The other big advantage my study had was a larger sample size. For instance, I got something similar to what you did for 2009, despite our different criteria and inputs: an average of 58% of market value for second-time arb-eligible super-twos; 62% of market value for second-time arb-eligible non-super-twos. So that’s pretty similar, as you found (my numbers are higher than yours because I didn’t exclude closers, probably). But when I added the 2010 data, the average for both years was 43% of market value for second-time arb-eligible super-twos and 55% for second-time arb-eligible non-super-twos, which looks more like the 40/60/80 scale that Tango suggests.

As with you, my numbers got a little wonky when looking at arb awards for players with 5+ years of service, mostly because the samples are so small.

This won’t affect your results much in this particular case, but you should keep your criteria the same for every sample of players. You included a closer and several players with less than 0.5 WAR in your sample of second-time arb-eligible super-twos. If you eliminate those, your result will be very close to what I got for 2009: 56%.

Anyway, the larger sample I used does reveal a pattern that approaches 20/40/60/80, but even in my study the samples for the last year probably weren’t big enough. I do think my study demonstrates that super-twos are paid according to years of service rather than times through arbitration (i.e., super-twos make less than what first-time arb-eligibles with 3+ years of service do, and previous super-twos with 3+ years of service make roughly the same as first-time arb-eligibles with 3+ years of service, etc.).

I only used two years of data because those were the only years for which I could easily get WAR projections, but I’d be interested to see what you find if you apply your method to multiple years of data.