Land transport — Algeria's public companies (2012-2022)

Activity of Algeria's public land-transport companies, quarter by quarter, 2012 to 2022. 1,605 observations, 43 quarters, eight companies: the national railway SNTF and its infrastructure subsidiaries INFRARFER and INFRARAIL, the Algiers metro (EMA), ETUSA for the capital's urban transport, Groupe LOGITRANS for freight. Passengers carried, tonnages, turnover, receivables, debt, payroll, headcount, fleet immobilisation rates.

This dataset did not exist, and here is why. These tables live in the Quarterly Statistical Bulletins, bilingual French-Arabic. Their hierarchy — which company, which activity, which indicator — is carried neither by the text nor by indentation: every label starts at the same position, 31.8 points from the edge. It is carried by the boldness of the type. Bold for a company or a heading, light italic for an indicator. And a company is told from a heading by what follows it: another bold title for one, its indicators for the other.

An extraction reading the text layer in flow order sees none of this. Ours tried that first: it produced thirty-nine "companies", among them "Source: Ministry of Transport" and "Average wait in the roadstead". This file has eight, all real.

A built-in check, unusually strong. Each row carries the quarter's value, the same quarter a year earlier, and the rate of change between them — computed by the office from those two numbers. Recomputing it verifies all three columns at once: a column shift, a misread value, a mismatched indicator, all break that ratio. Indicators failing it are removed one by one, without disqualifying their neighbours.

A second check, by overlap. Each quarter is published twice — once as the current quarter, once the following year as a comparison term. This attests the ATTRIBUTION and not merely the value: two bulletins agreeing on the same company's turnover did not misattribute it the same way. 179 keys where the two readings diverge are not published at all.

A third, on ambiguity. A key appearing twice in one table means two companies were merged — the layout did not deliver their boundary. The whole table is then refused.

What is not in the file. Port activity, for a structural reason rather than a reading one: its headings carry values of their own — "Total traffic" is both a total and the parent of the "Unloaded" and "Loaded" rows detailing it — and the hierarchy rule, which assumes a parent has no values, does not apply. "Unloaded" appears there under three parents. The five port tables are refused rather than half-published.

Ten observations on border movements are included — foreign visitors and Algerians resident abroad, 2018 to 2022 — showing the pandemic's tourism collapse: 1,933,778 foreign visitors in 2019, 66,995 in 2021, 757,380 in 2022.

Useful columns. colonne says whether a value comes from its first-hand publication or from a later bulletin's comparison column. variation_annuelle_pourcent carries the office's own published rate, the very one used for the check.