This article was originally published on February 20, 2018.

Fantasy basketball’s increasing popularity is not being disregarded. Having taken notice, the National Basketball Association (NBA) officially adopted a universal scoring system for fantasy leagues in a bid to make fantasy basketball even more accessible. While participation increases, however, so does the amount of money at stake. Fantasy’s increasing popularity is subjecting an ever-growing population to the possibility of monetary loss.

Naturally, participants in fantasy basketball are inclined to take measures in order to win, thereby protecting their money and earning bragging rights. NBA associate vice president of fantasy sports, Scott Kaufman-Ross, explains that there is overwhelming “data that shows how engaged fantasy players are, and how much live basketball they watch.” However, simply watching basketball is not always sufficient to gain an edge, and consequently fantasy participants often resort paying for predictions from professionals.

While it may be the case that this is a worthwhile investment in information, fantasy players should limit themselves to one professional. No one is any better than the others for making particular predictions.

Our exploratory analysis is founded on a predictor versus predictor comparison between predicted values and actual values. In the comparison we viewed over-predictions and under-predictions to be equally as faulty, and therefore made each statistic to be positive so it did not reflect whether the prediction was an overestimate or underestimate. Throughout this discussion, we refer to the metric as the “absolute residual.”

Below is a plot of the absolute residual of all four rankers across all the data. Note that the rankers are given coded names to hide their identity.

On average, all predictors are off by about seven points in their predictions. We see that all predictors have very similar interquartile ranges (IQR) as well for absolute residual.

One factor we thought would be valuable in differentiating predictors was each predictor’s performance for individual positions. This way, if certain predictors performed significantly better for any individual position, fantasy basketball players could better choose which ranking list to utilize.

As mentioned above, we studied the absolute residual for each expert under each of the five positions. We used the mean absolute residual and the interquartile range as our primary metrics. Again, we found very little variation in any position among the experts. For point guard, shooting guard, and small forward, all four experts produced mean absolute residuals between 6.7 and 7.2 points, with a maximum difference of .4 points for point guards. Similar results were found for power forwards and centers, although all four experts were off by slightly more on average for big men. These two positions produced a mean absolute residual range of 7.0 to 7.4 points. The experts’ interquartile range for all positions were very comparable, indicating a negligible difference in their rankings for all positions.

We conclude that no ranker is significantly better at predicting results for any position than any other ranker. Below is our full table of data for your convenience.

We next decided to do a comparison of predictions across teams. For this, we used the absolute residual for each predictor’s IQR, mean and median for each team. We then calculated the average of the IQR, mean and median by team. We used these statistics to find out whether or not certain rankers did better with different teams or not.

As with player comparisons, there was no significant difference between predictors when looking at players by teams. For each team there did not appear to be any significant difference in the best and worst performing ranker. The difference between the best and worst for IQR, mean and median never appeared to go above 1.5 point and infrequently rose over 1 point.

The chart above shows the average absolute residual for each of the four rankers across all thirty teams. The last column is the average across all rankers. For almost all of the teams, the predictors all scored generally about the same, not matter how good or bad their predictions turned out. For example, all the predictors struggled to accurately predict the scores of Denver Nuggets players. Each predictor’s IQR, mean and median for the Nuggets was in their bottom few scores, and many times was the predictor’s worst result in the category. As a result, the Nuggets scored the highest in all three categories.

As a result, we concluded that there is no real difference in selecting a predictor based on the team which a player plays on.

The final step of our analysis was to see if certain rankers were better at predicting the higher caliber players. As a DFS player, you likely care more about the predictions for players that you are spending more money on.

We broke players into four quartiles based on salary. The salary breakdowns were as follows:

1st Quartile: Salary < $3500

2nd Quartile: $3500 < Salary < $3900

3rd Quartile: $3900 < Salary < $5500

4th Quartile: $5500 < Salary

Below is a plot of the average absolute residual of the four quartiles of players. As we would expect, higher salary players have a higher absolute residual than lower salary players.

Examining the first quartile (lowest caliber) players, we see that across the four rankers, there is very little distinction in average absolute residual for predictions. The fifth boxplot on the right shows the average of all predictor of this tier of player. We see that no predictor is much better or worse than the average of them all.

We repeat the same comparison for the three other quartiles. For all three, we see no significant distinction between any individual predictor or the average.

In conclusion, we see that there is very little to no distinction between predictors in any facet. No predictor has an edge of another at the aggregate level, or with any team, position or tier of player. Our recommendation is NOT to dissuade DFS players from looking at predicted points. Our recommendation is to not take any predictor more seriously than another, and to understand quantitatively how “good” of a recommendation they are getting. If you’re looking to gain an edge in daily fantasy basketball, don’t waste your money on multiple professional predictors. Certainly, do not pay for advice.