Since the discovery that the majority of low-redshift galaxies exhibit some level of spiral structure, a number of theories have been proposed as to why these patterns exist. A popular explanation is a process known as swing amplification, yet there is no observational evidence to prove that such a mechanism is at play. By using a number of measured properties of galaxies, and scaling relations where there are no direct measurements, we model samples of SDSS and S4G spiral galaxies in terms of their relative halo, bulge, and disc mass and size. Using these models, we test predictions of swing amplification theory with respect to directly measured spiral arm numbers from Galaxy Zoo 2. We find that neither a universal cored nor cuspy inner dark matter profile can correctly predict observed numbers of arms in galaxies. However, by invoking a halo contraction/expansion model, a clear bimodality in the spiral galaxy population emerges. Approximately 40 per cent of unbarred spiral galaxies at z ≲ 0.1 and M* ≳ 1010 M⊙ have spiral arms that can be modelled by swing amplification. This population display a significant correlation between predicted and observed spiral arm numbers, evidence that they are swing amplified modes. The remainder are dominated by two-arm systems for which the model predicts significantly higher arm numbers. These are likely driven by tidal interactions or other mechanisms.