Give Me The Rock

this is fantasy basketball 

How Consistent are Player Stats from One Year to the Next?

Author Icon for Patrick

Categorized as: Author: Patrick, Fantasy Basketball, Fantasy Basketball Strategy
Posted on: December 6th, 2007

When I’m prepping for a draft and want to figure out how valuable a player is expected to be, 90% of the time I’ll take a look at his stats from the prior season and make an adjustment up or down depending on the guy’s age. If I’m really being thorough, maybe I look at the past two years. But is that right? Are players’ stats consistent enough from year to year to make accurate predictions that way?

Statistical analysis to the rescue! One of the simplest ways to look at this question is to examine the correlations of the major stat categories to see how related player stats are from one year to the next. The goal is to answer the question: how consistent are the box score stats from year to year?

To start with, I compiled data for every player in the NBA for every year they were in the league from 2003 to 2007. Guys who were in the league the entire time had 5 years of data, while guys who retired, were drafted, or came into or out of the league for other reasons had less. I then removed players who averaged less than 20 minutes a game for two seasons in a row. This means that a guy who averaged 10 minutes in 2003 and 10 minutes in 2004 had his 2003 and 2004 stats removed, but a guy who averaged 10 minutes in 2003 and 25 minutes in 2004 stayed in. I did this to increase variability in the data (journeymen who average 5 to 15 minutes a game over the course of a few years are remarkably consistent, in an “I score 2 points a game” kind of way). However, I did not want to mask any important increases or decreases in playing time, so guys who jumped back or forth over the 20 minute threshold remained in the data. Rookies are included in the analysis, but their first comparison year was obviously between their rookie and sophomore seasons.

If all that sounds incredibly boring, trust me, it was. But onto the “fun” part. I ran year to year pairwise correlations on 11 statistical categories (games, minutes, field and free throw percentage, rebounds, assist, steals, blocks, turnovers, and points scored) to see how related the categories were from one season to the next. The comparisons were made between the 2003 and 2004 seasons, the 2004 and 2005 seasons, and so on.

After all that, here are the results.

Five-Year Average (2002-2007)
Category Correlation Average Std Deviation
Blocks 0.900 0.5 0.6
Assists 0.847 2.5 1.9
Three Pointers Made 0.837 0.7 0.7
Rebounds 0.812 4.7 2.5
Points 0.802 11.4 5.9
Turnovers 0.747 1.6 0.8
Steals 0.739 0.9 0.4
Free Throw Percentage 0.620 74.5% 10.9%
Minutes 0.570 27.6 7.9
Field Goal Percentage 0.479 44.6% 5.4%
Games Played 0.190 66.8 17.4

The categories are listed from most to least correlated, meaning blocks stay the most constant from year to year, followed by assists and three pointers. Not surprisingly, all three are specialized stats that usually belong to a certain type of player. A six foot point guard who doesn’t block any shots isn’t suddenly going to grow eight inches and become a force in the paint, for example. So, yes, for the stats at the top end of list, you can get a really good idea of what a guy will do based on what he did last year.

While I’m not surprised to see that games played at the bottom of the list, I did not expect the correlation to be so low (0.19). Looking at the table, it’s a huge drop off from the second least correlated item (field goal percentage at 0.49) to games played. The results suggest that the number of games played (on average) in one year is barely related to the one before or after it. Put more simply, it would not be wise to predict how many games a guy will play next season based on the number he plays in this one.

I was also surprised to see both field goal and free throw percentage at the lower end of the correlation list. Before starting this analysis, I would have guessed they’d be at the top because percentages are not dependent on playing time. But compared to blocks, assists, etc, they are quite a bit less consistent.

For clarity, here are the year to year changes in blocks, points, and games played in graphical form (click to view larger image).

The shortcoming of this type of analysis is that it does not examine specific players to see if any types of players are more or less consistent than others, and if so, who they are. In fantasy, it’s pretty common to discuss a player’s durability or lack thereof as if we can easily predict how many games a guy will play in the future based on his past. While this analysis shows that this is generally not true, there are certain players who have been consistent (over the last five years, at least). Jason Terry, for example, played in either 80 or 81 games for all the five years that were included in the data. Is Terry an outlyer? A freak of nature? Did he have his bones replaced with an adamantium exoskeleton in a clandestine government project? Are there others like him? I don’t know. But if there is ever a part 2 to this analysis, I’ll take a look at that question.

Other Stuff Like This:

  • Cors

    Great work on the stat analysis there mate.
    Personally, i haven’t done stax of proper statistical analysis since Uni. So i may be way off here. But looking at your data, I think the high correlation for blocks and assists may be due to the high number of players that don’t do many of either. For example, there are a tonne of players that get little to no blocks year in year out and their production in this area will barely change (cos they aint getting any taller). So i think that sort of thing might be skewing your data a fair bit. I think the %’s is one of the least skewed bits of data, so it prob best represents what you’re trying to achieve.
    Hope this rant is helpful for future analysis stuff. Keep up the good work, luv yr site.

  • http://www.givemetherock.com Patrick

    Cors, this may be more a part 2 discussion, but I think you’re spot on with your commentary. It became clear to me when I looked at the graph of blocks vs. points. Blocks have a high level of correlation at the low end (i.e., players who do not block any shots) and most of the variability occurs at the high end (i.e., the players that we’d be interested in drafting in fantasy leagues for their blocks). On the other end of the spectrum, the variability in points per game is consistent across all players (as are percentages, I would assume). So, I think you’re absolutely right. Saying you could predict blocks or assists or threes is a little misleading. You could easily predict it for guys who don’t do any of it (the six foot guard with 0 blocks). But for centers who block more shots, they are much more inconsistent from year to year. Provided I can find the time, I’d love to explore the question more deeply.

    I appreciate the interesting comment.

  • http://www.givemetherock.com Patrick

    Also, if you look at the points chart, someone (who you can see as a dot on the bottom right) scored about 25 points a game one year and under 10 the next. Any guesses on who that was? I didn’t control for games played, only minutes, so it’s possible it could be a fluke.

  • Pingback: Give Me The Rock » Blog Archive » The GMTR Hot List: Top 10 Fantasy Shooting Guards

  • Pingback: Give Me The Rock » Blog Archive » Fantasy Basketball and Managing Injuries