Sources and method

Where each part of this site comes from, what the research adds, and what the UNHCR numbers can and cannot show. The Methodology at the bottom explains step by step how the information was processed, with artificial intelligence.

Primary source for the timeline and background

Garcia Arias, Jenny. 2024. “The Demography of Crisis-Driven Outflows from Venezuela.” Population and Development Review 50(3): 643–675 (September 2024). https://doi.org/10.1111/padr.12651

Open access under a Creative Commons Attribution License (stated on p. 643). The author is a Marie Curie Postdoctoral Fellow at INED (Paris) and an associate researcher at IIES, Universidad Católica Andrés Bello. The study estimates annual outflows from Venezuela by age and sex for 2011–2021 from official counts of Venezuelan-born people in seven main destinations.

What this site takes from it	Where in the article
The timeline figure (rows, markers, dashed periods, dots, phase bands) and the figure note and sources	Figure 1, p. 649
The four periods: precrisis 2011–2013, early phase 2014–2017, socioeconomic upheaval 2018–2020, COVID-19 pandemic 2021	p. 648
First, second and third emigration waves; 2013 economic controls; 2017 Supreme Court ruling	pp. 647–648, note 1 on p. 669
Hyperinflation, GDP fall and extreme poverty, 2018–2019	p. 644
Peak outflow in 2018–2019 and the COVID-19 period	pp. 655, 664
Darién Gap and transit caveats used in the Explorer notes	pp. 646, 665–666, 668

The background wording you see on the Timeline page follows the article, with citations omitted where marked. The timeline is a redrawn, interactive adaptation: marker positions were read off the printed figure, so dates are approximate. The author name spelled “Rousse” in the article text is “Rousset” in its reference list.

What the study found, for context

Each row is a finding or statement from the article, with its page. These measure outflows by country of birth, not the UNHCR counts used in the Explorer.

Finding	Page
Total outflow since 2011 is about 5.16 million (95% CI 5.13–5.19), equal to 18% of the estimated 2021 Venezuelan population. About 4.73 million (CI 4.65–4.80) left between 2015 and 2020.	p. 655
Nearly 1.70 million left in 2018–2019, the largest exodus. Nearly 300,000 more crossed between mid-2020 and mid-2021 despite COVID-19 mobility restrictions.	pp. 655, 664
Outflows moved from family migration with many children (precrisis and early phase) to working-age adults, especially ages 15 to 40. In 2019 people aged 50 or older became a significant share, especially older women during the pandemic.	pp. 655, 660
Women’s migration intensity exceeded men’s in every year. The gross migraproduction rate peaked in 2019 at 4.22 for men and 4.74 for women. Men’s larger share in 2014–2018 mostly reflects the composition of the population still in Venezuela.	pp. 658, 662
Boys under 15 were 19.1% of cumulative outflows in 2011–2021, girls 17.6%.	p. 663
Compared with the expected 2021 population, Venezuela has 20% fewer women of reproductive age and 17.8% fewer people aged 15–64. The share aged 60 or older is higher by 8% (men) and 10% (women).	p. 660
Seven destinations (Colombia, Ecuador, Peru, Chile, Mexico, United States, Spain) held about 90% of Venezuelans abroad in April 2022. Their share of Venezuelan-born people abroad rose from 72.1% in 2010 to 94.5% in 2022.	pp. 646, 653
Early emigrants headed mainly to the United States and Europe (Spain, Italy, Portugal). Within a decade, outflows of middle- and lower-income Venezuelans headed mainly to Colombia, Ecuador and Peru.	p. 662
Except for Colombia, 90% of displaced Venezuelans remained in the first country they reached. Colombia was a transit country for people heading south on foot.	p. 666
Family migration coincided with relaxed MERCOSUR entry rules. As outflows grew, most recipient countries imposed entry barriers, and the Venezuelan government closed borders and failed to issue valid passports to many citizens.	p. 663
Many Venezuelans now leave the South American countries where they first settled and cross the Darién Gap towards the US border. The author expects these flows to include a large share of children.	pp. 668–669
Return migration is reported as negligible. The study assumes none, and a test with 6% return changed its indicators by under 0.02%.	pp. 665–666
Limits: counts include Venezuelan-born people regardless of legal status, Colombian returnees, and second- and third-generation descendants who may hold Spanish or Colombian citizenship. Returnees to other countries are missing, and people in transit are likely undercounted.	pp. 646, 665–666

How this relates to the UNHCR numbers in the Explorer

Different measures. The article estimates annual outflows, 2011–2021, by country of birth and regardless of legal status, from censuses and administrative records in seven countries plus R4V and UN DESA totals. The Explorer shows UNHCR year-end counts of recorded people, 2020–2025, by host country. Do not add or subtract one from the other.
Only 2020 and 2021 overlap. For orientation, the Explorer’s 2021 total across destinations is 5.58 million; the article’s cumulative outflow estimate for 2011–2021 is 5.16 million (CI 5.13–5.19). The gap is expected given the different definitions.
Both warn about transit. The article says people in transit are undercounted (pp. 665–666). The Explorer flags Guatemala and Honduras for the same reason.

UNHCR data

UNHCR persons of concern demographics for people of Venezuelan origin: 350 rows, 2020–2025, 70 destination countries plus one record where Venezuela is its own country of asylum. The UNHCR footnotes file was supplied with it.
Values are year-end stocks. The extract has no population-type column; the only Venezuela-specific footnote says “Refugees includes people in a refugee-like situation” (2020–2025).
Sex and age are reported for 48% of people in 2020 and 77% in 2025. The US 2025 figure is partial, and Colombia 2023 has shifted age bands. The workbook’s Data quality sheet lists 14 issues and how each is handled.

Download the workbook (.xlsx)Download the cleaned data (.csv)

Indicator sources printed in Figure 1

As given in the note to Figure 1 (p. 649).

Methodology

Transparency note: this information was processed with artificial intelligence. An AI assistant (Claude, developed by Anthropic) read the UNHCR files and the article, cleaned and aggregated the data, classified destinations, flagged data-quality issues, built the workbook, the Explorer and the Timeline, and drafted the explanatory text, including this page, following instructions from the project owner. AI makes mistakes. The calculations have not been independently audited by a human analyst, so check any figure you plan to quote against the UNHCR data and the article, using the page references given.

Step by step

Receive the inputs. The project owner supplied the UNHCR persons of concern demographics file (350 data rows), the UNHCR footnotes file (375 rows), an image of the timeline figure with background text, and the article PDF. No other source was used: every UNHCR number comes from those files, and every figure attributed to the study comes from the article.
Profile and check the data. The AI read both files in full and ran integrity checks: no duplicate year-and-country rows, female and male age bands add up to their sex totals in every row, and the sex totals never exceed the overall total. It noted that the extract has no population-type column and that the total is often larger than the sex-and-age sum.
Read the footnotes. The footnotes file was searched for notes naming Venezuela and for notes on destination countries that appear in the data (for example the United States, Spain, Chile and Costa Rica). Those found are shown in the Data note column of the workbook and on its Footnotes sheet.
Clean without changing values. A trailing space in “Curacao ” was trimmed, long country names were shortened (UNHCR’s name is kept on the Lookups sheet), and the “Other” age columns were renamed “Age unknown” because they hold people with no reported age, not a third sex. The Explorer merges ages 5–11 and 12–17 into 5–17 because Colombia’s 2023 age bands are shifted. No source number was altered.
Set aside the in-country record. Six rows (one per year) have Venezuela as both country of origin and country of asylum. Their population type is not stated, so they stay in the data but are excluded from every “recorded abroad” total.
Group destinations into corridors. Each of the 70 destination countries was assigned to one of seven corridors: Andean region, Southern Cone & Brazil, Central America & Mexico, Caribbean & Guianas, North America, Europe and Rest of world. The first four make up Latin America and the Caribbean. This is a judgement made to look at flows within and outside the region, not an official classification. It can be edited on the Lookups sheet.
Calculate. Totals by year, country and corridor; shares of the total; the change in recorded stock against the previous year-end; and sex-and-age coverage (female plus male, divided by the total). Age and sex shares use only people whose sex and age are reported, and the coverage is shown beside them. In the workbook every figure is a live formula (SUMIFS, INDEX/MATCH, RANK). The Explorer computes the same sums in the browser from the embedded data.
Flag data-quality issues. Rows were flagged for UNHCR footnotes and for patterns the AI judged worth a warning: a total identical to the previous year, round-number estimates, very large year-to-year swings, and sex and age reported for under 25% of a total above 1,000. 81 of the 350 rows carry a note. The flags that are not UNHCR footnotes are labelled as analyst flags.
Verify the calculations. The workbook was recalculated in LibreOffice: 3,859 formulas, no errors. Headline numbers (totals by year, corridor totals, top destinations, coverage) were compared with a separate calculation in Python and matched.
Build the Explorer. The cleaned data is embedded in one HTML page and the charts are drawn as SVG. The seven corridor colours were checked with a colour-blind separation test in light and dark themes. Three light-theme colours fall below 3:1 contrast, so every chart also has direct labels, a legend and a table. The page was tested in a Chromium browser at desktop and phone widths, in light and dark mode, for script errors and sideways overflow.
Recreate the Timeline. Figure 1 was redrawn from the supplied image. Positions of markers, dots and dashed lines were measured by eye in image pixels and converted to years (2011 at 135 px, 106.4 px per year), so dates are approximate. The background text was matched to passages in the article (all 33 pages were read), quoted as printed with citations omitted where marked, and given page numbers. Items for which the article has no text were left blank instead of filled in.
Write the reading of the evidence. The Insights sheet and the Explorer notes describe patterns visible in the numbers, such as corridor shares and the 2025 changes. Where a pattern could have several causes, such as the Guatemala and Honduras swings, it is described as a possibility and set beside what the article says, not stated as fact. The article and UNHCR measure different things, so they are shown side by side and never combined.
Publish. The files were committed to this repository. The pages are static, with the data embedded, and load no external scripts or data.

What the AI did and what it did not

Done with AI

Reading and checking the files
Cleaning, aggregating and classifying the data
Flagging data-quality issues
Building the workbook, the Explorer and the Timeline
Drafting the text on every page, including this one

Not done with AI

The UNHCR data and footnotes (UNHCR)
The article and Figure 1 (J. Garcia Arias, 2024)
The question, the choice of sources and the structure of the site (project owner)

Limits of this method

The corridor grouping is a judgement call. A different grouping would change the corridor totals, not the country totals.
Timeline positions are estimates taken from an image.
The web links in the figure note were copied as printed. They were not opened or checked.
UNHCR data was used as supplied. It was not re-downloaded or compared with the UNHCR database.
The analysis scripts ran inside the AI session and are not published in this repository. The workbook formulas and the CSV allow every figure to be reproduced.
If you find an error, treat the linked sources as the authority and tell the project owner.

Prepared in October 2026.