Dashboards using:
- Slope charts (entry → outcome)
- Parity bands or index values (reference = 1.00 or 100)
- Interpretive annotations and caveats

Transparency Practices

Publish methodology docs and reference group choice
Display confidence intervals or small-n flags; apply suppression rules
Mark series breaks when coding/thresholds change

Typical Flow

Define population or cohort for comparison
- Specify denominator: population vs. prior-stage; set reference group
Compute stage-specific rates by group & reference
- Apply controls (age, severity, prior history, geography) as appropriate
Calculate disparity ratios or index values
- Show confidence intervals; flag small-n cells for caution/suppression
Review trend stability and mark series breaks
- Annotate coding or threshold changes; avoid false trend narratives
Publish findings with narrative context & transparency
- Use parity bands/slope charts; include methods, CIs, and suppression thresholds

Fields

Field	Type	Required	Codeset	Description
pipeline_place_id	uuid	✅	—	Unique identifier for this analytics record.
pipeline_stage_id	enum	✅	stages.yml#stage_key(8) intake detention adjudication disposition supervision commitment_and_placement reentry closure	One of the canonical stages (here: violation_response).
pipeline_place_key	enum	✅	pipeline_places.yml#place_key(45) intake pre_petition_diversion_and_deflection diversion youth_assessment_tools prosecutorial_screening status_offenses adjudication case_planning family_engagement case_timeliness language_access restorative_justice detention_screening community_atds electronic_monitoring court_appearance_and_fta initial_hearing pretrial_supervision probation intensive_supervision …and 25 more	Canonical key for this place (maps to route/slug).
youth_id	string		—	Rarely used; only if record is filtered to a specific individual (normally null).
staff_id	string		—	Analyst or approving staff (optional).
occurred_datetime	datetime	✅	—	Timestamp the review results were finalized.
jurisdiction_code	string	✅	—	County/parish/circuit or standardized local code.
legal_case_id	string		—	Not typical for analytics records; included for completeness.
source_system	string	✅	—	Origin system name (analytics pipeline or BI export).
source_file	string		—	Batch/file id if exported.
extract_run_id	string		—	ETL/analytics run id for lineage.
series_break_flag	boolean		—	Comparability break (coding/threshold/tool change).
series_break_reason	enum		series_breaks.yml#reason(4) definition_change vendor_change coverage_change policy_change	Reason for break when flagged.
cohort_key	enum	✅	cohort_types.yml#cohort(5) intake referral adjudication release stage_transit	Intake, referral, adjudication, release, or stage-transit cohort.
denominator_type_code	enum	✅	denominator_types.yml#type(2) population_based prior_stage	population_based or prior_stage (conditional).
analyzed_stage_key	enum	✅	stages.yml#stage_key(8) intake detention adjudication disposition supervision commitment_and_placement reentry closure	Stage being analyzed (here typically violation_response).
group_dimension_code	enum	✅	group_dimensions.yml#dimension(4) race_ethnicity gender geography rurality	Dimension being compared (race_ethnicity, gender, geography, rurality).
group_value_code	string	✅	—	Specific group value (e.g., black, hispanic_latino, female, rural, specific county code).
reference_group_code	string	✅	—	Reference group used for parity (e.g., white, male, metro).
period_start_date	date	✅	—	Start of the interval (month/quarter/year start per trend frame).
period_end_date	date	✅	—	End of the interval.
trend_interval_code	enum		❌ missing in _index.yml	monthly, quarterly, annual.
smoothing_method_code	enum		smoothing_methods.yml#method(5) none rolling_3 rolling_4 loess other	none, rolling_3, rolling_4, loess, other.
control_age_band_code	enum		age_bands.yml#band⚠️ using age_bands.yml#severity(6) status misdemeanor felony_low felony_high violent unknown	Age banding schema used (if controlled).
control_offense_severity_code	enum		❌ missing in _index.yml	Severity scheme used for normalization.
control_history_codes	array<string>		history_controls.yml#control(3) prior_referral prior_adjudication prior_placement	Prior history controls (referrals, adjudications, placements); semicolon-delimited.
geography_control_code	enum		geography_controls.yml#geo(2) site_fixed_effects urban_rural	Site fixed effects or urban_rural normalization.
reference_choice_note	string		—	Short note capturing reference-group rationale.
ci_method_code	enum		ci_methods.yml#method(5) wald wilson exact bootstrap other	Method used to compute confidence intervals (wald, wilson, exact, bootstrap, other).
suppression_policy_code	enum		❌ missing in _index.yml	Policy for small-n suppression/flagging.
small_n_flag	boolean		—	True if the cell is flagged for small-n caution.
suppressed_flag	boolean		—	True if the cell is suppressed per policy.
rate_numerator	integer		—	Count in stage for the group (for the selected frame).
rate_denominator	integer		—	Denominator count (population or prior-stage cohort size).
rate_value	number		—	Stage rate for the group (proportion/percentage as raw number).
reference_rate_value	number		—	Stage rate for the reference group.
rate_ratio_value	number		—	Group rate ÷ reference rate.
parity_index_value	number		—	Index scale value (e.g., 1.00 parity = 100 if using 100-based index).
parity_index_scale_code	enum		parity_scales.yml#scale(2) scale_1x scale_100x	scale_1x or scale_100x.
ci_low	number		—	Lower bound of confidence interval for rate_ratio or index.
ci_high	number		—	Upper bound of confidence interval for rate_ratio or index.
visualization_frame_code	enum		viz_frames.yml#frame(4) slope_chart parity_band index_trend small_multiples	slope_chart, parity_band, index_trend, small_multiples.
annotation_note_id	string		—	Reference to narrative/annotation stored elsewhere.
Download CSVwhat_we_track.csv

Data & Methods

Disparity metrics rely on standardized denominators and consistent classification fields (race, ethnicity, gender, geography). Cross-stage rate ratios are calculated using the relative rate index (RRI) or comparable parity measures. When definitions or population bases shift, we mark series breaks. Statistical significance is assessed via bootstrapped confidence intervals or Monte Carlo methods for small cells. Suppression and disclosure controls follow Suppression & Small-n. For analytic consistency, see Trend Interpretation & Baseline Selection and Cross-Jurisdiction Comparability.

EDORA • Learn