Cross Validated
2026-11-16 11:38 UTC
By JonasSoe
AI-113-20261116-social-media-57cddacc
How should I test whether monitoring dates are representative of daily precipitation when samples are clustered by site and date?
I have daily precipitation on a 10 × 10 km grid from 2011-2025, and trips from a monitoring programme. Each trip is linked to a grid cell and date, and precipitation is taken from that cell-day. I want to know whether monitoring dates are underrepresented on days with heavy rain (≥10 mm), and on the 0–3 and 0–5 days after heavy rain. Sampling is routine, but dates are flexible within calendar months. The data contain >100,000 trips, >75,000 unique cell-days, >3,500 sampling dates, roughly 500 cells and 2.7 million eligible cell-days. Multiple trips can occur in the same cell-day, and many cells are sampled on the same date. My current approach collapses to one row per cell-day, stratifies by cell × year × month, compares the observed share of heavy or after-heavy days with the reference share among all days in the same stratum, and uses a two-sided date-block permutation test. Within each year-month, it randomly assigns eligible dates to the observed sampling-date blocks, preserving which cells were sampled on each date. I use 10,000 permutations and Holm adjustment across the three exposures. My concern is that this test conditions on the observed number of sampling dates per month and the cells sampled on each date. It therefore tests whether the chosen dates are weather-representative, but not whether the number of trips or cells sampled per day is weather-dependent. There is also spatial correlation: cells sampled on the same date share the same weather system. My questi…
I have daily precipitation on a 10 × 10 km grid from 2011-2025, and trips from a monitoring programme. Each trip is linked to a grid cell and date, and precipitation is taken from that cell-day. I want to know whether monitoring dates are underrepresented on days with heavy rain (≥10 mm), and on the 0–3 and 0–5 days after heavy rain. Sampling is routine, but dates are flexible within calendar months. The data contain >100,000 trips, >75,000 unique cell-days, >3,500 sampling dates, roughly 500 cells and 2.7 million eligible cell-days. Multiple trips can occur in the same cell-day, and many cells are sampled on the same date. My current approach collapses to one row per cell-day, stratifies by cell × year × month, compares the observed share of heavy or after-heavy days with the reference share among all days in the same stratum, and uses a two-sided date-block permutation test. Within each year-month, it randomly assigns eligible dates to the observed sampling-date blocks, preserving which cells were sampled on each date. I use 10,000 permutations and Holm adjustment across the three exposures. My concern is that this test conditions on the observed number of sampling dates per month and the cells sampled on each date. It therefore tests whether the chosen dates are weather-representative, but not whether the number of trips or cells sampled per day is weather-dependent. There is also spatial correlation: cells sampled on the same date share the same weather system. My questi…
Full article content could not be extracted automatically. Read the original below.
Source:
Cross Validated
· stats.stackexchange.com