Despite the growing trend towards the use of big data methodologies, there is still limited application of such techniques to understand how spectrum is used in mobile networks. In this paper we analyse how low (<1 GHz) and high (>1 GHz) frequency spectrum is used in 4G networks in urban areas, in relation to eNodeB density, available bandwidth, Reference Signal Received Power (RSRP) and Reference Signal Received Quality (RSRQ). We present a method to analyse the strategies used by Mobile Network Operators (MNOs) to deal with traffic congestion, and the degree to which they must densify their networks depending on their spectrum portfolio. Using crowdsourced data from 2017 from a popular mobile app, we apply this method to Greater London. We find that the fraction of sites that fully use all available bands to the MNO range from 2% to 20%. Additionally, MNOs with large bandwidth use 42% fewer sites on average in dense urban environments. This difference decreases in suburban areas to 23% fewer sites. The lowest frequencies in each eNodeB tend to exhibit lower RSRP values, as they are often used to serve cell-edge users. These frequencies also show lower RSRQ values because of higher interference caused by neighbouring cells. Similarly, large (high frequency) bandwidth improves RSRQ as it allows for fewer users per MHz, which reduces interference and enables larger cell sizes. We conclude that in dense urban environments, the available bandwidth, rather than propagation properties, determines the preferred band for network deployment by MNOs. INDEX TERMS 4G mobile communication, radio spectrum management, crowdsourcing