tower by provider and when they were added in OpenCellID. Because the data does not have information on when the cell towers were built, the measure date is used as a proxy. Most of the cell towers were reported to the dataset in 2015 and 2016. Although some of the cell towers are probably older, the increase likely reflects the actual development in the telecom sector. The Burmese telecom market only opened up for foreign competition in 2014, and before that the state-owned MPT was a monopoly. As new firms entered the market in 2014, also MPT had to start expanding its network to remain competitive. When constructing the predicted cell phone coverage, I use all cell phone towers that were included in the dataset before September 2017. The strength of cell phone signal in a given location is primarily determined by distance to the cell tower and whether the receiver (i.e. mobile phone) is in line of sight of the cell tower. Obstructions, such as hills, buildings, or dense foliage, reduce the signal. I use a radio propagation model to predict where the signal is strong enough for cell phone reception. I apply the irregular terrain model (ITM), as introduced by Olken (2009), to calculate the predicted network coverage area. The model calculates predicted signal loss due to topography and distance between a transmitter and receiver. A number of validation studies have found that the ITM yields highly accurate predictions, and the model has been widely used in professional radio planning (Crabtree and Kern 2018). The surface area of Myanmar is first divided into 200m×200m cells, and then for each cell, the ITM predicts whether the signal is strong enough for reception (See Appendix A for a more detailed description of the coverage prediction). I aggregate these predictions to obtain share of each township with cell phone reception. I do this separately for MPT and for the combination of all other mobile network providers. The share of township with coverage from MPT is the main independent variable. Figure 4 maps the geographic variation in the predicted MPT cell phone coverage. The predicted coverage is very unevenly distributed, with fairly comprehensive coverage in the central parts of the country, and large peripheral areas with very poor cell phone coverage. Because I use crowdsourced data, it is likely that not all cell towers are included in the data, and there might be some error in the exact locations of the towers. I also need to approximate a number of technical parameters when conducting the coverage prediction. Measurement error in the independent variable may therefore bias the estimates towards zero. Measures of terrain elevation are obtained from NASA’s Shuttle Radar Topography Mission (SRTM), which has generated publicly available high resolution topographic data of the world (Jarvis et al. 2008). I use the one arc second resolution (approximately 30 meters at the equator) in the cell phone coverage calculation. In addition to elevation, the signal propagation model also takes into account how land use—e.g. water, forest, cropland—affects propagation. The land cover classification is obtained from the University of Maryland. 11