2010/11 Draft Review Part 2: Proportional Inter-round variance

In this series, we'll look at how valuable draft picks are by round and to each manager. Who does the best job of drafting, and how likely is any given manager to find a starter in each round?

PART TWO

In part one of this series, we looked at averages by round.  We saw that there was a general decline in the average points scored by players selected in each successive round and concluded that we did well in predicting the average fantasy points of skaters.

In this post, we'll look at the rate of selecting high level players in each round.  For this analysis, players who scored more than 80 fantasy points on the year are considered 'high level' or 'high impact.'  This is a nice round number that represents roughly 1 fantasy point per game over the season.  Exactly 1/3 of the skaters selected fall into this category (51 of 153 skaters taken), or an average of just over 4 players of this calibre per team.  For comparison, 10 players taken in the whole draft scored 100 fantasy points or more.

As we continue to delve into different ways of determining the value of draft picks and the value of a given draft pick relative to another, it's important to remember that with each pick choice become more constrained.  Often we imagine that process being one of the best players going first and the worst players going last.  Statistically, we expect to see a decline in the quality of players selected round to round, as well as to see the best players chosen earliest.

In reality, choice is made on a number of different variables (age, position, personal taste etc.) and is not always rational.  We see in the chart below that the number of players with >80 points selected in each round is a not a very clear downward trend.  This can be left as somewhat of a mystery for now, but keep in mind that some of these picks (eg. 3 of the first 4 picks overall) were chosen for their potential to be high level players in the long-term future, rather than for their likelihood of making an immediate fantasy impact.  

We see in the chart above that Round 10 was the most successful by this metric, followed by Rounds 2 through 5.  By a slightly adjusted metric, Round 1 is the clear winner, followed by Round 6 (shown below).

We see here the rate* of skaters taken in each round who scored 100 fantasy points or more.  Round 1 had three players taken of this caliber, while Round 6 had 2.  Only 5 other rounds contained players with 100 or more fantasy points, including the notorious Round 5.

You might remember from Part 1 that the average points of players selected in round 5 was under 57, or roughly 15 fewer than either round 4 or 6.  These charts reveal the highly sensitive nature of averages, as 5 of the 12 players selected in that round went on to be 'high level' skaters, versus only three of those from Round 6 scoring more than 80 points (though 2 of those scored more than 100).

The conflict in those two descriptions of Round 5 (average vs. # of high level players) can be explained through injuries. Four players selected in Round 5 had serious injuries that limited them to 30 or fewer fantasy points.  As always, injury is a major and mostly unpredictable factor in the success of a player.  With that in mind, there is a fair argument for using point-per-game averages to determine successful draft picks.  This, of course, suffers from the opposite sensitivity - is David Perron (13.9 points; 1.4 per game) as valuable as Shane Doan (101 pts; 1.4 per game)?

With these limitations (drafting for many reasons and injuries), this data cannot be understood as the likelihood of selecting a high impact player.  Instead, we can only understand it as description for now (at least until we look at some of the trends within rounds).

A predictive model can be applied, however, and we see below the rate of selecting high impact skaters as well as a prediction of that rate if we ignore alternative motivations for drafting.

The orange line is the average rate of selecting a high impact player over all 14 rounds (1/3).  The blue line is the actual rate of selecting those players.* Remember that some rounds had goalies selected as well, changing this graph slightly from the one above which simply counts the number of high impact players taken.  Finally, the black line is what we might expect to see if we were reasonably successful at predicting the incidence of high-impact players and selected them accordingly.

Interestingly, somewhere in round 4 this predictive model suggests we should start selecting high impact players at a lower than average rate.  More simply, the likelihood of selecting a high-level player after round 4 is below average.  To re-phrase again, picks in the top four rounds can be seen as categorically (read: in a different category) better than those that come in the rounds to follow.

Taken as a whole, however, these graphs do not suggest a lot of hope for winning using the draft.  On an average draft, a manager should expect to get between 4 and 5 players who score more than 1.0 fantasy points per game over a season, including between 0 and 1 player who reaches 100 fantasy points.  Assuming 8 high impact keepers, that's between 12 and 13 skaters of high impact on your roster.

Still, with the unexpected variability in how many high impact players are taken in each round, we haven't yet resolved how valuable having high round picks can be in achieving an above average draft.  Going by this data alone, we might conclude that Round 10 is the most valuable round, though our predictive model and Part One show a more logical connection between round of selection and performance.  Next we'll look at the heart of each round (ignoring the first and last few picks) to see if there is an appreciable difference over the course of the draft.


*Rate (in this case) is the number of high impact players taken divided by the total number of players taken.

Upcoming in this series:

- Part 3: Inter-Round Variance Using Thirdiles and Interquartiles

- Part 4: Drafted versus Undrafted Skaters and Goalies

- Part 5: Inner-Round Variance by Round

- Part 6: Inner-Round Variance by Manager