Chase Stuart
Footballguy
Last year, in this column, I decided to find out. I'll cut and paste the relevant parts, and make some 2008 updates, to make this as short and informative as possible.
How much weight should we give to our preseason projections relative to the actual season-to-date results, when predicting the remainder of the season results? I won't pretend that it's an easy question, or one that has a definitive answer. From 2000-2006, Footballguys.com projected exactly 100 RBs to score 100 FPs or more in a season, using the standard scoring system of 1 point per 10 yards, 6 points per TD, and zero points per reception. With a sample that large, we can decipher how to use preseason projections and in-season results to predict remainder of season results. For example, in 2001, a rookie Travis Henry wasn't very good, but was expected to shoulder more of the load in 2002. Footballguys.com projected him to score 11.4 FP/G in '02, but in week one he exploded for 178 yards and 3 TDs against the Jets. While he wouldn't average 35.8 FP/G for the season, what would have been a good prediction for the rest of the year? It turns out he averaged 14.9 FP/G the rest of the year; so a formula of six parts preseason projections and one part week one projections would have predicted the right result. (6/7 * 11.4 + 1/7 * 35.8 = 14.9)
We can perform a regression analysis to figure out how much weight should be placed on the preseason projections and how much weight on the week one results, in order to predict the remaining fantasy points production. Since we have a large sample of 100 RBs, we can be confident in the results of this analysis.
It turns out that the "correct" formula is 5.02 parts projections, and 1 part week one results. This feels about right to me. Right now, LT is projected to average 19.0 FP/G, Adrian Peterson is projected at 16.0 FP/G, and Marshawn Lynch 13.1 FP/G. What week one results would make us project all three RBs as even for the rest of the year? Tomlinson scoring 9 points, Peterson scoring 24 points, and Lynch scoring 39 points. So if LT had 30 total yards and one TD, and Lynch had 210 total yards and three TDs, you might be indifferent to trading Lynch for Tomlinson, or, of course, vice-versa.
Below shows the breakdowns for the RBs through X weeks of the season:
After week Proj Wt Reg Wt1 5.02 1.002 3.09 1.003 2.27 1.004 2.00 1.005 1.66 1.006 1.20 1.007 1.00 1.428 1.00 2.26It's worth noting that even after six weeks, preseason projections still matter more than the regular season results. Willie Green was projected to score 9.5 FP/G as a rookie, and was averaging just 2.1 FP/G through six games; he averaged 12.4 FP/G the rest of the way. In 2005, Tomlinson was averaging over 26 FP/G after six games, but scored only 160 FPs over his last ten games, much more in line with his projection of 18.6 FP/G. And preseason projections aren't worthless after eight weeks, either - they just become less and less important. Your remainder-of-season projections shouldn't be based off binary analysis: preseason projections don't either matter or don't. Rather, it's a sliding scale, where they become less and less valuable each week.How do things look at wide receiver? The table below displays the results of the same type of regression analysis, performed on the 197 WRs projected to score 80 or more FPs.
After week Proj Wt Reg Wt1 6.45 1.002 3.15 1.003 2.61 1.004 1.91 1.005 1.54 1.006 1.22 1.007 1.04 1.008 1.00 1.07Preseason projections seem to matter a bit more here, and that's not terribly surprising. Wide receivers get fewer targets than RBs get carries and targets, which means a really good or bad start could be the result of a small sample. Therefore, you should be less quick to give up on your wide receivers. Even though I'm saying not to weigh week one heavily, it still shouldn't be ignored. Remember that in 2001, Randy Moss was projected to be the top WR by just about everyone, coming off a ridiculous 1437/15 season. He had just 28 yards in week one, and 2001 ended up being the worst season of his young, four-year career. In stark contrast, though, is Marvin Harrison of the same year. He would gain just 35 yards in week one, but would finish the season with over 1500 yards and 15 TDs. So always remember not to trade your studs after just a bad week.You shouldn't even give up on your studs after five games, unless they've tanked. Once again, remember we're dealing with a continuum, and there's no on/off switch as to when the regular season starts to matter and the preseason projections stop. But even after five games, the preseason projections still matter a bit more than the regular season projections. Lee Evans in 2005 is a good example; he had just 3.4 FP/G through five games, despite being projected to score 8.3 FP/G. He would average 9.4 FP/G the rest of the year. Remember our 2001 Randy Moss, who was bad in week one? Well he wasn't good in weeks two, three, four or five, either. But in the last 11 games he would haul in 9 TDs and 900+ receiving yards, nearly doubling his per-game output from the first five weeks.
So what's it all mean? The two tables above can be a rough guide for you when thinking about making trades or deciding who to start in a given week. But they should just be used as a guide, and not as the formula to end all formulas.
One thing to remember is that strength of schedule matters. When using large samples, strength of schedule (and other random variables) aren't relevant to the analysis. However, strength of schedule matters a lot when we're analyzing our current, individual players.
Additionally, not all fantasy points are created equally. A player with a couple of TDs and not a lot of touches is probably going to disappoint in the future. Likewise, if Marion Barber III has 240 total yards through the first three weeks of the season but no scores, that's probably a better sign than if he had 80 total yards and three scores. Along the same lines, yards per carry, touchdowns, targets and yards are all relevant information to have, and tell you more about a player than merely fantasy points per game.
Finally, there are some times when you can just throw out the preseason projections. Obviously if a player gets hurt that changes everything, but there are subtler events that could cause a revision. If we see Ocho Cinco look great in week one, he'll deserve a big spike because he's presumably healthy. The same goes with Ronnie Brown. If DeAngelo Williams, Ahman Green or Chris Perry get 25 carries in week one, then an bump is in order. For established players, this won't be the case. With younger players or guys on new teams, we've got less reason to be confident in our projections. In those cases, it makes sense to weigh what you see this year a bit more than when watching LaDainian Tomlinson.
How much weight should we give to our preseason projections relative to the actual season-to-date results, when predicting the remainder of the season results? I won't pretend that it's an easy question, or one that has a definitive answer. From 2000-2006, Footballguys.com projected exactly 100 RBs to score 100 FPs or more in a season, using the standard scoring system of 1 point per 10 yards, 6 points per TD, and zero points per reception. With a sample that large, we can decipher how to use preseason projections and in-season results to predict remainder of season results. For example, in 2001, a rookie Travis Henry wasn't very good, but was expected to shoulder more of the load in 2002. Footballguys.com projected him to score 11.4 FP/G in '02, but in week one he exploded for 178 yards and 3 TDs against the Jets. While he wouldn't average 35.8 FP/G for the season, what would have been a good prediction for the rest of the year? It turns out he averaged 14.9 FP/G the rest of the year; so a formula of six parts preseason projections and one part week one projections would have predicted the right result. (6/7 * 11.4 + 1/7 * 35.8 = 14.9)
We can perform a regression analysis to figure out how much weight should be placed on the preseason projections and how much weight on the week one results, in order to predict the remaining fantasy points production. Since we have a large sample of 100 RBs, we can be confident in the results of this analysis.
It turns out that the "correct" formula is 5.02 parts projections, and 1 part week one results. This feels about right to me. Right now, LT is projected to average 19.0 FP/G, Adrian Peterson is projected at 16.0 FP/G, and Marshawn Lynch 13.1 FP/G. What week one results would make us project all three RBs as even for the rest of the year? Tomlinson scoring 9 points, Peterson scoring 24 points, and Lynch scoring 39 points. So if LT had 30 total yards and one TD, and Lynch had 210 total yards and three TDs, you might be indifferent to trading Lynch for Tomlinson, or, of course, vice-versa.
Below shows the breakdowns for the RBs through X weeks of the season:
After week Proj Wt Reg Wt1 5.02 1.002 3.09 1.003 2.27 1.004 2.00 1.005 1.66 1.006 1.20 1.007 1.00 1.428 1.00 2.26It's worth noting that even after six weeks, preseason projections still matter more than the regular season results. Willie Green was projected to score 9.5 FP/G as a rookie, and was averaging just 2.1 FP/G through six games; he averaged 12.4 FP/G the rest of the way. In 2005, Tomlinson was averaging over 26 FP/G after six games, but scored only 160 FPs over his last ten games, much more in line with his projection of 18.6 FP/G. And preseason projections aren't worthless after eight weeks, either - they just become less and less important. Your remainder-of-season projections shouldn't be based off binary analysis: preseason projections don't either matter or don't. Rather, it's a sliding scale, where they become less and less valuable each week.How do things look at wide receiver? The table below displays the results of the same type of regression analysis, performed on the 197 WRs projected to score 80 or more FPs.
After week Proj Wt Reg Wt1 6.45 1.002 3.15 1.003 2.61 1.004 1.91 1.005 1.54 1.006 1.22 1.007 1.04 1.008 1.00 1.07Preseason projections seem to matter a bit more here, and that's not terribly surprising. Wide receivers get fewer targets than RBs get carries and targets, which means a really good or bad start could be the result of a small sample. Therefore, you should be less quick to give up on your wide receivers. Even though I'm saying not to weigh week one heavily, it still shouldn't be ignored. Remember that in 2001, Randy Moss was projected to be the top WR by just about everyone, coming off a ridiculous 1437/15 season. He had just 28 yards in week one, and 2001 ended up being the worst season of his young, four-year career. In stark contrast, though, is Marvin Harrison of the same year. He would gain just 35 yards in week one, but would finish the season with over 1500 yards and 15 TDs. So always remember not to trade your studs after just a bad week.You shouldn't even give up on your studs after five games, unless they've tanked. Once again, remember we're dealing with a continuum, and there's no on/off switch as to when the regular season starts to matter and the preseason projections stop. But even after five games, the preseason projections still matter a bit more than the regular season projections. Lee Evans in 2005 is a good example; he had just 3.4 FP/G through five games, despite being projected to score 8.3 FP/G. He would average 9.4 FP/G the rest of the year. Remember our 2001 Randy Moss, who was bad in week one? Well he wasn't good in weeks two, three, four or five, either. But in the last 11 games he would haul in 9 TDs and 900+ receiving yards, nearly doubling his per-game output from the first five weeks.
So what's it all mean? The two tables above can be a rough guide for you when thinking about making trades or deciding who to start in a given week. But they should just be used as a guide, and not as the formula to end all formulas.
One thing to remember is that strength of schedule matters. When using large samples, strength of schedule (and other random variables) aren't relevant to the analysis. However, strength of schedule matters a lot when we're analyzing our current, individual players.
Additionally, not all fantasy points are created equally. A player with a couple of TDs and not a lot of touches is probably going to disappoint in the future. Likewise, if Marion Barber III has 240 total yards through the first three weeks of the season but no scores, that's probably a better sign than if he had 80 total yards and three scores. Along the same lines, yards per carry, touchdowns, targets and yards are all relevant information to have, and tell you more about a player than merely fantasy points per game.
Finally, there are some times when you can just throw out the preseason projections. Obviously if a player gets hurt that changes everything, but there are subtler events that could cause a revision. If we see Ocho Cinco look great in week one, he'll deserve a big spike because he's presumably healthy. The same goes with Ronnie Brown. If DeAngelo Williams, Ahman Green or Chris Perry get 25 carries in week one, then an bump is in order. For established players, this won't be the case. With younger players or guys on new teams, we've got less reason to be confident in our projections. In those cases, it makes sense to weigh what you see this year a bit more than when watching LaDainian Tomlinson.
This is great analysis Chase. We should do more of this kind of work. 
