Issue Intelligence

Every flag root-caused to a layer · A1 questionnaire · A2 field · B firm do-file · C our check · D structural

  

Data Quality Overview

L2PHL "Listening to the Philippines" Panel Survey · Rounds 1–N · Questionnaire-grounded checks

New in v3: Each module now has a deep-dive page showing every question, its presence across rounds, questionnaire changes (new/dropped/renamed/skip-logic), and data quality issues linked to the question. Click any module card or the module links in the sidebar to explore.

🚦 RAG Status Criteria

Module-level status (cards below)
🔴 Red Skip violations > 100 or mandatory missing > 100 or worst variable ≥ 30% missing
🟡 Yellow Any skip/mandatory/OOR violations or worst variable ≥ 10% missing
🟢 Green Zero violations and all variables < 10% missing
Variable-level status (heatmap cells)
🔴 Red ≥ 15% missing in any round
🟡 Yellow ≥ 5% missing in any round
🟢 Green < 5% missing in all rounds

Module max-missing uses only unconditional variables (not gated sub-items with small denominators). Skip & mandatory counts are summed across all rounds.

Module-Level Quality Summary Click a card to view details

Per-round strip: ! open firm issue   open  · closed   clean

Sample Counts per Round

Questionnaire Change Summary by Round

Panel Structure & Household Tracking

Participation across rounds · Attrition bias · PSU coverage vs targets · In / out households

Sample Funnel — Cumulative Attrition

R1 baseline HHs tracked through all rounds (blue = retained from R1, orange = new entries)

Retention Rate & Effective Sample

% of previous-round HHs retained. Dashed line = call interval violation rate.

▼ 🔬 Attrition Bias Analysis

Round-to-round selection bias: for each transition (R1→R2, R2→R3, …) are households that stay systematically different from those that drop out? Characteristics compared using data from the previous round.

Regional Retention Rate by Transition (R1→R2)
Click a transition tab above to update — sorted lowest to highest
Sample Composition Drift
Each region's share of total sample: R1 vs latest complete round (pp change)
Retention Rate Heatmap — Region × Transition

% of previous-round HHs retained per transition, by region. Red <50%, orange 50–65%, yellow 65–75%, green ≥75%.

Households per Round
Retained from R1 vs new entries
Participation Pattern Distribution
Top patterns (1=present, 0=absent per round)

▼ 🏠 Household Panel Tracker Full HH participation matrix

Every household across all rounds — when they joined, when they left, and their full participation history. Green = interviewed, Red = absent.

▼ 📍 PSU Coverage vs Targets Urban: 6 HH/PSU  |  Rural: 7 HH/PSU

▼ 🗺️ Household Counts by Region, Urban/Rural & Round

▼ 🔄 Round-by-Round In / Out Summary

Tracked relative to R1 baseline. "New" = households not seen in R1.

▼ 🔁 Leavers vs. New Entries — Are Replacements Representative?

For each transition, compares households that left (using their last-observed characteristics) against households that entered (using first-observed characteristics). If replacements differ from leavers, the survey composition is shifting.

% Urban: Leavers vs New Entries
Are new entries more/less urban than those who left?
Mean HH Size: Leavers vs New Entries
Are new households larger/smaller?
Regional Composition: Leavers vs New Entries

Each region's share of leavers vs share of new entries per transition. Large differences indicate regional replacement bias.

▼ 👥 Attrition Composition — Who Stays, Who Leaves, Who Is New?

Urban/Rural split of retained, dropped, and new-entry households per round transition.

Urban/Rural Mix by Group
% Urban among retained vs dropped vs new (each transition)
Volume by Group per Transition
Absolute HH counts

▼ ⚠️ PSU Problem Tracker Under-Target Only

PSUs that are below their household target in at least one round. Sorted by most rounds under target. Over-target is not flagged.

▼ 🔗 Within-PSU Refusal Clustering Risk Field Action Required

Cross-reference of under-target PSUs with round-to-round attrition rates by region. Regions flagged HIGH show both elevated attrition AND PSU coverage problems — two failure modes that compound each other and cannot both be fixed by weighting alone.

▼ 📅 Call Interval Tracker

Days between consecutive interviews for the same household. Minimum required interval is 20 days. Intervals of 20–29 days are permissible with supervisor approval; anything under 20 days is flagged as a violation.

Households Called Too Early (<20 days since last interview)

Operator Performance QC

Performance metrics per interviewer (int_id), derived from M00 passport and cross-module checks. Flags are independent indicators — a red operator should be reviewed, not automatically penalised. Short-circuit rates show the % of interviews where a skip-heavy gate answer was recorded (M03 SH1=2 "no shock", M04 A1=2 "not employed", M06 F1=2 "no bank account") compared with the fleet average. Unusually high rates may indicate that the operator is guiding respondents toward shorter-survey answers.

▼ 📊 Fleet Analytics Charts Volume, flags, duration, scatter

Interviews & Operators by Round

Flag Distribution

Duration vs Skip Violations

Each bubble = one operator. Size = interview count. Color = RAG status.

Module Duration — Fleet Distribution

Mean duration per module across all operators (fleet P10–P90 range).

▼ 📋 Operator Table & Detail

Filter:

All Questionnaire Changes by Round

Every new question, dropped question, wording change, skip-logic change, and code-list change across R2–RN

Changes are detected by comparing questionnaire workbooks for each consecutive round pair. The "presence" column shows which rounds include each variable.