Beatpaths was really fun during the entire 2005 NFL Season, but there’s been one niggling issue that I’ve been trying to figure out how to address for the start of the 2006 NFL Season.

The beatpaths graphs don’t really make a lot of sense until each team has a few games under their belts so you can really see how things start to inter-relate.

So the question becomes, how to get meaningful graphs for the first few weeks of the season?

Possible solutions:

- Start with the full 2005 season, just adding the 2006 results in until week 4 or 5 or so, at which point 2005 gets removed.
- Start with the full 2005 season, but phase out 1/4 of the games for each of the first four weeks
- Start with the 2006 preseason but phase them out quickly
- Something more advanced where a 2006 game outcome would somehow cancel out “all similar games” from 2005 (common opponents, etc).
- Do nothing and just start fresh with 2006 and let things get interesting over time.

I would suggest avoiding using the preseason games, as most people feel they really don’t give a true indication of a team.

Doing something more tricky, while interesting, would probably be confusing. It’s difficult enough explaining how your beatpaths work without throwing a bunch of special criteria on top.

Taking the previous season as a starting point, especially for weeks 1 and 2, makes sense to me. That’s how people tend to start looking at team strengths at the beginning of a season anyway. As for how to phase those out, if you remove those games all at once, after week 4 for example, then your graph is likely to shift dramatically between the week with 2005 games and without. If you phase them out gradually, removing the oldest games each week, that’ll keep the graph shifts gradual as well.

Another possibility would be to always have a sliding window that’s the length of a full season, so that the pre-season 2006 graph is the final 2005 graph, after week 1 of 2006 you drop week 1 of the previous year, and so on until at the end of the 2006 season the graph just reflects 2006. I’m not saying that’s a good idea, but it is a possibility.

My suggestion would be to use the 2005 season, and drop the 1/4 games each week. Either that or just start fresh and watch the new season graph grow.

I agree with JD, to me the best seems to be use last season and drop 1/4 of if each week the first four weeks. Gives a good starting point but after 4 weeks we usually know which teams have falled alot and which have been surprises.

Glad to see Beatpaths is comign back, and glad I still check the RSS. ðŸ™‚

I enjoy watching how it goes starting from week 1 where there are just 16 straight, non-connected beatpaths and then seeing it slowly devolve into the organized chaos that the season becomes.

That said, if you want to make it more *interesting*, I agree with the use-last-season-and-drop-1/4-games-each-week solution.

I agree that it is fun to watch the figure evolve. However, if you do wish to use prior year data, I would recommend using a weighted value for last year’s data. For example, weighting last year 50% and using the current year data. I think this would be better than using just last years data because teams do change in the offseason, draft, trades, etc. So there is some continuity between years, but it is not 100% perfect.

Preseason should mean NOTHING. I kind of like it starting as a new season.

I am with WookieBH. Start fresh, let the chaos evolve.

I think the best thing is to show graphs accurately from the start and not use graphs based on preseason or last year. It will beinteresting to see if the Super Bowl winner was at the top of the graph for the entire season. Or we willse who can stay on top (or tied to a top spot) the longest. OUr guesses as to any other way of starting are no less accurate than haivng week 1 with 16 teams tied for first and 16 tied for last.

So what did you decide?

I advocate starting where you left off in 2005 and mixing in new data as it replaces last year’s games (or some similar weighting). I think that magnifying each of the first four games to equal a quarter of last season is heavy-handed, to say the least. Especially since there may still be some tweaking going on early in the season, so these games would be least likely to be indicative of “season’s average” play.

After all, it would be difficult to numerically quantify how much each team has changed since then (i.e. 5% of the player have been traded or injured, 1 coaching change, 2 assistant coaching changes, etc.)

As for the “start fresh” methodology, I don’t agree at all. It’s not like the teams have completely disbanded their players and coaching staff, and arbitrarily reformed into new teams, so there’s really no “chaos” whatsoever. If so, you could make your picks by flipping a coin, which is essentially what you’re doing until you get a sufficiently representative set of data. That could be halfway through the season!

A better analogy would be some sort of “Delta” factor that would have to be calculated on a per-team basis. For example, a major coaching staff change, benching Roethlisberger, or a key playmaker being traded on or off a team. i.e. a high delta factor means the team has changed considerably, and last year’s data shouldn’t be heavily weighted.

As for what algorithm to use when mixing in fresh data…I don’t have any definitive answers, but possibly: canceling out last year’s beatloops or dropping a 2005 game each week into 2006 (creating the equivalent of the market’s 52-week moving average and recalculating beatloops accordingly).

However, neither of these systems will be perfect at weighting this year’s games over last year’s, though the canceling method seems more likely to reach that goal quickly. Though you need an applied mathematician to prove that ðŸ™‚