Discussion

As always, the analysis in this chapter is based on modeling decisions, and modeling is almost always an iterative process. In general, you want to start with something simple that yields an approximate answer, identify likely sources of error, and look for opportunities for improvement.

In this example, I would consider these options:

  • I chose a prior based on the average goals per game for each team. But this statistic is averaged across all opponents. Against a particular opponent, we might expect more variability. For example, if the team with the best offense plays the team with the worst defense, the expected goals per game might be several standard deviations above the mean.

  • For data I used only the first four games of the championship series. If the same teams played each other during the regular season, I could use the results from those games as well. One complication is that the composition of teams changes during the season due to trades and injuries. So it might be best to give more weight to recent games.

  • To take advantage of all available information, we could use results from all regular season games to estimate each team’s goal scoring rate, possibly adjusted by estimating an additional factor for each pairwise match-up. This approach would be more complicated, but it is still feasible.

For the first option, we could use the results from the regular season to estimate the variability across all pairwise match-ups. Thanks to Dirk Hoag at http://forechecker.blogspot.com/ ...

Get Think Bayes now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.