Interpreting Clinic Performance Averages with Missing and Skewed Data
The health ministry's performance dashboard reports an average outpatient waiting time of 42 minutes across 120 rural clinics, and an average bed occupancy of 68%. The minister has cited these figures to claim that the system is 'broadly adequate.' A closer review finds that the waiting-time median is 28 minutes, but 15 clinics report waits above 95 minutes; eight clinics submitted zero minutes for every month, which staff say means 'data not collected.' Bed occupancy ranges from 31% to 104%, and three clinics with the highest occupancy serve districts with rapidly growing populations. The dashboard target is 90% of patients seen within 30 minutes. The ministry must decide which clinics receive a 60-million equipment fund and whether to publish a revised performance rating. You are the statistics adviser asked to assess the dashboard and recommend allocation criteria.
1. What is wrong with using the dashboard averages, and what data-quality actions are needed?
Read the worked answer
The dashboard mixes mean and median and treats missing data as zero. A mean wait of 42 minutes is higher than the median of 28, which suggests a long right tail: 15 clinics above 95 minutes pull the average up. Eight clinics reporting zero every month is not a real zero; it is missing data. Including those zeros biases the average downward, while the high outliers bias it upward; the direction depends on counts. Bed occupancy of 68% hides a range of 31% to 104%; occupancy above 100% means bed counts or reporting are wrong, or patients are being placed outside licensed beds. A single national average cannot support a claim that the system is 'broadly adequate'. Data actions: validate the eight zero-reporting clinics; audit the 15 high-wait clinics; reconcile bed capacity and admissions; publish medians, 90th percentiles, and the proportion meeting the 30-minute target; show clinic-level data with denominators and missing-data flags.
2. Recommend allocation criteria for the 60-million equipment fund and decide whether to publish a revised rating now.
Read the worked answer
Allocation of the 60 million should be rule-based and separate data quality from performance. Reserve 8-10 million for data systems, training and audit in the eight zero-reporting clinics. Allocate the remainder using a composite: validated waiting-time performance (not raw mean), occupancy pressure, population growth, distance and access, and readiness to use equipment. Give priority to clinics with high validated waits and high occupancy in growing districts; do not reward missing data with equipment, but fund their data systems. Require maintenance and staffing plans. Publish provisional ratings with caveats, not a final star rating, until the data are validated. If a rating is published now, label it provisional, show data-quality flags, and correct the minister's broad claim. The revised rating should use the target measure: percentage of patients seen within 30 minutes, plus median wait and occupancy.