# Review-length filtering by store and star rating — methodology

Snapshot queried September 5, 2026. Publisher: BigIdeasDB. These files contain aggregate research results only, with no customer or reviewer identities, raw review text, or internal row identifiers. Exact counts are supplied for reproducibility; article prose uses rounded counts. Percentages are calculated from exact counts. This is an observational convenience corpus, not a representative sample of the entire market. The query date is not the date of every underlying observation. No trend, causality, purchase intent, or current vendor capability is established by these tables.

## Population and computation
All 136,923 stored mobile review records at query time, without a date restriction. Join each review to its app profile to obtain store type; absent store information becomes unknown. All scores were integers 1–5 and all texts were nonempty. Measure PostgreSQL character length of text after trimming surrounding whitespace. Group by store and score; calculate record count, nonempty-text count, median text character length, count at least 200 characters, and distinct linked apps. Retained percent = count at least 200 characters / records * 100. The cutoff is a hypothetical analysis filter, not a recommendation or quality score.

## Coverage audit and exclusions
125,398 records have an app link and appear in the Apple App Store or Google Play Store buckets. Another 11,525 records have no app link: retain them as unknown in the CSV but exclude them from store-specific comparisons. Source-review references are populated; there were zero repeated app-plus-source-review groups among linked records. This does not rule out copied text or the same person reviewing several apps.

68,643 records have no review date. Among dated records, dates span July 12, 2011 to April 26, 2026; none are after the query date. This is a historical corpus, not a September 2026 review sample. Selection and ingestion can influence text-length distributions. Do not infer that one store or star group is inherently more informative, or that these results generalize to all reviews. Text length is not a measure of truth or usefulness.

## Key calculation
Google Play one-star: 13,384 / 31,718 = 42.20% retained. Five-star: 11,295 / 13,495 = 83.70% retained. A 200-character cutoff changes the composition of this stored corpus, excluding a greater share of its one-star records. Other cutoffs may produce different results.

## CSV fields
store and score: grouping labels; records: all records in bucket; with_text: nonempty trimmed text; median_text_characters: median of nonempty text lengths; text_at_least_200_characters: retained records; apps: distinct linked app profiles (not additive across ratings); retained_percent: retained/records*100; snapshot_date: query date. Unknown-store rows have zero linked apps.

## Source page
https://bigideasdb.com/customer-review-analysis#original-research