How we remove outliers from Amaravati asking-price comparisons

What the research shows

• Property Desk calculation: 687 normalized property records become 627 pricing observations after 60 documented exclusions. The original inventory records remain unchanged.

  [1]

• Property Desk calculation: The statistical screen removes 42 records outside a use-specific 1.5-IQR boundary for total asking amount, advertised area or asking rate. It is applied once after the area-quality screen.

  [1]

• Property Desk calculation: The median asking rate is ₹42,000 per sq yd before the area/outlier screens and ₹42,000 after them; the median total ask changes from 148.5 Lacs to 140.4 Lacs.

  [1]

Property Desk research context

Amaravati Property Desk analysed advertised asking prices captured on 27 August 2026. The figures describe that snapshot and are not completed-sale prices.

Realistic records still need a defined comparison group

The dataset contains property advertisements with captured asking amounts and source context. A very large plot or a high-total offer can be a legitimate property and still sit outside a comparison intended to describe the more typical advertised size and price mix. An outlier flag is a statistical selection decision, not a finding that the property or advertisement is false.

The exclusions happen in a fixed order

• Start with 687 normalized records representing 922 source ads.
• Exclude 4 source exceptions: the imported code/type exceptions and the record carrying a cancellation remark.
• Exclude 14 unresolved advertised-versus-recorded area differences above 1% of advertised area.
• Calculate the outlier boundaries on the remaining 669 records, separately for advertised residential and commercial use.
• Remove the 42 records outside any boundary, leaving 627: 400 residential and 227 commercial.

What the 1.5-IQR rule means

For each advertised use, we calculate the 25th percentile, Q1, and 75th percentile, Q3. Their difference is the interquartile range, or IQR. A value below Q1 minus 1.5 times IQR, or above Q3 plus 1.5 times IQR, is outside the comparison. Percentiles use linear interpolation at position (record count minus one) times the percentile. Values exactly on a boundary remain included.

We apply the same rule to three quantities: the total asking amount, the advertised area and the asking rate calculated from that area. A record outside any one boundary is excluded once. The boundaries are not repeatedly recomputed to keep removing more records, and no values are replaced with a boundary value.

The actual boundaries used for this capture

• Residential: total asking amount above 588 Lacs, area above 1,650 sq yd, or unit rate outside ₹13,000–₹69,000 per sq yd.
• Commercial: total asking amount above 444.8 Lacs, area above 900 sq yd, or unit rate outside ₹20,500–₹72,500 per sq yd.

The calculated lower total-price and area boundaries are negative, so no positive record is excluded at those lower boundaries. In this capture, the statistical removals arise from the total-price and area tails; no record is removed for unit rate alone. The method therefore does not discard a quotation simply because it is the cheapest or most expensive rate in its village.

What changes, and what remains outside the conclusion

The overall median rate remains ₹42,000 per sq yd after these screens. The median total moves from 148.5 Lacs to 140.4 Lacs. Reporting both the rule and its effect makes the choice visible. The filtered figures describe the retained cohort; they should not be used to value a large plot that was excluded by design.

Every retained property has equal weight. Locality labels are not merged without evidence, budget ceilings are inclusive, and medians for groups below five records are withheld from the articles. Prices are asking amounts from one captured source, not registered sales or a historical price series. This method improves comparability without converting an advertisement into verified title, approval or availability.

How to read these figures

Unless a paragraph explicitly describes the original source or an excluded example, pricing and budget figures use 627 retained records: 400 residential and 227 commercial. We excluded 4 source exceptions, 14 unresolved area differences above 1% and 42 outliers using one-pass 1.5-IQR boundaries within advertised use on total ask, area and unit rate. Original records remain intact. Each retained property has equal weight; budget ceilings are inclusive; localities retain the source labels; group price medians require at least five records. Totals shown in Lacs and Cr are rounded to at most two decimals.

The import keeps the lowest positive advertised amount for each exact-code group. A median is the middle observation; the middle-half range runs from the 25th to 75th percentile. The figures describe this filtered, single-source capture. Current seller availability, authority, title, boundaries, approval and access require separate checks.

Archived research examples: 27 August 2026

These sixteen examples are transcribed from the retained research snapshot used for these articles. References identify records in that dated research set, not current listings, current owners or selling availability. The original 687-record snapshot has SHA-256 d846265e0a46e24288118c594b89a8db0ce19e1e6e24badabf34cfc099638d87. Its public calculation summary is linked in the sources. No owner names, telephone numbers or identity documents are included.