Do robots erode the tax base or enlarge the denominator? Revenue, output, and trust in 149 countries, 2002-2023

This deposit contains the datasets and Stata code needed to reproduce every table and figure in "Do robots erode the tax base or enlarge the denominator? Revenue, output, and trust in 149 countries, 2002-2023". The article asks whether robotization erodes the tax base. It separates the tax-to-GDP ratio into its two parts, real revenue per capita and real output per capita, and estimates the association of each with a robot stock built from trade data, using a Mundlak correlated random effects decomposition and fixed effects with Driscoll-Kraay standard errors on a panel of 149 countries. The robot measure is constructed rather than licensed. It is built from UN Comtrade imports under HS 847950, screened on unit value to exclude consignments inconsistent with industrial robot prices, netted of re-exports, accumulated with ten per cent annual depreciation and scaled by the labour force. Flags for imputed quantities and carry-forward years are retained in the deposited files so the screen and the accumulation can be audited, and alternative stocks under five and fifteen per cent depreciation and a wider unit-value band are supplied for the robustness suite. Because the measure captures imports, countries that produce robots domestically sit outside the identified variation. Two do-files run in order. 01_build_master_panel.do assembles the data from raw sources and produces master_panel.dta and est_panel_0223.dta. 02_replication_tables_figures.do reproduces Table 1, Tables 4 to 11 and Figures 1 to 5 from est_panel_0223.dta, which is supplied here, so it can be run on its own. Software is Stata 19.5; the user-written commands required are xtscc, estout, ivreg2, ranktest, avar, xtivreg2 and wbopendata. Set the working directory to the deposit folder before running; the code contains no cd commands. Files: est_panel_0223.dta, the estimation file, 2,677 observations on 149 countries; master_panel.dta, the full 1996-2023 build for 192 countries; robot_stock.dta, robot_panel.dta, robot_alt.dta and robot_bandB.dta, the constructed robot series; rmean_fig1.dta for Figure 1; legatum_trust.dta and manuf_share.dta as merged inputs; the two do-files; and a README documenting the construction and the coding decisions. Four third-party datasets are not redistributed here because they are large, licensed to their publishers and freely obtainable at source: the UN Comtrade extracts for HS 847950, the UNU-WIDER Government Revenue Dataset 2025, the World Development Indicators series used as controls, and the Worldwide Governance Indicators. The World Values Survey trust file is likewise not redistributed under its terms of use; the derived country-year measure is already merged into the supplied panels, so the estimation runs without it. The README names each source with the filename the code expects. The panel comprises every economy for which the robot stock, the revenue series and the controls can all be constructed, rather than a random sample from a larger register. Standard errors are read under a superpopulation interpretation following Gill (1999) and Abadie, Athey, Imbens and Wooldridge (2020). Additional notes field Derived in part from UN Comtrade, the UNU-WIDER Government Revenue Dataset, the World Bank World Development Indicators and Worldwide Governance Indicators, the World Values Survey and the Legatum Prosperity Index. Those sources are not redistributed and remain subject to the terms of their publishers. Creators.