Methodology
How Habitat Studies sources, treats and calculates the metrics published inside the Dubai Real Estate Workspace.
Data source
Habitat Studies relies on transaction records sourced from the Dubai Land Department (DLD) monthly release. Files are externally cleaned before ingestion.
Data coverage
Coverage begins January 2025 onward. Every published month reflects transactions registered with DLD during that period.
Update frequency
Datasets are refreshed on a monthly cadence. The platform stores each release as an independent, replaceable batch.
Dataset treatment
The platform does not clean, deduplicate or silently modify uploaded records. Fields arrive as provided by the external cleaning pipeline and are stored verbatim.
PPSF methodology
PPSF (Price Per Square Foot) uses the value supplied in the source dataset. Missing or non-positive values are excluded from PPSF aggregations.
Median vs average
Median values are computed across the complete set of filtered transaction records. Monthly medians are never averaged together — each KPI recomputes from the underlying rows.
Off-plan classification
Off-plan / Ready classification uses the IS_OFFPLAN_EN field as delivered by DLD.
Freehold classification
Freehold / Non-freehold classification uses the IS_FREE_HOLD_EN field as delivered by DLD.
Known limitations
Some transactions may have blank developer or master project references. These records are retained but may not appear in developer- or project-filtered views.
Area Master (STANDARD_AREA)
All user-facing area outputs use STANDARD_AREA derived from the Area Master mapping. Transactions retain the raw AREA_EN verbatim as delivered by DLD. Join: Transactions.AREA_EN → Area_Master.RAW_AREA → Area_Master.STANDARD_AREA. Fallback: if AREA_EN has no master entry, STANDARD_AREA = AREA_EN.
Habitat Studies Peer Markets (V1)
Each area's feature vector has 9 z-standardized dimensions with fixed weights: Median PPSF (30%), Median Ticket (20%), Off-Plan Share (15%), log Transaction Volume (15%), and 5 classified unit-mix shares — Studio, 1 BR, 2 BR, 3 BR, 4 BR+ (20% split evenly, 4% each). The 'Other' bucket is excluded from the feature vector so incomplete classification does not distort similarity. Distance: weighted Euclidean in z-space (weights sum to 1.00). Similarity score (fixed global calibration): score = round(100 · exp(−d² / 2)). Reference points — d 0.00→100, 0.25→97, 0.50→88, 0.75→76, 1.00→61, 1.25→46, 1.50→32, 2.00→14, 3.00→1. 100 is reserved for effectively identical feature vectors. Eligibility rules (strict, no mean imputation): the area must (1) have ≥20 transactions in the selected period, (2) present all four core metrics (median_ppsf, median_ticket, offplan_share, transactions), and (3) have unit-mix completeness ≥30% (share of transactions classified into a room category). Ineligible areas are dropped entirely from the peer pool. The current area is always excluded from its own peer list. The peer_average shown in the Area Benchmark is computed over the top-3 similar peers only. This is a data-similarity model — not an investment recommendation.
Rental Data Management — R2 Analytical Foundation (v1.0.0-2026-02-16)
Rental analytics are deterministic, read-time aggregations computed directly over the frozen rental_transactions store. No materialised views, no caches, no AI estimation. Universe · Residential is the default published universe. Commercial is a strict opt-in view. All other USAGE_EN values (Industrial, Storage, Educational, Health, blank) are excluded from analytics by design. Source integrity · Raw DLD records are preserved verbatim. There is no source deduplication, no heuristic value trimming, and no outlier removal. All registered contracts remain eligible for contract-volume counts, including multi-property leases (TOTAL_PROPERTIES > 1). Financial driver · ANNUAL_AMOUNT is the sole rental financial driver. All price and rate metrics are medians only — no means, no trimmed means, no imputation. Per-unit gate · Per-unit metrics (median annual rent, median unit size, median rent per ft²) apply TOTAL_PROPERTIES = 1 exclusively. Contract-volume counts include TOTAL_PROPERTIES > 1; per-unit samples do not. All / New / Renewed · VERSION_EN drives the New vs Renewed split. All, New and Renewed are independently queryable; New + Renewed = All exactly, because these are the only two values in the source. Geography · Only canonical Areas resolved from the approved Area Master appear in analytical outputs. Rentals whose AREA_EN does not resolve to an approved standard_area are silently excluded and surfaced in records_dropped_non_canonical. This is enforced entirely at read time — the R1 rental_transactions rows are never mutated. Thresholds · Area-level medians are suppressed below n = 20 contracts; rent-per-ft² medians require n = 30. Rent-per-ft² is computed as ANNUAL_AMOUNT / (ACTUAL_AREA × 10.7639) and is reported only for Flat and Villa sub-types. Sales isolation · Rental and Sales pipelines are strictly independent. Rental analytics never read from `transactions`; Sales analytics never read from `rental_transactions`. Coverage endpoints are reported separately per dataset_family. Active publication · Only rows with is_active = True in the currently published batch are counted. Superseded, replaced, archived and validation-failed batches contribute nothing. Reference edition — 2026-06 · 50,499 active rows · 42,391 Residential · 7,996 Commercial · 99.27 % canonical Area coverage (post-governance 2026-02-16, 10 Rental Areas approved into canonical Area Master) · 88.57 % single-unit contracts.
Last methodology update
This methodology page reflects Phase 1 of the Dubai Real Estate Workspace, extended with the Area Intelligence Sprint (Phase 1-6) and the Rental Data Management R2 Analytical Foundation.