[Arnold]

When searching for a trend, does it make sense to include in your sample your entire database of games (>10 seasons), or do you throw everything away but the last 3 seasons or so?

To clarify, I'm talking about common trends, not team specific trends.

[diogee]

I usually go no more than 2 years due to the constant changes made within every team.

[ChuteBoxe]

2-3 years at most.

[Ganchrow]

When searching for a trend, does it make sense to include in your sample your entire database of games (>10 seasons), or do you throw everything away but the last 3 seasons or so?

To clarify, I'm talking about common trends, not team specific trends.

Why dont you give an example of the type of "trend" for which you're searching?

[LT Profits]

Since you are talking about league-wide trends, the larger the sample the better, as long as it is a logical trend and not someting like "teams that win on Tuesdays on articficail turf when the manager wears pink underwear". If you are researching something like how NBA teams do the next game after covering the spread by 10 points or more last game, then that is relavent froim year-to-year and I see nothing wrong with going back as far as possible.

As for team trends, probably two years maximum.

[Arilou]

Team trends tend to be fake, but if you are looking anyway it tends to be about 2 years but vary it team-to-team depending on their history and what has or hasn't changed. In terms of universal stuff, you should go back until the last major rules change that invalidates the data, so you have about two decades in the NFL and two years in the NHL, etc, but make sure to preserve some of your data out of sample.

[Justin7]

I usually go no more than 2 years due to the constant changes made within every team.

I agree. Also, the market recognizes trends quickly, making anything older than 2 years dangerous to follow.

[Arnold]

Why dont you give an example of the type of "trend" for which you're searching?

I keep a bunch of variables in my database for each game played. Things like how many days a team rested before the game, what's their winning/losing streak, whether they played at home last game, being favorites/underdogs, by how many points won/lost last game, etc. Then with this data I try to find something relevant. I noticed in the NHL, the home advantage isn't as intense anymore. In the 80's it was 60% SU. In the 90's it was 56.5%. In 00's, it is 55.5%. Whenever I search for an NHL SU trend, I get very different results if I exclude the 80's and even the 90's. I don't think the recent rule changes had anything to do with it.

Since you are talking about league-wide trends, the larger the sample the better, as long as it is a logical trend and not someting like "teams that win on Tuesdays on articficail turf when the manager wears pink underwear".

Believe me, I don't look for these silly things I don't even keep this data to start with. I don't really care if it was Tuesday or Sunday, or what color someone's underwear is.

If you are researching something like how NBA teams do the next game after covering the spread by 10 points or more last game, then that is relavent froim year-to-year and I see nothing wrong with going back as far as possible.

I was thinking maybe some trends have a life time? Meaning, they are good only for the past few years or so. Because if you include your entire database in your search, then you could get a 500-500 results based on your search. If you include only the past few years, then you might get 300-200. Maybe it's random, I can't say. But how would one know?

[Data]

I noticed in the NHL, the home advantage isn't as intense anymore. In the 80's it was 60% SU. In the 90's it was 56.5%. In 00's, it is 55.5%. Whenever I search for an NHL SU trend, I get very different results if I exclude the 80's and even the 90's. I don't think the recent rule changes had anything to do with it.

I am impressed, I have never seen this angle posted in the forums. I noticed the same effect in NBA and I think it is due to growing quality of officiating.

[JEMFS]

In college football, due to graduation and coaching turnover I go back no more than 3 years.