## Data-quality findings for `raw.Website_Analytics` The data is structurally clean, but it has several real quality problems, mostly impossible or implausible traffic values. ### Impossible or inconsistent values - **Entrances exceed page views in 24,648 of 62,496 rows (about 39%).** An entrance is a page view that starts a session, so entrances can never exceed page views. The share is highest on: - brgov.com: 9,211 of 17,595 rows - www.brla.gov: 11,830 of 24,597 rows - city.brla.gov: 243 of 404 rows - gis.brla.gov: 87 of 147 rows - my.brla.gov: 64 of 191 rows The maximum entrances (1,680,247) on gis.brla.gov is far above its maximum page views (110,198). That points to a metric-definition or aggregation mismatch, such as entrances and views counted over different scopes or periods, rather than random noise. Treat entrances as unreliable relative to page views. - **Zero page views in 6 rows** (budget.brla.gov: 4, my.brla.gov: 2). Rows for pages that exist but received no views are odd in a traffic log. - **Zero average time on page in 7,319 rows (about 12%).** This is probably "no measurable time" (for example single-page sessions) recorded as 0 rather than null, so it will drag down averages. ### Extreme outliers - **city.brla.gov has a maximum of 228,326,194 page views** and 227,064,076 entrances in one row. Its 404 rows have a median of 2,088, and the overall median is 8. The overall 99th percentile is about 64,542, so this one row is roughly 3,500 times the 99th percentile. It is almost certainly a root or aggregate row, or a tracking artifact. - **www.brla.gov's maximum of 17.6M page views** against a median of 19 is also extreme. - **Heavy skew everywhere.** The overall median is only 8 views, and per-site medians range from 2 to 56,615. Means and sums will be dominated by a handful of rows. - **Maximum time on page is 2,794 seconds (about 47 minutes).** This is plausible for an idle open tab but probably an outlier. ### Duplicates and format issues - **21 keys on (WEBSITE, YEAR, PAGE URL) repeat, covering 42 rows (max 2 per key).** This breaks the stated grain of one row per website, year and page. - **Format issues on www.brla.gov:** two rows have a whitespace problem and one has a non-http URL. These checks were flagged only on that site, and I did not inspect the actual values. - **The site list is mixed.** The 11 sites include an external ArcGIS host (data.ebrgis.opendata.arcgis.com), and the tiny sites (checkbook.brla.gov has 3 rows, 311.brla.gov has 35) have too few rows for stable statistics. ### What checked out clean - There are no negative values, no unique views above page views, and no negative times. - There are no nulls in page path, URL, bounce rate or exit rate. - No bounce or exit rates were flagged as out of range, so I found no sign of mixed percentage scales. - Years run 2014–2024 with no gaps in range. ### Suggested handling 1. Exclude or flag rows where entrances exceed page views before computing bounce or entry-page analyses. 2. Cap or exclude the city.brla.gov extreme row, and use medians or percentiles instead of sums. 3. Treat zero time on page as missing. 4. Deduplicate the 21 repeated keys. 5. Trim whitespace and normalize URLs. The supporting queries are the profile checks behind steps 0–3 above: the overall profile, the per-site breakdown, the duplicate-key check, and the per-site entrances-versus-views comparison.