Numeric verification checks (96)
| Paper | ID | Status | Scope | Check | Computation | Result |
|---|---|---|---|---|---|---|
| Gil (2019) | V1 | pass | internal | Each city column in Tabla 1 sums to 100%. p. 17 |
Madrid 57.63+14.56+10.25+17.56; Barcelona 49.95+21.73+13.69+14.63; NY 31.77+17.84+14.98+35.41; Amsterdam 32.38+23.00+19.04+25.58; Paris 30.50+17.33+16.06+36.11; London 35.73+21.58+14.92+27.77 | All six columns sum to exactly 100.00%. |
| Gil (2019) | V2 | minor discrepancy | internal | Share of hosts responding within 24 hours (cumulative) vs the claim that response times are 'very short'. p. 17 |
Sum of first three rows per city. | Madrid 82.44%, Barcelona 85.37%, New York 64.59%, Amsterdam 74.42%, Paris 63.89%, London 72.23%. One-hour response is the modal category only in Madrid and Barcelona; in NY and Paris 'Otro' is the largest category. |
| Gil (2019) | V3 | not checkable | internal | Implied interview count from quote labels vs 'decenas'. pp. 7-16 |
Highest labels cited: Nueva York-15, San Francisco-16, Madrid-15, Barcelona-1. | Consistent with at least ~47 interviews, i.e. 'dozens'; exact N unknown. |
| Gil & Sequera (2020) | V1 | pass | internal | Central District share. p. 7 |
8,759 / 18,332 | 47.78% — matches. |
| Gil & Sequera (2020) | V2 | pass | internal | District-weighted reconstruction of city-wide shares. pp. 7-11 |
0.4778 x Central + 0.5222 x Rest for entire apts (76.3/51), commercial (80.07/71.64), multi-host (68.85/51.57) | 63.09%, 75.67%, 59.83% vs reported 63.0%, 75.7%, 59.8% — consistent. |
| Gil & Sequera (2020) | V3 | fail | internal | 'Nearly 10 points greater than the Madrid average'. p. 9 |
80.07 − 75.7 | 4.37 points (8.43 vs rest of city). |
| Gil & Sequera (2020) | V4 | pass | internal | Host-band shares sum and >20 band listings. pp. 12-13 |
74.50+21.19+3.65+0.66; 40.17+27.7+17.11+15.02; 2,753/18,332 | 100.00%; 100.00%; 15.02% — matches. Implied number of hosts ≈ 65/0.0066 ≈ 9,850. |
| Gil & Sequera (2020) | V5 | pass | internal | Band-level district and accommodation shares reproduce totals. p. 13 |
Σ band listing share x band central share; Σ band share x band entire share | Central = 47.78% (reported 47.78%); multi-host share of central = 68.85% (reported 68.85%); entire apts = 63.06% (reported 63.0%). |
| Gil & Sequera (2020) | V6 | pass | internal | 2017 comparison: point changes vs relative growth. p. 13 |
2.92/(17.11−2.92); 6.61/(15.02−6.61); 4.84/(40.17+4.84) | 20.58%, 78.6%, 10.75% — all match. Implied 2-5 band change: −4.69 points (not reported). |
| Gil & Sequera (2020) | V7 | pass | internal | 'Less than 5% of hosts control over 30% of listings'. p. 13 |
3.65+0.66; 17.11+15.02 | 4.31% of hosts, 32.13% of listings — matches. |
| Gil (2021) | V1 | fail | internal | Sample size after exclusions. p. 6 |
7,654 − 451 − 46 | = 7,157, but 7,203 is used throughout (7,203 = 7,654 − 451). |
| Gil (2021) | V2 | pass | internal | P2P vs commercial listing shares. p. 9 |
1,745/7,203; 5,458/7,203 | 24.23% / 75.77% — matches. |
| Gil (2021) | V3 | minor discrepancy | internal | Nights shares from reported counts. p. 10 |
133,831 / (133,831 + 1,096,993) | 10.87% p2p / 89.13% commercial; text on p. 10 reports 11.8% / 88.2% (conclusion p. 20 correctly uses 10.87%). |
| Gil (2021) | V4 | pass | internal | Accommodation types reconcile to total. p. 11 |
4,920 + 2,251 + 32 shared rooms | = 7,203; 68.30% / 31.25% / 0.44% — matches. |
| Gil (2021) | V5 | minor discrepancy | internal | Host-band shares. pp. 12-13 |
Hosts 2,986+1,048+118+15 = 4,167; shares 71.66 / 25.15 / 2.83 / 0.36% | Text gives 71.91% for single-listing hosts (bands then sum to 100.25%); the complement 28.34% is correct. |
| Gil (2021) | V6 | pass | internal | Listing-band shares and averages. pp. 12-13 |
2,986+2,534+1,023+660 = 7,203; 41.45 / 35.18 / 14.20 / 9.16%; means 2.42, 8.67, 44.0 | All match except 9.16% vs reported 9.14%. |
| Gil (2021) | V7 | fail | internal | 'Hosts with six or more listings hold almost a third of listings'. p. 12 |
(1,023 + 660) / 7,203 | 23.4% — closer to a quarter than a third. |
| Gil (2021) | V8 | pass | internal | District shares. p. 13 |
1,551/7,203; 1,233/7,203 | 21.53% and 17.12% (sum 38.65%) — matches. |
| Gil (2021) | V9 | pass | internal | Rent-gap annualisation. p. 15 |
€585 x 12 | €7,020 > €7,000 — consistent. |
| Gil (2021) | V10 | pass | internal | Revenue shares complement. p. 19 |
67.58% + 32.45% | 100.03% — rounding. |
| Gil (2021) | V11 | not checkable | internal | Implied Ciutat Vella population. p. 18 |
6,165 / 0.2277 | ≈ 27,075 residents — plausible for the district; not otherwise checkable. |
| Gil & Martínez (2021) | V1 | fail | internal | Table 2 totals. p. 10 |
Σ row differences; Σ price − Σ cost | Both give €17,654,238 vs reported €17,799,057 (−€144,819). |
| Gil & Martínez (2021) | V2 | fail | internal | Table 2 per-home total. p. 10 |
17,654,238 / 2,935 | €6,015 vs reported €5,394. |
| Gil & Martínez (2021) | V3 | minor discrepancy | internal | Table 2 implied price per home by municipality. p. 10 |
selling price / units per row | €68,391 in every municipality — uniform allocation, not observed prices. Row-level per-home differences are otherwise correct. |
| Gil & Martínez (2021) | V4 | pass | internal | Table 2 unit and cost totals. p. 10 |
Σ units; Σ cost; Σ price | 2,935; €183,072,193; €200,726,431 — match. |
| Gil & Martínez (2021) | V5 | pass | internal | Table 3 SOCIMI counts. p. 12 |
2+5+8+17+20+21+4 − 4 exclusions | 73 — matches cumulative column. |
| Gil & Martínez (2021) | V6 | pass | internal | Madrid social housing sold. p. 9 |
1,861 + 2,935 | 4,796 ≈ 'nearly 5,000' — consistent. |
| Gil & Martínez (2021) | V7 | pass | internal | NPL reduction. p. 14 |
1 − 61.5/197.2 | 68.8% ≈ 'nearly 70%' — consistent. |
| Martínez & Gil (2021) | V1 | minor discrepancy | internal | Average evictions 2009-2015 vs footnote repossession figures. p. 4; fn. 11 |
64,902 x 0.8 ≈ 51,900 home repossessions in 2015 (a peak year); 40,740 x 0.8 ≈ 32,600 in 2011 | An 80,000/year average would require other years far above the figures given; inconsistent unless a broader definition is used (not stated). |
| Martínez & Gil (2021) | V2 | fail | cross paper | Cross-corpus eviction totals. p. 4 |
80,000 x 7 (2009-2015) = 560,000 vs 409,000 for 2009-2018 (gil_martinez_2023_state_led, p. 11) | Same authors report incompatible magnitudes in different papers. |
| Martínez & Gil (2021) | V3 | pass | internal | Vacancy rate vs absolute numbers. p. 4 |
3.5M / 0.14 = 25M; 7.1M / 0.28 = 25.4M | Both imply a ~25M dwelling stock — internally consistent. |
| Martínez & Gil (2022) | V1 | fail | internal | Black-card spending plausibility. p. 10 |
€4,000,000,000 / 87 | ≈ €46 million per card holder — implausible for personal expense cards; likely an error. |
| Martínez & Gil (2022) | V2 | pass | internal | Raval agreement arithmetic. p. 13 |
6 regularised + 3 repossessed | = 9 apartments — consistent. |
| Martínez & Gil (2022) | V3 | not checkable | internal | Sareb average price per property. p. 6 |
€50.781bn / 352,000 | ≈ €144,000 per property; the paper says acquisitions were 'above market prices' — not checkable from the paper's data. |
| Martínez & Gil (2022) | V4 | minor discrepancy | cross paper | Cross-corpus consistency of post-2013 figures. p. 7 |
Compare with gil_martinez_2023_state_led | 73 SOCIMIs, house prices +31% (Madrid 56%, Catalonia 45%) identical; rents '+40% in some cities' here vs '+50%' there; 1,860 vs 1,861 homes. |
| Sequera et al. (2022) | V1 | not checkable | internal | Implied count of former tourist apartments. p. 3 |
1,796 x 0.7361 | ≈ 1,322 listings; consistent figure but derivation not documented. |
| Sequera et al. (2022) | V2 | pass | internal | Listing series internal logic. p. 3 |
1,300 → 4,000 (+208%) → 4,600 (+15%) → 1,900 (−59%) | Pattern consistent with a temporary switch; September 2021 level still 46% above 2019. |
| Sequera et al. (2022) | V3 | minor discrepancy | cross paper | Cross-corpus consistency on pandemic switching. p. 3 |
Compare with gil_2024_assetization_valencia ('STRs did not return to the residential market') and gil_vidal_2026_midterm_rental ('temporarily redirected to residential or mid-term segments') | This paper documents a temporary rise of STR units on residential portals, mostly under seasonal contracts — reconcilable with both, but the 2023 paper's (gil_2024_assetization_valencia) unqualified statement understates the temporary switch. |
| Gil (2023) | V1 | pass | internal | Displacement arithmetic. p. 13 |
3,663 x 2.47 | 9,047.6 — matches. |
| Gil (2023) | V2 | minor discrepancy | internal | Homes listed >60 days vs stated share. pp. 10, 13 |
74.75% x 4,920 | 3,678, vs 3,663 used (difference 15; threshold >60 vs ≥60 could explain). |
| Gil (2023) | V3 | pass | internal | Table 2 price per bedroom. p. 11, Table 2 |
price / bedrooms (studio = 1) | 59.70, 65.22, 38.78, 28.52, 27.69 — all match. |
| Gil (2023) | V4 | minor discrepancy | internal | Table 2 price per guest. p. 11, Table 2 |
price / guests | 21.55, 20.70, 17.16, 15.04, 15.49 vs reported 24.82, 23.10, 17.74, 15.36, 15.49. Consistent only if per-guest is a mean of listing-level ratios (legitimate but then inconsistent with the per-bedroom column, which is a ratio of means). Not documented. |
| Gil (2023) | V5 | pass | internal | Guests per bedroom by size. p. 12 |
4.52/2; 5.69/3; 7.15/4 | 2.26, 1.90, 1.79 — match. |
| Gil (2023) | V6 | minor discrepancy | internal | Listings by guests-per-bedroom categories. p. 12 |
40% + 30% + 20% | 90% — 10% of listings unaccounted for. |
| Gil (2023) | V7 | pass | internal | Four-person group example. p. 11 |
110.74 − 77.55 | €33.19 saving (text: '€30') — consistent. |
| Gil (2023) | V8 | pass | cross paper | Revenue shares vs Gil (2021) on the same data. p. 12 |
52.79/67.53/32.47 (here) vs 52.81/67.58/32.45 (2021) | Very close; differences consistent with changed revenue formula. Shares sum to 100.00%. |
| Gil (2023) | V9 | not checkable | internal | Implied total STR revenue and Poblats Marítims population. pp. 10, 14 |
€30m / 0.518; 5,345 / 0.0959 | ≈ €58m total; ≈ 55,700 residents — plausible, not otherwise checkable. |
| Gil (2023) | V10 | fail | internal | Sample size after exclusions (carried over). p. 9 |
7,654 − 451 − 46 | 7,157 ≠ 7,203. |
| Gutiérrez-Cueli et al. (2023) | V1 | pass | internal | Survey response rate. p. 3 |
94 / (150 − 30 inactive) | 78% of active groups — good coverage for an organisational survey. |
| Gutiérrez-Cueli et al. (2023) | V2 | not checkable | internal | Survey proportions base. pp. 10, 17 |
Percentages vs n = 94 (lower for some items) | Item-level n not reported; percentages cannot be converted to counts. |
| Gutiérrez-Cueli et al. (2023) | V3 | pass | cross paper | Background statistics consistency with other corpus papers. pp. 8-9 |
Compare with gil_martinez_2023_state_led (4M homes 2000-06, prices +123%, household debt +172%) | Identical figures and sources (López & Rodríguez 2010; Colau & Alemany 2012) — consistent. |
| Gil & Palomera (2024) | V1 | pass | internal | Structure of the legislative narrative. p. 13 |
'Four phases' vs episodes analysed | The four phases correspond to the four legislative episodes (2019 LAU, COVID measures, Catalan law, Right to Housing Law), each following the same three-step sequence — consistent. |
| Gil & Palomera (2024) | V2 | pass | internal | Catalan parliamentary majority. p. 18 |
ERC 32 + Podemos 8 + CUP 4 + Junts 34 | 78 MPs — a majority of the 135-seat Parliament of Catalonia (analyst note: chamber size is general knowledge, not from the paper). |
| Gil & Palomera (2024) | V3 | not checkable | internal | National coalition arithmetic as described. pp. 12-13 |
PSOE-led group 120 + UP 35 + left opposition 24 | 179 — the paper's account is internally consistent with the left bloc being a minority within it. |
| Vidal et al. (2024) | V1 | pass | internal | 'Tenant households earn only 55% of what landlord households earn' (ex-rent). p. 9 |
22,183/39,931; 18,457/33,602 | 55.6% (mean), 54.9% (median) — matches. |
| Vidal et al. (2024) | V2 | pass | internal | Document count in Table 2. p. 7 |
3+2+2+2 + (2 joint + 2 additional Chamber docs) + 1+1+1 | 16 unique documents — matches 'A total of 16 documents'. |
| Vidal et al. (2024) | V3 | pass | internal | Poverty shares of all households. p. 9 |
6.5% x 9%; 15% x 9% | 0.59%, 1.35% — match 0.6% and 1.3%. |
| Vidal et al. (2024) | V4 | not checkable | internal | Implied total Spanish pension-fund assets. p. 10 |
€150m / 0.0014 | ≈ €107bn — plausible, not otherwise checkable. |
| Vidal et al. (2024) | V5 | pass | cross paper | Cross-corpus: tenant at-risk-of-poverty rate. p. 9 |
36.3% (LCS 2021, here) vs 30.16% (LCS 2023, gil_vidal_2025 results) and 36.28% (gil_vidal_2025 conclusions) | The 2025 paper's conclusion figure matches this paper's 2021 estimate, suggesting it was carried over from this earlier analysis. |
| Gil et al. (2025) | V1 | pass | internal | Table 1 totals and shares. p. 11 |
Dwellings 78,383 + 121,979 + 2,031 = 202,393; sub-bands sum to 121,979; landlords 10,035+6,224+2,578+869+1,175 = 20,881; total landlords 99,264 | 38.73% / 60.27% / 79.0% / 21.0% — all shares and ratios (5.84, 3.64, 7.53, 12.64, 41.57) reproduce. |
| Gil et al. (2025) | V2 | pass | internal | Income ratios. pp. 12-14 |
45,977/22,851; 45,977/30,433; 38,162/22,851 | 2.01x, 1.51x, 1.67x — match 'double', '~50% higher', '67% higher'. |
| Gil et al. (2025) | V3 | pass | internal | Table 2 internal consistency. p. 14 |
Share of rents ∝ share of households x mean rent; monthly = annual/12 | Implied rent shares 16.88 / 9.75 / 11.89 / 61.48% (reported 16.87 / 9.75 / 11.89 / 61.49); monthly 287 / 459 / 600 / 1,476 — consistent. |
| Gil et al. (2025) | V4 | pass | internal | Poverty shares of all households. p. 15 |
6.47% x 9.3%; 7.63% x 9.3%; 9.5% x 9.3% | 0.60%, 0.71%, 0.88% — match 0.6 / 0.7 / 0.9%. |
| Gil et al. (2025) | V5 | fail | internal | Tenant poverty rate: results vs conclusion. pp. 15-16 |
Compare | 30.16% vs 36.28%; landlords 6.47% vs 6.48%. |
| Gil et al. (2025) | V6 | fail | internal | 'Majority are multi-property landlords'. pp. 9, 11 |
Table 1 landlord shares | Single-dwelling landlords are 79% of landlords; the majority claim holds for dwellings (60.3%), not landlords. |
| Vidal et al. (2025) | V1 | pass | internal | NYC rent-regulated stock implied. p. 4 |
100,000 / 0.10 | ≈ 1 million rent-regulated units — consistent with the claim as stated. |
| Vidal et al. (2025) | V2 | pass | cross paper | Cross-corpus consistency on Spanish policy facts. pp. 9-10 |
Compare with gil_palomera_2024 and gil_vidal_2026 (2019 lease extension to 7 years for legal persons; Catalonia only region applying 2023 controls) | Consistent. |
| Gil et al. (2026a) | V1 | pass | internal | Household growth rates from shares and property counts. pp. 10-13 |
Implied households = single-property holdings / share (10,059,534/0.583; 7,810,630/0.409) → 17.25M, 19.10M; then 0-property and 2+ counts. | +63.1%, +53.2%, −22.4% vs reported +63.4%, +53.5%, −22.4% — consistent; implied household totals plausible. |
| Gil et al. (2026a) | V2 | pass | internal | 'Multiplied by more than four'. p. 10 |
625,883 / 137,943 | 4.54x — matches. |
| Gil et al. (2026a) | V3 | fail | internal | Multi-owner share among owning households in EFF vs Catastro majority claim. pp. 8, 12 |
2008: 26.8/(58.3+26.8); 2022: 37.1/(40.9+37.1) | 31.5% → 47.6% in EFF (minority), vs 46.1% → 51.7% in Catastro (majority). |
| Gil et al. (2026a) | V4 | minor discrepancy | internal | Total properties held by households (EFF) vs 'two million new properties'. pp. 9-10 |
Σ 2008 bands = 21,784,420; Σ 2022 bands = 28,030,347 | +6.25M in EFF (2008-2022) vs ~+2M in Catastro (2008-2025). Sources are not reconciled. |
| Gil et al. (2026a) | V5 | fail | internal | Owner-occupation vs 'owner' category. pp. 6-7, 15 |
63.85% owners + (9.79% − 1.4% landlords who rent) | ≈72% of households live in their own home in 2022, not 63.9%. |
| Gil et al. (2026a) | V6 | pass | internal | Tenure categories sum. p. 7 |
78.99+11.92+3.38; 63.85+19.17+9.79 | 94.3% and 92.8% — remainder (other tenures such as free use) not shown; acceptable. |
| Gil et al. (2026a) | V7 | pass | internal | 'Landlords almost tripled'. p. 4 |
9.79 / 3.38 | 2.9x — matches. |
| Gil et al. (2026b) | V1 | pass | internal | Whole-market composition. pp. 9-10 |
0.472 x 2,570,845 / (2,570,845 + 223,552 + 300,000) | 39.2% single landlords — matches; private multi 43.9%, legal+public 16.9% ('~17%'). |
| Gil et al. (2026b) | V2 | pass | internal | Corporate 8% share. p. 9 |
223,552 / (2,570,845 + 223,552) | 8.0% of the private market — consistent. |
| Gil et al. (2026b) | V3 | not checkable | internal | Composition excluding public housing. pp. 9-10 |
1,213,439 / 2,794,397 | 43.4% single vs 56.6% multi + corporate. |
| Gil et al. (2026b) | V4 | pass | internal | 2016-2023 growth consistency. pp. 11-12 |
Total growth 2,570,845/1,900,883 = 1.352; solve 1.304s + 1.399(1−s) = 1.352 | 2016 single share ≈ 49.0% → 47.2% in 2023 — consistent with reported figures; share shift 1.7 points. |
| Gil et al. (2026b) | V5 | pass | internal | City shares sum. p. 8 |
Each pair sums to 100% | All eight cities sum to 100%. |
| Gil et al. (2026b) | V6 | minor discrepancy | cross paper | Coverage vs tenant households (cross-corpus). p. 6 |
20.2% x ~19.1M households (implied by gil_concentracion_inmobiliaria EFF figures) | ≈3.86M renting households vs 3.09M dwellings counted — ≈0.76M gap (undeclared/other). |
| Gil et al. (2026c) | V1 | pass | internal | Income ratios vs tenants. pp. 8, 18 |
32,120/21,335; 50,959/21,335; 80,375/21,335 | 1.51, 2.39, 3.77 — match ('51% superior', '2.39 veces', '3.77 veces'). |
| Gil et al. (2026c) | V2 | pass | internal | Wealth ratios vs tenants. p. 10 |
193,919/2,217; 407,975/2,217; 996,826/2,217 | 87.5 ('casi 90'), 184.0, 449.6 — match. |
| Gil et al. (2026c) | V3 | pass | internal | Age-gap and tenure-gap absolute differences. p. 20 |
221,254 − 20,069; 407,975 − 2,217; 996,826 − 2,217 | 201,185 ('unos 200.000'), 405,758, 994,609 — match. |
| Gil et al. (2026c) | V4 | pass | internal | 'Eleven times less wealth'. p. 29 |
223,960/20,069; 221,254/20,069 | 11.2 and 11.0 — match. |
| Gil et al. (2026c) | V5 | minor discrepancy | internal | Gini increase. pp. 4, 25 |
0.591/0.501; 0.591/0.513 | +18.0% from 2005; +15.2% from 2002. |
| Gil et al. (2026c) | V6 | pass | internal | Top-share complements. p. 4 |
65.4 + 34.5; 58.1 + 41.9 | 99.9% and 100.0% — rounding. |
| Gil et al. (2026c) | V7 | pass | internal | Youth ownership drop. p. 15 |
69.3 − 31.8 | 37.5 points — matches '>37 points'. |
| Gil et al. (2026c) | V8 | pass | cross paper | Cross-report consistency of owner-occupation (vs Report 2). p. 17 |
Age-specific 2022 ownership rates range 31.8-84.0%; any population-weighted average lies far above 63.85%. | Consistent with an overall owner-occupation rate of roughly 72-75%, supporting the audit finding that Report 2's 63.9% 'propietarios' excludes owner-occupying landlords. |
| Gil et al. (2026d) | V1 | pass | internal | Barcelona MTR ratio and total listings change. p. 5 |
0.0476 x 3.6888 / 0.2327; (0.0476 x 3.6888 + 0.2327)/1.0476 − 1 | 75.46% (reported 75.45%); −61.03% (reported −61.02%) — consistent. |
| Gil et al. (2026d) | V2 | minor discrepancy | internal | Madrid MTR share. p. 5 |
Solve for initial ratio from MTR +73.89%, LTR −58.58%, total −53.15% | Implied 2019 ratio 4.27% → 2024 ratio 17.94% of residential listings (matches 17.97%) = 15.2% of total listings; text says 'of total listings'. |
| Gil et al. (2026d) | V3 | pass | internal | Málaga MTR growth. p. 6 |
458/51 − 1 | +798.0% — matches. |
| Gil et al. (2026d) | V4 | pass | internal | Rent gaps and price growth from reported €/m2. pp. 6-9 |
MTR/LTR − 1 and end/start − 1 per city | Madrid 44.32%; Barcelona 33.93% → 16.71%, +25.38% / +43.89%; Málaga 20.25% → 10.02%, +47.34% / +61.03% — all match; Madrid 2019 gap back-computed 42.18% (reported 42.17%). |
| Gil et al. (2026d) | V5 | fail | internal | 'Compression of 17 percentage points in three years' (Barcelona). p. 9 |
53.17 − 16.71 (Dec 2021 → Dec 2024); 33.93 − 16.71 (2019 → 2024) | 36.46 points in three years or 17.22 points in five years. |
| Gil et al. (2026d) | V6 | pass | internal | Leads: per-listing vs total. p. 9 |
71.4/9 = 7.93x per listing; x (1 − 0.7673) listings | Total residential leads ≈ 1.85x — per-listing growth mostly due to fewer listings. |
| Gil et al. (2026d) | V7 | pass | internal | Madrid foreign searches relative increase. p. 9 |
12.22/9.62 | +27% — matches. |
| Gil et al. (2026d) | V8 | pass | internal | Quarter count. p. 4 |
Q1-2019 to Q4-2024 | 24 quarters — matches; simple regressions use N = 23 (unexplained drop of one observation). |