Effects from Bit Resolution on Biquad

I attached four plots characterizing the precision of the transfer function for different bit resolutions (assuming a 10MHz sampling rate and a critical frequency of 10Hz). In the meeting yesterday, Lee mentioned how the lines were not normalized, so I updated the graph to have normalized lines (Figure 1). However, in that graph, all the lines (excluding the 18 bit one) are aligned perfectly which is odd and still something I am trying to understand. Regardless, I included the non-normalized version for reference (Figure 2). I also included a graph showing the ratio between the float version and truncated versions to see if there were noticeable deviations due to different bit resolution, but there were not (Figure 3). For the remaining graph, I changed the bit resolution for all of the coefficients rather than just the b coefficients (Figure 4) to explore what that would look like.