Context. Water vapor is the main source of atmospheric opacity for millimeter and sub-millimeter astronomy. Hence, several studies seek to effectively characterize it for site-testing purposes. In this task, reanalysis databases are quickly becoming a popular alternative to on-site measurements due to easy accessibility and the versatility of the data they provide. Aims. In the framework of validating the use of reanalysis data as a site-testing oriented tool, we perform a statistical comparison of atmospheric water vapor values obtainable from the MERRA-2 database with ground-based microwave radiometer measurements taken at two astronomical sites in Chile: Llano de Chajnantor, Atacama, and Cerro Paranal, Antofagasta. Methods. The MERRA-2 data were interpolated both vertically (across pressure levels) and geographically (latitude-longitude). For each site, different plots were generated: a direct temporal variation plot (to visually compare the data variation over time between both sources); a PWV versus PWV plot, fitting a linear fit through robust linear regression and calculating both the Pearson (r) and Spearman (ρ) correlation coefficients in order to look for correlations between both data sources; a histogram showing the distribution of the differences between the MERRA-2 data and the water vapor measurements (defined as APWV = PWVMERRA-2 − PWVsite), along with its standard deviation (σ), mean (µ), and median values, with the aim of better appreciating the similarities of the data sources over time; and a CDF plot to compare both data distributions disregarding time stamps. Finally, millimeter and sub-millimeter transmittance curves were created through the am atmospheric modeling software, which uses ozone and temperature data along with the verified water vapor data for the two studied sites as well as three other sites of interest for the next-generation Event Horizon Telescope: Las Campanas Observatory near La Serena, Chile; Valle Nevado, located near Santiago, Chile; and the General Bernardo O’Higgins base, located in Antarctica. Results. The interpolated MERRA-2 PWV values are highly correlated with the ground-based PWV values, with a Pearson coefficient greater than 0.9 and a Spearman coefficient higher than 0.85. However, their dependence is not linear, as PWVAPEX = m * PWV, with m being higher than 0.9 in both cases. The difference histograms show an almost zero-centered distribution for Llano de Chajnantor, with a µ value of −0.021 and a median value of −0.007. On the other hand, in Cerro Paranal, the difference histogram is slightly offset toward positive values, with µ value of 0.171 and a median value of 0.256. This offset is most likely due to the strong winds present in the site’s location, close to the Pacific Ocean. The transmittance curves show different performances depending on the site studied, with Cerro Chajnantor being the highest overall transmittance and Antarctica the lowest. Additionally, the transmittance profiles estimated for Cerro Chajnantor and Cerro Paranal were scaled using the PWV measurements, providing differences of less than 12% to the model data profiles. Results obtained at the Valle Nevado site suggest promising atmospheric conditions for stronomic observations in the millimeter and sub-millimeter range. Conclusions. The results we obtained show that the atmospheric water vapor estimation using MERRA-2 data can be used for site testing of new sites by evaluating the millimeter–sub-millimeter transmittance profile through vertical pressure correction and averaging the closest grid points to the site. This new method opens the door for future site-testing studies using MERRA-2 and potentially other reanalysis databases (e.g., ERA5) as reliable sources of information.